* [PATCH 1 of 39] IB/ipath - Name zero counter offsets so it's clear they aren't counters
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 2 of 39] IB/ipath - update copyrights and other strings to reflect new company name Bryan O'Sullivan
` (38 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 28e3d8204fdb -r addf90abc724 drivers/infiniband/hw/ipath/ipath_mad.c
--- a/drivers/infiniband/hw/ipath/ipath_mad.c Fri Jun 23 22:47:27 2006 +0700
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:25 2006 -0700
@@ -215,7 +215,7 @@ static int recv_subn_get_portinfo(struct
/* P_KeyViolations are counted by hardware. */
pip->pkey_violations =
cpu_to_be16((ipath_layer_get_cr_errpkey(dev->dd) -
- dev->n_pkey_violations) & 0xFFFF);
+ dev->z_pkey_violations) & 0xFFFF);
pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
/* Only the hardware GUID is supported for now */
pip->guid_cap = 1;
@@ -389,7 +389,7 @@ static int recv_subn_set_portinfo(struct
* later.
*/
if (pip->pkey_violations == 0)
- dev->n_pkey_violations =
+ dev->z_pkey_violations =
ipath_layer_get_cr_errpkey(dev->dd);
if (pip->qkey_violations == 0)
@@ -844,18 +844,18 @@ static int recv_pma_get_portcounters(str
ipath_layer_get_counters(dev->dd, &cntrs);
/* Adjust counters for any resets done. */
- cntrs.symbol_error_counter -= dev->n_symbol_error_counter;
+ cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
cntrs.link_error_recovery_counter -=
- dev->n_link_error_recovery_counter;
- cntrs.link_downed_counter -= dev->n_link_downed_counter;
+ dev->z_link_error_recovery_counter;
+ cntrs.link_downed_counter -= dev->z_link_downed_counter;
cntrs.port_rcv_errors += dev->rcv_errors;
- cntrs.port_rcv_errors -= dev->n_port_rcv_errors;
- cntrs.port_rcv_remphys_errors -= dev->n_port_rcv_remphys_errors;
- cntrs.port_xmit_discards -= dev->n_port_xmit_discards;
- cntrs.port_xmit_data -= dev->n_port_xmit_data;
- cntrs.port_rcv_data -= dev->n_port_rcv_data;
- cntrs.port_xmit_packets -= dev->n_port_xmit_packets;
- cntrs.port_rcv_packets -= dev->n_port_rcv_packets;
+ cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
+ cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
+ cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
+ cntrs.port_xmit_data -= dev->z_port_xmit_data;
+ cntrs.port_rcv_data -= dev->z_port_rcv_data;
+ cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
+ cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
memset(pmp->data, 0, sizeof(pmp->data));
@@ -928,10 +928,10 @@ static int recv_pma_get_portcounters_ext
&rpkts, &xwait);
/* Adjust counters for any resets done. */
- swords -= dev->n_port_xmit_data;
- rwords -= dev->n_port_rcv_data;
- spkts -= dev->n_port_xmit_packets;
- rpkts -= dev->n_port_rcv_packets;
+ swords -= dev->z_port_xmit_data;
+ rwords -= dev->z_port_rcv_data;
+ spkts -= dev->z_port_xmit_packets;
+ rpkts -= dev->z_port_rcv_packets;
memset(pmp->data, 0, sizeof(pmp->data));
@@ -967,37 +967,37 @@ static int recv_pma_set_portcounters(str
ipath_layer_get_counters(dev->dd, &cntrs);
if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
- dev->n_symbol_error_counter = cntrs.symbol_error_counter;
+ dev->z_symbol_error_counter = cntrs.symbol_error_counter;
if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
- dev->n_link_error_recovery_counter =
+ dev->z_link_error_recovery_counter =
cntrs.link_error_recovery_counter;
if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
- dev->n_link_downed_counter = cntrs.link_downed_counter;
+ dev->z_link_downed_counter = cntrs.link_downed_counter;
if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
- dev->n_port_rcv_errors =
+ dev->z_port_rcv_errors =
cntrs.port_rcv_errors + dev->rcv_errors;
if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
- dev->n_port_rcv_remphys_errors =
+ dev->z_port_rcv_remphys_errors =
cntrs.port_rcv_remphys_errors;
if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
- dev->n_port_xmit_discards = cntrs.port_xmit_discards;
+ dev->z_port_xmit_discards = cntrs.port_xmit_discards;
if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
- dev->n_port_xmit_data = cntrs.port_xmit_data;
+ dev->z_port_xmit_data = cntrs.port_xmit_data;
if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
- dev->n_port_rcv_data = cntrs.port_rcv_data;
+ dev->z_port_rcv_data = cntrs.port_rcv_data;
if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
- dev->n_port_xmit_packets = cntrs.port_xmit_packets;
+ dev->z_port_xmit_packets = cntrs.port_xmit_packets;
if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
- dev->n_port_rcv_packets = cntrs.port_rcv_packets;
+ dev->z_port_rcv_packets = cntrs.port_rcv_packets;
return recv_pma_get_portcounters(pmp, ibdev, port);
}
@@ -1014,16 +1014,16 @@ static int recv_pma_set_portcounters_ext
&rpkts, &xwait);
if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
- dev->n_port_xmit_data = swords;
+ dev->z_port_xmit_data = swords;
if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
- dev->n_port_rcv_data = rwords;
+ dev->z_port_rcv_data = rwords;
if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
- dev->n_port_xmit_packets = spkts;
+ dev->z_port_xmit_packets = spkts;
if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
- dev->n_port_rcv_packets = rpkts;
+ dev->z_port_rcv_packets = rpkts;
if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
dev->n_unicast_xmit = 0;
@@ -1285,18 +1285,18 @@ int ipath_process_mad(struct ib_device *
ipath_layer_get_counters(to_idev(ibdev)->dd, &cntrs);
dev->rcv_errors++;
- dev->n_symbol_error_counter = cntrs.symbol_error_counter;
- dev->n_link_error_recovery_counter =
+ dev->z_symbol_error_counter = cntrs.symbol_error_counter;
+ dev->z_link_error_recovery_counter =
cntrs.link_error_recovery_counter;
- dev->n_link_downed_counter = cntrs.link_downed_counter;
- dev->n_port_rcv_errors = cntrs.port_rcv_errors + 1;
- dev->n_port_rcv_remphys_errors =
+ dev->z_link_downed_counter = cntrs.link_downed_counter;
+ dev->z_port_rcv_errors = cntrs.port_rcv_errors + 1;
+ dev->z_port_rcv_remphys_errors =
cntrs.port_rcv_remphys_errors;
- dev->n_port_xmit_discards = cntrs.port_xmit_discards;
- dev->n_port_xmit_data = cntrs.port_xmit_data;
- dev->n_port_rcv_data = cntrs.port_rcv_data;
- dev->n_port_xmit_packets = cntrs.port_xmit_packets;
- dev->n_port_rcv_packets = cntrs.port_rcv_packets;
+ dev->z_port_xmit_discards = cntrs.port_xmit_discards;
+ dev->z_port_xmit_data = cntrs.port_xmit_data;
+ dev->z_port_rcv_data = cntrs.port_rcv_data;
+ dev->z_port_xmit_packets = cntrs.port_xmit_packets;
+ dev->z_port_rcv_packets = cntrs.port_rcv_packets;
}
switch (in_mad->mad_hdr.mgmt_class) {
case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
diff -r 28e3d8204fdb -r addf90abc724 drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Fri Jun 23 22:47:27 2006 +0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
@@ -646,7 +646,7 @@ static int ipath_query_port(struct ib_de
props->max_msg_sz = 4096;
props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd);
props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) -
- dev->n_pkey_violations;
+ dev->z_pkey_violations;
props->qkey_viol_cntr = dev->qkey_violations;
props->active_width = IB_WIDTH_4X;
/* See rate_show() */
diff -r 28e3d8204fdb -r addf90abc724 drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Fri Jun 23 22:47:27 2006 +0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
@@ -442,17 +442,17 @@ struct ipath_ibdev {
u64 n_unicast_rcv; /* total unicast packets received */
u64 n_multicast_xmit; /* total multicast packets sent */
u64 n_multicast_rcv; /* total multicast packets received */
- u64 n_symbol_error_counter; /* starting count for PMA */
- u64 n_link_error_recovery_counter; /* starting count for PMA */
- u64 n_link_downed_counter; /* starting count for PMA */
- u64 n_port_rcv_errors; /* starting count for PMA */
- u64 n_port_rcv_remphys_errors; /* starting count for PMA */
- u64 n_port_xmit_discards; /* starting count for PMA */
- u64 n_port_xmit_data; /* starting count for PMA */
- u64 n_port_rcv_data; /* starting count for PMA */
- u64 n_port_xmit_packets; /* starting count for PMA */
- u64 n_port_rcv_packets; /* starting count for PMA */
- u32 n_pkey_violations; /* starting count for PMA */
+ u64 z_symbol_error_counter; /* starting count for PMA */
+ u64 z_link_error_recovery_counter; /* starting count for PMA */
+ u64 z_link_downed_counter; /* starting count for PMA */
+ u64 z_port_rcv_errors; /* starting count for PMA */
+ u64 z_port_rcv_remphys_errors; /* starting count for PMA */
+ u64 z_port_xmit_discards; /* starting count for PMA */
+ u64 z_port_xmit_data; /* starting count for PMA */
+ u64 z_port_rcv_data; /* starting count for PMA */
+ u64 z_port_xmit_packets; /* starting count for PMA */
+ u64 z_port_rcv_packets; /* starting count for PMA */
+ u32 z_pkey_violations; /* starting count for PMA */
u32 n_rc_resends;
u32 n_rc_acks;
u32 n_rc_qacks;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 2 of 39] IB/ipath - update copyrights and other strings to reflect new company name
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 1 of 39] IB/ipath - Name zero counter offsets so it's clear they aren't counters Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 3 of 39] IB/ipath - Share more common code between RC and UC protocols Bryan O'Sullivan
` (37 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/Kconfig
--- a/drivers/infiniband/hw/ipath/Kconfig Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/Kconfig Thu Jun 29 14:33:25 2006 -0700
@@ -1,16 +1,16 @@ config IPATH_CORE
config IPATH_CORE
- tristate "PathScale InfiniPath Driver"
+ tristate "QLogic InfiniPath Driver"
depends on 64BIT && PCI_MSI && NET
---help---
- This is a low-level driver for PathScale InfiniPath host channel
+ This is a low-level driver for QLogic InfiniPath host channel
adapters (HCAs) based on the HT-400 and PE-800 chips.
config INFINIBAND_IPATH
- tristate "PathScale InfiniPath Verbs Driver"
+ tristate "QLogic InfiniPath Verbs Driver"
depends on IPATH_CORE && INFINIBAND
---help---
This is a driver that provides InfiniBand verbs support for
- PathScale InfiniPath host channel adapters (HCAs). This
+ QLogic InfiniPath host channel adapters (HCAs). This
allows these devices to be used with both kernel upper level
protocols such as IP-over-InfiniBand as well as with userspace
applications (in conjunction with InfiniBand userspace access).
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/Makefile
--- a/drivers/infiniband/hw/ipath/Makefile Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/Makefile Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,4 @@ EXTRA_CFLAGS += -DIPATH_IDSTR='"PathScal
-EXTRA_CFLAGS += -DIPATH_IDSTR='"PathScale kernel.org driver"' \
+EXTRA_CFLAGS += -DIPATH_IDSTR='"QLogic kernel.org driver"' \
-DIPATH_KERN_TYPE=0
obj-$(CONFIG_IPATH_CORE) += ipath_core.o
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_common.h
--- a/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
@@ -38,7 +39,7 @@
* to communicate between kernel and user code.
*/
-/* This is the IEEE-assigned OUI for PathScale, Inc. */
+/* This is the IEEE-assigned OUI for QLogic, Inc. InfiniPath */
#define IPATH_SRC_OUI_1 0x00
#define IPATH_SRC_OUI_2 0x11
#define IPATH_SRC_OUI_3 0x75
@@ -342,9 +343,9 @@ struct ipath_base_info {
/*
* Similarly, this is the kernel version going back to the user. It's
* slightly different, in that we want to tell if the driver was built as
- * part of a PathScale release, or from the driver from OpenIB, kernel.org,
+ * part of a QLogic release, or from the driver from OpenIB, kernel.org,
* or a standard distribution, for support reasons. The high bit is 0 for
- * non-PathScale, and 1 for PathScale-built/supplied.
+ * non-QLogic, and 1 for QLogic-built/supplied.
*
* It's returned by the driver to the user code during initialization in the
* spi_sw_version field of ipath_base_info, so the user code can in turn
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_cq.c
--- a/drivers/infiniband/hw/ipath/ipath_cq.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_cq.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_debug.h
--- a/drivers/infiniband/hw/ipath/ipath_debug.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_debug.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_diag.c
--- a/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
@@ -52,7 +53,7 @@ const char *ipath_get_unit_name(int unit
EXPORT_SYMBOL_GPL(ipath_get_unit_name);
-#define DRIVER_LOAD_MSG "PathScale " IPATH_DRV_NAME " loaded: "
+#define DRIVER_LOAD_MSG "QLogic " IPATH_DRV_NAME " loaded: "
#define PFX IPATH_DRV_NAME ": "
/*
@@ -74,8 +75,8 @@ EXPORT_SYMBOL_GPL(ipath_debug);
EXPORT_SYMBOL_GPL(ipath_debug);
MODULE_LICENSE("GPL");
-MODULE_AUTHOR("PathScale <support@pathscale.com>");
-MODULE_DESCRIPTION("Pathscale InfiniPath driver");
+MODULE_AUTHOR("QLogic <support@pathscale.com>");
+MODULE_DESCRIPTION("QLogic InfiniPath driver");
const char *ipath_ibcstatus_str[] = {
"Disabled",
@@ -452,7 +453,7 @@ static int __devinit ipath_init_one(stru
ipath_init_pe800_funcs(dd);
break;
default:
- ipath_dev_err(dd, "Found unknown PathScale deviceid 0x%x, "
+ ipath_dev_err(dd, "Found unknown QLogic deviceid 0x%x, "
"failing\n", ent->device);
return -ENODEV;
}
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_eeprom.c
--- a/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_fs.c
--- a/drivers/infiniband/hw/ipath/ipath_fs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_fs.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_ht400.c
--- a/drivers/infiniband/hw/ipath/ipath_ht400.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ht400.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_init_chip.c
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,6 +1,7 @@
#ifndef _IPATH_KERNEL_H
#define _IPATH_KERNEL_H
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_keys.c
--- a/drivers/infiniband/hw/ipath/ipath_keys.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_keys.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_layer.h
--- a/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_mad.c
--- a/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_mr.c
--- a/drivers/infiniband/hw/ipath/ipath_mr.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mr.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_pe800.c
--- a/drivers/infiniband/hw/ipath/ipath_pe800.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_pe800.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
@@ -44,7 +45,7 @@
/*
* This file contains all the chip-specific register information and
- * access functions for the PathScale PE800, the PCI-Express chip.
+ * access functions for the QLogic InfiniPath PE800, the PCI-Express chip.
*
* This lists the InfiniPath PE800 registers, in the actual chip layout.
* This structure should never be directly accessed.
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_rc.c
--- a/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_registers.h
--- a/drivers/infiniband/hw/ipath/ipath_registers.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_registers.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_ruc.c
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_srq.c
--- a/drivers/infiniband/hw/ipath/ipath_srq.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_srq.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_stats.c
--- a/drivers/infiniband/hw/ipath/ipath_stats.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_stats.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_sysfs.c
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_uc.c
--- a/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_ud.c
--- a/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_user_pages.c
--- a/drivers/infiniband/hw/ipath/ipath_user_pages.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_user_pages.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
@@ -56,8 +57,8 @@ MODULE_PARM_DESC(debug, "Verbs debug mas
MODULE_PARM_DESC(debug, "Verbs debug mask");
MODULE_LICENSE("GPL");
-MODULE_AUTHOR("PathScale <support@pathscale.com>");
-MODULE_DESCRIPTION("Pathscale InfiniPath driver");
+MODULE_AUTHOR("QLogic <support@pathscale.com>");
+MODULE_DESCRIPTION("QLogic InfiniPath driver");
const int ib_ipath_state_ops[IB_QPS_ERR + 1] = {
[IB_QPS_RESET] = 0,
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ipath_wc_x86_64.c
--- a/drivers/infiniband/hw/ipath/ipath_wc_x86_64.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_wc_x86_64.c Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/ips_common.h
--- a/drivers/infiniband/hw/ipath/ips_common.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ips_common.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,6 +1,7 @@
#ifndef IPS_COMMON_H
#define IPS_COMMON_H
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
diff -r addf90abc724 -r f7c82500b9c7 drivers/infiniband/hw/ipath/verbs_debug.h
--- a/drivers/infiniband/hw/ipath/verbs_debug.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/verbs_debug.h Thu Jun 29 14:33:25 2006 -0700
@@ -1,4 +1,5 @@
/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 3 of 39] IB/ipath - Share more common code between RC and UC protocols
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 1 of 39] IB/ipath - Name zero counter offsets so it's clear they aren't counters Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 2 of 39] IB/ipath - update copyrights and other strings to reflect new company name Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 4 of 39] IB/ipath - fix an indenting problem Bryan O'Sullivan
` (36 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r f7c82500b9c7 -r ebf646d10db0 drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
@@ -709,9 +709,7 @@ struct ib_qp *ipath_create_qp(struct ib_
spin_lock_init(&qp->r_rq.lock);
atomic_set(&qp->refcount, 0);
init_waitqueue_head(&qp->wait);
- tasklet_init(&qp->s_task,
- init_attr->qp_type == IB_QPT_RC ?
- ipath_do_rc_send : ipath_do_uc_send,
+ tasklet_init(&qp->s_task, ipath_do_ruc_send,
(unsigned long)qp);
INIT_LIST_HEAD(&qp->piowait);
INIT_LIST_HEAD(&qp->timerwait);
@@ -896,9 +894,9 @@ void ipath_get_credit(struct ipath_qp *q
* as many packets as we like. Otherwise, we have to
* honor the credit field.
*/
- if (credit == IPS_AETH_CREDIT_INVAL) {
+ if (credit == IPS_AETH_CREDIT_INVAL)
qp->s_lsn = (u32) -1;
- } else if (qp->s_lsn != (u32) -1) {
+ else if (qp->s_lsn != (u32) -1) {
/* Compute new LSN (i.e., MSN + credit) */
credit = (aeth + credit_table[credit]) & IPS_MSN_MASK;
if (ipath_cmp24(credit, qp->s_lsn) > 0)
diff -r f7c82500b9c7 -r ebf646d10db0 drivers/infiniband/hw/ipath/ipath_rc.c
--- a/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
@@ -73,9 +73,9 @@ static void ipath_init_restart(struct ip
* Return bth0 if constructed; otherwise, return 0.
* Note the QP s_lock must be held.
*/
-static inline u32 ipath_make_rc_ack(struct ipath_qp *qp,
- struct ipath_other_headers *ohdr,
- u32 pmtu)
+u32 ipath_make_rc_ack(struct ipath_qp *qp,
+ struct ipath_other_headers *ohdr,
+ u32 pmtu)
{
struct ipath_sge_state *ss;
u32 hwords;
@@ -96,8 +96,7 @@ static inline u32 ipath_make_rc_ack(stru
if (len > pmtu) {
len = pmtu;
qp->s_ack_state = OP(RDMA_READ_RESPONSE_FIRST);
- }
- else
+ } else
qp->s_ack_state = OP(RDMA_READ_RESPONSE_ONLY);
qp->s_rdma_len -= len;
bth0 = qp->s_ack_state << 24;
@@ -177,9 +176,9 @@ static inline u32 ipath_make_rc_ack(stru
* Return 1 if constructed; otherwise, return 0.
* Note the QP s_lock must be held.
*/
-static inline int ipath_make_rc_req(struct ipath_qp *qp,
- struct ipath_other_headers *ohdr,
- u32 pmtu, u32 *bth0p, u32 *bth2p)
+int ipath_make_rc_req(struct ipath_qp *qp,
+ struct ipath_other_headers *ohdr,
+ u32 pmtu, u32 *bth0p, u32 *bth2p)
{
struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
struct ipath_sge_state *ss;
@@ -497,160 +496,33 @@ done:
return 0;
}
-static inline void ipath_make_rc_grh(struct ipath_qp *qp,
- struct ib_global_route *grh,
- u32 nwords)
-{
- struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
-
- /* GRH header size in 32-bit words. */
- qp->s_hdrwords += 10;
- qp->s_hdr.u.l.grh.version_tclass_flow =
- cpu_to_be32((6 << 28) |
- (grh->traffic_class << 20) |
- grh->flow_label);
- qp->s_hdr.u.l.grh.paylen =
- cpu_to_be16(((qp->s_hdrwords - 12) + nwords +
- SIZE_OF_CRC) << 2);
- /* next_hdr is defined by C8-7 in ch. 8.4.1 */
- qp->s_hdr.u.l.grh.next_hdr = 0x1B;
- qp->s_hdr.u.l.grh.hop_limit = grh->hop_limit;
- /* The SGID is 32-bit aligned. */
- qp->s_hdr.u.l.grh.sgid.global.subnet_prefix = dev->gid_prefix;
- qp->s_hdr.u.l.grh.sgid.global.interface_id =
- ipath_layer_get_guid(dev->dd);
- qp->s_hdr.u.l.grh.dgid = grh->dgid;
-}
-
/**
- * ipath_do_rc_send - perform a send on an RC QP
- * @data: contains a pointer to the QP
+ * send_rc_ack - Construct an ACK packet and send it
+ * @qp: a pointer to the QP
*
- * Process entries in the send work queue until credit or queue is
- * exhausted. Only allow one CPU to send a packet per QP (tasklet).
- * Otherwise, after we drop the QP s_lock, two threads could send
- * packets out of order.
+ * This is called from ipath_rc_rcv() and only uses the receive
+ * side QP state.
+ * Note that RDMA reads are handled in the send side QP state and tasklet.
*/
-void ipath_do_rc_send(unsigned long data)
-{
- struct ipath_qp *qp = (struct ipath_qp *)data;
- struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
- unsigned long flags;
- u16 lrh0;
- u32 nwords;
- u32 extra_bytes;
- u32 bth0;
- u32 bth2;
- u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
- struct ipath_other_headers *ohdr;
-
- if (test_and_set_bit(IPATH_S_BUSY, &qp->s_flags))
- goto bail;
-
- if (unlikely(qp->remote_ah_attr.dlid ==
- ipath_layer_get_lid(dev->dd))) {
- struct ib_wc wc;
-
- /*
- * Pass in an uninitialized ib_wc to be consistent with
- * other places where ipath_ruc_loopback() is called.
- */
- ipath_ruc_loopback(qp, &wc);
- goto clear;
- }
-
- ohdr = &qp->s_hdr.u.oth;
- if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
- ohdr = &qp->s_hdr.u.l.oth;
-
-again:
- /* Check for a constructed packet to be sent. */
- if (qp->s_hdrwords != 0) {
- /*
- * If no PIO bufs are available, return. An interrupt will
- * call ipath_ib_piobufavail() when one is available.
- */
- _VERBS_INFO("h %u %p\n", qp->s_hdrwords, &qp->s_hdr);
- _VERBS_INFO("d %u %p %u %p %u %u %u %u\n", qp->s_cur_size,
- qp->s_cur_sge->sg_list,
- qp->s_cur_sge->num_sge,
- qp->s_cur_sge->sge.vaddr,
- qp->s_cur_sge->sge.sge_length,
- qp->s_cur_sge->sge.length,
- qp->s_cur_sge->sge.m,
- qp->s_cur_sge->sge.n);
- if (ipath_verbs_send(dev->dd, qp->s_hdrwords,
- (u32 *) &qp->s_hdr, qp->s_cur_size,
- qp->s_cur_sge)) {
- ipath_no_bufs_available(qp, dev);
- goto bail;
- }
- dev->n_unicast_xmit++;
- /* Record that we sent the packet and s_hdr is empty. */
- qp->s_hdrwords = 0;
- }
-
- /*
- * The lock is needed to synchronize between setting
- * qp->s_ack_state, resend timer, and post_send().
- */
- spin_lock_irqsave(&qp->s_lock, flags);
-
- /* Sending responses has higher priority over sending requests. */
- if (qp->s_ack_state != OP(ACKNOWLEDGE) &&
- (bth0 = ipath_make_rc_ack(qp, ohdr, pmtu)) != 0)
- bth2 = qp->s_ack_psn++ & IPS_PSN_MASK;
- else if (!ipath_make_rc_req(qp, ohdr, pmtu, &bth0, &bth2))
- goto done;
-
- spin_unlock_irqrestore(&qp->s_lock, flags);
-
- /* Construct the header. */
- extra_bytes = (4 - qp->s_cur_size) & 3;
- nwords = (qp->s_cur_size + extra_bytes) >> 2;
- lrh0 = IPS_LRH_BTH;
- if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
- ipath_make_rc_grh(qp, &qp->remote_ah_attr.grh, nwords);
- lrh0 = IPS_LRH_GRH;
- }
- lrh0 |= qp->remote_ah_attr.sl << 4;
- qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
- qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
- qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords +
- SIZE_OF_CRC);
- qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
- bth0 |= ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
- bth0 |= extra_bytes << 20;
- ohdr->bth[0] = cpu_to_be32(bth0);
- ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
- ohdr->bth[2] = cpu_to_be32(bth2);
-
- /* Check for more work to do. */
- goto again;
-
-done:
- spin_unlock_irqrestore(&qp->s_lock, flags);
-clear:
- clear_bit(IPATH_S_BUSY, &qp->s_flags);
-bail:
- return;
-}
-
static void send_rc_ack(struct ipath_qp *qp)
{
struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
u16 lrh0;
u32 bth0;
+ u32 hwords;
+ struct ipath_ib_header hdr;
struct ipath_other_headers *ohdr;
/* Construct the header. */
- ohdr = &qp->s_hdr.u.oth;
+ ohdr = &hdr.u.oth;
lrh0 = IPS_LRH_BTH;
/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4. */
- qp->s_hdrwords = 6;
+ hwords = 6;
if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
- ipath_make_rc_grh(qp, &qp->remote_ah_attr.grh, 0);
- ohdr = &qp->s_hdr.u.l.oth;
+ hwords += ipath_make_grh(dev, &hdr.u.l.grh,
+ &qp->remote_ah_attr.grh,
+ hwords, 0);
+ ohdr = &hdr.u.l.oth;
lrh0 = IPS_LRH_GRH;
}
bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
@@ -658,15 +530,14 @@ static void send_rc_ack(struct ipath_qp
if (qp->s_ack_state >= OP(COMPARE_SWAP)) {
bth0 |= IB_OPCODE_ATOMIC_ACKNOWLEDGE << 24;
ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->s_ack_atomic);
- qp->s_hdrwords += sizeof(ohdr->u.at.atomic_ack_eth) / 4;
- }
- else
+ hwords += sizeof(ohdr->u.at.atomic_ack_eth) / 4;
+ } else
bth0 |= OP(ACKNOWLEDGE) << 24;
lrh0 |= qp->remote_ah_attr.sl << 4;
- qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
- qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
- qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + SIZE_OF_CRC);
- qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
+ hdr.lrh[0] = cpu_to_be16(lrh0);
+ hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+ hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
+ hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
ohdr->bth[0] = cpu_to_be32(bth0);
ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
ohdr->bth[2] = cpu_to_be32(qp->s_ack_psn & IPS_PSN_MASK);
@@ -674,8 +545,7 @@ static void send_rc_ack(struct ipath_qp
/*
* If we can send the ACK, clear the ACK state.
*/
- if (ipath_verbs_send(dev->dd, qp->s_hdrwords, (u32 *) &qp->s_hdr,
- 0, NULL) == 0) {
+ if (ipath_verbs_send(dev->dd, hwords, (u32 *) &hdr, 0, NULL) == 0) {
qp->s_ack_state = OP(ACKNOWLEDGE);
dev->n_rc_qacks++;
dev->n_unicast_xmit++;
@@ -805,7 +675,7 @@ bail:
* @qp: the QP
* @psn: the packet sequence number to restart at
*
- * This is called from ipath_rc_rcv() to process an incoming RC ACK
+ * This is called from ipath_rc_rcv_resp() to process an incoming RC ACK
* for the given QP.
* Called at interrupt level with the QP s_lock held.
*/
@@ -1231,18 +1101,12 @@ static inline void ipath_rc_rcv_resp(str
* ICRC (4).
*/
if (unlikely(tlen <= (hdrsize + pad + 8))) {
- /*
- * XXX Need to generate an error CQ
- * entry.
- */
+ /* XXX Need to generate an error CQ entry. */
goto ack_done;
}
tlen -= hdrsize + pad + 8;
if (unlikely(tlen != qp->s_len)) {
- /*
- * XXX Need to generate an error CQ
- * entry.
- */
+ /* XXX Need to generate an error CQ entry. */
goto ack_done;
}
if (!header_in_data)
@@ -1384,7 +1248,7 @@ static inline int ipath_rc_rcv_error(str
case OP(COMPARE_SWAP):
case OP(FETCH_ADD):
/*
- * Check for the PSN of the last atomic operations
+ * Check for the PSN of the last atomic operation
* performed and resend the result if found.
*/
if ((psn & IPS_PSN_MASK) != qp->r_atomic_psn) {
@@ -1454,11 +1318,6 @@ void ipath_rc_rcv(struct ipath_ibdev *de
} else
psn = be32_to_cpu(ohdr->bth[2]);
}
- /*
- * The opcode is in the low byte when its in network order
- * (top byte when in host order).
- */
- opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
/*
* Process responses (ACKs) before anything else. Note that the
@@ -1466,6 +1325,7 @@ void ipath_rc_rcv(struct ipath_ibdev *de
* queue rather than the expected receive packet sequence number.
* In other words, this QP is the requester.
*/
+ opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
ipath_rc_rcv_resp(dev, ohdr, data, tlen, qp, opcode, psn,
diff -r f7c82500b9c7 -r ebf646d10db0 drivers/infiniband/hw/ipath/ipath_ruc.c
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:25 2006 -0700
@@ -32,6 +32,7 @@
*/
#include "ipath_verbs.h"
+#include "ips_common.h"
/*
* Convert the AETH RNR timeout code into the number of milliseconds.
@@ -188,7 +189,6 @@ bail:
/**
* ipath_ruc_loopback - handle UC and RC lookback requests
* @sqp: the loopback QP
- * @wc: the work completion entry
*
* This is called from ipath_do_uc_send() or ipath_do_rc_send() to
* forward a WQE addressed to the same HCA.
@@ -197,13 +197,14 @@ bail:
* receive interrupts since this is a connected protocol and all packets
* will pass through here.
*/
-void ipath_ruc_loopback(struct ipath_qp *sqp, struct ib_wc *wc)
+static void ipath_ruc_loopback(struct ipath_qp *sqp)
{
struct ipath_ibdev *dev = to_idev(sqp->ibqp.device);
struct ipath_qp *qp;
struct ipath_swqe *wqe;
struct ipath_sge *sge;
unsigned long flags;
+ struct ib_wc wc;
u64 sdata;
qp = ipath_lookup_qpn(&dev->qp_table, sqp->remote_qpn);
@@ -234,8 +235,8 @@ again:
wqe = get_swqe_ptr(sqp, sqp->s_last);
spin_unlock_irqrestore(&sqp->s_lock, flags);
- wc->wc_flags = 0;
- wc->imm_data = 0;
+ wc.wc_flags = 0;
+ wc.imm_data = 0;
sqp->s_sge.sge = wqe->sg_list[0];
sqp->s_sge.sg_list = wqe->sg_list + 1;
@@ -243,8 +244,8 @@ again:
sqp->s_len = wqe->length;
switch (wqe->wr.opcode) {
case IB_WR_SEND_WITH_IMM:
- wc->wc_flags = IB_WC_WITH_IMM;
- wc->imm_data = wqe->wr.imm_data;
+ wc.wc_flags = IB_WC_WITH_IMM;
+ wc.imm_data = wqe->wr.imm_data;
/* FALLTHROUGH */
case IB_WR_SEND:
spin_lock_irqsave(&qp->r_rq.lock, flags);
@@ -255,7 +256,7 @@ again:
if (qp->ibqp.qp_type == IB_QPT_UC)
goto send_comp;
if (sqp->s_rnr_retry == 0) {
- wc->status = IB_WC_RNR_RETRY_EXC_ERR;
+ wc.status = IB_WC_RNR_RETRY_EXC_ERR;
goto err;
}
if (sqp->s_rnr_retry_cnt < 7)
@@ -270,8 +271,8 @@ again:
break;
case IB_WR_RDMA_WRITE_WITH_IMM:
- wc->wc_flags = IB_WC_WITH_IMM;
- wc->imm_data = wqe->wr.imm_data;
+ wc.wc_flags = IB_WC_WITH_IMM;
+ wc.imm_data = wqe->wr.imm_data;
spin_lock_irqsave(&qp->r_rq.lock, flags);
if (!ipath_get_rwqe(qp, 1))
goto rnr_nak;
@@ -285,20 +286,20 @@ again:
wqe->wr.wr.rdma.rkey,
IB_ACCESS_REMOTE_WRITE))) {
acc_err:
- wc->status = IB_WC_REM_ACCESS_ERR;
+ wc.status = IB_WC_REM_ACCESS_ERR;
err:
- wc->wr_id = wqe->wr.wr_id;
- wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
- wc->vendor_err = 0;
- wc->byte_len = 0;
- wc->qp_num = sqp->ibqp.qp_num;
- wc->src_qp = sqp->remote_qpn;
- wc->pkey_index = 0;
- wc->slid = sqp->remote_ah_attr.dlid;
- wc->sl = sqp->remote_ah_attr.sl;
- wc->dlid_path_bits = 0;
- wc->port_num = 0;
- ipath_sqerror_qp(sqp, wc);
+ wc.wr_id = wqe->wr.wr_id;
+ wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
+ wc.vendor_err = 0;
+ wc.byte_len = 0;
+ wc.qp_num = sqp->ibqp.qp_num;
+ wc.src_qp = sqp->remote_qpn;
+ wc.pkey_index = 0;
+ wc.slid = sqp->remote_ah_attr.dlid;
+ wc.sl = sqp->remote_ah_attr.sl;
+ wc.dlid_path_bits = 0;
+ wc.port_num = 0;
+ ipath_sqerror_qp(sqp, &wc);
goto done;
}
break;
@@ -374,22 +375,22 @@ again:
goto send_comp;
if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
- wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+ wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
else
- wc->opcode = IB_WC_RECV;
- wc->wr_id = qp->r_wr_id;
- wc->status = IB_WC_SUCCESS;
- wc->vendor_err = 0;
- wc->byte_len = wqe->length;
- wc->qp_num = qp->ibqp.qp_num;
- wc->src_qp = qp->remote_qpn;
+ wc.opcode = IB_WC_RECV;
+ wc.wr_id = qp->r_wr_id;
+ wc.status = IB_WC_SUCCESS;
+ wc.vendor_err = 0;
+ wc.byte_len = wqe->length;
+ wc.qp_num = qp->ibqp.qp_num;
+ wc.src_qp = qp->remote_qpn;
/* XXX do we know which pkey matched? Only needed for GSI. */
- wc->pkey_index = 0;
- wc->slid = qp->remote_ah_attr.dlid;
- wc->sl = qp->remote_ah_attr.sl;
- wc->dlid_path_bits = 0;
+ wc.pkey_index = 0;
+ wc.slid = qp->remote_ah_attr.dlid;
+ wc.sl = qp->remote_ah_attr.sl;
+ wc.dlid_path_bits = 0;
/* Signal completion event if the solicited bit is set. */
- ipath_cq_enter(to_icq(qp->ibqp.recv_cq), wc,
+ ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
wqe->wr.send_flags & IB_SEND_SOLICITED);
send_comp:
@@ -397,19 +398,19 @@ send_comp:
if (!test_bit(IPATH_S_SIGNAL_REQ_WR, &sqp->s_flags) ||
(wqe->wr.send_flags & IB_SEND_SIGNALED)) {
- wc->wr_id = wqe->wr.wr_id;
- wc->status = IB_WC_SUCCESS;
- wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
- wc->vendor_err = 0;
- wc->byte_len = wqe->length;
- wc->qp_num = sqp->ibqp.qp_num;
- wc->src_qp = 0;
- wc->pkey_index = 0;
- wc->slid = 0;
- wc->sl = 0;
- wc->dlid_path_bits = 0;
- wc->port_num = 0;
- ipath_cq_enter(to_icq(sqp->ibqp.send_cq), wc, 0);
+ wc.wr_id = wqe->wr.wr_id;
+ wc.status = IB_WC_SUCCESS;
+ wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
+ wc.vendor_err = 0;
+ wc.byte_len = wqe->length;
+ wc.qp_num = sqp->ibqp.qp_num;
+ wc.src_qp = 0;
+ wc.pkey_index = 0;
+ wc.slid = 0;
+ wc.sl = 0;
+ wc.dlid_path_bits = 0;
+ wc.port_num = 0;
+ ipath_cq_enter(to_icq(sqp->ibqp.send_cq), &wc, 0);
}
/* Update s_last now that we are finished with the SWQE */
@@ -455,11 +456,11 @@ void ipath_no_bufs_available(struct ipat
}
/**
- * ipath_post_rc_send - post RC and UC sends
+ * ipath_post_ruc_send - post RC and UC sends
* @qp: the QP to post on
* @wr: the work request to send
*/
-int ipath_post_rc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
+int ipath_post_ruc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
{
struct ipath_swqe *wqe;
unsigned long flags;
@@ -534,13 +535,149 @@ int ipath_post_rc_send(struct ipath_qp *
qp->s_head = next;
spin_unlock_irqrestore(&qp->s_lock, flags);
- if (qp->ibqp.qp_type == IB_QPT_UC)
- ipath_do_uc_send((unsigned long) qp);
- else
- ipath_do_rc_send((unsigned long) qp);
+ ipath_do_ruc_send((unsigned long) qp);
ret = 0;
bail:
return ret;
}
+
+/**
+ * ipath_make_grh - construct a GRH header
+ * @dev: a pointer to the ipath device
+ * @hdr: a pointer to the GRH header being constructed
+ * @grh: the global route address to send to
+ * @hwords: the number of 32 bit words of header being sent
+ * @nwords: the number of 32 bit words of data being sent
+ *
+ * Return the size of the header in 32 bit words.
+ */
+u32 ipath_make_grh(struct ipath_ibdev *dev, struct ib_grh *hdr,
+ struct ib_global_route *grh, u32 hwords, u32 nwords)
+{
+ hdr->version_tclass_flow =
+ cpu_to_be32((6 << 28) |
+ (grh->traffic_class << 20) |
+ grh->flow_label);
+ hdr->paylen = cpu_to_be16((hwords - 2 + nwords + SIZE_OF_CRC) << 2);
+ /* next_hdr is defined by C8-7 in ch. 8.4.1 */
+ hdr->next_hdr = 0x1B;
+ hdr->hop_limit = grh->hop_limit;
+ /* The SGID is 32-bit aligned. */
+ hdr->sgid.global.subnet_prefix = dev->gid_prefix;
+ hdr->sgid.global.interface_id = ipath_layer_get_guid(dev->dd);
+ hdr->dgid = grh->dgid;
+
+ /* GRH header size in 32-bit words. */
+ return sizeof(struct ib_grh) / sizeof(u32);
+}
+
+/**
+ * ipath_do_ruc_send - perform a send on an RC or UC QP
+ * @data: contains a pointer to the QP
+ *
+ * Process entries in the send work queue until credit or queue is
+ * exhausted. Only allow one CPU to send a packet per QP (tasklet).
+ * Otherwise, after we drop the QP s_lock, two threads could send
+ * packets out of order.
+ */
+void ipath_do_ruc_send(unsigned long data)
+{
+ struct ipath_qp *qp = (struct ipath_qp *)data;
+ struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
+ unsigned long flags;
+ u16 lrh0;
+ u32 nwords;
+ u32 extra_bytes;
+ u32 bth0;
+ u32 bth2;
+ u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+ struct ipath_other_headers *ohdr;
+
+ if (test_and_set_bit(IPATH_S_BUSY, &qp->s_flags))
+ goto bail;
+
+ if (unlikely(qp->remote_ah_attr.dlid ==
+ ipath_layer_get_lid(dev->dd))) {
+ ipath_ruc_loopback(qp);
+ goto clear;
+ }
+
+ ohdr = &qp->s_hdr.u.oth;
+ if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+ ohdr = &qp->s_hdr.u.l.oth;
+
+again:
+ /* Check for a constructed packet to be sent. */
+ if (qp->s_hdrwords != 0) {
+ /*
+ * If no PIO bufs are available, return. An interrupt will
+ * call ipath_ib_piobufavail() when one is available.
+ */
+ if (ipath_verbs_send(dev->dd, qp->s_hdrwords,
+ (u32 *) &qp->s_hdr, qp->s_cur_size,
+ qp->s_cur_sge)) {
+ ipath_no_bufs_available(qp, dev);
+ goto bail;
+ }
+ dev->n_unicast_xmit++;
+ /* Record that we sent the packet and s_hdr is empty. */
+ qp->s_hdrwords = 0;
+ }
+
+ /*
+ * The lock is needed to synchronize between setting
+ * qp->s_ack_state, resend timer, and post_send().
+ */
+ spin_lock_irqsave(&qp->s_lock, flags);
+
+ /* Sending responses has higher priority over sending requests. */
+ if (qp->s_ack_state != IB_OPCODE_RC_ACKNOWLEDGE &&
+ (bth0 = ipath_make_rc_ack(qp, ohdr, pmtu)) != 0)
+ bth2 = qp->s_ack_psn++ & IPS_PSN_MASK;
+ else if (!((qp->ibqp.qp_type == IB_QPT_RC) ?
+ ipath_make_rc_req(qp, ohdr, pmtu, &bth0, &bth2) :
+ ipath_make_uc_req(qp, ohdr, pmtu, &bth0, &bth2))) {
+ /*
+ * Clear the busy bit before unlocking to avoid races with
+ * adding new work queue items and then failing to process
+ * them.
+ */
+ clear_bit(IPATH_S_BUSY, &qp->s_flags);
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+ goto bail;
+ }
+
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+
+ /* Construct the header. */
+ extra_bytes = (4 - qp->s_cur_size) & 3;
+ nwords = (qp->s_cur_size + extra_bytes) >> 2;
+ lrh0 = IPS_LRH_BTH;
+ if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+ qp->s_hdrwords += ipath_make_grh(dev, &qp->s_hdr.u.l.grh,
+ &qp->remote_ah_attr.grh,
+ qp->s_hdrwords, nwords);
+ lrh0 = IPS_LRH_GRH;
+ }
+ lrh0 |= qp->remote_ah_attr.sl << 4;
+ qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
+ qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+ qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords +
+ SIZE_OF_CRC);
+ qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
+ bth0 |= ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
+ bth0 |= extra_bytes << 20;
+ ohdr->bth[0] = cpu_to_be32(bth0);
+ ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+ ohdr->bth[2] = cpu_to_be32(bth2);
+
+ /* Check for more work to do. */
+ goto again;
+
+clear:
+ clear_bit(IPATH_S_BUSY, &qp->s_flags);
+bail:
+ return;
+}
diff -r f7c82500b9c7 -r ebf646d10db0 drivers/infiniband/hw/ipath/ipath_uc.c
--- a/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:25 2006 -0700
@@ -62,90 +62,40 @@ static void complete_last_send(struct ip
}
/**
- * ipath_do_uc_send - do a send on a UC queue
- * @data: contains a pointer to the QP to send on
- *
- * Process entries in the send work queue until the queue is exhausted.
- * Only allow one CPU to send a packet per QP (tasklet).
- * Otherwise, after we drop the QP lock, two threads could send
- * packets out of order.
- * This is similar to ipath_do_rc_send() below except we don't have
- * timeouts or resends.
+ * ipath_make_uc_req - construct a request packet (SEND, RDMA write)
+ * @qp: a pointer to the QP
+ * @ohdr: a pointer to the IB header being constructed
+ * @pmtu: the path MTU
+ * @bth0p: pointer to the BTH opcode word
+ * @bth2p: pointer to the BTH PSN word
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ * Note the QP s_lock must be held and interrupts disabled.
*/
-void ipath_do_uc_send(unsigned long data)
+int ipath_make_uc_req(struct ipath_qp *qp,
+ struct ipath_other_headers *ohdr,
+ u32 pmtu, u32 *bth0p, u32 *bth2p)
{
- struct ipath_qp *qp = (struct ipath_qp *)data;
- struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
struct ipath_swqe *wqe;
- unsigned long flags;
- u16 lrh0;
u32 hwords;
- u32 nwords;
- u32 extra_bytes;
u32 bth0;
- u32 bth2;
- u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
u32 len;
- struct ipath_other_headers *ohdr;
struct ib_wc wc;
- if (test_and_set_bit(IPATH_S_BUSY, &qp->s_flags))
- goto bail;
-
- if (unlikely(qp->remote_ah_attr.dlid ==
- ipath_layer_get_lid(dev->dd))) {
- /* Pass in an uninitialized ib_wc to save stack space. */
- ipath_ruc_loopback(qp, &wc);
- clear_bit(IPATH_S_BUSY, &qp->s_flags);
- goto bail;
- }
-
- ohdr = &qp->s_hdr.u.oth;
- if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
- ohdr = &qp->s_hdr.u.l.oth;
-
-again:
- /* Check for a constructed packet to be sent. */
- if (qp->s_hdrwords != 0) {
- /*
- * If no PIO bufs are available, return.
- * An interrupt will call ipath_ib_piobufavail()
- * when one is available.
- */
- if (ipath_verbs_send(dev->dd, qp->s_hdrwords,
- (u32 *) &qp->s_hdr,
- qp->s_cur_size,
- qp->s_cur_sge)) {
- ipath_no_bufs_available(qp, dev);
- goto bail;
- }
- dev->n_unicast_xmit++;
- /* Record that we sent the packet and s_hdr is empty. */
- qp->s_hdrwords = 0;
- }
-
- lrh0 = IPS_LRH_BTH;
+ if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK))
+ goto done;
+
/* header size in 32-bit words LRH+BTH = (8+12)/4. */
hwords = 5;
-
- /*
- * The lock is needed to synchronize between
- * setting qp->s_ack_state and post_send().
- */
- spin_lock_irqsave(&qp->s_lock, flags);
-
- if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK))
- goto done;
-
- bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
-
- /* Send a request. */
+ bth0 = 0;
+
+ /* Get the next send request. */
wqe = get_swqe_ptr(qp, qp->s_last);
switch (qp->s_state) {
default:
/*
- * Signal the completion of the last send (if there is
- * one).
+ * Signal the completion of the last send
+ * (if there is one).
*/
if (qp->s_last != qp->s_tail)
complete_last_send(qp, wqe, &wc);
@@ -258,61 +208,16 @@ again:
}
break;
}
- bth2 = qp->s_next_psn++ & IPS_PSN_MASK;
qp->s_len -= len;
- bth0 |= qp->s_state << 24;
-
- spin_unlock_irqrestore(&qp->s_lock, flags);
-
- /* Construct the header. */
- extra_bytes = (4 - len) & 3;
- nwords = (len + extra_bytes) >> 2;
- if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
- /* Header size in 32-bit words. */
- hwords += 10;
- lrh0 = IPS_LRH_GRH;
- qp->s_hdr.u.l.grh.version_tclass_flow =
- cpu_to_be32((6 << 28) |
- (qp->remote_ah_attr.grh.traffic_class
- << 20) |
- qp->remote_ah_attr.grh.flow_label);
- qp->s_hdr.u.l.grh.paylen =
- cpu_to_be16(((hwords - 12) + nwords +
- SIZE_OF_CRC) << 2);
- /* next_hdr is defined by C8-7 in ch. 8.4.1 */
- qp->s_hdr.u.l.grh.next_hdr = 0x1B;
- qp->s_hdr.u.l.grh.hop_limit =
- qp->remote_ah_attr.grh.hop_limit;
- /* The SGID is 32-bit aligned. */
- qp->s_hdr.u.l.grh.sgid.global.subnet_prefix =
- dev->gid_prefix;
- qp->s_hdr.u.l.grh.sgid.global.interface_id =
- ipath_layer_get_guid(dev->dd);
- qp->s_hdr.u.l.grh.dgid = qp->remote_ah_attr.grh.dgid;
- }
qp->s_hdrwords = hwords;
qp->s_cur_sge = &qp->s_sge;
qp->s_cur_size = len;
- lrh0 |= qp->remote_ah_attr.sl << 4;
- qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
- /* DEST LID */
- qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
- qp->s_hdr.lrh[2] = cpu_to_be16(hwords + nwords + SIZE_OF_CRC);
- qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
- bth0 |= extra_bytes << 20;
- ohdr->bth[0] = cpu_to_be32(bth0);
- ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
- ohdr->bth[2] = cpu_to_be32(bth2);
-
- /* Check for more work to do. */
- goto again;
+ *bth0p = bth0 | (qp->s_state << 24);
+ *bth2p = qp->s_next_psn++ & IPS_PSN_MASK;
+ return 1;
done:
- spin_unlock_irqrestore(&qp->s_lock, flags);
- clear_bit(IPATH_S_BUSY, &qp->s_flags);
-
-bail:
- return;
+ return 0;
}
/**
@@ -536,12 +441,13 @@ void ipath_uc_rcv(struct ipath_ibdev *de
if (qp->r_len != 0) {
u32 rkey = be32_to_cpu(reth->rkey);
u64 vaddr = be64_to_cpu(reth->vaddr);
+ int ok;
/* Check rkey */
- if (unlikely(!ipath_rkey_ok(
- dev, &qp->r_sge, qp->r_len,
- vaddr, rkey,
- IB_ACCESS_REMOTE_WRITE))) {
+ ok = ipath_rkey_ok(dev, &qp->r_sge, qp->r_len,
+ vaddr, rkey,
+ IB_ACCESS_REMOTE_WRITE);
+ if (unlikely(!ok)) {
dev->n_pkt_drops++;
goto done;
}
@@ -559,8 +465,7 @@ void ipath_uc_rcv(struct ipath_ibdev *de
}
if (opcode == OP(RDMA_WRITE_ONLY))
goto rdma_last;
- else if (opcode ==
- OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+ else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
goto rdma_last_imm;
/* FALLTHROUGH */
case OP(RDMA_WRITE_MIDDLE):
@@ -593,9 +498,9 @@ void ipath_uc_rcv(struct ipath_ibdev *de
dev->n_pkt_drops++;
goto done;
}
- if (qp->r_reuse_sge) {
+ if (qp->r_reuse_sge)
qp->r_reuse_sge = 0;
- } else if (!ipath_get_rwqe(qp, 1)) {
+ else if (!ipath_get_rwqe(qp, 1)) {
dev->n_pkt_drops++;
goto done;
}
diff -r f7c82500b9c7 -r ebf646d10db0 drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
@@ -194,7 +194,7 @@ static int ipath_post_send(struct ib_qp
switch (qp->ibqp.qp_type) {
case IB_QPT_UC:
case IB_QPT_RC:
- err = ipath_post_rc_send(qp, wr);
+ err = ipath_post_ruc_send(qp, wr);
break;
case IB_QPT_SMI:
diff -r f7c82500b9c7 -r ebf646d10db0 drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
@@ -581,10 +581,6 @@ void ipath_sqerror_qp(struct ipath_qp *q
void ipath_get_credit(struct ipath_qp *qp, u32 aeth);
-void ipath_do_rc_send(unsigned long data);
-
-void ipath_do_uc_send(unsigned long data);
-
void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int sig);
int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss,
@@ -597,7 +593,7 @@ void ipath_copy_sge(struct ipath_sge_sta
void ipath_skip_sge(struct ipath_sge_state *ss, u32 length);
-int ipath_post_rc_send(struct ipath_qp *qp, struct ib_send_wr *wr);
+int ipath_post_ruc_send(struct ipath_qp *qp, struct ib_send_wr *wr);
void ipath_uc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
int has_grh, void *data, u32 tlen, struct ipath_qp *qp);
@@ -679,7 +675,19 @@ void ipath_insert_rnr_queue(struct ipath
int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only);
-void ipath_ruc_loopback(struct ipath_qp *sqp, struct ib_wc *wc);
+u32 ipath_make_grh(struct ipath_ibdev *dev, struct ib_grh *hdr,
+ struct ib_global_route *grh, u32 hwords, u32 nwords);
+
+void ipath_do_ruc_send(unsigned long data);
+
+u32 ipath_make_rc_ack(struct ipath_qp *qp, struct ipath_other_headers *ohdr,
+ u32 pmtu);
+
+int ipath_make_rc_req(struct ipath_qp *qp, struct ipath_other_headers *ohdr,
+ u32 pmtu, u32 *bth0p, u32 *bth2p);
+
+int ipath_make_uc_req(struct ipath_qp *qp, struct ipath_other_headers *ohdr,
+ u32 pmtu, u32 *bth0p, u32 *bth2p);
extern const enum ib_wc_opcode ib_ipath_wc_opcode[];
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 4 of 39] IB/ipath - fix an indenting problem
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (2 preceding siblings ...)
2006-06-29 21:40 ` [PATCH 3 of 39] IB/ipath - Share more common code between RC and UC protocols Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 5 of 39] IB/ipath - fix shared receive queues for RC Bryan O'Sullivan
` (35 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r ebf646d10db0 -r c93c2b42d279 drivers/infiniband/hw/ipath/ipath_rc.c
--- a/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
@@ -1053,32 +1053,32 @@ static inline void ipath_rc_rcv_resp(str
goto ack_done;
}
rdma_read:
- if (unlikely(qp->s_state != OP(RDMA_READ_REQUEST)))
- goto ack_done;
- if (unlikely(tlen != (hdrsize + pmtu + 4)))
- goto ack_done;
- if (unlikely(pmtu >= qp->s_len))
- goto ack_done;
- /* We got a response so update the timeout. */
- if (unlikely(qp->s_last == qp->s_tail ||
- get_swqe_ptr(qp, qp->s_last)->wr.opcode !=
- IB_WR_RDMA_READ))
- goto ack_done;
- spin_lock(&dev->pending_lock);
- if (qp->s_rnr_timeout == 0 && !list_empty(&qp->timerwait))
- list_move_tail(&qp->timerwait,
- &dev->pending[dev->pending_index]);
- spin_unlock(&dev->pending_lock);
- /*
- * Update the RDMA receive state but do the copy w/o holding the
- * locks and blocking interrupts. XXX Yet another place that
- * affects relaxed RDMA order since we don't want s_sge modified.
- */
- qp->s_len -= pmtu;
- qp->s_last_psn = psn;
- spin_unlock_irqrestore(&qp->s_lock, flags);
- ipath_copy_sge(&qp->s_sge, data, pmtu);
- goto bail;
+ if (unlikely(qp->s_state != OP(RDMA_READ_REQUEST)))
+ goto ack_done;
+ if (unlikely(tlen != (hdrsize + pmtu + 4)))
+ goto ack_done;
+ if (unlikely(pmtu >= qp->s_len))
+ goto ack_done;
+ /* We got a response so update the timeout. */
+ if (unlikely(qp->s_last == qp->s_tail ||
+ get_swqe_ptr(qp, qp->s_last)->wr.opcode !=
+ IB_WR_RDMA_READ))
+ goto ack_done;
+ spin_lock(&dev->pending_lock);
+ if (qp->s_rnr_timeout == 0 && !list_empty(&qp->timerwait))
+ list_move_tail(&qp->timerwait,
+ &dev->pending[dev->pending_index]);
+ spin_unlock(&dev->pending_lock);
+ /*
+ * Update the RDMA receive state but do the copy w/o holding the
+ * locks and blocking interrupts. XXX Yet another place that
+ * affects relaxed RDMA order since we don't want s_sge modified.
+ */
+ qp->s_len -= pmtu;
+ qp->s_last_psn = psn;
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+ ipath_copy_sge(&qp->s_sge, data, pmtu);
+ goto bail;
case OP(RDMA_READ_RESPONSE_LAST):
/* ACKs READ req. */
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 5 of 39] IB/ipath - fix shared receive queues for RC
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (3 preceding siblings ...)
2006-06-29 21:40 ` [PATCH 4 of 39] IB/ipath - fix an indenting problem Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 6 of 39] IB/ipath - Allow diags on any unit Bryan O'Sullivan
` (34 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r c93c2b42d279 -r e4f29a4e0c0f drivers/infiniband/hw/ipath/ipath_rc.c
--- a/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:25 2006 -0700
@@ -257,7 +257,7 @@ int ipath_make_rc_req(struct ipath_qp *q
break;
case IB_WR_RDMA_WRITE:
- if (newreq)
+ if (newreq && qp->s_lsn != (u32) -1)
qp->s_lsn++;
/* FALLTHROUGH */
case IB_WR_RDMA_WRITE_WITH_IMM:
@@ -283,8 +283,7 @@ int ipath_make_rc_req(struct ipath_qp *q
else {
qp->s_state =
OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
- /* Immediate data comes
- * after RETH */
+ /* Immediate data comes after RETH */
ohdr->u.rc.imm_data = wqe->wr.imm_data;
hwords += 1;
if (wqe->wr.send_flags & IB_SEND_SOLICITED)
@@ -304,7 +303,8 @@ int ipath_make_rc_req(struct ipath_qp *q
qp->s_state = OP(RDMA_READ_REQUEST);
hwords += sizeof(ohdr->u.rc.reth) / 4;
if (newreq) {
- qp->s_lsn++;
+ if (qp->s_lsn != (u32) -1)
+ qp->s_lsn++;
/*
* Adjust s_next_psn to count the
* expected number of responses.
@@ -335,7 +335,8 @@ int ipath_make_rc_req(struct ipath_qp *q
wqe->wr.wr.atomic.compare_add);
hwords += sizeof(struct ib_atomic_eth) / 4;
if (newreq) {
- qp->s_lsn++;
+ if (qp->s_lsn != (u32) -1)
+ qp->s_lsn++;
wqe->lpsn = wqe->psn;
}
if (++qp->s_cur == qp->s_size)
@@ -553,6 +554,88 @@ static void send_rc_ack(struct ipath_qp
}
/**
+ * reset_psn - reset the QP state to send starting from PSN
+ * @qp: the QP
+ * @psn: the packet sequence number to restart at
+ *
+ * This is called from ipath_rc_rcv() to process an incoming RC ACK
+ * for the given QP.
+ * Called at interrupt level with the QP s_lock held.
+ */
+static void reset_psn(struct ipath_qp *qp, u32 psn)
+{
+ u32 n = qp->s_last;
+ struct ipath_swqe *wqe = get_swqe_ptr(qp, n);
+ u32 opcode;
+
+ qp->s_cur = n;
+
+ /*
+ * If we are starting the request from the beginning,
+ * let the normal send code handle initialization.
+ */
+ if (ipath_cmp24(psn, wqe->psn) <= 0) {
+ qp->s_state = OP(SEND_LAST);
+ goto done;
+ }
+
+ /* Find the work request opcode corresponding to the given PSN. */
+ opcode = wqe->wr.opcode;
+ for (;;) {
+ int diff;
+
+ if (++n == qp->s_size)
+ n = 0;
+ if (n == qp->s_tail)
+ break;
+ wqe = get_swqe_ptr(qp, n);
+ diff = ipath_cmp24(psn, wqe->psn);
+ if (diff < 0)
+ break;
+ qp->s_cur = n;
+ /*
+ * If we are starting the request from the beginning,
+ * let the normal send code handle initialization.
+ */
+ if (diff == 0) {
+ qp->s_state = OP(SEND_LAST);
+ goto done;
+ }
+ opcode = wqe->wr.opcode;
+ }
+
+ /*
+ * Set the state to restart in the middle of a request.
+ * Don't change the s_sge, s_cur_sge, or s_cur_size.
+ * See ipath_do_rc_send().
+ */
+ switch (opcode) {
+ case IB_WR_SEND:
+ case IB_WR_SEND_WITH_IMM:
+ qp->s_state = OP(RDMA_READ_RESPONSE_FIRST);
+ break;
+
+ case IB_WR_RDMA_WRITE:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ qp->s_state = OP(RDMA_READ_RESPONSE_LAST);
+ break;
+
+ case IB_WR_RDMA_READ:
+ qp->s_state = OP(RDMA_READ_RESPONSE_MIDDLE);
+ break;
+
+ default:
+ /*
+ * This case shouldn't happen since its only
+ * one PSN per req.
+ */
+ qp->s_state = OP(SEND_LAST);
+ }
+done:
+ qp->s_psn = psn;
+}
+
+/**
* ipath_restart_rc - back up requester to resend the last un-ACKed request
* @qp: the QP to restart
* @psn: packet sequence number for the request
@@ -564,7 +647,6 @@ void ipath_restart_rc(struct ipath_qp *q
{
struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
struct ipath_ibdev *dev;
- u32 n;
/*
* If there are no requests pending, we are done.
@@ -606,130 +688,13 @@ void ipath_restart_rc(struct ipath_qp *q
else
dev->n_rc_resends += (int)qp->s_psn - (int)psn;
- /*
- * If we are starting the request from the beginning, let the normal
- * send code handle initialization.
- */
- qp->s_cur = qp->s_last;
- if (ipath_cmp24(psn, wqe->psn) <= 0) {
- qp->s_state = OP(SEND_LAST);
- qp->s_psn = wqe->psn;
- } else {
- n = qp->s_cur;
- for (;;) {
- if (++n == qp->s_size)
- n = 0;
- if (n == qp->s_tail) {
- if (ipath_cmp24(psn, qp->s_next_psn) >= 0) {
- qp->s_cur = n;
- wqe = get_swqe_ptr(qp, n);
- }
- break;
- }
- wqe = get_swqe_ptr(qp, n);
- if (ipath_cmp24(psn, wqe->psn) < 0)
- break;
- qp->s_cur = n;
- }
- qp->s_psn = psn;
-
- /*
- * Reset the state to restart in the middle of a request.
- * Don't change the s_sge, s_cur_sge, or s_cur_size.
- * See ipath_do_rc_send().
- */
- switch (wqe->wr.opcode) {
- case IB_WR_SEND:
- case IB_WR_SEND_WITH_IMM:
- qp->s_state = OP(RDMA_READ_RESPONSE_FIRST);
- break;
-
- case IB_WR_RDMA_WRITE:
- case IB_WR_RDMA_WRITE_WITH_IMM:
- qp->s_state = OP(RDMA_READ_RESPONSE_LAST);
- break;
-
- case IB_WR_RDMA_READ:
- qp->s_state =
- OP(RDMA_READ_RESPONSE_MIDDLE);
- break;
-
- default:
- /*
- * This case shouldn't happen since its only
- * one PSN per req.
- */
- qp->s_state = OP(SEND_LAST);
- }
- }
+ reset_psn(qp, psn);
done:
tasklet_hi_schedule(&qp->s_task);
bail:
return;
-}
-
-/**
- * reset_psn - reset the QP state to send starting from PSN
- * @qp: the QP
- * @psn: the packet sequence number to restart at
- *
- * This is called from ipath_rc_rcv_resp() to process an incoming RC ACK
- * for the given QP.
- * Called at interrupt level with the QP s_lock held.
- */
-static void reset_psn(struct ipath_qp *qp, u32 psn)
-{
- struct ipath_swqe *wqe;
- u32 n;
-
- n = qp->s_cur;
- wqe = get_swqe_ptr(qp, n);
- for (;;) {
- if (++n == qp->s_size)
- n = 0;
- if (n == qp->s_tail) {
- if (ipath_cmp24(psn, qp->s_next_psn) >= 0) {
- qp->s_cur = n;
- wqe = get_swqe_ptr(qp, n);
- }
- break;
- }
- wqe = get_swqe_ptr(qp, n);
- if (ipath_cmp24(psn, wqe->psn) < 0)
- break;
- qp->s_cur = n;
- }
- qp->s_psn = psn;
-
- /*
- * Set the state to restart in the middle of a
- * request. Don't change the s_sge, s_cur_sge, or
- * s_cur_size. See ipath_do_rc_send().
- */
- switch (wqe->wr.opcode) {
- case IB_WR_SEND:
- case IB_WR_SEND_WITH_IMM:
- qp->s_state = OP(RDMA_READ_RESPONSE_FIRST);
- break;
-
- case IB_WR_RDMA_WRITE:
- case IB_WR_RDMA_WRITE_WITH_IMM:
- qp->s_state = OP(RDMA_READ_RESPONSE_LAST);
- break;
-
- case IB_WR_RDMA_READ:
- qp->s_state = OP(RDMA_READ_RESPONSE_MIDDLE);
- break;
-
- default:
- /*
- * This case shouldn't happen since its only
- * one PSN per req.
- */
- qp->s_state = OP(SEND_LAST);
- }
}
/**
@@ -738,7 +703,7 @@ static void reset_psn(struct ipath_qp *q
* @psn: the packet sequence number of the ACK
* @opcode: the opcode of the request that resulted in the ACK
*
- * This is called from ipath_rc_rcv() to process an incoming RC ACK
+ * This is called from ipath_rc_rcv_resp() to process an incoming RC ACK
* for the given QP.
* Called at interrupt level with the QP s_lock held.
* Returns 1 if OK, 0 if current operation should be aborted (NAK).
@@ -877,22 +842,12 @@ static int do_rc_ack(struct ipath_qp *qp
if (qp->s_last == qp->s_tail)
goto bail;
- /* The last valid PSN seen is the previous request's. */
- qp->s_last_psn = wqe->psn - 1;
+ /* The last valid PSN is the previous PSN. */
+ qp->s_last_psn = psn - 1;
dev->n_rc_resends += (int)qp->s_psn - (int)psn;
- /*
- * If we are starting the request from the beginning, let
- * the normal send code handle initialization.
- */
- qp->s_cur = qp->s_last;
- wqe = get_swqe_ptr(qp, qp->s_cur);
- if (ipath_cmp24(psn, wqe->psn) <= 0) {
- qp->s_state = OP(SEND_LAST);
- qp->s_psn = wqe->psn;
- } else
- reset_psn(qp, psn);
+ reset_psn(qp, psn);
qp->s_rnr_timeout =
ib_ipath_rnr_table[(aeth >> IPS_AETH_CREDIT_SHIFT) &
@@ -1070,9 +1025,10 @@ static inline void ipath_rc_rcv_resp(str
&dev->pending[dev->pending_index]);
spin_unlock(&dev->pending_lock);
/*
- * Update the RDMA receive state but do the copy w/o holding the
- * locks and blocking interrupts. XXX Yet another place that
- * affects relaxed RDMA order since we don't want s_sge modified.
+ * Update the RDMA receive state but do the copy w/o
+ * holding the locks and blocking interrupts.
+ * XXX Yet another place that affects relaxed RDMA order
+ * since we don't want s_sge modified.
*/
qp->s_len -= pmtu;
qp->s_last_psn = psn;
@@ -1119,9 +1075,12 @@ static inline void ipath_rc_rcv_resp(str
if (do_rc_ack(qp, aeth, psn, OP(RDMA_READ_RESPONSE_LAST))) {
/*
* Change the state so we contimue
- * processing new requests.
+ * processing new requests and wake up the
+ * tasklet if there are posted sends.
*/
qp->s_state = OP(SEND_LAST);
+ if (qp->s_tail != qp->s_head)
+ tasklet_hi_schedule(&qp->s_task);
}
goto ack_done;
}
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 6 of 39] IB/ipath - Allow diags on any unit
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (4 preceding siblings ...)
2006-06-29 21:40 ` [PATCH 5 of 39] IB/ipath - fix shared receive queues for RC Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 7 of 39] IB/ipath - update some comments and fix typos Bryan O'Sullivan
` (33 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
There is no longer a /dev/ipath_diag file; instead, there's
/dev/ipath_diag0, 1, etc.
It's still not possible to have diags run on more than one unit at a time,
but that's easy to fix at some point.
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r e4f29a4e0c0f -r 600ceb6aeb8c drivers/infiniband/hw/ipath/ipath_diag.c
--- a/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:25 2006 -0700
@@ -67,18 +67,20 @@ static struct file_operations diag_file_
.release = ipath_diag_release
};
-static struct cdev *diag_cdev;
-static struct class_device *diag_class_dev;
-
-int ipath_diag_init(void)
-{
- return ipath_cdev_init(IPATH_DIAG_MINOR, "ipath_diag",
- &diag_file_ops, &diag_cdev, &diag_class_dev);
-}
-
-void ipath_diag_cleanup(void)
-{
- ipath_cdev_cleanup(&diag_cdev, &diag_class_dev);
+int ipath_diag_add(struct ipath_devdata *dd)
+{
+ char name[16];
+
+ snprintf(name, sizeof(name), "ipath_diag%d", dd->ipath_unit);
+
+ return ipath_cdev_init(IPATH_DIAG_MINOR_BASE + dd->ipath_unit, name,
+ &diag_file_ops, &dd->diag_cdev,
+ &dd->diag_class_dev);
+}
+
+void ipath_diag_remove(struct ipath_devdata *dd)
+{
+ ipath_cdev_cleanup(&dd->diag_cdev, &dd->diag_class_dev);
}
/**
@@ -102,8 +104,7 @@ static int ipath_read_umem64(struct ipat
int ret;
/* not very efficient, but it works for now */
- if (reg_addr < dd->ipath_kregbase ||
- reg_end > dd->ipath_kregend) {
+ if (reg_addr < dd->ipath_kregbase || reg_end > dd->ipath_kregend) {
ret = -EINVAL;
goto bail;
}
@@ -140,8 +141,7 @@ static int ipath_write_umem64(struct ipa
int ret;
/* not very efficient, but it works for now */
- if (reg_addr < dd->ipath_kregbase ||
- reg_end > dd->ipath_kregend) {
+ if (reg_addr < dd->ipath_kregbase || reg_end > dd->ipath_kregend) {
ret = -EINVAL;
goto bail;
}
@@ -240,59 +240,45 @@ bail:
static int ipath_diag_open(struct inode *in, struct file *fp)
{
+ int unit = iminor(in) - IPATH_DIAG_MINOR_BASE;
struct ipath_devdata *dd;
- int unit = 0; /* XXX this is bogus */
- unsigned long flags;
- int ret;
-
- dd = ipath_lookup(unit);
+ int ret;
mutex_lock(&ipath_mutex);
- spin_lock_irqsave(&ipath_devs_lock, flags);
if (ipath_diag_inuse) {
ret = -EBUSY;
goto bail;
}
- list_for_each_entry(dd, &ipath_dev_list, ipath_list) {
- /*
- * we need at least one infinipath device to be present
- * (don't use INITTED, because we want to be able to open
- * even if device is in freeze mode, which cleared INITTED).
- * There is a small amount of risk to this, which is why we
- * also verify kregbase is set.
- */
-
- if (!(dd->ipath_flags & IPATH_PRESENT) ||
- !dd->ipath_kregbase)
- continue;
-
- ipath_diag_inuse = 1;
- diag_set_link = 0;
- ret = 0;
- goto bail;
- }
-
- ret = -ENODEV;
-
-bail:
- spin_unlock_irqrestore(&ipath_devs_lock, flags);
+ dd = ipath_lookup(unit);
+
+ if (dd == NULL || !(dd->ipath_flags & IPATH_PRESENT) ||
+ !dd->ipath_kregbase) {
+ ret = -ENODEV;
+ goto bail;
+ }
+
+ fp->private_data = dd;
+ ipath_diag_inuse = 1;
+ diag_set_link = 0;
+ ret = 0;
/* Only expose a way to reset the device if we
make it into diag mode. */
- if (ret == 0)
- ipath_expose_reset(&dd->pcidev->dev);
-
+ ipath_expose_reset(&dd->pcidev->dev);
+
+bail:
mutex_unlock(&ipath_mutex);
return ret;
}
-static int ipath_diag_release(struct inode *i, struct file *f)
+static int ipath_diag_release(struct inode *in, struct file *fp)
{
mutex_lock(&ipath_mutex);
ipath_diag_inuse = 0;
+ fp->private_data = NULL;
mutex_unlock(&ipath_mutex);
return 0;
}
@@ -300,16 +286,9 @@ static ssize_t ipath_diag_read(struct fi
static ssize_t ipath_diag_read(struct file *fp, char __user *data,
size_t count, loff_t *off)
{
- int unit = 0; /* XXX provide for reads on other units some day */
- struct ipath_devdata *dd;
+ struct ipath_devdata *dd = fp->private_data;
void __iomem *kreg_base;
ssize_t ret;
-
- dd = ipath_lookup(unit);
- if (!dd) {
- ret = -ENODEV;
- goto bail;
- }
kreg_base = dd->ipath_kregbase;
@@ -329,23 +308,16 @@ static ssize_t ipath_diag_read(struct fi
ret = count;
}
-bail:
return ret;
}
static ssize_t ipath_diag_write(struct file *fp, const char __user *data,
size_t count, loff_t *off)
{
- int unit = 0; /* XXX this is bogus */
- struct ipath_devdata *dd;
+ struct ipath_devdata *dd = fp->private_data;
void __iomem *kreg_base;
ssize_t ret;
- dd = ipath_lookup(unit);
- if (!dd) {
- ret = -ENODEV;
- goto bail;
- }
kreg_base = dd->ipath_kregbase;
if (count == 0)
@@ -364,6 +336,5 @@ static ssize_t ipath_diag_write(struct f
ret = count;
}
-bail:
- return ret;
-}
+ return ret;
+}
diff -r e4f29a4e0c0f -r 600ceb6aeb8c drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -546,6 +546,7 @@ static int __devinit ipath_init_one(stru
ipath_device_create_group(&pdev->dev, dd);
ipathfs_add_device(dd);
ipath_user_add(dd);
+ ipath_diag_add(dd);
ipath_layer_add(dd);
goto bail;
@@ -578,8 +579,9 @@ static void __devexit ipath_remove_one(s
return;
dd = pci_get_drvdata(pdev);
- ipath_layer_del(dd);
- ipath_user_del(dd);
+ ipath_layer_remove(dd);
+ ipath_diag_remove(dd);
+ ipath_user_remove(dd);
ipathfs_remove_device(dd);
ipath_device_remove_group(&pdev->dev, dd);
ipath_cdbg(VERBOSE, "Releasing pci memory regions, dd %p, "
diff -r e4f29a4e0c0f -r 600ceb6aeb8c drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
@@ -1443,16 +1443,16 @@ done:
static int ipath_open(struct inode *in, struct file *fp)
{
- int ret, minor;
+ int ret, user_minor;
mutex_lock(&ipath_mutex);
- minor = iminor(in);
+ user_minor = iminor(in) - IPATH_USER_MINOR_BASE;
ipath_cdbg(VERBOSE, "open on dev %lx (minor %d)\n",
- (long)in->i_rdev, minor);
-
- if (minor)
- ret = find_free_port(minor - 1, fp);
+ (long)in->i_rdev, user_minor);
+
+ if (user_minor)
+ ret = find_free_port(user_minor - 1, fp);
else
ret = find_best_unit(fp);
@@ -1860,19 +1860,12 @@ int ipath_user_add(struct ipath_devdata
"error %d\n", -ret);
goto bail;
}
- ret = ipath_diag_init();
- if (ret < 0) {
- ipath_dev_err(dd, "Unable to set up diag support: "
- "error %d\n", -ret);
- goto bail_sma;
- }
-
ret = init_cdev(0, "ipath", &ipath_file_ops, &wildcard_cdev,
&wildcard_class_dev);
if (ret < 0) {
ipath_dev_err(dd, "Could not create wildcard "
"minor: error %d\n", -ret);
- goto bail_diag;
+ goto bail_sma;
}
atomic_set(&user_setup, 1);
@@ -1881,31 +1874,28 @@ int ipath_user_add(struct ipath_devdata
snprintf(name, sizeof(name), "ipath%d", dd->ipath_unit);
ret = init_cdev(dd->ipath_unit + 1, name, &ipath_file_ops,
- &dd->cdev, &dd->class_dev);
+ &dd->user_cdev, &dd->user_class_dev);
if (ret < 0)
ipath_dev_err(dd, "Could not create user minor %d, %s\n",
dd->ipath_unit + 1, name);
goto bail;
-bail_diag:
- ipath_diag_cleanup();
bail_sma:
user_cleanup();
bail:
return ret;
}
-void ipath_user_del(struct ipath_devdata *dd)
-{
- cleanup_cdev(&dd->cdev, &dd->class_dev);
+void ipath_user_remove(struct ipath_devdata *dd)
+{
+ cleanup_cdev(&dd->user_cdev, &dd->user_class_dev);
if (atomic_dec_return(&user_count) == 0) {
if (atomic_read(&user_setup) == 0)
goto bail;
cleanup_cdev(&wildcard_cdev, &wildcard_class_dev);
- ipath_diag_cleanup();
user_cleanup();
atomic_set(&user_setup, 0);
diff -r e4f29a4e0c0f -r 600ceb6aeb8c drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
@@ -355,8 +355,10 @@ struct ipath_devdata {
char *ipath_freezemsg;
/* pci access data structure */
struct pci_dev *pcidev;
- struct cdev *cdev;
- struct class_device *class_dev;
+ struct cdev *user_cdev;
+ struct cdev *diag_cdev;
+ struct class_device *user_class_dev;
+ struct class_device *diag_class_dev;
/* timer used to prevent stats overflow, error throttling, etc. */
struct timer_list ipath_stats_timer;
/* check for stale messages in rcv queue */
@@ -538,7 +540,7 @@ extern int __ipath_verbs_rcv(struct ipat
extern int __ipath_verbs_rcv(struct ipath_devdata *, void *, void *, u32);
void ipath_layer_add(struct ipath_devdata *);
-void ipath_layer_del(struct ipath_devdata *);
+void ipath_layer_remove(struct ipath_devdata *);
int ipath_init_chip(struct ipath_devdata *, int);
int ipath_enable_wc(struct ipath_devdata *dd);
@@ -552,14 +554,14 @@ void ipath_cdev_cleanup(struct cdev **cd
void ipath_cdev_cleanup(struct cdev **cdevp,
struct class_device **class_devp);
-int ipath_diag_init(void);
-void ipath_diag_cleanup(void);
+int ipath_diag_add(struct ipath_devdata *);
+void ipath_diag_remove(struct ipath_devdata *);
void ipath_diag_bringup_link(struct ipath_devdata *);
extern wait_queue_head_t ipath_sma_state_wait;
int ipath_user_add(struct ipath_devdata *dd);
-void ipath_user_del(struct ipath_devdata *dd);
+void ipath_user_remove(struct ipath_devdata *dd);
struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd, gfp_t);
@@ -843,9 +845,10 @@ extern struct mutex ipath_mutex;
#define IPATH_DRV_NAME "ipath_core"
#define IPATH_MAJOR 233
+#define IPATH_USER_MINOR_BASE 0
#define IPATH_SMA_MINOR 128
-#define IPATH_DIAG_MINOR 129
-#define IPATH_NMINORS 130
+#define IPATH_DIAG_MINOR_BASE 129
+#define IPATH_NMINORS 255
#define ipath_dev_err(dd,fmt,...) \
do { \
diff -r e4f29a4e0c0f -r 600ceb6aeb8c drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
@@ -404,7 +404,7 @@ void ipath_layer_add(struct ipath_devdat
mutex_unlock(&ipath_layer_mutex);
}
-void ipath_layer_del(struct ipath_devdata *dd)
+void ipath_layer_remove(struct ipath_devdata *dd)
{
mutex_lock(&ipath_layer_mutex);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 7 of 39] IB/ipath - update some comments and fix typos
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (5 preceding siblings ...)
2006-06-29 21:40 ` [PATCH 6 of 39] IB/ipath - Allow diags on any unit Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:40 ` [PATCH 8 of 39] IB/ipath - remove some duplicate code Bryan O'Sullivan
` (32 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 600ceb6aeb8c -r 8f08597cacd2 drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
@@ -723,13 +723,8 @@ u64 ipath_read_kreg64_port(const struct
* @port: port number
*
* Return the contents of a register that is virtualized to be per port.
- * Prints a debug message and returns -1 on errors (not distinguishable from
- * valid contents at runtime; we may add a separate error variable at some
- * point).
- *
- * This is normally not used by the kernel, but may be for debugging, and
- * has a different implementation than user mode, which is why it's not in
- * _common.h.
+ * Returns -1 on errors (not distinguishable from valid contents at
+ * runtime; we may add a separate error variable at some point).
*/
static inline u32 ipath_read_ureg32(const struct ipath_devdata *dd,
ipath_ureg regno, int port)
diff -r 600ceb6aeb8c -r 8f08597cacd2 drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
@@ -885,7 +885,7 @@ static void copy_io(u32 __iomem *piobuf,
/**
* ipath_verbs_send - send a packet from the verbs layer
* @dd: the infinipath device
- * @hdrwords: the number of works in the header
+ * @hdrwords: the number of words in the header
* @hdr: the packet header
* @len: the length of the packet in bytes
* @ss: the SGE to send
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 8 of 39] IB/ipath - remove some duplicate code
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (6 preceding siblings ...)
2006-06-29 21:40 ` [PATCH 7 of 39] IB/ipath - update some comments and fix typos Bryan O'Sullivan
@ 2006-06-29 21:40 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 9 of 39] IB/ipath - don't allow resources to be created with illegal values Bryan O'Sullivan
` (31 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 8f08597cacd2 -r 081142011371 drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
@@ -511,9 +511,6 @@ int ipath_modify_qp(struct ib_qp *ibqp,
if (attr_mask & IB_QP_QKEY)
qp->qkey = attr->qkey;
- if (attr_mask & IB_QP_PKEY_INDEX)
- qp->s_pkey_index = attr->pkey_index;
-
qp->state = new_state;
spin_unlock(&qp->s_lock);
spin_unlock_irqrestore(&qp->r_rq.lock, flags);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 9 of 39] IB/ipath - don't allow resources to be created with illegal values
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (7 preceding siblings ...)
2006-06-29 21:40 ` [PATCH 8 of 39] IB/ipath - remove some duplicate code Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 10 of 39] IB/ipath - fix some memory leaks on failure paths Bryan O'Sullivan
` (30 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 081142011371 -r ac81d2563bba drivers/infiniband/hw/ipath/ipath_mr.c
--- a/drivers/infiniband/hw/ipath/ipath_mr.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mr.c Thu Jun 29 14:33:25 2006 -0700
@@ -169,6 +169,11 @@ struct ib_mr *ipath_reg_user_mr(struct i
struct ib_umem_chunk *chunk;
int n, m, i;
struct ib_mr *ret;
+
+ if (region->length == 0) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail;
+ }
n = 0;
list_for_each_entry(chunk, ®ion->chunk_list, list)
diff -r 081142011371 -r ac81d2563bba drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
@@ -667,6 +667,14 @@ struct ib_qp *ipath_create_qp(struct ib_
goto bail;
}
+ if (init_attr->cap.max_send_sge +
+ init_attr->cap.max_recv_sge +
+ init_attr->cap.max_send_wr +
+ init_attr->cap.max_recv_wr == 0) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail;
+ }
+
switch (init_attr->qp_type) {
case IB_QPT_UC:
case IB_QPT_RC:
diff -r 081142011371 -r ac81d2563bba drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
@@ -788,6 +788,17 @@ static struct ib_ah *ipath_create_ah(str
if (ah_attr->dlid >= IPS_MULTICAST_LID_BASE &&
ah_attr->dlid != IPS_PERMISSIVE_LID &&
!(ah_attr->ah_flags & IB_AH_GRH)) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail;
+ }
+
+ if (ah_attr->dlid == 0) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail;
+ }
+
+ if (ah_attr->port_num != 1 ||
+ ah_attr->port_num > pd->device->phys_port_cnt) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 10 of 39] IB/ipath - fix some memory leaks on failure paths
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (8 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 9 of 39] IB/ipath - don't allow resources to be created with illegal values Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 11 of 39] IB/ipath - return an error for unknown multicast GID Bryan O'Sullivan
` (29 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r ac81d2563bba -r 160e5cf91761 drivers/infiniband/hw/ipath/ipath_init_chip.c
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:25 2006 -0700
@@ -115,6 +115,7 @@ static int create_port0_egr(struct ipath
"eager TID %u\n", e);
while (e != 0)
dev_kfree_skb(skbs[--e]);
+ vfree(skbs);
ret = -ENOMEM;
goto bail;
}
diff -r ac81d2563bba -r 160e5cf91761 drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
@@ -692,6 +692,7 @@ struct ib_qp *ipath_create_qp(struct ib_
case IB_QPT_GSI:
qp = kmalloc(sizeof(*qp), GFP_KERNEL);
if (!qp) {
+ vfree(swq);
ret = ERR_PTR(-ENOMEM);
goto bail;
}
@@ -702,6 +703,7 @@ struct ib_qp *ipath_create_qp(struct ib_
qp->r_rq.wq = vmalloc(qp->r_rq.size * sz);
if (!qp->r_rq.wq) {
kfree(qp);
+ vfree(swq);
ret = ERR_PTR(-ENOMEM);
goto bail;
}
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 11 of 39] IB/ipath - return an error for unknown multicast GID
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (9 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 10 of 39] IB/ipath - fix some memory leaks on failure paths Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 12 of 39] IB/ipath - report correct device identification information in /sys Bryan O'Sullivan
` (28 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 160e5cf91761 -r 1e1f3da0e78d drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c Thu Jun 29 14:33:25 2006 -0700
@@ -273,7 +273,7 @@ int ipath_multicast_detach(struct ib_qp
while (1) {
if (n == NULL) {
spin_unlock_irqrestore(&mcast_lock, flags);
- ret = 0;
+ ret = -EINVAL;
goto bail;
}
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 12 of 39] IB/ipath - report correct device identification information in /sys
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (10 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 11 of 39] IB/ipath - return an error for unknown multicast GID Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 13 of 39] IB/ipath - enforce device resource limits Bryan O'Sullivan
` (27 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 1e1f3da0e78d -r 21d5d64750ac drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:25 2006 -0700
@@ -341,18 +341,26 @@ u32 ipath_layer_get_nguid(struct ipath_d
EXPORT_SYMBOL_GPL(ipath_layer_get_nguid);
-int ipath_layer_query_device(struct ipath_devdata *dd, u32 * vendor,
- u32 * boardrev, u32 * majrev, u32 * minrev)
-{
- *vendor = dd->ipath_vendorid;
- *boardrev = dd->ipath_boardrev;
- *majrev = dd->ipath_majrev;
- *minrev = dd->ipath_minrev;
-
- return 0;
-}
-
-EXPORT_SYMBOL_GPL(ipath_layer_query_device);
+u32 ipath_layer_get_majrev(struct ipath_devdata *dd)
+{
+ return dd->ipath_majrev;
+}
+
+EXPORT_SYMBOL_GPL(ipath_layer_get_majrev);
+
+u32 ipath_layer_get_minrev(struct ipath_devdata *dd)
+{
+ return dd->ipath_minrev;
+}
+
+EXPORT_SYMBOL_GPL(ipath_layer_get_minrev);
+
+u32 ipath_layer_get_pcirev(struct ipath_devdata *dd)
+{
+ return dd->ipath_pcirev;
+}
+
+EXPORT_SYMBOL_GPL(ipath_layer_get_pcirev);
u32 ipath_layer_get_flags(struct ipath_devdata *dd)
{
@@ -374,6 +382,13 @@ u16 ipath_layer_get_deviceid(struct ipat
}
EXPORT_SYMBOL_GPL(ipath_layer_get_deviceid);
+
+u32 ipath_layer_get_vendorid(struct ipath_devdata *dd)
+{
+ return dd->ipath_vendorid;
+}
+
+EXPORT_SYMBOL_GPL(ipath_layer_get_vendorid);
u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd)
{
diff -r 1e1f3da0e78d -r 21d5d64750ac drivers/infiniband/hw/ipath/ipath_layer.h
--- a/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:25 2006 -0700
@@ -144,11 +144,13 @@ int ipath_layer_set_guid(struct ipath_de
int ipath_layer_set_guid(struct ipath_devdata *, __be64 guid);
__be64 ipath_layer_get_guid(struct ipath_devdata *);
u32 ipath_layer_get_nguid(struct ipath_devdata *);
-int ipath_layer_query_device(struct ipath_devdata *, u32 * vendor,
- u32 * boardrev, u32 * majrev, u32 * minrev);
+u32 ipath_layer_get_majrev(struct ipath_devdata *);
+u32 ipath_layer_get_minrev(struct ipath_devdata *);
+u32 ipath_layer_get_pcirev(struct ipath_devdata *);
u32 ipath_layer_get_flags(struct ipath_devdata *dd);
struct device *ipath_layer_get_device(struct ipath_devdata *dd);
u16 ipath_layer_get_deviceid(struct ipath_devdata *dd);
+u32 ipath_layer_get_vendorid(struct ipath_devdata *);
u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd);
u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd);
int ipath_layer_enable_timer(struct ipath_devdata *dd);
diff -r 1e1f3da0e78d -r 21d5d64750ac drivers/infiniband/hw/ipath/ipath_mad.c
--- a/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:25 2006 -0700
@@ -85,7 +85,7 @@ static int recv_subn_get_nodeinfo(struct
{
struct nodeinfo *nip = (struct nodeinfo *)&smp->data;
struct ipath_devdata *dd = to_idev(ibdev)->dd;
- u32 vendor, boardid, majrev, minrev;
+ u32 vendor, majrev, minrev;
if (smp->attr_mod)
smp->status |= IB_SMP_INVALID_FIELD;
@@ -105,9 +105,11 @@ static int recv_subn_get_nodeinfo(struct
nip->port_guid = nip->sys_guid;
nip->partition_cap = cpu_to_be16(ipath_layer_get_npkeys(dd));
nip->device_id = cpu_to_be16(ipath_layer_get_deviceid(dd));
- ipath_layer_query_device(dd, &vendor, &boardid, &majrev, &minrev);
+ majrev = ipath_layer_get_majrev(dd);
+ minrev = ipath_layer_get_minrev(dd);
nip->revision = cpu_to_be32((majrev << 16) | minrev);
nip->local_port_num = port;
+ vendor = ipath_layer_get_vendorid(dd);
nip->vendor_id[0] = 0;
nip->vendor_id[1] = vendor >> 8;
nip->vendor_id[2] = vendor;
diff -r 1e1f3da0e78d -r 21d5d64750ac drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
@@ -568,18 +568,15 @@ static int ipath_query_device(struct ib_
struct ib_device_attr *props)
{
struct ipath_ibdev *dev = to_idev(ibdev);
- u32 vendor, boardrev, majrev, minrev;
memset(props, 0, sizeof(*props));
props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
IB_DEVICE_SYS_IMAGE_GUID;
- ipath_layer_query_device(dev->dd, &vendor, &boardrev,
- &majrev, &minrev);
- props->vendor_id = vendor;
- props->vendor_part_id = boardrev;
- props->hw_ver = boardrev << 16 | majrev << 8 | minrev;
+ props->vendor_id = ipath_layer_get_vendorid(dev->dd);
+ props->vendor_part_id = ipath_layer_get_deviceid(dev->dd);
+ props->hw_ver = ipath_layer_get_pcirev(dev->dd);
props->sys_image_guid = dev->sys_image_guid;
@@ -1121,11 +1118,8 @@ static ssize_t show_rev(struct class_dev
{
struct ipath_ibdev *dev =
container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
- int vendor, boardrev, majrev, minrev;
-
- ipath_layer_query_device(dev->dd, &vendor, &boardrev,
- &majrev, &minrev);
- return sprintf(buf, "%d.%d\n", majrev, minrev);
+
+ return sprintf(buf, "%x\n", ipath_layer_get_pcirev(dev->dd));
}
static ssize_t show_hca(struct class_device *cdev, char *buf)
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 13 of 39] IB/ipath - enforce device resource limits
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (11 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 12 of 39] IB/ipath - report correct device identification information in /sys Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 14 of 39] IB/ipath - removed unused field ipath_kregvirt from struct ipath_devdata Bryan O'Sullivan
` (26 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
These limits are somewhat artificial in that we don't actually have any
device limits. However, the verbs layer expects that such limits exist
and are enforced, so we make up arbitrary (but sensible) limits.
Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 21d5d64750ac -r a94e9f9c9c23 drivers/infiniband/hw/ipath/ipath_cq.c
--- a/drivers/infiniband/hw/ipath/ipath_cq.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_cq.c Thu Jun 29 14:33:25 2006 -0700
@@ -158,9 +158,20 @@ struct ib_cq *ipath_create_cq(struct ib_
struct ib_ucontext *context,
struct ib_udata *udata)
{
+ struct ipath_ibdev *dev = to_idev(ibdev);
struct ipath_cq *cq;
struct ib_wc *wc;
struct ib_cq *ret;
+
+ if (entries > ib_ipath_max_cqes) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail;
+ }
+
+ if (dev->n_cqs_allocated == ib_ipath_max_cqs) {
+ ret = ERR_PTR(-ENOMEM);
+ goto bail;
+ }
/*
* Need to use vmalloc() if we want to support large #s of
@@ -197,6 +208,8 @@ struct ib_cq *ipath_create_cq(struct ib_
ret = &cq->ibcq;
+ dev->n_cqs_allocated++;
+
bail:
return ret;
}
@@ -211,9 +224,11 @@ bail:
*/
int ipath_destroy_cq(struct ib_cq *ibcq)
{
+ struct ipath_ibdev *dev = to_idev(ibcq->device);
struct ipath_cq *cq = to_icq(ibcq);
tasklet_kill(&cq->comptask);
+ dev->n_cqs_allocated--;
vfree(cq->queue);
kfree(cq);
diff -r 21d5d64750ac -r a94e9f9c9c23 drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:25 2006 -0700
@@ -661,8 +661,10 @@ struct ib_qp *ipath_create_qp(struct ib_
size_t sz;
struct ib_qp *ret;
- if (init_attr->cap.max_send_sge > 255 ||
- init_attr->cap.max_recv_sge > 255) {
+ if (init_attr->cap.max_send_sge > ib_ipath_max_sges ||
+ init_attr->cap.max_recv_sge > ib_ipath_max_sges ||
+ init_attr->cap.max_send_wr > ib_ipath_max_qp_wrs ||
+ init_attr->cap.max_recv_wr > ib_ipath_max_qp_wrs) {
ret = ERR_PTR(-ENOMEM);
goto bail;
}
diff -r 21d5d64750ac -r a94e9f9c9c23 drivers/infiniband/hw/ipath/ipath_srq.c
--- a/drivers/infiniband/hw/ipath/ipath_srq.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_srq.c Thu Jun 29 14:33:25 2006 -0700
@@ -126,11 +126,23 @@ struct ib_srq *ipath_create_srq(struct i
struct ib_srq_init_attr *srq_init_attr,
struct ib_udata *udata)
{
+ struct ipath_ibdev *dev = to_idev(ibpd->device);
struct ipath_srq *srq;
u32 sz;
struct ib_srq *ret;
- if (srq_init_attr->attr.max_sge < 1) {
+ if (dev->n_srqs_allocated == ib_ipath_max_srqs) {
+ ret = ERR_PTR(-ENOMEM);
+ goto bail;
+ }
+
+ if (srq_init_attr->attr.max_wr == 0) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail;
+ }
+
+ if ((srq_init_attr->attr.max_sge > ib_ipath_max_srq_sges) ||
+ (srq_init_attr->attr.max_wr > ib_ipath_max_srq_wrs)) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
@@ -165,6 +177,8 @@ struct ib_srq *ipath_create_srq(struct i
ret = &srq->ibsrq;
+ dev->n_srqs_allocated++;
+
bail:
return ret;
}
@@ -182,24 +196,26 @@ int ipath_modify_srq(struct ib_srq *ibsr
unsigned long flags;
int ret;
- if (attr_mask & IB_SRQ_LIMIT) {
- spin_lock_irqsave(&srq->rq.lock, flags);
- srq->limit = attr->srq_limit;
- spin_unlock_irqrestore(&srq->rq.lock, flags);
- }
+ if (attr_mask & IB_SRQ_MAX_WR)
+ if ((attr->max_wr > ib_ipath_max_srq_wrs) ||
+ (attr->max_sge > srq->rq.max_sge)) {
+ ret = -EINVAL;
+ goto bail;
+ }
+
+ if (attr_mask & IB_SRQ_LIMIT)
+ if (attr->srq_limit >= srq->rq.size) {
+ ret = -EINVAL;
+ goto bail;
+ }
+
if (attr_mask & IB_SRQ_MAX_WR) {
- u32 size = attr->max_wr + 1;
struct ipath_rwqe *wq, *p;
- u32 n;
- u32 sz;
-
- if (attr->max_sge < srq->rq.max_sge) {
- ret = -EINVAL;
- goto bail;
- }
+ u32 sz, size, n;
sz = sizeof(struct ipath_rwqe) +
attr->max_sge * sizeof(struct ipath_sge);
+ size = attr->max_wr + 1;
wq = vmalloc(size * sz);
if (!wq) {
ret = -ENOMEM;
@@ -243,6 +259,11 @@ int ipath_modify_srq(struct ib_srq *ibsr
spin_unlock_irqrestore(&srq->rq.lock, flags);
}
+ if (attr_mask & IB_SRQ_LIMIT) {
+ spin_lock_irqsave(&srq->rq.lock, flags);
+ srq->limit = attr->srq_limit;
+ spin_unlock_irqrestore(&srq->rq.lock, flags);
+ }
ret = 0;
bail:
@@ -266,7 +287,9 @@ int ipath_destroy_srq(struct ib_srq *ibs
int ipath_destroy_srq(struct ib_srq *ibsrq)
{
struct ipath_srq *srq = to_isrq(ibsrq);
-
+ struct ipath_ibdev *dev = to_idev(ibsrq->device);
+
+ dev->n_srqs_allocated--;
vfree(srq->rq.wq);
kfree(srq);
diff -r 21d5d64750ac -r a94e9f9c9c23 drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:25 2006 -0700
@@ -55,6 +55,59 @@ unsigned int ib_ipath_debug; /* debug ma
unsigned int ib_ipath_debug; /* debug mask */
module_param_named(debug, ib_ipath_debug, uint, S_IWUSR | S_IRUGO);
MODULE_PARM_DESC(debug, "Verbs debug mask");
+
+static unsigned int ib_ipath_max_pds = 0xFFFF;
+module_param_named(max_pds, ib_ipath_max_pds, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_pds,
+ "Maximum number of protection domains to support");
+
+static unsigned int ib_ipath_max_ahs = 0xFFFF;
+module_param_named(max_ahs, ib_ipath_max_ahs, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
+
+unsigned int ib_ipath_max_cqes = 0x2FFFF;
+module_param_named(max_cqes, ib_ipath_max_cqes, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_cqes,
+ "Maximum number of completion queue entries to support");
+
+unsigned int ib_ipath_max_cqs = 0x1FFFF;
+module_param_named(max_cqs, ib_ipath_max_cqs, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
+
+unsigned int ib_ipath_max_qp_wrs = 0x3FFF;
+module_param_named(max_qp_wrs, ib_ipath_max_qp_wrs, uint,
+ S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
+
+unsigned int ib_ipath_max_sges = 0x60;
+module_param_named(max_sges, ib_ipath_max_sges, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
+
+unsigned int ib_ipath_max_mcast_grps = 16384;
+module_param_named(max_mcast_grps, ib_ipath_max_mcast_grps, uint,
+ S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_mcast_grps,
+ "Maximum number of multicast groups to support");
+
+unsigned int ib_ipath_max_mcast_qp_attached = 16;
+module_param_named(max_mcast_qp_attached, ib_ipath_max_mcast_qp_attached,
+ uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_mcast_qp_attached,
+ "Maximum number of attached QPs to support");
+
+unsigned int ib_ipath_max_srqs = 1024;
+module_param_named(max_srqs, ib_ipath_max_srqs, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
+
+unsigned int ib_ipath_max_srq_sges = 128;
+module_param_named(max_srq_sges, ib_ipath_max_srq_sges,
+ uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
+
+unsigned int ib_ipath_max_srq_wrs = 0x1FFFF;
+module_param_named(max_srq_wrs, ib_ipath_max_srq_wrs,
+ uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
MODULE_LICENSE("GPL");
MODULE_AUTHOR("QLogic <support@pathscale.com>");
@@ -581,24 +634,25 @@ static int ipath_query_device(struct ib_
props->sys_image_guid = dev->sys_image_guid;
props->max_mr_size = ~0ull;
- props->max_qp = 0xffff;
- props->max_qp_wr = 0xffff;
- props->max_sge = 255;
- props->max_cq = 0xffff;
- props->max_cqe = 0xffff;
- props->max_mr = 0xffff;
- props->max_pd = 0xffff;
+ props->max_qp = dev->qp_table.max;
+ props->max_qp_wr = ib_ipath_max_qp_wrs;
+ props->max_sge = ib_ipath_max_sges;
+ props->max_cq = ib_ipath_max_cqs;
+ props->max_ah = ib_ipath_max_ahs;
+ props->max_cqe = ib_ipath_max_cqes;
+ props->max_mr = dev->lk_table.max;
+ props->max_pd = ib_ipath_max_pds;
props->max_qp_rd_atom = 1;
props->max_qp_init_rd_atom = 1;
/* props->max_res_rd_atom */
- props->max_srq = 0xffff;
- props->max_srq_wr = 0xffff;
- props->max_srq_sge = 255;
+ props->max_srq = ib_ipath_max_srqs;
+ props->max_srq_wr = ib_ipath_max_srq_wrs;
+ props->max_srq_sge = ib_ipath_max_srq_sges;
/* props->local_ca_ack_delay */
props->atomic_cap = IB_ATOMIC_HCA;
props->max_pkeys = ipath_layer_get_npkeys(dev->dd);
- props->max_mcast_grp = 0xffff;
- props->max_mcast_qp_attach = 0xffff;
+ props->max_mcast_grp = ib_ipath_max_mcast_grps;
+ props->max_mcast_qp_attach = ib_ipath_max_mcast_qp_attached;
props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
props->max_mcast_grp;
@@ -741,8 +795,21 @@ static struct ib_pd *ipath_alloc_pd(stru
struct ib_ucontext *context,
struct ib_udata *udata)
{
+ struct ipath_ibdev *dev = to_idev(ibdev);
struct ipath_pd *pd;
struct ib_pd *ret;
+
+ /*
+ * This is actually totally arbitrary. Some correctness tests
+ * assume there's a maximum number of PDs that can be allocated.
+ * We don't actually have this limit, but we fail the test if
+ * we allow allocations of more than we report for this value.
+ */
+
+ if (dev->n_pds_allocated == ib_ipath_max_pds) {
+ ret = ERR_PTR(-ENOMEM);
+ goto bail;
+ }
pd = kmalloc(sizeof *pd, GFP_KERNEL);
if (!pd) {
@@ -750,6 +817,8 @@ static struct ib_pd *ipath_alloc_pd(stru
goto bail;
}
+ dev->n_pds_allocated++;
+
/* ib_alloc_pd() will initialize pd->ibpd. */
pd->user = udata != NULL;
@@ -762,6 +831,9 @@ static int ipath_dealloc_pd(struct ib_pd
static int ipath_dealloc_pd(struct ib_pd *ibpd)
{
struct ipath_pd *pd = to_ipd(ibpd);
+ struct ipath_ibdev *dev = to_idev(ibpd->device);
+
+ dev->n_pds_allocated--;
kfree(pd);
@@ -780,6 +852,12 @@ static struct ib_ah *ipath_create_ah(str
{
struct ipath_ah *ah;
struct ib_ah *ret;
+ struct ipath_ibdev *dev = to_idev(pd->device);
+
+ if (dev->n_ahs_allocated == ib_ipath_max_ahs) {
+ ret = ERR_PTR(-ENOMEM);
+ goto bail;
+ }
/* A multicast address requires a GRH (see ch. 8.4.1). */
if (ah_attr->dlid >= IPS_MULTICAST_LID_BASE &&
@@ -794,7 +872,7 @@ static struct ib_ah *ipath_create_ah(str
goto bail;
}
- if (ah_attr->port_num != 1 ||
+ if (ah_attr->port_num < 1 ||
ah_attr->port_num > pd->device->phys_port_cnt) {
ret = ERR_PTR(-EINVAL);
goto bail;
@@ -806,6 +884,8 @@ static struct ib_ah *ipath_create_ah(str
goto bail;
}
+ dev->n_ahs_allocated++;
+
/* ib_create_ah() will initialize ah->ibah. */
ah->attr = *ah_attr;
@@ -823,7 +903,10 @@ bail:
*/
static int ipath_destroy_ah(struct ib_ah *ibah)
{
+ struct ipath_ibdev *dev = to_idev(ibah->device);
struct ipath_ah *ah = to_iah(ibah);
+
+ dev->n_ahs_allocated--;
kfree(ah);
diff -r 21d5d64750ac -r a94e9f9c9c23 drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:25 2006 -0700
@@ -149,6 +149,7 @@ struct ipath_mcast {
struct list_head qp_list;
wait_queue_head_t wait;
atomic_t refcount;
+ int n_attached;
};
/* Memory region */
@@ -432,6 +433,11 @@ struct ipath_ibdev {
__be64 sys_image_guid; /* in network order */
__be64 gid_prefix; /* in network order */
__be64 mkey;
+ u32 n_pds_allocated; /* number of PDs allocated for device */
+ u32 n_ahs_allocated; /* number of AHs allocated for device */
+ u32 n_cqs_allocated; /* number of CQs allocated for device */
+ u32 n_srqs_allocated; /* number of SRQs allocated for device */
+ u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
u64 ipath_sword; /* total dwords sent (sample result) */
u64 ipath_rword; /* total dwords received (sample result) */
u64 ipath_spkts; /* total packets sent (sample result) */
@@ -697,6 +703,24 @@ extern const int ib_ipath_state_ops[];
extern unsigned int ib_ipath_lkey_table_size;
+extern unsigned int ib_ipath_max_cqes;
+
+extern unsigned int ib_ipath_max_cqs;
+
+extern unsigned int ib_ipath_max_qp_wrs;
+
+extern unsigned int ib_ipath_max_sges;
+
+extern unsigned int ib_ipath_max_mcast_grps;
+
+extern unsigned int ib_ipath_max_mcast_qp_attached;
+
+extern unsigned int ib_ipath_max_srqs;
+
+extern unsigned int ib_ipath_max_srq_sges;
+
+extern unsigned int ib_ipath_max_srq_wrs;
+
extern const u32 ib_ipath_rnr_table[];
#endif /* IPATH_VERBS_H */
diff -r 21d5d64750ac -r a94e9f9c9c23 drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c Thu Jun 29 14:33:25 2006 -0700
@@ -93,6 +93,7 @@ static struct ipath_mcast *ipath_mcast_a
INIT_LIST_HEAD(&mcast->qp_list);
init_waitqueue_head(&mcast->wait);
atomic_set(&mcast->refcount, 0);
+ mcast->n_attached = 0;
bail:
return mcast;
@@ -158,7 +159,8 @@ bail:
* the table but the QP was added. Return ESRCH if the QP was already
* attached and neither structure was added.
*/
-static int ipath_mcast_add(struct ipath_mcast *mcast,
+static int ipath_mcast_add(struct ipath_ibdev *dev,
+ struct ipath_mcast *mcast,
struct ipath_mcast_qp *mqp)
{
struct rb_node **n = &mcast_tree.rb_node;
@@ -189,16 +191,28 @@ static int ipath_mcast_add(struct ipath_
/* Search the QP list to see if this is already there. */
list_for_each_entry_rcu(p, &tmcast->qp_list, list) {
if (p->qp == mqp->qp) {
- spin_unlock_irqrestore(&mcast_lock, flags);
ret = ESRCH;
goto bail;
}
}
+ if (tmcast->n_attached == ib_ipath_max_mcast_qp_attached) {
+ ret = ENOMEM;
+ goto bail;
+ }
+
+ tmcast->n_attached++;
+
list_add_tail_rcu(&mqp->list, &tmcast->qp_list);
- spin_unlock_irqrestore(&mcast_lock, flags);
ret = EEXIST;
goto bail;
}
+
+ if (dev->n_mcast_grps_allocated == ib_ipath_max_mcast_grps) {
+ ret = ENOMEM;
+ goto bail;
+ }
+
+ dev->n_mcast_grps_allocated++;
list_add_tail_rcu(&mqp->list, &mcast->qp_list);
@@ -206,17 +220,18 @@ static int ipath_mcast_add(struct ipath_
rb_link_node(&mcast->rb_node, pn, n);
rb_insert_color(&mcast->rb_node, &mcast_tree);
+ ret = 0;
+
+bail:
spin_unlock_irqrestore(&mcast_lock, flags);
- ret = 0;
-
-bail:
return ret;
}
int ipath_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
{
struct ipath_qp *qp = to_iqp(ibqp);
+ struct ipath_ibdev *dev = to_idev(ibqp->device);
struct ipath_mcast *mcast;
struct ipath_mcast_qp *mqp;
int ret;
@@ -236,7 +251,7 @@ int ipath_multicast_attach(struct ib_qp
ret = -ENOMEM;
goto bail;
}
- switch (ipath_mcast_add(mcast, mqp)) {
+ switch (ipath_mcast_add(dev, mcast, mqp)) {
case ESRCH:
/* Neither was used: can't attach the same QP twice. */
ipath_mcast_qp_free(mqp);
@@ -246,6 +261,12 @@ int ipath_multicast_attach(struct ib_qp
case EEXIST: /* The mcast wasn't used */
ipath_mcast_free(mcast);
break;
+ case ENOMEM:
+ /* Exceeded the maximum number of mcast groups. */
+ ipath_mcast_qp_free(mqp);
+ ipath_mcast_free(mcast);
+ ret = -ENOMEM;
+ goto bail;
default:
break;
}
@@ -259,6 +280,7 @@ int ipath_multicast_detach(struct ib_qp
int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
{
struct ipath_qp *qp = to_iqp(ibqp);
+ struct ipath_ibdev *dev = to_idev(ibqp->device);
struct ipath_mcast *mcast = NULL;
struct ipath_mcast_qp *p, *tmp;
struct rb_node *n;
@@ -297,6 +319,7 @@ int ipath_multicast_detach(struct ib_qp
* link until we are sure there are no list walkers.
*/
list_del_rcu(&p->list);
+ mcast->n_attached--;
/* If this was the last attached QP, remove the GID too. */
if (list_empty(&mcast->qp_list)) {
@@ -320,6 +343,7 @@ int ipath_multicast_detach(struct ib_qp
atomic_dec(&mcast->refcount);
wait_event(mcast->wait, !atomic_read(&mcast->refcount));
ipath_mcast_free(mcast);
+ dev->n_mcast_grps_allocated--;
}
ret = 0;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 14 of 39] IB/ipath - removed unused field ipath_kregvirt from struct ipath_devdata
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (12 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 13 of 39] IB/ipath - enforce device resource limits Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 15 of 39] IB/ipath - print better debug info when handling 32/64-bit DMA mask problems Bryan O'Sullivan
` (25 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r a94e9f9c9c23 -r e43b4df874a9 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -496,10 +496,8 @@ static int __devinit ipath_init_one(stru
((void __iomem *)dd->ipath_kregbase + len);
dd->ipath_physaddr = addr; /* used for io_remap, etc. */
/* for user mmap */
- dd->ipath_kregvirt = (u64 __iomem *) phys_to_virt(addr);
- ipath_cdbg(VERBOSE, "mapped io addr %llx to kregbase %p "
- "kregvirt %p\n", addr, dd->ipath_kregbase,
- dd->ipath_kregvirt);
+ ipath_cdbg(VERBOSE, "mapped io addr %llx to kregbase %p\n",
+ addr, dd->ipath_kregbase);
/*
* clear ipath_flags here instead of in ipath_init_chip as it is set
@@ -1809,7 +1807,6 @@ static void cleanup_device(struct ipath_
* re-init
*/
dd->ipath_kregbase = NULL;
- dd->ipath_kregvirt = NULL;
dd->ipath_uregbase = 0;
dd->ipath_sregbase = 0;
dd->ipath_cregbase = 0;
diff -r a94e9f9c9c23 -r e43b4df874a9 drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
@@ -158,11 +158,6 @@ struct ipath_devdata {
unsigned long ipath_physaddr;
/* base of memory alloced for ipath_kregbase, for free */
u64 *ipath_kregalloc;
- /*
- * version of kregbase that doesn't have high bits set (for 32 bit
- * programs, so mmap64 44 bit works)
- */
- u64 __iomem *ipath_kregvirt;
/*
* virtual address where port0 rcvhdrqtail updated for this unit.
* only written to by the chip, not the driver.
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 15 of 39] IB/ipath - print better debug info when handling 32/64-bit DMA mask problems
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (13 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 14 of 39] IB/ipath - removed unused field ipath_kregvirt from struct ipath_devdata Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 16 of 39] IB/ipath - enable freeze mode when shutting down device Bryan O'Sullivan
` (24 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r e43b4df874a9 -r 125471ee6c68 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -425,12 +425,29 @@ static int __devinit ipath_init_one(stru
*/
ret = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
if (ret) {
- dev_info(&pdev->dev, "pci_set_dma_mask unit %u "
- "fails: %d\n", dd->ipath_unit, ret);
+ dev_info(&pdev->dev,
+ "Unable to set DMA mask for unit %u: %d\n",
+ dd->ipath_unit, ret);
goto bail_regions;
}
- else
+ else {
ipath_dbg("No 64bit DMA mask, used 32 bit mask\n");
+ ret = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
+ if (ret)
+ dev_info(&pdev->dev,
+ "Unable to set DMA consistent mask "
+ "for unit %u: %d\n",
+ dd->ipath_unit, ret);
+
+ }
+ }
+ else {
+ ret = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
+ if (ret)
+ dev_info(&pdev->dev,
+ "Unable to set DMA consistent mask "
+ "for unit %u: %d\n",
+ dd->ipath_unit, ret);
}
pci_set_master(pdev);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 16 of 39] IB/ipath - enable freeze mode when shutting down device
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (14 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 15 of 39] IB/ipath - print better debug info when handling 32/64-bit DMA mask problems Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 17 of 39] IB/ipath - use more appropriate gfp flags Bryan O'Sullivan
` (23 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 125471ee6c68 -r fd5e733f02ac drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -1656,7 +1656,7 @@ void ipath_shutdown_device(struct ipath_
/* disable IBC */
dd->ipath_control &= ~INFINIPATH_C_LINKENABLE;
ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
- dd->ipath_control);
+ dd->ipath_control | INFINIPATH_C_FREEZEMODE);
/*
* clear SerdesEnable and turn the leds off; do this here because
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 17 of 39] IB/ipath - use more appropriate gfp flags
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (15 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 16 of 39] IB/ipath - enable freeze mode when shutting down device Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-30 0:02 ` Andrew Morton
2006-06-29 21:41 ` [PATCH 18 of 39] IB/ipath - use vmalloc to allocate struct ipath_devdata Bryan O'Sullivan
` (22 subsequent siblings)
39 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
This helps us to survive better when memory is fragmented.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r fd5e733f02ac -r 9d943b828776 drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
@@ -705,6 +705,15 @@ static int ipath_create_user_egr(struct
unsigned e, egrcnt, alloced, egrperchunk, chunk, egrsize, egroff;
size_t size;
int ret;
+ gfp_t gfp_flags;
+
+ /*
+ * GFP_USER, but without GFP_FS, so buffer cache can be
+ * coalesced (we hope); otherwise, even at order 4,
+ * heavy filesystem activity makes these fail, and we can
+ * use compound pages.
+ */
+ gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
egrcnt = dd->ipath_rcvegrcnt;
/* TID number offset for this port */
@@ -721,10 +730,8 @@ static int ipath_create_user_egr(struct
* memory pressure (creating large files and then copying them over
* NFS while doing lots of MPI jobs), we hit some allocation
* failures, even though we can sleep... (2.6.10) Still get
- * failures at 64K. 32K is the lowest we can go without waiting
- * more memory again. It seems likely that the coalescing in
- * free_pages, etc. still has issues (as it has had previously
- * during 2.6.x development).
+ * failures at 64K. 32K is the lowest we can go without wasting
+ * additional memory.
*/
size = 0x8000;
alloced = ALIGN(egrsize * egrcnt, size);
@@ -745,12 +752,6 @@ static int ipath_create_user_egr(struct
goto bail_rcvegrbuf;
}
for (e = 0; e < pd->port_rcvegrbuf_chunks; e++) {
- /*
- * GFP_USER, but without GFP_FS, so buffer cache can be
- * coalesced (we hope); otherwise, even at order 4,
- * heavy filesystem activity makes these fail
- */
- gfp_t gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
pd->port_rcvegrbuf[e] = dma_alloc_coherent(
&dd->pcidev->dev, size, &pd->port_rcvegrbuf_phys[e],
@@ -1167,9 +1168,10 @@ static int ipath_mmap(struct file *fp, s
ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;
- ipath_cdbg(MM, "ushare: pgaddr %llx vm_start=%lx, vmlen %lx\n",
+ ipath_cdbg(MM, "pgaddr %llx vm_start=%lx len %lx port %u:%u\n",
(unsigned long long) pgaddr, vma->vm_start,
- vma->vm_end - vma->vm_start);
+ vma->vm_end - vma->vm_start, dd->ipath_unit,
+ pd->port_port);
if (pgaddr == ureg)
ret = mmap_ureg(vma, dd, ureg);
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 17 of 39] IB/ipath - use more appropriate gfp flags
2006-06-29 21:41 ` [PATCH 17 of 39] IB/ipath - use more appropriate gfp flags Bryan O'Sullivan
@ 2006-06-30 0:02 ` Andrew Morton
0 siblings, 0 replies; 65+ messages in thread
From: Andrew Morton @ 2006-06-30 0:02 UTC (permalink / raw)
To: Bryan O'Sullivan; +Cc: rdreier, mst, openib-general, linux-kernel, netdev
"Bryan O'Sullivan" <bos@pathscale.com> wrote:
>
> diff -r fd5e733f02ac -r 9d943b828776 drivers/infiniband/hw/ipath/ipath_file_ops.c
> --- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
> +++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
> @@ -705,6 +705,15 @@ static int ipath_create_user_egr(struct
> unsigned e, egrcnt, alloced, egrperchunk, chunk, egrsize, egroff;
> size_t size;
> int ret;
> + gfp_t gfp_flags;
> +
> + /*
> + * GFP_USER, but without GFP_FS, so buffer cache can be
> + * coalesced (we hope); otherwise, even at order 4,
> + * heavy filesystem activity makes these fail, and we can
> + * use compound pages.
> + */
> + gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
Yes, GFP_NOFS|_GFP_COMP is reasonably strong - we can do swapout but not
file pageout.
I expect you'll find that a full GFP_KERNEL is OK here. The ~__GFP_FS is
used to prevent the vm scanner from calling into ->writepage() and getting
stuck on locks which the __alloc_pages() caller already holds.
But ipathfs doesn't even implement ->writepage(), so I don't see any
problem with setting __GFP_FS. If you're getting into trouble there then
I'd recommend giving it a try - it will make memory reclaim more
successful, especially with ext3, where a ->writepage often cleans the page
synchronously without doing any IO.
That being said, order-4 allocations will be fairly reliably unreliable.
^ permalink raw reply [flat|nested] 65+ messages in thread
* [PATCH 18 of 39] IB/ipath - use vmalloc to allocate struct ipath_devdata
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (16 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 17 of 39] IB/ipath - use more appropriate gfp flags Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 19 of 39] IB/ipath - memory management cleanups Bryan O'Sullivan
` (21 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
This is not a DMA target, so no need to use dma_alloc_coherent on it.
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 9d943b828776 -r 9c072f8e7e68 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -171,14 +171,13 @@ static void ipath_free_devdata(struct pc
list_del(&dd->ipath_list);
spin_unlock_irqrestore(&ipath_devs_lock, flags);
}
- dma_free_coherent(&pdev->dev, sizeof(*dd), dd, dd->ipath_dma_addr);
+ vfree(dd);
}
static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev)
{
unsigned long flags;
struct ipath_devdata *dd;
- dma_addr_t dma_addr;
int ret;
if (!idr_pre_get(&unit_table, GFP_KERNEL)) {
@@ -186,15 +185,12 @@ static struct ipath_devdata *ipath_alloc
goto bail;
}
- dd = dma_alloc_coherent(&pdev->dev, sizeof(*dd), &dma_addr,
- GFP_KERNEL);
-
+ dd = vmalloc(sizeof(*dd));
if (!dd) {
dd = ERR_PTR(-ENOMEM);
goto bail;
}
-
- dd->ipath_dma_addr = dma_addr;
+ memset(dd, 0, sizeof(*dd));
dd->ipath_unit = -1;
spin_lock_irqsave(&ipath_devs_lock, flags);
diff -r 9d943b828776 -r 9c072f8e7e68 drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
@@ -163,7 +163,6 @@ struct ipath_devdata {
* only written to by the chip, not the driver.
*/
volatile __le64 *ipath_hdrqtailptr;
- dma_addr_t ipath_dma_addr;
/* ipath_cfgports pointers */
struct ipath_portdata **ipath_pd;
/* sk_buffs used by port 0 eager receive queue */
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 19 of 39] IB/ipath - memory management cleanups
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (17 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 18 of 39] IB/ipath - use vmalloc to allocate struct ipath_devdata Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 20 of 39] IB/ipath - reduce overhead on receive interrupts Bryan O'Sullivan
` (20 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
Made in-memory rcvhdrq tail update be in dma_alloc'ed memory, not
random user or special kernel (needed for ppc, also "just the right
thing to do").
Some cleanups to make unexpected link transitions less likely to produce
complaints about packet errors, and also to not leave SMA packets stuck
and unable to go out.
A few other random debug and comment cleanups.
Always init rcvhdrq head/tail registers to 0, to avoid race conditions
(should have been that way some time ago).
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 9c072f8e7e68 -r 1e8837473193 drivers/infiniband/hw/ipath/ipath_common.h
--- a/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:25 2006 -0700
@@ -311,6 +311,9 @@ struct ipath_base_info {
__u32 spi_rcv_egrchunksize;
/* total size of mmap to cover full rcvegrbuffers */
__u32 spi_rcv_egrbuftotlen;
+ __u32 spi_filler_for_align;
+ /* address of readonly memory copy of the rcvhdrq tail register. */
+ __u64 spi_rcvhdr_tailaddr;
} __attribute__ ((aligned(8)));
@@ -380,13 +383,7 @@ struct ipath_user_info {
*/
__u32 spu_rcvhdrsize;
- /*
- * cache line aligned (64 byte) user address to
- * which the rcvhdrtail register will be written by infinipath
- * whenever it changes, so that no chip registers are read in
- * the performance path.
- */
- __u64 spu_rcvhdraddr;
+ __u64 spu_unused; /* kept for compatible layout */
/*
* address of struct base_info to write to
diff -r 9c072f8e7e68 -r 1e8837473193 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -131,14 +131,6 @@ static struct pci_driver ipath_driver =
.id_table = ipath_pci_tbl,
};
-/*
- * This is where port 0's rcvhdrtail register is written back; we also
- * want nothing else sharing the cache line, so make it a cache line
- * in size. Used for all units.
- */
-volatile __le64 *ipath_port0_rcvhdrtail;
-dma_addr_t ipath_port0_rcvhdrtail_dma;
-static int port0_rcvhdrtail_refs;
static inline void read_bars(struct ipath_devdata *dd, struct pci_dev *dev,
u32 *bar0, u32 *bar1)
@@ -268,47 +260,6 @@ int ipath_count_units(int *npresentp, in
return nunits;
}
-static int init_port0_rcvhdrtail(struct pci_dev *pdev)
-{
- int ret;
-
- mutex_lock(&ipath_mutex);
-
- if (!ipath_port0_rcvhdrtail) {
- ipath_port0_rcvhdrtail =
- dma_alloc_coherent(&pdev->dev,
- IPATH_PORT0_RCVHDRTAIL_SIZE,
- &ipath_port0_rcvhdrtail_dma,
- GFP_KERNEL);
-
- if (!ipath_port0_rcvhdrtail) {
- ret = -ENOMEM;
- goto bail;
- }
- }
- port0_rcvhdrtail_refs++;
- ret = 0;
-
-bail:
- mutex_unlock(&ipath_mutex);
-
- return ret;
-}
-
-static void cleanup_port0_rcvhdrtail(struct pci_dev *pdev)
-{
- mutex_lock(&ipath_mutex);
-
- if (!--port0_rcvhdrtail_refs) {
- dma_free_coherent(&pdev->dev, IPATH_PORT0_RCVHDRTAIL_SIZE,
- (void *) ipath_port0_rcvhdrtail,
- ipath_port0_rcvhdrtail_dma);
- ipath_port0_rcvhdrtail = NULL;
- }
-
- mutex_unlock(&ipath_mutex);
-}
-
/*
* These next two routines are placeholders in case we don't have per-arch
* code for controlling write combining. If explicit control of write
@@ -333,20 +284,12 @@ static int __devinit ipath_init_one(stru
u32 bar0 = 0, bar1 = 0;
u8 rev;
- ret = init_port0_rcvhdrtail(pdev);
- if (ret < 0) {
- printk(KERN_ERR IPATH_DRV_NAME
- ": Could not allocate port0_rcvhdrtail: error %d\n",
- -ret);
- goto bail;
- }
-
dd = ipath_alloc_devdata(pdev);
if (IS_ERR(dd)) {
ret = PTR_ERR(dd);
printk(KERN_ERR IPATH_DRV_NAME
": Could not allocate devdata: error %d\n", -ret);
- goto bail_rcvhdrtail;
+ goto bail;
}
ipath_cdbg(VERBOSE, "initializing unit #%u\n", dd->ipath_unit);
@@ -574,9 +517,6 @@ bail_devdata:
bail_devdata:
ipath_free_devdata(pdev, dd);
-bail_rcvhdrtail:
- cleanup_port0_rcvhdrtail(pdev);
-
bail:
return ret;
}
@@ -608,7 +548,6 @@ static void __devexit ipath_remove_one(s
pci_disable_device(pdev);
ipath_free_devdata(pdev, dd);
- cleanup_port0_rcvhdrtail(pdev);
}
/* general driver use */
@@ -1383,26 +1322,20 @@ bail:
* @dd: the infinipath device
* @pd: the port data
*
- * this *must* be physically contiguous memory, and for now,
- * that limits it to what kmalloc can do.
+ * this must be contiguous memory (from an i/o perspective), and must be
+ * DMA'able (which means for some systems, it will go through an IOMMU,
+ * or be forced into a low address range).
*/
int ipath_create_rcvhdrq(struct ipath_devdata *dd,
struct ipath_portdata *pd)
{
- int ret = 0, amt;
-
- amt = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
- sizeof(u32), PAGE_SIZE);
+ int ret = 0;
+
if (!pd->port_rcvhdrq) {
- /*
- * not using REPEAT isn't viable; at 128KB, we can easily
- * fail this. The problem with REPEAT is we can block here
- * "forever". There isn't an inbetween, unfortunately. We
- * could reduce the risk by never freeing the rcvhdrq except
- * at unload, but even then, the first time a port is used,
- * we could delay for some time...
- */
+ dma_addr_t phys_hdrqtail;
gfp_t gfp_flags = GFP_USER | __GFP_COMP;
+ int amt = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
+ sizeof(u32), PAGE_SIZE);
pd->port_rcvhdrq = dma_alloc_coherent(
&dd->pcidev->dev, amt, &pd->port_rcvhdrq_phys,
@@ -1415,6 +1348,16 @@ int ipath_create_rcvhdrq(struct ipath_de
ret = -ENOMEM;
goto bail;
}
+ pd->port_rcvhdrtail_kvaddr = dma_alloc_coherent(
+ &dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail, GFP_KERNEL);
+ if (!pd->port_rcvhdrtail_kvaddr) {
+ ipath_dev_err(dd, "attempt to allocate 1 page "
+ "for port %u rcvhdrqtailaddr failed\n",
+ pd->port_port);
+ ret = -ENOMEM;
+ goto bail;
+ }
+ pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail;
pd->port_rcvhdrq_size = amt;
@@ -1424,20 +1367,28 @@ int ipath_create_rcvhdrq(struct ipath_de
(unsigned long) pd->port_rcvhdrq_phys,
(unsigned long) pd->port_rcvhdrq_size,
pd->port_port);
- } else {
- /*
- * clear for security, sanity, and/or debugging, each
- * time we reuse
- */
- memset(pd->port_rcvhdrq, 0, amt);
- }
+
+ ipath_cdbg(VERBOSE, "port %d hdrtailaddr, %llx physical\n",
+ pd->port_port,
+ (unsigned long long) phys_hdrqtail);
+ }
+ else
+ ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; "
+ "hdrtailaddr@%p %llx physical\n",
+ pd->port_port, pd->port_rcvhdrq,
+ pd->port_rcvhdrq_phys, pd->port_rcvhdrtail_kvaddr,
+ (unsigned long long)pd->port_rcvhdrqtailaddr_phys);
+
+ /* clear for security and sanity on each use */
+ memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size);
+ memset((void *)pd->port_rcvhdrtail_kvaddr, 0, PAGE_SIZE);
/*
* tell chip each time we init it, even if we are re-using previous
- * memory (we zero it at process close)
- */
- ipath_cdbg(VERBOSE, "writing port %d rcvhdraddr as %lx\n",
- pd->port_port, (unsigned long) pd->port_rcvhdrq_phys);
+ * memory (we zero the register at process close)
+ */
+ ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdrtailaddr,
+ pd->port_port, pd->port_rcvhdrqtailaddr_phys);
ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdraddr,
pd->port_port, pd->port_rcvhdrq_phys);
@@ -1525,15 +1476,27 @@ void ipath_set_ib_lstate(struct ipath_de
[INFINIPATH_IBCC_LINKCMD_ARMED] = "ARMED",
[INFINIPATH_IBCC_LINKCMD_ACTIVE] = "ACTIVE"
};
+ int linkcmd = (which >> INFINIPATH_IBCC_LINKCMD_SHIFT) &
+ INFINIPATH_IBCC_LINKCMD_MASK;
+
ipath_cdbg(SMA, "Trying to move unit %u to %s, current ltstate "
"is %s\n", dd->ipath_unit,
- what[(which >> INFINIPATH_IBCC_LINKCMD_SHIFT) &
- INFINIPATH_IBCC_LINKCMD_MASK],
+ what[linkcmd],
ipath_ibcstatus_str[
(ipath_read_kreg64
(dd, dd->ipath_kregs->kr_ibcstatus) >>
INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) &
INFINIPATH_IBCS_LINKTRAININGSTATE_MASK]);
+ /* flush all queued sends when going to DOWN or INIT, to be sure that
+ * they don't block SMA and other MAD packets */
+ if (!linkcmd || linkcmd == INFINIPATH_IBCC_LINKCMD_INIT) {
+ ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
+ INFINIPATH_S_ABORT);
+ ipath_disarm_piobufs(dd, dd->ipath_lastport_piobuf,
+ (unsigned)(dd->ipath_piobcnt2k +
+ dd->ipath_piobcnt4k) -
+ dd->ipath_lastport_piobuf);
+ }
ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
dd->ipath_ibcctrl | which);
@@ -1681,60 +1644,54 @@ void ipath_shutdown_device(struct ipath_
/**
* ipath_free_pddata - free a port's allocated data
* @dd: the infinipath device
- * @port: the port
- * @freehdrq: free the port data structure if true
- *
- * when closing, free up any allocated data for a port, if the
- * reference count goes to zero
- * Note: this also optionally frees the portdata itself!
- * Any changes here have to be matched up with the reinit case
- * of ipath_init_chip(), which calls this routine on reinit after reset.
- */
-void ipath_free_pddata(struct ipath_devdata *dd, u32 port, int freehdrq)
-{
- struct ipath_portdata *pd = dd->ipath_pd[port];
-
+ * @pd: the portdata structure
+ *
+ * free up any allocated data for a port
+ * This should not touch anything that would affect a simultaneous
+ * re-allocation of port data, because it is called after ipath_mutex
+ * is released (and can be called from reinit as well).
+ * It should never change any chip state, or global driver state.
+ * (The only exception to global state is freeing the port0 port0_skbs.)
+ */
+void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd)
+{
if (!pd)
return;
- if (freehdrq)
- /*
- * only clear and free portdata if we are going to also
- * release the hdrq, otherwise we leak the hdrq on each
- * open/close cycle
- */
- dd->ipath_pd[port] = NULL;
- if (freehdrq && pd->port_rcvhdrq) {
+
+ if (pd->port_rcvhdrq) {
ipath_cdbg(VERBOSE, "free closed port %d rcvhdrq @ %p "
"(size=%lu)\n", pd->port_port, pd->port_rcvhdrq,
(unsigned long) pd->port_rcvhdrq_size);
dma_free_coherent(&dd->pcidev->dev, pd->port_rcvhdrq_size,
pd->port_rcvhdrq, pd->port_rcvhdrq_phys);
pd->port_rcvhdrq = NULL;
- }
- if (port && pd->port_rcvegrbuf) {
- /* always free this */
- if (pd->port_rcvegrbuf) {
- unsigned e;
-
- for (e = 0; e < pd->port_rcvegrbuf_chunks; e++) {
- void *base = pd->port_rcvegrbuf[e];
- size_t size = pd->port_rcvegrbuf_size;
-
- ipath_cdbg(VERBOSE, "egrbuf free(%p, %lu), "
- "chunk %u/%u\n", base,
- (unsigned long) size,
- e, pd->port_rcvegrbuf_chunks);
- dma_free_coherent(
- &dd->pcidev->dev, size, base,
- pd->port_rcvegrbuf_phys[e]);
- }
- vfree(pd->port_rcvegrbuf);
- pd->port_rcvegrbuf = NULL;
- vfree(pd->port_rcvegrbuf_phys);
- pd->port_rcvegrbuf_phys = NULL;
- }
+ if (pd->port_rcvhdrtail_kvaddr) {
+ dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+ (void *)pd->port_rcvhdrtail_kvaddr,
+ pd->port_rcvhdrqtailaddr_phys);
+ pd->port_rcvhdrtail_kvaddr = NULL;
+ }
+ }
+ if (pd->port_port && pd->port_rcvegrbuf) {
+ unsigned e;
+
+ for (e = 0; e < pd->port_rcvegrbuf_chunks; e++) {
+ void *base = pd->port_rcvegrbuf[e];
+ size_t size = pd->port_rcvegrbuf_size;
+
+ ipath_cdbg(VERBOSE, "egrbuf free(%p, %lu), "
+ "chunk %u/%u\n", base,
+ (unsigned long) size,
+ e, pd->port_rcvegrbuf_chunks);
+ dma_free_coherent(&dd->pcidev->dev, size,
+ base, pd->port_rcvegrbuf_phys[e]);
+ }
+ vfree(pd->port_rcvegrbuf);
+ pd->port_rcvegrbuf = NULL;
+ vfree(pd->port_rcvegrbuf_phys);
+ pd->port_rcvegrbuf_phys = NULL;
pd->port_rcvegrbuf_chunks = 0;
- } else if (port == 0 && dd->ipath_port0_skbs) {
+ } else if (pd->port_port == 0 && dd->ipath_port0_skbs) {
unsigned e;
struct sk_buff **skbs = dd->ipath_port0_skbs;
@@ -1746,10 +1703,8 @@ void ipath_free_pddata(struct ipath_devd
dev_kfree_skb(skbs[e]);
vfree(skbs);
}
- if (freehdrq) {
- kfree(pd->port_tid_pg_list);
- kfree(pd);
- }
+ kfree(pd->port_tid_pg_list);
+ kfree(pd);
}
static int __init infinipath_init(void)
@@ -1874,10 +1829,14 @@ static void cleanup_device(struct ipath_
/*
* free any resources still in use (usually just kernel ports)
- * at unload
- */
- for (port = 0; port < dd->ipath_cfgports; port++)
- ipath_free_pddata(dd, port, 1);
+ * at unload; we do for portcnt, not cfgports, because cfgports
+ * could have changed while we were loaded.
+ */
+ for (port = 0; port < dd->ipath_portcnt; port++) {
+ struct ipath_portdata *pd = dd->ipath_pd[port];
+ dd->ipath_pd[port] = NULL;
+ ipath_free_pddata(dd, pd);
+ }
kfree(dd->ipath_pd);
/*
* debuggability, in case some cleanup path tries to use it
diff -r 9c072f8e7e68 -r 1e8837473193 drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:25 2006 -0700
@@ -123,6 +123,7 @@ static int ipath_get_base_info(struct ip
* on to yet another method of dealing with this
*/
kinfo->spi_rcvhdr_base = (u64) pd->port_rcvhdrq_phys;
+ kinfo->spi_rcvhdr_tailaddr = (u64)pd->port_rcvhdrqtailaddr_phys;
kinfo->spi_rcv_egrbufs = (u64) pd->port_rcvegr_phys;
kinfo->spi_pioavailaddr = (u64) dd->ipath_pioavailregs_phys;
kinfo->spi_status = (u64) kinfo->spi_pioavailaddr +
@@ -785,11 +786,12 @@ static int ipath_create_user_egr(struct
bail_rcvegrbuf_phys:
for (e = 0; e < pd->port_rcvegrbuf_chunks &&
- pd->port_rcvegrbuf[e]; e++)
+ pd->port_rcvegrbuf[e]; e++) {
dma_free_coherent(&dd->pcidev->dev, size,
pd->port_rcvegrbuf[e],
pd->port_rcvegrbuf_phys[e]);
+ }
vfree(pd->port_rcvegrbuf_phys);
pd->port_rcvegrbuf_phys = NULL;
bail_rcvegrbuf:
@@ -804,10 +806,7 @@ static int ipath_do_user_init(struct ipa
{
int ret = 0;
struct ipath_devdata *dd = pd->port_dd;
- u64 physaddr, uaddr, off, atmp;
- struct page *pagep;
u32 head32;
- u64 head;
/* for now, if major version is different, bail */
if ((uinfo->spu_userversion >> 16) != IPATH_USER_SWMAJOR) {
@@ -831,54 +830,6 @@ static int ipath_do_user_init(struct ipa
}
/* for now we do nothing with rcvhdrcnt: uinfo->spu_rcvhdrcnt */
-
- /* set up for the rcvhdr Q tail register writeback to user memory */
- if (!uinfo->spu_rcvhdraddr ||
- !access_ok(VERIFY_WRITE, (u64 __user *) (unsigned long)
- uinfo->spu_rcvhdraddr, sizeof(u64))) {
- ipath_dbg("Port %d rcvhdrtail addr %llx not valid\n",
- pd->port_port,
- (unsigned long long) uinfo->spu_rcvhdraddr);
- ret = -EINVAL;
- goto done;
- }
-
- off = offset_in_page(uinfo->spu_rcvhdraddr);
- uaddr = PAGE_MASK & (unsigned long) uinfo->spu_rcvhdraddr;
- ret = ipath_get_user_pages_nocopy(uaddr, &pagep);
- if (ret) {
- dev_info(&dd->pcidev->dev, "Failed to lookup and lock "
- "address %llx for rcvhdrtail: errno %d\n",
- (unsigned long long) uinfo->spu_rcvhdraddr, -ret);
- goto done;
- }
- ipath_stats.sps_pagelocks++;
- pd->port_rcvhdrtail_uaddr = uaddr;
- pd->port_rcvhdrtail_pagep = pagep;
- pd->port_rcvhdrtail_kvaddr =
- page_address(pagep);
- pd->port_rcvhdrtail_kvaddr += off;
- physaddr = page_to_phys(pagep) + off;
- ipath_cdbg(VERBOSE, "port %d user addr %llx hdrtailaddr, %llx "
- "physical (off=%llx)\n",
- pd->port_port,
- (unsigned long long) uinfo->spu_rcvhdraddr,
- (unsigned long long) physaddr, (unsigned long long) off);
- ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdrtailaddr,
- pd->port_port, physaddr);
- atmp = ipath_read_kreg64_port(dd,
- dd->ipath_kregs->kr_rcvhdrtailaddr,
- pd->port_port);
- if (physaddr != atmp) {
- ipath_dev_err(dd,
- "Catastrophic software error, "
- "RcvHdrTailAddr%u written as %llx, "
- "read back as %llx\n", pd->port_port,
- (unsigned long long) physaddr,
- (unsigned long long) atmp);
- ret = -EINVAL;
- goto done;
- }
/* for right now, kernel piobufs are at end, so port 1 is at 0 */
pd->port_piobufs = dd->ipath_piobufbase +
@@ -898,26 +849,18 @@ static int ipath_do_user_init(struct ipa
ret = ipath_create_user_egr(pd);
if (ret)
goto done;
- /* enable receives now */
- /* atomically set enable bit for this port */
- set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
- &dd->ipath_rcvctrl);
/*
- * set the head registers for this port to the current values
+ * set the eager head register for this port to the current values
* of the tail pointers, since we don't know if they were
* updated on last use of the port.
*/
- head32 = ipath_read_ureg32(dd, ur_rcvhdrtail, pd->port_port);
- head = (u64) head32;
- ipath_write_ureg(dd, ur_rcvhdrhead, head, pd->port_port);
head32 = ipath_read_ureg32(dd, ur_rcvegrindextail, pd->port_port);
ipath_write_ureg(dd, ur_rcvegrindexhead, head32, pd->port_port);
dd->ipath_lastegrheads[pd->port_port] = -1;
dd->ipath_lastrcvhdrqtails[pd->port_port] = -1;
- ipath_cdbg(VERBOSE, "Wrote port%d head %llx, egrhead %x from "
- "tail regs\n", pd->port_port,
- (unsigned long long) head, head32);
+ ipath_cdbg(VERBOSE, "Wrote port%d egrhead %x from tail regs\n",
+ pd->port_port, head32);
pd->port_tidcursor = 0; /* start at beginning after open */
/*
* now enable the port; the tail registers will be written to memory
@@ -926,13 +869,62 @@ static int ipath_do_user_init(struct ipa
* transition from 0 to 1, so clear it first, then set it as part of
* enabling the port. This will (very briefly) affect any other
* open ports, but it shouldn't be long enough to be an issue.
+ * We explictly set the in-memory copy to 0 beforehand, so we don't
+ * have to wait to be sure the DMA update has happened.
*/
+ *pd->port_rcvhdrtail_kvaddr = 0ULL;
+ set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
+ &dd->ipath_rcvctrl);
ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
dd->ipath_rcvctrl & ~INFINIPATH_R_TAILUPD);
ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
dd->ipath_rcvctrl);
-
done:
+ return ret;
+}
+
+
+/* common code for the mappings on dma_alloc_coherent mem */
+static int ipath_mmap_mem(struct vm_area_struct *vma,
+ struct ipath_portdata *pd, unsigned len,
+ int write_ok, dma_addr_t addr, char *what)
+{
+ struct ipath_devdata *dd = pd->port_dd;
+ unsigned pfn = (unsigned long)addr >> PAGE_SHIFT;
+ int ret;
+
+ if ((vma->vm_end - vma->vm_start) > len) {
+ dev_info(&dd->pcidev->dev,
+ "FAIL on %s: len %lx > %x\n", what,
+ vma->vm_end - vma->vm_start, len);
+ ret = -EFAULT;
+ goto bail;
+ }
+
+ if (!write_ok) {
+ if (vma->vm_flags & VM_WRITE) {
+ dev_info(&dd->pcidev->dev,
+ "%s must be mapped readonly\n", what);
+ ret = -EPERM;
+ goto bail;
+ }
+
+ /* don't allow them to later change with mprotect */
+ vma->vm_flags &= ~VM_MAYWRITE;
+ }
+
+ ret = remap_pfn_range(vma, vma->vm_start, pfn,
+ len, vma->vm_page_prot);
+ if (ret)
+ dev_info(&dd->pcidev->dev,
+ "%s port%u mmap of %lx, %x bytes r%c failed: %d\n",
+ what, pd->port_port, (unsigned long)addr, len,
+ write_ok?'w':'o', ret);
+ else
+ ipath_cdbg(VERBOSE, "%s port%u mmaped %lx, %x bytes r%c\n",
+ what, pd->port_port, (unsigned long)addr, len,
+ write_ok?'w':'o');
+bail:
return ret;
}
@@ -942,8 +934,11 @@ static int mmap_ureg(struct vm_area_stru
unsigned long phys;
int ret;
- /* it's the real hardware, so io_remap works */
-
+ /*
+ * This is real hardware, so use io_remap. This is the mechanism
+ * for the user process to update the head registers for their port
+ * in the chip.
+ */
if ((vma->vm_end - vma->vm_start) > PAGE_SIZE) {
dev_info(&dd->pcidev->dev, "FAIL mmap userreg: reqlen "
"%lx > PAGE\n", vma->vm_end - vma->vm_start);
@@ -969,10 +964,11 @@ static int mmap_piobufs(struct vm_area_s
int ret;
/*
- * When we map the PIO buffers, we want to map them as writeonly, no
- * read possible.
+ * When we map the PIO buffers in the chip, we want to map them as
+ * writeonly, no read possible. This prevents access to previous
+ * process data, and catches users who might try to read the i/o
+ * space due to a bug.
*/
-
if ((vma->vm_end - vma->vm_start) >
(dd->ipath_pbufsport * dd->ipath_palign)) {
dev_info(&dd->pcidev->dev, "FAIL mmap piobufs: "
@@ -983,11 +979,10 @@ static int mmap_piobufs(struct vm_area_s
}
phys = dd->ipath_physaddr + pd->port_piobufs;
+
/*
- * Do *NOT* mark this as non-cached (PWT bit), or we don't get the
+ * Don't mark this as non-cached, or we don't get the
* write combining behavior we want on the PIO buffers!
- * vma->vm_page_prot =
- * pgprot_noncached(vma->vm_page_prot);
*/
if (vma->vm_flags & VM_READ) {
@@ -999,8 +994,7 @@ static int mmap_piobufs(struct vm_area_s
}
/* don't allow them to later change to readable with mprotect */
-
- vma->vm_flags &= ~VM_MAYWRITE;
+ vma->vm_flags &= ~VM_MAYREAD;
vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
ret = io_remap_pfn_range(vma, vma->vm_start, phys >> PAGE_SHIFT,
@@ -1018,11 +1012,6 @@ static int mmap_rcvegrbufs(struct vm_are
size_t total_size, i;
dma_addr_t *phys;
int ret;
-
- if (!pd->port_rcvegrbuf) {
- ret = -EFAULT;
- goto bail;
- }
size = pd->port_rcvegrbuf_size;
total_size = pd->port_rcvegrbuf_chunks * size;
@@ -1041,12 +1030,11 @@ static int mmap_rcvegrbufs(struct vm_are
ret = -EPERM;
goto bail;
}
+ /* don't allow them to later change to writeable with mprotect */
+ vma->vm_flags &= ~VM_MAYWRITE;
start = vma->vm_start;
phys = pd->port_rcvegrbuf_phys;
-
- /* don't allow them to later change to writeable with mprotect */
- vma->vm_flags &= ~VM_MAYWRITE;
for (i = 0; i < pd->port_rcvegrbuf_chunks; i++, start += size) {
ret = remap_pfn_range(vma, start, phys[i] >> PAGE_SHIFT,
@@ -1056,78 +1044,6 @@ static int mmap_rcvegrbufs(struct vm_are
}
ret = 0;
-bail:
- return ret;
-}
-
-static int mmap_rcvhdrq(struct vm_area_struct *vma,
- struct ipath_portdata *pd)
-{
- struct ipath_devdata *dd = pd->port_dd;
- size_t total_size;
- int ret;
-
- /*
- * kmalloc'ed memory, physically contiguous; this is from
- * spi_rcvhdr_base; we allow user to map read-write so they can
- * write hdrq entries to allow protocol code to directly poll
- * whether a hdrq entry has been written.
- */
- total_size = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
- sizeof(u32), PAGE_SIZE);
- if ((vma->vm_end - vma->vm_start) > total_size) {
- dev_info(&dd->pcidev->dev,
- "FAIL on rcvhdrq: reqlen %lx > actual %lx\n",
- vma->vm_end - vma->vm_start,
- (unsigned long) total_size);
- ret = -EFAULT;
- goto bail;
- }
-
- ret = remap_pfn_range(vma, vma->vm_start,
- pd->port_rcvhdrq_phys >> PAGE_SHIFT,
- vma->vm_end - vma->vm_start,
- vma->vm_page_prot);
-bail:
- return ret;
-}
-
-static int mmap_pioavailregs(struct vm_area_struct *vma,
- struct ipath_portdata *pd)
-{
- struct ipath_devdata *dd = pd->port_dd;
- int ret;
-
- /*
- * when we map the PIO bufferavail registers, we want to map them as
- * readonly, no write possible.
- *
- * kmalloc'ed memory, physically contiguous, one page only, readonly
- */
-
- if ((vma->vm_end - vma->vm_start) > PAGE_SIZE) {
- dev_info(&dd->pcidev->dev, "FAIL on pioavailregs_dma: "
- "reqlen %lx > actual %lx\n",
- vma->vm_end - vma->vm_start,
- (unsigned long) PAGE_SIZE);
- ret = -EFAULT;
- goto bail;
- }
-
- if (vma->vm_flags & VM_WRITE) {
- dev_info(&dd->pcidev->dev,
- "Can't map pioavailregs as writable (flags=%lx)\n",
- vma->vm_flags);
- ret = -EPERM;
- goto bail;
- }
-
- /* don't allow them to later change with mprotect */
- vma->vm_flags &= ~VM_MAYWRITE;
-
- ret = remap_pfn_range(vma, vma->vm_start,
- dd->ipath_pioavailregs_phys >> PAGE_SHIFT,
- PAGE_SIZE, vma->vm_page_prot);
bail:
return ret;
}
@@ -1151,6 +1067,7 @@ static int ipath_mmap(struct file *fp, s
pd = port_fp(fp);
dd = pd->port_dd;
+
/*
* This is the ipath_do_user_init() code, mapping the shared buffers
* into the user process. The address referred to by vm_pgoff is the
@@ -1160,29 +1077,59 @@ static int ipath_mmap(struct file *fp, s
pgaddr = vma->vm_pgoff << PAGE_SHIFT;
/*
- * note that ureg does *NOT* have the kregvirt as part of it, to be
- * sure that for 32 bit programs, we don't end up trying to map a >
- * 44 address. Has to match ipath_get_base_info() code that sets
- * __spi_uregbase
+ * Must fit in 40 bits for our hardware; some checked elsewhere,
+ * but we'll be paranoid. Check for 0 is mostly in case one of the
+ * allocations failed, but user called mmap anyway. We want to catch
+ * that before it can match.
*/
-
+ if (!pgaddr || pgaddr >= (1ULL<<40)) {
+ ipath_dev_err(dd, "Bad phys addr %llx, start %lx, end %lx\n",
+ (unsigned long long)pgaddr, vma->vm_start, vma->vm_end);
+ return -EINVAL;
+ }
+
+ /* just the offset of the port user registers, not physical addr */
ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;
- ipath_cdbg(MM, "pgaddr %llx vm_start=%lx len %lx port %u:%u\n",
+ ipath_cdbg(MM, "ushare: pgaddr %llx vm_start=%lx, vmlen %lx\n",
(unsigned long long) pgaddr, vma->vm_start,
- vma->vm_end - vma->vm_start, dd->ipath_unit,
- pd->port_port);
-
- if (pgaddr == ureg)
+ vma->vm_end - vma->vm_start);
+
+ if (vma->vm_start & (PAGE_SIZE-1)) {
+ ipath_dev_err(dd,
+ "vm_start not aligned: %lx, end=%lx phys %lx\n",
+ vma->vm_start, vma->vm_end, (unsigned long)pgaddr);
+ ret = -EINVAL;
+ }
+ else if (pgaddr == ureg)
ret = mmap_ureg(vma, dd, ureg);
else if (pgaddr == pd->port_piobufs)
ret = mmap_piobufs(vma, dd, pd);
else if (pgaddr == (u64) pd->port_rcvegr_phys)
ret = mmap_rcvegrbufs(vma, pd);
- else if (pgaddr == (u64) pd->port_rcvhdrq_phys)
- ret = mmap_rcvhdrq(vma, pd);
+ else if (pgaddr == (u64) pd->port_rcvhdrq_phys) {
+ /*
+ * The rcvhdrq itself; readonly except on HT-400 (so have
+ * to allow writable mapping), multiple pages, contiguous
+ * from an i/o perspective.
+ */
+ unsigned total_size =
+ ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize
+ * sizeof(u32), PAGE_SIZE);
+ ret = ipath_mmap_mem(vma, pd, total_size, 1,
+ pd->port_rcvhdrq_phys,
+ "rcvhdrq");
+ }
+ else if (pgaddr == (u64)pd->port_rcvhdrqtailaddr_phys)
+ /* in-memory copy of rcvhdrq tail register */
+ ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
+ pd->port_rcvhdrqtailaddr_phys,
+ "rcvhdrq tail");
else if (pgaddr == dd->ipath_pioavailregs_phys)
- ret = mmap_pioavailregs(vma, pd);
+ /* in-memory copy of pioavail registers */
+ ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
+ dd->ipath_pioavailregs_phys,
+ "pioavail registers");
else
ret = -EINVAL;
@@ -1539,14 +1486,6 @@ static int ipath_close(struct inode *in,
}
if (dd->ipath_kregbase) {
- if (pd->port_rcvhdrtail_uaddr) {
- pd->port_rcvhdrtail_uaddr = 0;
- pd->port_rcvhdrtail_kvaddr = NULL;
- ipath_release_user_pages_on_close(
- &pd->port_rcvhdrtail_pagep, 1);
- pd->port_rcvhdrtail_pagep = NULL;
- ipath_stats.sps_pageunlocks++;
- }
ipath_write_kreg_port(
dd, dd->ipath_kregs->kr_rcvhdrtailaddr,
port, 0ULL);
@@ -1583,9 +1522,9 @@ static int ipath_close(struct inode *in,
dd->ipath_f_clear_tids(dd, pd->port_port);
- ipath_free_pddata(dd, pd->port_port, 0);
-
+ dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */
mutex_unlock(&ipath_mutex);
+ ipath_free_pddata(dd, pd); /* after releasing the mutex */
return ret;
}
@@ -1905,3 +1844,4 @@ bail:
bail:
return;
}
+
diff -r 9c072f8e7e68 -r 1e8837473193 drivers/infiniband/hw/ipath/ipath_init_chip.c
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:25 2006 -0700
@@ -411,17 +411,8 @@ static int init_pioavailregs(struct ipat
/* and its length */
dd->ipath_freezelen = L1_CACHE_BYTES - sizeof(dd->ipath_statusp[0]);
- if (dd->ipath_unit * 64 > (IPATH_PORT0_RCVHDRTAIL_SIZE - 64)) {
- ipath_dev_err(dd, "unit %u too large for port 0 "
- "rcvhdrtail buffer size\n", dd->ipath_unit);
- ret = -ENODEV;
- }
- else
- ret = 0;
-
- /* so we can get current tail in ipath_kreceive(), per chip */
- dd->ipath_hdrqtailptr = &ipath_port0_rcvhdrtail[
- dd->ipath_unit * (64 / sizeof(*ipath_port0_rcvhdrtail))];
+ ret = 0;
+
done:
return ret;
}
@@ -654,7 +645,7 @@ int ipath_init_chip(struct ipath_devdata
{
int ret = 0, i;
u32 val32, kpiobufs;
- u64 val, atmp;
+ u64 val;
struct ipath_portdata *pd = NULL; /* keep gcc4 happy */
ret = init_housekeeping(dd, &pd, reinit);
@@ -777,24 +768,6 @@ int ipath_init_chip(struct ipath_devdata
goto done;
}
- val = ipath_port0_rcvhdrtail_dma + dd->ipath_unit * 64;
-
- /* verify that the alignment requirement was met */
- ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdrtailaddr,
- 0, val);
- atmp = ipath_read_kreg64_port(
- dd, dd->ipath_kregs->kr_rcvhdrtailaddr, 0);
- if (val != atmp) {
- ipath_dev_err(dd, "Catastrophic software error, "
- "RcvHdrTailAddr0 written as %llx, "
- "read back as %llx from %x\n",
- (unsigned long long) val,
- (unsigned long long) atmp,
- dd->ipath_kregs->kr_rcvhdrtailaddr);
- ret = -EINVAL;
- goto done;
- }
-
ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvbthqp, IPATH_KD_QP);
/*
@@ -845,12 +818,18 @@ int ipath_init_chip(struct ipath_devdata
* re-init, the simplest way to handle this is to free
* existing, and re-allocate.
*/
- if (reinit)
- ipath_free_pddata(dd, 0, 0);
+ if (reinit) {
+ struct ipath_portdata *pd = dd->ipath_pd[0];
+ dd->ipath_pd[0] = NULL;
+ ipath_free_pddata(dd, pd);
+ }
dd->ipath_f_tidtemplate(dd);
ret = ipath_create_rcvhdrq(dd, pd);
- if (!ret)
+ if (!ret) {
+ dd->ipath_hdrqtailptr =
+ (volatile __le64 *)pd->port_rcvhdrtail_kvaddr;
ret = create_port0_egr(dd);
+ }
if (ret)
ipath_dev_err(dd, "failed to allocate port 0 (kernel) "
"rcvhdrq and/or egr bufs\n");
diff -r 9c072f8e7e68 -r 1e8837473193 drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
@@ -37,6 +37,7 @@
#include "ips_common.h"
#include "ipath_layer.h"
+/* These are all rcv-related errors which we want to count for stats */
#define E_SUM_PKTERRS \
(INFINIPATH_E_RHDRLEN | INFINIPATH_E_RBADTID | \
INFINIPATH_E_RBADVERSION | INFINIPATH_E_RHDR | \
@@ -45,12 +46,25 @@
INFINIPATH_E_RFORMATERR | INFINIPATH_E_RUNSUPVL | \
INFINIPATH_E_RUNEXPCHAR | INFINIPATH_E_REBP)
+/* These are all send-related errors which we want to count for stats */
#define E_SUM_ERRS \
(INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM | \
INFINIPATH_E_SDROPPEDDATAPKT | INFINIPATH_E_SDROPPEDSMPPKT | \
INFINIPATH_E_SMAXPKTLEN | INFINIPATH_E_SUNSUPVL | \
INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SPKTLEN | \
INFINIPATH_E_INVALIDADDR)
+
+/*
+ * these are errors that can occur when the link changes state while
+ * a packet is being sent or received. This doesn't cover things
+ * like EBP or VCRC that can be the result of a sending having the
+ * link change state, so we receive a "known bad" packet.
+ */
+#define E_SUM_LINK_PKTERRS \
+ (INFINIPATH_E_SDROPPEDDATAPKT | INFINIPATH_E_SDROPPEDSMPPKT | \
+ INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SPKTLEN | \
+ INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RMINPKTLEN | \
+ INFINIPATH_E_RUNEXPCHAR)
static u64 handle_e_sum_errs(struct ipath_devdata *dd, ipath_err_t errs)
{
@@ -101,9 +115,7 @@ static u64 handle_e_sum_errs(struct ipat
if (ipath_debug & __IPATH_PKTDBG)
printk("\n");
}
- if ((errs & (INFINIPATH_E_SDROPPEDDATAPKT |
- INFINIPATH_E_SDROPPEDSMPPKT |
- INFINIPATH_E_SMINPKTLEN)) &&
+ if ((errs & E_SUM_LINK_PKTERRS) &&
!(dd->ipath_flags & IPATH_LINKACTIVE)) {
/*
* This can happen when SMA is trying to bring the link
@@ -112,11 +124,9 @@ static u64 handle_e_sum_errs(struct ipat
* valid. We don't want to confuse people, so we just
* don't print them, except at debug
*/
- ipath_dbg("Ignoring pktsend errors %llx, because not "
- "yet active\n", (unsigned long long) errs);
- ignore_this_time = INFINIPATH_E_SDROPPEDDATAPKT |
- INFINIPATH_E_SDROPPEDSMPPKT |
- INFINIPATH_E_SMINPKTLEN;
+ ipath_dbg("Ignoring packet errors %llx, because link not "
+ "ACTIVE\n", (unsigned long long) errs);
+ ignore_this_time = errs & E_SUM_LINK_PKTERRS;
}
return ignore_this_time;
@@ -157,7 +167,29 @@ static void handle_e_ibstatuschanged(str
*/
val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
lstate = val & IPATH_IBSTATE_MASK;
- if (lstate == IPATH_IBSTATE_INIT || lstate == IPATH_IBSTATE_ARM ||
+
+ /*
+ * this is confusing enough when it happens that I want to always put it
+ * on the console and in the logs. If it was a requested state change,
+ * we'll have already cleared the flags, so we won't print this warning
+ */
+ if ((lstate != IPATH_IBSTATE_ARM && lstate != IPATH_IBSTATE_ACTIVE)
+ && (dd->ipath_flags & (IPATH_LINKARMED | IPATH_LINKACTIVE))) {
+ dev_info(&dd->pcidev->dev, "Link state changed from %s to %s\n",
+ (dd->ipath_flags & IPATH_LINKARMED) ? "ARM" : "ACTIVE",
+ ib_linkstate(lstate));
+ /*
+ * Flush all queued sends when link went to DOWN or INIT,
+ * to be sure that they don't block SMA and other MAD packets
+ */
+ ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
+ INFINIPATH_S_ABORT);
+ ipath_disarm_piobufs(dd, dd->ipath_lastport_piobuf,
+ (unsigned)(dd->ipath_piobcnt2k +
+ dd->ipath_piobcnt4k) -
+ dd->ipath_lastport_piobuf);
+ }
+ else if (lstate == IPATH_IBSTATE_INIT || lstate == IPATH_IBSTATE_ARM ||
lstate == IPATH_IBSTATE_ACTIVE) {
/*
* only print at SMA if there is a change, debug if not
@@ -380,6 +412,19 @@ static void handle_errors(struct ipath_d
if (errs & E_SUM_ERRS)
ignore_this_time = handle_e_sum_errs(dd, errs);
+ else if ((errs & E_SUM_LINK_PKTERRS) &&
+ !(dd->ipath_flags & IPATH_LINKACTIVE)) {
+ /*
+ * This can happen when SMA is trying to bring the link
+ * up, but the IB link changes state at the "wrong" time.
+ * The IB logic then complains that the packet isn't
+ * valid. We don't want to confuse people, so we just
+ * don't print them, except at debug
+ */
+ ipath_dbg("Ignoring packet errors %llx, because link not "
+ "ACTIVE\n", (unsigned long long) errs);
+ ignore_this_time = errs & E_SUM_LINK_PKTERRS;
+ }
if (supp_msgs == 250000) {
/*
diff -r 9c072f8e7e68 -r 1e8837473193 drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:25 2006 -0700
@@ -62,9 +62,7 @@ struct ipath_portdata {
/* rcvhdrq base, needs mmap before useful */
void *port_rcvhdrq;
/* kernel virtual address where hdrqtail is updated */
- u64 *port_rcvhdrtail_kvaddr;
- /* page * used for uaddr */
- struct page *port_rcvhdrtail_pagep;
+ volatile __le64 *port_rcvhdrtail_kvaddr;
/*
* temp buffer for expected send setup, allocated at open, instead
* of each setup call
@@ -79,11 +77,7 @@ struct ipath_portdata {
dma_addr_t port_rcvegr_phys;
/* mmap of hdrq, must fit in 44 bits */
dma_addr_t port_rcvhdrq_phys;
- /*
- * the actual user address that we ipath_mlock'ed, so we can
- * ipath_munlock it at close
- */
- unsigned long port_rcvhdrtail_uaddr;
+ dma_addr_t port_rcvhdrqtailaddr_phys;
/*
* number of opens on this instance (0 or 1; ignoring forks, dup,
* etc. for now)
@@ -515,11 +509,6 @@ struct ipath_devdata {
u8 ipath_lmc;
};
-extern volatile __le64 *ipath_port0_rcvhdrtail;
-extern dma_addr_t ipath_port0_rcvhdrtail_dma;
-
-#define IPATH_PORT0_RCVHDRTAIL_SIZE PAGE_SIZE
-
extern struct list_head ipath_dev_list;
extern spinlock_t ipath_devs_lock;
extern struct ipath_devdata *ipath_lookup(int unit);
@@ -579,7 +568,7 @@ void ipath_disarm_piobufs(struct ipath_d
unsigned cnt);
int ipath_create_rcvhdrq(struct ipath_devdata *, struct ipath_portdata *);
-void ipath_free_pddata(struct ipath_devdata *, u32, int);
+void ipath_free_pddata(struct ipath_devdata *, struct ipath_portdata *);
int ipath_parse_ushort(const char *str, unsigned short *valp);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 20 of 39] IB/ipath - reduce overhead on receive interrupts
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (18 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 19 of 39] IB/ipath - memory management cleanups Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 21 of 39] IB/ipath - fixed bug 9776 for real. The problem was that I was updating Bryan O'Sullivan
` (19 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Also count the number of interrupts where that works (fastrcvint).
On any interrupt where the port0 head and tail registers are not equal,
just call the ipath_kreceive code without reading the interrupt status,
thus saving the approximately 0.25usec processor stall waiting for the
read to return. If any other interrupt bits are set, or head==tail,
take the normal path, but that has been reordered to handle read ahead
of pioavail. Also no longer call ipath_kreceive() from ipath_qcheck(),
because that just seems to make things worse, and isn't really buying
us anything, these days.
Also no longer loop in ipath_kreceive(); better to not hold things off
too long (I saw many cases where we would loop 4-8 times, and handle
thousands (up to 3500) in a single call).
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 1e8837473193 -r 8bc865893a11 drivers/infiniband/hw/ipath/ipath_common.h
--- a/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:25 2006 -0700
@@ -97,8 +97,8 @@ struct infinipath_stats {
__u64 sps_hwerrs;
/* number of times IB link changed state unexpectedly */
__u64 sps_iblink;
- /* no longer used; left for compatibility */
- __u64 sps_unused3;
+ /* kernel receive interrupts that didn't read intstat */
+ __u64 sps_fastrcvint;
/* number of kernel (port0) packets received */
__u64 sps_port0pkts;
/* number of "ethernet" packets sent by driver */
diff -r 1e8837473193 -r 8bc865893a11 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
@@ -888,12 +888,7 @@ void ipath_kreceive(struct ipath_devdata
(u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
goto done;
-gotmore:
- /*
- * read only once at start. If in flood situation, this helps
- * performance slightly. If more arrive while we are processing,
- * we'll come back here and do them
- */
+ /* read only once at start for performance */
hdrqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);
for (i = 0, l = dd->ipath_port0head; l != hdrqtail; i++) {
@@ -1022,10 +1017,6 @@ gotmore:
dd->ipath_port0head = l;
- if (hdrqtail != (u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
- /* more arrived while we handled first batch */
- goto gotmore;
-
if (pkttot > ipath_stats.sps_maxpkts_call)
ipath_stats.sps_maxpkts_call = pkttot;
ipath_stats.sps_port0pkts += pkttot;
diff -r 1e8837473193 -r 8bc865893a11 drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
@@ -539,10 +539,10 @@ static void handle_errors(struct ipath_d
continue;
if (hd == (tl + 1) ||
(!hd && tl == dd->ipath_hdrqlast)) {
+ if (i == 0)
+ chkerrpkts = 1;
dd->ipath_lastrcvhdrqtails[i] = tl;
pd->port_hdrqfull++;
- if (i == 0)
- chkerrpkts = 1;
}
}
}
@@ -724,7 +724,12 @@ set:
dd->ipath_sendctrl);
}
-static void handle_rcv(struct ipath_devdata *dd, u32 istat)
+/*
+ * Handle receive interrupts for user ports; this means a user
+ * process was waiting for a packet to arrive, and didn't want
+ * to poll
+ */
+static void handle_urcv(struct ipath_devdata *dd, u32 istat)
{
u64 portr;
int i;
@@ -734,22 +739,17 @@ static void handle_rcv(struct ipath_devd
infinipath_i_rcvavail_mask)
| ((istat >> INFINIPATH_I_RCVURG_SHIFT) &
infinipath_i_rcvurg_mask);
- for (i = 0; i < dd->ipath_cfgports; i++) {
+ for (i = 1; i < dd->ipath_cfgports; i++) {
struct ipath_portdata *pd = dd->ipath_pd[i];
- if (portr & (1 << i) && pd &&
- pd->port_cnt) {
- if (i == 0)
- ipath_kreceive(dd);
- else if (test_bit(IPATH_PORT_WAITING_RCV,
- &pd->port_flag)) {
- int rcbit;
- clear_bit(IPATH_PORT_WAITING_RCV,
- &pd->port_flag);
- rcbit = i + INFINIPATH_R_INTRAVAIL_SHIFT;
- clear_bit(1UL << rcbit, &dd->ipath_rcvctrl);
- wake_up_interruptible(&pd->port_wait);
- rcvdint = 1;
- }
+ if (portr & (1 << i) && pd && pd->port_cnt &&
+ test_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag)) {
+ int rcbit;
+ clear_bit(IPATH_PORT_WAITING_RCV,
+ &pd->port_flag);
+ rcbit = i + INFINIPATH_R_INTRAVAIL_SHIFT;
+ clear_bit(1UL << rcbit, &dd->ipath_rcvctrl);
+ wake_up_interruptible(&pd->port_wait);
+ rcvdint = 1;
}
}
if (rcvdint) {
@@ -767,14 +767,17 @@ irqreturn_t ipath_intr(int irq, void *da
struct ipath_devdata *dd = data;
u32 istat;
ipath_err_t estat = 0;
+ irqreturn_t ret;
+ u32 p0bits;
static unsigned unexpected = 0;
- irqreturn_t ret;
-
- if(!(dd->ipath_flags & IPATH_PRESENT)) {
- /* this is mostly so we don't try to touch the chip while
- * it is being reset */
- /*
- * This return value is perhaps odd, but we do not want the
+ static const u32 port0rbits = (1U<<INFINIPATH_I_RCVAVAIL_SHIFT) |
+ (1U<<INFINIPATH_I_RCVURG_SHIFT);
+
+ ipath_stats.sps_ints++;
+
+ if (!(dd->ipath_flags & IPATH_PRESENT)) {
+ /*
+ * This return value is not great, but we do not want the
* interrupt core code to remove our interrupt handler
* because we don't appear to be handling an interrupt
* during a chip reset.
@@ -782,6 +785,50 @@ irqreturn_t ipath_intr(int irq, void *da
return IRQ_HANDLED;
}
+ /*
+ * this needs to be flags&initted, not statusp, so we keep
+ * taking interrupts even after link goes down, etc.
+ * Also, we *must* clear the interrupt at some point, or we won't
+ * take it again, which can be real bad for errors, etc...
+ */
+
+ if (!(dd->ipath_flags & IPATH_INITTED)) {
+ ipath_bad_intr(dd, &unexpected);
+ ret = IRQ_NONE;
+ goto bail;
+ }
+
+ /*
+ * We try to avoid reading the interrupt status register, since
+ * that's a PIO read, and stalls the processor for up to about
+ * ~0.25 usec. The idea is that if we processed a port0 packet,
+ * we blindly clear the port 0 receive interrupt bits, and nothing
+ * else, then return. If other interrupts are pending, the chip
+ * will re-interrupt us as soon as we write the intclear register.
+ * We then won't process any more kernel packets (if not the 2nd
+ * time, then the 3rd or 4th) and we'll then handle the other
+ * interrupts. We clear the interrupts first so that we don't
+ * lose intr for later packets that arrive while we are processing.
+ */
+ if (dd->ipath_port0head !=
+ (u32)le64_to_cpu(*dd->ipath_hdrqtailptr)) {
+ u32 oldhead = dd->ipath_port0head;
+ if (dd->ipath_flags & IPATH_GPIO_INTR) {
+ ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
+ (u64) (1 << 2));
+ p0bits = port0rbits | INFINIPATH_I_GPIO;
+ }
+ else
+ p0bits = port0rbits;
+ ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, p0bits);
+ ipath_kreceive(dd);
+ if (oldhead != dd->ipath_port0head) {
+ ipath_stats.sps_fastrcvint++;
+ goto done;
+ }
+ istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
+ }
+
istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
if (unlikely(!istat)) {
ipath_stats.sps_nullintr++;
@@ -795,31 +842,17 @@ irqreturn_t ipath_intr(int irq, void *da
goto bail;
}
- ipath_stats.sps_ints++;
-
- /*
- * this needs to be flags&initted, not statusp, so we keep
- * taking interrupts even after link goes down, etc.
- * Also, we *must* clear the interrupt at some point, or we won't
- * take it again, which can be real bad for errors, etc...
- */
-
- if (!(dd->ipath_flags & IPATH_INITTED)) {
- ipath_bad_intr(dd, &unexpected);
- ret = IRQ_NONE;
- goto bail;
- }
if (unexpected)
unexpected = 0;
- ipath_cdbg(VERBOSE, "intr stat=0x%x\n", istat);
-
- if (istat & ~infinipath_i_bitsextant)
+ if (unlikely(istat & ~infinipath_i_bitsextant))
ipath_dev_err(dd,
"interrupt with unknown interrupts %x set\n",
istat & (u32) ~ infinipath_i_bitsextant);
-
- if (istat & INFINIPATH_I_ERROR) {
+ else
+ ipath_cdbg(VERBOSE, "intr stat=0x%x\n", istat);
+
+ if (unlikely(istat & INFINIPATH_I_ERROR)) {
ipath_stats.sps_errints++;
estat = ipath_read_kreg64(dd,
dd->ipath_kregs->kr_errorstatus);
@@ -837,7 +870,14 @@ irqreturn_t ipath_intr(int irq, void *da
handle_errors(dd, estat);
}
+ p0bits = port0rbits;
if (istat & INFINIPATH_I_GPIO) {
+ /*
+ * Packets are available in the port 0 rcv queue.
+ * Eventually this needs to be generalized to check
+ * IPATH_GPIO_INTR, and the specific GPIO bit, if
+ * GPIO interrupts are used for anything else.
+ */
if (unlikely(!(dd->ipath_flags & IPATH_GPIO_INTR))) {
u32 gpiostatus;
gpiostatus = ipath_read_kreg32(
@@ -851,14 +891,7 @@ irqreturn_t ipath_intr(int irq, void *da
/* Clear GPIO status bit 2 */
ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
(u64) (1 << 2));
-
- /*
- * Packets are available in the port 0 rcv queue.
- * Eventually this needs to be generalized to check
- * IPATH_GPIO_INTR, and the specific GPIO bit, if
- * GPIO interrupts are used for anything else.
- */
- ipath_kreceive(dd);
+ p0bits |= INFINIPATH_I_GPIO;
}
}
@@ -871,6 +904,25 @@ irqreturn_t ipath_intr(int irq, void *da
*/
ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, istat);
+ /*
+ * we check for both transition from empty to non-empty, and urgent
+ * packets (those with the interrupt bit set in the header), and
+ * if enabled, the GPIO bit 2 interrupt used for port0 on some
+ * HT-400 boards.
+ * Do this before checking for pio buffers available, since
+ * receives can overflow; piobuf waiters can afford a few
+ * extra cycles, since they were waiting anyway.
+ */
+ if (istat & p0bits) {
+ ipath_kreceive(dd);
+ istat &= ~port0rbits;
+ }
+ if (istat & ((infinipath_i_rcvavail_mask <<
+ INFINIPATH_I_RCVAVAIL_SHIFT)
+ | (infinipath_i_rcvurg_mask <<
+ INFINIPATH_I_RCVURG_SHIFT)))
+ handle_urcv(dd, istat);
+
if (istat & INFINIPATH_I_SPIOBUFAVAIL) {
clear_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
@@ -882,17 +934,7 @@ irqreturn_t ipath_intr(int irq, void *da
handle_layer_pioavail(dd);
}
- /*
- * we check for both transition from empty to non-empty, and urgent
- * packets (those with the interrupt bit set in the header)
- */
-
- if (istat & ((infinipath_i_rcvavail_mask <<
- INFINIPATH_I_RCVAVAIL_SHIFT)
- | (infinipath_i_rcvurg_mask <<
- INFINIPATH_I_RCVURG_SHIFT)))
- handle_rcv(dd, istat);
-
+done:
ret = IRQ_HANDLED;
bail:
diff -r 1e8837473193 -r 8bc865893a11 drivers/infiniband/hw/ipath/ipath_stats.c
--- a/drivers/infiniband/hw/ipath/ipath_stats.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_stats.c Thu Jun 29 14:33:25 2006 -0700
@@ -186,7 +186,6 @@ static void ipath_qcheck(struct ipath_de
dd->ipath_port0head,
(unsigned long long)
ipath_stats.sps_port0pkts);
- ipath_kreceive(dd);
}
dd->ipath_lastport0rcv_cnt = ipath_stats.sps_port0pkts;
}
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 21 of 39] IB/ipath - fixed bug 9776 for real. The problem was that I was updating
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (19 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 20 of 39] IB/ipath - reduce overhead on receive interrupts Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 22 of 39] IB/ipath - fix lost interrupts on HT-400 Bryan O'Sullivan
` (18 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
the head register multiple times in the rcvhdrq processing loop,
and setting the counter on each update. Since that meant that
the tail register was ahead of head for all but the last update,
we would get extra interrupts. The fix was to not write the
counter value except on the last update.
I also changed to update rcvhdrhead and rcvegrindexhead at most
every 16 packets, if there were lots of packets in the queue
(and of course, on the last packet, regardless).
I also made some small cleanups while debugging this.
With these changes, xeon/monty typically sees two openib packets
per interrupt on sdp and ipoib, opteron/monty is about 1.25 pkts/intr.
I'm seeing about 3800 Mbit/s monty/xeon, and 5000-5100 opteron/monty
with netperf sdp. Netpipe doesn't show as good as that, peaking
at about 4400 on opteron/monty sdp. Plain ipoib xeon is about 2100+
netperf, opteron 2900+, at 128KB
Signed-off-by: olson@eng-12.pathscale.com
Signed-off-by: Bryan O'Sullivan <bos@pathscale.com>
diff -r 8bc865893a11 -r 1a4350d895c9 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -870,7 +870,7 @@ void ipath_kreceive(struct ipath_devdata
const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize; /* words */
u32 etail = -1, l, hdrqtail;
struct ips_message_header *hdr;
- u32 eflags, i, etype, tlen, pkttot = 0;
+ u32 eflags, i, etype, tlen, pkttot = 0, updegr=0;
static u64 totcalls; /* stats, may eventually remove */
char emsg[128];
@@ -884,14 +884,14 @@ void ipath_kreceive(struct ipath_devdata
if (test_and_set_bit(0, &dd->ipath_rcv_pending))
goto bail;
- if (dd->ipath_port0head ==
- (u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
+ l = dd->ipath_port0head;
+ if (l == (u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
goto done;
/* read only once at start for performance */
hdrqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);
- for (i = 0, l = dd->ipath_port0head; l != hdrqtail; i++) {
+ for (i = 0; l != hdrqtail; i++) {
u32 qp;
u8 *bthbytes;
@@ -1002,15 +1002,26 @@ void ipath_kreceive(struct ipath_devdata
l += rsize;
if (l >= maxcnt)
l = 0;
+ if (etype != RCVHQ_RCV_TYPE_EXPECTED)
+ updegr = 1;
/*
- * update for each packet, to help prevent overflows if we
- * have lots of packets.
+ * update head regs on last packet, and every 16 packets.
+ * Reduce bus traffic, while still trying to prevent
+ * rcvhdrq overflows, for when the queue is nearly full
*/
- (void)ipath_write_ureg(dd, ur_rcvhdrhead,
- dd->ipath_rhdrhead_intr_off | l, 0);
- if (etype != RCVHQ_RCV_TYPE_EXPECTED)
- (void)ipath_write_ureg(dd, ur_rcvegrindexhead,
- etail, 0);
+ if (l == hdrqtail || (i && !(i&0xf))) {
+ u64 lval;
+ if (l == hdrqtail) /* want interrupt only on last */
+ lval = dd->ipath_rhdrhead_intr_off | l;
+ else
+ lval = l;
+ (void)ipath_write_ureg(dd, ur_rcvhdrhead, lval, 0);
+ if (updegr) {
+ (void)ipath_write_ureg(dd, ur_rcvegrindexhead,
+ etail, 0);
+ updegr = 0;
+ }
+ }
}
pkttot += i;
diff -r 8bc865893a11 -r 1a4350d895c9 drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:25 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
@@ -383,7 +383,7 @@ static unsigned handle_frequent_errors(s
return supp_msgs;
}
-static void handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
+static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
{
char msg[512];
u64 ignore_this_time = 0;
@@ -480,7 +480,7 @@ static void handle_errors(struct ipath_d
INFINIPATH_E_IBSTATUSCHANGED);
}
if (!errs)
- return;
+ return 0;
if (!noprint)
/*
@@ -604,9 +604,7 @@ static void handle_errors(struct ipath_d
wake_up_interruptible(&ipath_sma_state_wait);
}
- if (chkerrpkts)
- /* process possible error packets in hdrq */
- ipath_kreceive(dd);
+ return chkerrpkts;
}
/* this is separate to allow for better optimization of ipath_intr() */
@@ -765,10 +763,10 @@ irqreturn_t ipath_intr(int irq, void *da
irqreturn_t ipath_intr(int irq, void *data, struct pt_regs *regs)
{
struct ipath_devdata *dd = data;
- u32 istat;
+ u32 istat, chk0rcv = 0;
ipath_err_t estat = 0;
irqreturn_t ret;
- u32 p0bits;
+ u32 p0bits, oldhead;
static unsigned unexpected = 0;
static const u32 port0rbits = (1U<<INFINIPATH_I_RCVAVAIL_SHIFT) |
(1U<<INFINIPATH_I_RCVURG_SHIFT);
@@ -810,9 +808,8 @@ irqreturn_t ipath_intr(int irq, void *da
* interrupts. We clear the interrupts first so that we don't
* lose intr for later packets that arrive while we are processing.
*/
- if (dd->ipath_port0head !=
- (u32)le64_to_cpu(*dd->ipath_hdrqtailptr)) {
- u32 oldhead = dd->ipath_port0head;
+ oldhead = dd->ipath_port0head;
+ if (oldhead != (u32) le64_to_cpu(*dd->ipath_hdrqtailptr)) {
if (dd->ipath_flags & IPATH_GPIO_INTR) {
ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
(u64) (1 << 2));
@@ -830,6 +827,8 @@ irqreturn_t ipath_intr(int irq, void *da
}
istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
+ p0bits = port0rbits;
+
if (unlikely(!istat)) {
ipath_stats.sps_nullintr++;
ret = IRQ_NONE; /* not our interrupt, or already handled */
@@ -867,10 +866,11 @@ irqreturn_t ipath_intr(int irq, void *da
ipath_dev_err(dd, "Read of error status failed "
"(all bits set); ignoring\n");
else
- handle_errors(dd, estat);
- }
-
- p0bits = port0rbits;
+ if (handle_errors(dd, estat))
+ /* force calling ipath_kreceive() */
+ chk0rcv = 1;
+ }
+
if (istat & INFINIPATH_I_GPIO) {
/*
* Packets are available in the port 0 rcv queue.
@@ -892,8 +892,10 @@ irqreturn_t ipath_intr(int irq, void *da
ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
(u64) (1 << 2));
p0bits |= INFINIPATH_I_GPIO;
- }
- }
+ chk0rcv = 1;
+ }
+ }
+ chk0rcv |= istat & p0bits;
/*
* clear the ones we will deal with on this round
@@ -905,18 +907,16 @@ irqreturn_t ipath_intr(int irq, void *da
ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, istat);
/*
- * we check for both transition from empty to non-empty, and urgent
- * packets (those with the interrupt bit set in the header), and
- * if enabled, the GPIO bit 2 interrupt used for port0 on some
- * HT-400 boards.
- * Do this before checking for pio buffers available, since
- * receives can overflow; piobuf waiters can afford a few
- * extra cycles, since they were waiting anyway.
- */
- if (istat & p0bits) {
+ * handle port0 receive before checking for pio buffers available,
+ * since receives can overflow; piobuf waiters can afford a few
+ * extra cycles, since they were waiting anyway, and user's waiting
+ * for receive are at the bottom.
+ */
+ if (chk0rcv) {
ipath_kreceive(dd);
istat &= ~port0rbits;
}
+
if (istat & ((infinipath_i_rcvavail_mask <<
INFINIPATH_I_RCVAVAIL_SHIFT)
| (infinipath_i_rcvurg_mask <<
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 22 of 39] IB/ipath - fix lost interrupts on HT-400
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (20 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 21 of 39] IB/ipath - fixed bug 9776 for real. The problem was that I was updating Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 23 of 39] IB/ipath - disallow send of invalid packet sizes over UD Bryan O'Sullivan
` (17 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Do an extra check to see if in-memory tail changed while processing
packets, and if so, going back through the loop again (but only once
per call to ipath_kreceive()). In practice, this seems to be enough
to guarantee that if we crossed the clearing of an interrupt at start of
ipath_intr with a scheduled tail register update, that we'll process the
"extra" packet that lost the interrupt because we cleared it just as it
was about to arrive.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 1a4350d895c9 -r 811021b6c112 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -870,7 +870,7 @@ void ipath_kreceive(struct ipath_devdata
const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize; /* words */
u32 etail = -1, l, hdrqtail;
struct ips_message_header *hdr;
- u32 eflags, i, etype, tlen, pkttot = 0, updegr=0;
+ u32 eflags, i, etype, tlen, pkttot = 0, updegr=0, reloop=0;
static u64 totcalls; /* stats, may eventually remove */
char emsg[128];
@@ -885,9 +885,11 @@ void ipath_kreceive(struct ipath_devdata
goto bail;
l = dd->ipath_port0head;
- if (l == (u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
+ hdrqtail = (u32) le64_to_cpu(*dd->ipath_hdrqtailptr);
+ if (l == hdrqtail)
goto done;
+reloop:
/* read only once at start for performance */
hdrqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);
@@ -1011,7 +1013,7 @@ void ipath_kreceive(struct ipath_devdata
*/
if (l == hdrqtail || (i && !(i&0xf))) {
u64 lval;
- if (l == hdrqtail) /* want interrupt only on last */
+ if (l == hdrqtail) /* PE-800 interrupt only on last */
lval = dd->ipath_rhdrhead_intr_off | l;
else
lval = l;
@@ -1021,6 +1023,23 @@ void ipath_kreceive(struct ipath_devdata
etail, 0);
updegr = 0;
}
+ }
+ }
+
+ if (!dd->ipath_rhdrhead_intr_off && !reloop) {
+ /* HT-400 workaround; we can have a race clearing chip
+ * interrupt with another interrupt about to be delivered,
+ * and can clear it before it is delivered on the GPIO
+ * workaround. By doing the extra check here for the
+ * in-memory tail register updating while we were doing
+ * earlier packets, we "almost" guarantee we have covered
+ * that case.
+ */
+ u32 hqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);
+ if (hqtail != hdrqtail) {
+ hdrqtail = hqtail;
+ reloop = 1; /* loop 1 extra time at most */
+ goto reloop;
}
}
diff -r 1a4350d895c9 -r 811021b6c112 drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
@@ -766,7 +766,7 @@ irqreturn_t ipath_intr(int irq, void *da
u32 istat, chk0rcv = 0;
ipath_err_t estat = 0;
irqreturn_t ret;
- u32 p0bits, oldhead;
+ u32 oldhead, curtail;
static unsigned unexpected = 0;
static const u32 port0rbits = (1U<<INFINIPATH_I_RCVAVAIL_SHIFT) |
(1U<<INFINIPATH_I_RCVURG_SHIFT);
@@ -809,15 +809,16 @@ irqreturn_t ipath_intr(int irq, void *da
* lose intr for later packets that arrive while we are processing.
*/
oldhead = dd->ipath_port0head;
- if (oldhead != (u32) le64_to_cpu(*dd->ipath_hdrqtailptr)) {
+ curtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);
+ if (oldhead != curtail) {
if (dd->ipath_flags & IPATH_GPIO_INTR) {
ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
(u64) (1 << 2));
- p0bits = port0rbits | INFINIPATH_I_GPIO;
+ istat = port0rbits | INFINIPATH_I_GPIO;
}
else
- p0bits = port0rbits;
- ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, p0bits);
+ istat = port0rbits;
+ ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, istat);
ipath_kreceive(dd);
if (oldhead != dd->ipath_port0head) {
ipath_stats.sps_fastrcvint++;
@@ -827,7 +828,6 @@ irqreturn_t ipath_intr(int irq, void *da
}
istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
- p0bits = port0rbits;
if (unlikely(!istat)) {
ipath_stats.sps_nullintr++;
@@ -890,19 +890,19 @@ irqreturn_t ipath_intr(int irq, void *da
else {
/* Clear GPIO status bit 2 */
ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
- (u64) (1 << 2));
- p0bits |= INFINIPATH_I_GPIO;
+ (u64) (1 << 2));
chk0rcv = 1;
}
}
- chk0rcv |= istat & p0bits;
-
- /*
- * clear the ones we will deal with on this round
- * We clear it early, mostly for receive interrupts, so we
- * know the chip will have seen this by the time we process
- * the queue, and will re-interrupt if necessary. The processor
- * itself won't take the interrupt again until we return.
+ chk0rcv |= istat & port0rbits;
+
+ /*
+ * Clear the interrupt bits we found set, unless they are receive
+ * related, in which case we already cleared them above, and don't
+ * want to clear them again, because we might lose an interrupt.
+ * Clear it early, so we "know" know the chip will have seen this by
+ * the time we process the queue, and will re-interrupt if necessary.
+ * The processor itself won't take the interrupt again until we return.
*/
ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, istat);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 23 of 39] IB/ipath - disallow send of invalid packet sizes over UD
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (21 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 22 of 39] IB/ipath - fix lost interrupts on HT-400 Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 24 of 39] IB/ipath - don't confuse the max message size with the MTU Bryan O'Sullivan
` (16 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 811021b6c112 -r 8e39364c2402 drivers/infiniband/hw/ipath/ipath_ud.c
--- a/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:26 2006 -0700
@@ -274,6 +274,11 @@ int ipath_post_ud_send(struct ipath_qp *
}
len += wr->sg_list[i].length;
ss.num_sge++;
+ }
+ /* Check for invalid packet size. */
+ if (len > ipath_layer_get_ibmtu(dev->dd)) {
+ ret = -EINVAL;
+ goto bail;
}
extra_bytes = (4 - len) & 3;
nwords = (len + extra_bytes) >> 2;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 24 of 39] IB/ipath - don't confuse the max message size with the MTU
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (22 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 23 of 39] IB/ipath - disallow send of invalid packet sizes over UD Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 25 of 39] IB/ipath - removed redundant statements Bryan O'Sullivan
` (15 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 8e39364c2402 -r e952aedb0e94 drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
@@ -695,7 +695,7 @@ static int ipath_query_port(struct ib_de
ipath_layer_get_lastibcstat(dev->dd) & 0xf];
props->port_cap_flags = dev->port_cap_flags;
props->gid_tbl_len = 1;
- props->max_msg_sz = 4096;
+ props->max_msg_sz = 0x80000000;
props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd);
props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) -
dev->z_pkey_violations;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 25 of 39] IB/ipath - removed redundant statements
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (23 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 24 of 39] IB/ipath - don't confuse the max message size with the MTU Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 26 of 39] IB/ipath - check for valid LID and multicast LIDs Bryan O'Sullivan
` (14 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
The tail register read became redundant as the result of earlier receive
interrupt bug fixes.
Drop another unneeded register read.
And another line that got duplicated.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r e952aedb0e94 -r 4c581c37bb95 drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -890,9 +890,6 @@ void ipath_kreceive(struct ipath_devdata
goto done;
reloop:
- /* read only once at start for performance */
- hdrqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);
-
for (i = 0; l != hdrqtail; i++) {
u32 qp;
u8 *bthbytes;
diff -r e952aedb0e94 -r 4c581c37bb95 drivers/infiniband/hw/ipath/ipath_ht400.c
--- a/drivers/infiniband/hw/ipath/ipath_ht400.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ht400.c Thu Jun 29 14:33:26 2006 -0700
@@ -1573,7 +1573,6 @@ void ipath_init_ht400_funcs(struct ipath
dd->ipath_f_reset = ipath_setup_ht_reset;
dd->ipath_f_get_boardname = ipath_ht_boardname;
dd->ipath_f_init_hwerrors = ipath_ht_init_hwerrors;
- dd->ipath_f_init_hwerrors = ipath_ht_init_hwerrors;
dd->ipath_f_early_init = ipath_ht_early_init;
dd->ipath_f_handle_hwerrors = ipath_ht_handle_hwerrors;
dd->ipath_f_quiet_serdes = ipath_ht_quiet_serdes;
diff -r e952aedb0e94 -r 4c581c37bb95 drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
@@ -824,7 +824,6 @@ irqreturn_t ipath_intr(int irq, void *da
ipath_stats.sps_fastrcvint++;
goto done;
}
- istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
}
istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 26 of 39] IB/ipath - check for valid LID and multicast LIDs
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (24 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 25 of 39] IB/ipath - removed redundant statements Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 27 of 39] IB/ipath - fixes to performance get counters for IB compliance Bryan O'Sullivan
` (13 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 4c581c37bb95 -r eef7f8021500 drivers/infiniband/hw/ipath/ipath_sysfs.c
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
@@ -280,7 +280,7 @@ static ssize_t store_lid(struct device *
if (ret < 0)
goto invalid;
- if (lid == 0 || lid >= 0xc000) {
+ if (lid == 0 || lid >= IPS_MULTICAST_LID_BASE) {
ret = -EINVAL;
goto invalid;
}
@@ -314,7 +314,7 @@ static ssize_t store_mlid(struct device
int ret;
ret = ipath_parse_ushort(buf, &mlid);
- if (ret < 0)
+ if (ret < 0 || mlid < IPS_MULTICAST_LID_BASE)
goto invalid;
unit = dd->ipath_unit;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 27 of 39] IB/ipath - fixes to performance get counters for IB compliance
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (25 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 26 of 39] IB/ipath - check for valid LID and multicast LIDs Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering Bryan O'Sullivan
` (12 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
This patch fixes some problems uncovered during IB compliance
testing to return the right values for error counters returned
by the Performance Get Counters packet.
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -460,6 +460,8 @@ static int __devinit ipath_init_one(stru
* by ipath_setup_htconfig.
*/
dd->ipath_flags = 0;
+ dd->ipath_lli_counter = 0;
+ dd->ipath_lli_errors = 0;
if (dd->ipath_f_bus(dd, pdev))
ipath_dev_err(dd, "Failed to setup config space; "
@@ -942,6 +944,18 @@ reloop:
"tlen=%x opcode=%x egridx=%x: %s\n",
eflags, l, etype, tlen, bthbytes[0],
ips_get_index((__le32 *) rc), emsg);
+ /* Count local link integrity errors. */
+ if (eflags & (INFINIPATH_RHF_H_ICRCERR |
+ INFINIPATH_RHF_H_VCRCERR)) {
+ u8 n = (dd->ipath_ibcctrl >>
+ INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
+ INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
+
+ if (++dd->ipath_lli_counter > n) {
+ dd->ipath_lli_counter = 0;
+ dd->ipath_lli_errors++;
+ }
+ }
} else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
int ret = __ipath_verbs_rcv(dd, rc + 1,
ebuf, tlen);
@@ -949,6 +963,9 @@ reloop:
ipath_cdbg(VERBOSE,
"received IB packet, "
"not SMA (QP=%x)\n", qp);
+ if (dd->ipath_lli_counter)
+ dd->ipath_lli_counter--;
+
} else if (etype == RCVHQ_RCV_TYPE_EAGER) {
if (qp == IPATH_KD_QP &&
bthbytes[0] == ipath_layer_rcv_opcode &&
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
@@ -262,6 +262,7 @@ static void handle_e_ibstatuschanged(str
| IPATH_LINKACTIVE |
IPATH_LINKARMED);
*dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
+ dd->ipath_lli_counter = 0;
if (!noprint) {
if (((dd->ipath_lastibcstat >>
INFINIPATH_IBCS_LINKSTATE_SHIFT) &
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:26 2006 -0700
@@ -507,6 +507,11 @@ struct ipath_devdata {
u8 ipath_pci_cacheline;
/* LID mask control */
u8 ipath_lmc;
+
+ /* local link integrity counter */
+ u32 ipath_lli_counter;
+ /* local link integrity errors */
+ u32 ipath_lli_errors;
};
extern struct list_head ipath_dev_list;
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:26 2006 -0700
@@ -1032,19 +1032,22 @@ int ipath_layer_get_counters(struct ipat
ipath_snap_cntr(dd, dd->ipath_cregs->cr_ibsymbolerrcnt);
cntrs->link_error_recovery_counter =
ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkerrrecovcnt);
+ /*
+ * The link downed counter counts when the other side downs the
+ * connection. We add in the number of times we downed the link
+ * due to local link integrity errors to compensate.
+ */
cntrs->link_downed_counter =
ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkdowncnt);
cntrs->port_rcv_errors =
ipath_snap_cntr(dd, dd->ipath_cregs->cr_rxdroppktcnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvovflcnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_portovflcnt) +
- ipath_snap_cntr(dd, dd->ipath_cregs->cr_errrcvflowctrlcnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_err_rlencnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_invalidrlencnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) +
- ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlinkcnt) +
ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt);
cntrs->port_rcv_remphys_errors =
ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt);
@@ -1058,6 +1061,8 @@ int ipath_layer_get_counters(struct ipat
ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
cntrs->port_rcv_packets =
ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
+ cntrs->local_link_integrity_errors = dd->ipath_lli_errors;
+ cntrs->excessive_buffer_overrun_errors = 0; /* XXX */
ret = 0;
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_layer.h
--- a/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:26 2006 -0700
@@ -55,6 +55,8 @@ struct ipath_layer_counters {
u64 port_rcv_data;
u64 port_xmit_packets;
u64 port_rcv_packets;
+ u32 local_link_integrity_errors;
+ u32 excessive_buffer_overrun_errors;
};
/*
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_mad.c
--- a/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:26 2006 -0700
@@ -613,6 +613,9 @@ struct ib_pma_portcounters {
#define IB_PMA_SEL_PORT_RCV_ERRORS __constant_htons(0x0008)
#define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS __constant_htons(0x0010)
#define IB_PMA_SEL_PORT_XMIT_DISCARDS __constant_htons(0x0040)
+#define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS __constant_htons(0x0200)
+#define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS __constant_htons(0x0400)
+#define IB_PMA_SEL_PORT_VL15_DROPPED __constant_htons(0x0800)
#define IB_PMA_SEL_PORT_XMIT_DATA __constant_htons(0x1000)
#define IB_PMA_SEL_PORT_RCV_DATA __constant_htons(0x2000)
#define IB_PMA_SEL_PORT_XMIT_PACKETS __constant_htons(0x4000)
@@ -859,6 +862,10 @@ static int recv_pma_get_portcounters(str
cntrs.port_rcv_data -= dev->z_port_rcv_data;
cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
+ cntrs.local_link_integrity_errors -=
+ dev->z_local_link_integrity_errors;
+ cntrs.excessive_buffer_overrun_errors -=
+ dev->z_excessive_buffer_overrun_errors;
memset(pmp->data, 0, sizeof(pmp->data));
@@ -896,6 +903,16 @@ static int recv_pma_get_portcounters(str
else
p->port_xmit_discards =
cpu_to_be16((u16)cntrs.port_xmit_discards);
+ if (cntrs.local_link_integrity_errors > 0xFUL)
+ cntrs.local_link_integrity_errors = 0xFUL;
+ if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
+ cntrs.excessive_buffer_overrun_errors = 0xFUL;
+ p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
+ cntrs.excessive_buffer_overrun_errors;
+ if (dev->n_vl15_dropped > 0xFFFFUL)
+ p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
+ else
+ p->vl15_dropped = cpu_to_be16((u16)dev->n_vl15_dropped);
if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
else
@@ -989,6 +1006,17 @@ static int recv_pma_set_portcounters(str
if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
dev->z_port_xmit_discards = cntrs.port_xmit_discards;
+
+ if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
+ dev->z_local_link_integrity_errors =
+ cntrs.local_link_integrity_errors;
+
+ if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
+ dev->z_excessive_buffer_overrun_errors =
+ cntrs.excessive_buffer_overrun_errors;
+
+ if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED)
+ dev->n_vl15_dropped = 0;
if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
dev->z_port_xmit_data = cntrs.port_xmit_data;
@@ -1275,32 +1303,8 @@ int ipath_process_mad(struct ib_device *
struct ib_wc *in_wc, struct ib_grh *in_grh,
struct ib_mad *in_mad, struct ib_mad *out_mad)
{
- struct ipath_ibdev *dev = to_idev(ibdev);
int ret;
- /*
- * Snapshot current HW counters to "clear" them.
- * This should be done when the driver is loaded except that for
- * some reason we get a zillion errors when brining up the link.
- */
- if (dev->rcv_errors == 0) {
- struct ipath_layer_counters cntrs;
-
- ipath_layer_get_counters(to_idev(ibdev)->dd, &cntrs);
- dev->rcv_errors++;
- dev->z_symbol_error_counter = cntrs.symbol_error_counter;
- dev->z_link_error_recovery_counter =
- cntrs.link_error_recovery_counter;
- dev->z_link_downed_counter = cntrs.link_downed_counter;
- dev->z_port_rcv_errors = cntrs.port_rcv_errors + 1;
- dev->z_port_rcv_remphys_errors =
- cntrs.port_rcv_remphys_errors;
- dev->z_port_xmit_discards = cntrs.port_xmit_discards;
- dev->z_port_xmit_data = cntrs.port_xmit_data;
- dev->z_port_rcv_data = cntrs.port_rcv_data;
- dev->z_port_xmit_packets = cntrs.port_xmit_packets;
- dev->z_port_rcv_packets = cntrs.port_rcv_packets;
- }
switch (in_mad->mad_hdr.mgmt_class) {
case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
case IB_MGMT_CLASS_SUBN_LID_ROUTED:
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_ud.c
--- a/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:26 2006 -0700
@@ -560,7 +560,16 @@ void ipath_ud_rcv(struct ipath_ibdev *de
spin_lock_irqsave(&rq->lock, flags);
if (rq->tail == rq->head) {
spin_unlock_irqrestore(&rq->lock, flags);
- dev->n_pkt_drops++;
+ /*
+ * Count VL15 packets dropped due to no receive buffer.
+ * Otherwise, count them as buffer overruns since usually,
+ * the HW will be able to receive packets even if there are
+ * no QPs with posted receive buffers.
+ */
+ if (qp->ibqp.qp_num == 0)
+ dev->n_vl15_dropped++;
+ else
+ dev->rcv_errors++;
goto bail;
}
/* Silently drop packets which are too big. */
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
@@ -981,6 +981,7 @@ static int ipath_verbs_register_sysfs(st
*/
static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd)
{
+ struct ipath_layer_counters cntrs;
struct ipath_ibdev *idev;
struct ib_device *dev;
int ret;
@@ -1030,6 +1031,25 @@ static void *ipath_register_ib_device(in
idev->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS;
idev->pma_counter_select[5] = IB_PMA_PORT_XMIT_WAIT;
idev->link_width_enabled = 3; /* 1x or 4x */
+
+ /* Snapshot current HW counters to "clear" them. */
+ ipath_layer_get_counters(dd, &cntrs);
+ idev->z_symbol_error_counter = cntrs.symbol_error_counter;
+ idev->z_link_error_recovery_counter =
+ cntrs.link_error_recovery_counter;
+ idev->z_link_downed_counter = cntrs.link_downed_counter;
+ idev->z_port_rcv_errors = cntrs.port_rcv_errors;
+ idev->z_port_rcv_remphys_errors =
+ cntrs.port_rcv_remphys_errors;
+ idev->z_port_xmit_discards = cntrs.port_xmit_discards;
+ idev->z_port_xmit_data = cntrs.port_xmit_data;
+ idev->z_port_rcv_data = cntrs.port_rcv_data;
+ idev->z_port_xmit_packets = cntrs.port_xmit_packets;
+ idev->z_port_rcv_packets = cntrs.port_rcv_packets;
+ idev->z_local_link_integrity_errors =
+ cntrs.local_link_integrity_errors;
+ idev->z_excessive_buffer_overrun_errors =
+ cntrs.excessive_buffer_overrun_errors;
/*
* The system image GUID is supposed to be the same for all
diff -r eef7f8021500 -r 7d22a8963bda drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:26 2006 -0700
@@ -460,6 +460,8 @@ struct ipath_ibdev {
u64 z_port_xmit_packets; /* starting count for PMA */
u64 z_port_rcv_packets; /* starting count for PMA */
u32 z_pkey_violations; /* starting count for PMA */
+ u32 z_local_link_integrity_errors; /* starting count for PMA */
+ u32 z_excessive_buffer_overrun_errors; /* starting count for PMA */
u32 n_rc_resends;
u32 n_rc_acks;
u32 n_rc_qacks;
@@ -469,6 +471,7 @@ struct ipath_ibdev {
u32 n_other_naks;
u32 n_timeouts;
u32 n_pkt_drops;
+ u32 n_vl15_dropped;
u32 n_wqe_errs;
u32 n_rdma_dup_busy;
u32 n_piowait;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (26 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 27 of 39] IB/ipath - fixes to performance get counters for IB compliance Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-30 0:07 ` Andrew Morton
2006-06-29 21:41 ` [PATCH 29 of 39] IB/ipath - RC receive interrupt performance changes Bryan O'Sullivan
` (11 subsequent siblings)
39 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
The mb() prevents the compiler from reordering on this function, with some versions
of gcc and -Os optimization. The result is random failures in the EEPROM read
without this change.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 7d22a8963bda -r 5f3c0b2d446d drivers/infiniband/hw/ipath/ipath_eeprom.c
--- a/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:26 2006 -0700
@@ -186,6 +186,7 @@ bail:
*/
static void i2c_wait_for_writes(struct ipath_devdata *dd)
{
+ mb();
(void)ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
}
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering
2006-06-29 21:41 ` [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering Bryan O'Sullivan
@ 2006-06-30 0:07 ` Andrew Morton
2006-06-30 21:08 ` Roland Dreier
0 siblings, 1 reply; 65+ messages in thread
From: Andrew Morton @ 2006-06-30 0:07 UTC (permalink / raw)
To: Bryan O'Sullivan; +Cc: rdreier, mst, openib-general, linux-kernel, netdev
"Bryan O'Sullivan" <bos@pathscale.com> wrote:
>
> The mb() prevents the compiler from reordering on this function, with some versions
> of gcc and -Os optimization. The result is random failures in the EEPROM read
> without this change.
>
>
> Signed-off-by: Dave Olson <dave.olson@qlogic.com>
> Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
>
> diff -r 7d22a8963bda -r 5f3c0b2d446d drivers/infiniband/hw/ipath/ipath_eeprom.c
> --- a/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:26 2006 -0700
> +++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:26 2006 -0700
> @@ -186,6 +186,7 @@ bail:
> */
> static void i2c_wait_for_writes(struct ipath_devdata *dd)
> {
> + mb();
> (void)ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
> }
>
That's a bit weird. I wouldn't have expected the compiler to muck around
with a readl().
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering
2006-06-30 0:07 ` Andrew Morton
@ 2006-06-30 21:08 ` Roland Dreier
0 siblings, 0 replies; 65+ messages in thread
From: Roland Dreier @ 2006-06-30 21:08 UTC (permalink / raw)
To: Andrew Morton
Cc: Bryan O'Sullivan, mst, openib-general, linux-kernel, netdev
> > static void i2c_wait_for_writes(struct ipath_devdata *dd)
> > {
> > + mb();
> > (void)ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
> > }
> That's a bit weird. I wouldn't have expected the compiler to muck around
> with a readl().
I never liked this patch. The last time it came up there were
conflicting answers about whether it was a code generation bug or a
real issue talking to hardware or what. At the least I think this
merits a big comment explain what's going on -- and even better would
be really understanding the bug that's being fixed so that we're
confident it is indeed a real fix.
- R.
^ permalink raw reply [flat|nested] 65+ messages in thread
* [PATCH 29 of 39] IB/ipath - RC receive interrupt performance changes
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (27 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 30 of 39] IB/ipath - purge sps_lid and sps_mlid arrays Bryan O'Sullivan
` (10 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
This patch separates QP state used for sending and receiving
RC packets so the processing in the receive interrupt handler
can be done mostly without locks being held. ACK packets are
now sent without requiring synchronization with the send tasklet.
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 5f3c0b2d446d -r 1bef8244297a drivers/infiniband/hw/ipath/ipath_keys.c
--- a/drivers/infiniband/hw/ipath/ipath_keys.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_keys.c Thu Jun 29 14:33:26 2006 -0700
@@ -121,6 +121,7 @@ int ipath_lkey_ok(struct ipath_lkey_tabl
struct ib_sge *sge, int acc)
{
struct ipath_mregion *mr;
+ unsigned n, m;
size_t off;
int ret;
@@ -152,20 +153,22 @@ int ipath_lkey_ok(struct ipath_lkey_tabl
}
off += mr->offset;
+ m = 0;
+ n = 0;
+ while (off >= mr->map[m]->segs[n].length) {
+ off -= mr->map[m]->segs[n].length;
+ n++;
+ if (n >= IPATH_SEGSZ) {
+ m++;
+ n = 0;
+ }
+ }
isge->mr = mr;
- isge->m = 0;
- isge->n = 0;
- while (off >= mr->map[isge->m]->segs[isge->n].length) {
- off -= mr->map[isge->m]->segs[isge->n].length;
- isge->n++;
- if (isge->n >= IPATH_SEGSZ) {
- isge->m++;
- isge->n = 0;
- }
- }
- isge->vaddr = mr->map[isge->m]->segs[isge->n].vaddr + off;
- isge->length = mr->map[isge->m]->segs[isge->n].length - off;
+ isge->vaddr = mr->map[m]->segs[n].vaddr + off;
+ isge->length = mr->map[m]->segs[n].length - off;
isge->sge_length = sge->length;
+ isge->m = m;
+ isge->n = n;
ret = 1;
@@ -190,6 +193,7 @@ int ipath_rkey_ok(struct ipath_ibdev *de
struct ipath_lkey_table *rkt = &dev->lk_table;
struct ipath_sge *sge = &ss->sge;
struct ipath_mregion *mr;
+ unsigned n, m;
size_t off;
int ret;
@@ -207,20 +211,22 @@ int ipath_rkey_ok(struct ipath_ibdev *de
}
off += mr->offset;
+ m = 0;
+ n = 0;
+ while (off >= mr->map[m]->segs[n].length) {
+ off -= mr->map[m]->segs[n].length;
+ n++;
+ if (n >= IPATH_SEGSZ) {
+ m++;
+ n = 0;
+ }
+ }
sge->mr = mr;
- sge->m = 0;
- sge->n = 0;
- while (off >= mr->map[sge->m]->segs[sge->n].length) {
- off -= mr->map[sge->m]->segs[sge->n].length;
- sge->n++;
- if (sge->n >= IPATH_SEGSZ) {
- sge->m++;
- sge->n = 0;
- }
- }
- sge->vaddr = mr->map[sge->m]->segs[sge->n].vaddr + off;
- sge->length = mr->map[sge->m]->segs[sge->n].length - off;
+ sge->vaddr = mr->map[m]->segs[n].vaddr + off;
+ sge->length = mr->map[m]->segs[n].length - off;
sge->sge_length = len;
+ sge->m = m;
+ sge->n = n;
ss->sg_list = NULL;
ss->num_sge = 1;
diff -r 5f3c0b2d446d -r 1bef8244297a drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:26 2006 -0700
@@ -333,10 +333,11 @@ static void ipath_reset_qp(struct ipath_
qp->remote_qpn = 0;
qp->qkey = 0;
qp->qp_access_flags = 0;
+ clear_bit(IPATH_S_BUSY, &qp->s_flags);
qp->s_hdrwords = 0;
qp->s_psn = 0;
qp->r_psn = 0;
- atomic_set(&qp->msn, 0);
+ qp->r_msn = 0;
if (qp->ibqp.qp_type == IB_QPT_RC) {
qp->s_state = IB_OPCODE_RC_SEND_LAST;
qp->r_state = IB_OPCODE_RC_SEND_LAST;
@@ -345,7 +346,8 @@ static void ipath_reset_qp(struct ipath_
qp->r_state = IB_OPCODE_UC_SEND_LAST;
}
qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
- qp->s_nak_state = 0;
+ qp->r_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
+ qp->r_nak_state = 0;
qp->s_rnr_timeout = 0;
qp->s_head = 0;
qp->s_tail = 0;
@@ -363,10 +365,10 @@ static void ipath_reset_qp(struct ipath_
* @qp: the QP to put into an error state
*
* Flushes both send and receive work queues.
- * QP r_rq.lock and s_lock should be held.
- */
-
-static void ipath_error_qp(struct ipath_qp *qp)
+ * QP s_lock should be held and interrupts disabled.
+ */
+
+void ipath_error_qp(struct ipath_qp *qp)
{
struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
struct ib_wc wc;
@@ -409,12 +411,14 @@ static void ipath_error_qp(struct ipath_
qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
wc.opcode = IB_WC_RECV;
+ spin_lock(&qp->r_rq.lock);
while (qp->r_rq.tail != qp->r_rq.head) {
wc.wr_id = get_rwqe_ptr(&qp->r_rq, qp->r_rq.tail)->wr_id;
if (++qp->r_rq.tail >= qp->r_rq.size)
qp->r_rq.tail = 0;
ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
}
+ spin_unlock(&qp->r_rq.lock);
}
/**
@@ -434,8 +438,7 @@ int ipath_modify_qp(struct ib_qp *ibqp,
unsigned long flags;
int ret;
- spin_lock_irqsave(&qp->r_rq.lock, flags);
- spin_lock(&qp->s_lock);
+ spin_lock_irqsave(&qp->s_lock, flags);
cur_state = attr_mask & IB_QP_CUR_STATE ?
attr->cur_qp_state : qp->state;
@@ -506,31 +509,19 @@ int ipath_modify_qp(struct ib_qp *ibqp,
}
if (attr_mask & IB_QP_MIN_RNR_TIMER)
- qp->s_min_rnr_timer = attr->min_rnr_timer;
+ qp->r_min_rnr_timer = attr->min_rnr_timer;
if (attr_mask & IB_QP_QKEY)
qp->qkey = attr->qkey;
qp->state = new_state;
- spin_unlock(&qp->s_lock);
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
-
- /*
- * If QP1 changed to the RTS state, try to move to the link to INIT
- * even if it was ACTIVE so the SM will reinitialize the SMA's
- * state.
- */
- if (qp->ibqp.qp_num == 1 && new_state == IB_QPS_RTS) {
- struct ipath_ibdev *dev = to_idev(ibqp->device);
-
- ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKDOWN);
- }
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+
ret = 0;
goto bail;
inval:
- spin_unlock(&qp->s_lock);
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
+ spin_unlock_irqrestore(&qp->s_lock, flags);
ret = -EINVAL;
bail:
@@ -564,7 +555,7 @@ int ipath_query_qp(struct ib_qp *ibqp, s
attr->sq_draining = 0;
attr->max_rd_atomic = 1;
attr->max_dest_rd_atomic = 1;
- attr->min_rnr_timer = qp->s_min_rnr_timer;
+ attr->min_rnr_timer = qp->r_min_rnr_timer;
attr->port_num = 1;
attr->timeout = 0;
attr->retry_cnt = qp->s_retry_cnt;
@@ -591,16 +582,12 @@ int ipath_query_qp(struct ib_qp *ibqp, s
* @qp: the queue pair to compute the AETH for
*
* Returns the AETH.
- *
- * The QP s_lock should be held.
*/
__be32 ipath_compute_aeth(struct ipath_qp *qp)
{
- u32 aeth = atomic_read(&qp->msn) & IPS_MSN_MASK;
-
- if (qp->s_nak_state) {
- aeth |= qp->s_nak_state << IPS_AETH_CREDIT_SHIFT;
- } else if (qp->ibqp.srq) {
+ u32 aeth = qp->r_msn & IPS_MSN_MASK;
+
+ if (qp->ibqp.srq) {
/*
* Shared receive queues don't generate credits.
* Set the credit field to the invalid value.
diff -r 5f3c0b2d446d -r 1bef8244297a drivers/infiniband/hw/ipath/ipath_rc.c
--- a/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:26 2006 -0700
@@ -42,7 +42,7 @@
* @qp: the QP who's SGE we're restarting
* @wqe: the work queue to initialize the QP's SGE from
*
- * The QP s_lock should be held.
+ * The QP s_lock should be held and interrupts disabled.
*/
static void ipath_init_restart(struct ipath_qp *qp, struct ipath_swqe *wqe)
{
@@ -77,7 +77,6 @@ u32 ipath_make_rc_ack(struct ipath_qp *q
struct ipath_other_headers *ohdr,
u32 pmtu)
{
- struct ipath_sge_state *ss;
u32 hwords;
u32 len;
u32 bth0;
@@ -91,7 +90,7 @@ u32 ipath_make_rc_ack(struct ipath_qp *q
*/
switch (qp->s_ack_state) {
case OP(RDMA_READ_REQUEST):
- ss = &qp->s_rdma_sge;
+ qp->s_cur_sge = &qp->s_rdma_sge;
len = qp->s_rdma_len;
if (len > pmtu) {
len = pmtu;
@@ -108,7 +107,7 @@ u32 ipath_make_rc_ack(struct ipath_qp *q
qp->s_ack_state = OP(RDMA_READ_RESPONSE_MIDDLE);
/* FALLTHROUGH */
case OP(RDMA_READ_RESPONSE_MIDDLE):
- ss = &qp->s_rdma_sge;
+ qp->s_cur_sge = &qp->s_rdma_sge;
len = qp->s_rdma_len;
if (len > pmtu)
len = pmtu;
@@ -127,41 +126,50 @@ u32 ipath_make_rc_ack(struct ipath_qp *q
* We have to prevent new requests from changing
* the r_sge state while a ipath_verbs_send()
* is in progress.
- * Changing r_state allows the receiver
- * to continue processing new packets.
- * We do it here now instead of above so
- * that we are sure the packet was sent before
- * changing the state.
- */
- qp->r_state = OP(RDMA_READ_RESPONSE_LAST);
+ */
qp->s_ack_state = OP(ACKNOWLEDGE);
- return 0;
+ bth0 = 0;
+ goto bail;
case OP(COMPARE_SWAP):
case OP(FETCH_ADD):
- ss = NULL;
+ qp->s_cur_sge = NULL;
len = 0;
- qp->r_state = OP(SEND_LAST);
- qp->s_ack_state = OP(ACKNOWLEDGE);
- bth0 = IB_OPCODE_ATOMIC_ACKNOWLEDGE << 24;
+ /*
+ * Set the s_ack_state so the receive interrupt handler
+ * won't try to send an ACK (out of order) until this one
+ * is actually sent.
+ */
+ qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+ bth0 = OP(ATOMIC_ACKNOWLEDGE) << 24;
ohdr->u.at.aeth = ipath_compute_aeth(qp);
- ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->s_ack_atomic);
+ ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->r_atomic_data);
hwords += sizeof(ohdr->u.at) / 4;
break;
default:
/* Send a regular ACK. */
- ss = NULL;
+ qp->s_cur_sge = NULL;
len = 0;
- qp->s_ack_state = OP(ACKNOWLEDGE);
- bth0 = qp->s_ack_state << 24;
- ohdr->u.aeth = ipath_compute_aeth(qp);
+ /*
+ * Set the s_ack_state so the receive interrupt handler
+ * won't try to send an ACK (out of order) until this one
+ * is actually sent.
+ */
+ qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+ bth0 = OP(ACKNOWLEDGE) << 24;
+ if (qp->s_nak_state)
+ ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPS_MSN_MASK) |
+ (qp->s_nak_state <<
+ IPS_AETH_CREDIT_SHIFT));
+ else
+ ohdr->u.aeth = ipath_compute_aeth(qp);
hwords++;
}
qp->s_hdrwords = hwords;
- qp->s_cur_sge = ss;
qp->s_cur_size = len;
+bail:
return bth0;
}
@@ -174,7 +182,7 @@ u32 ipath_make_rc_ack(struct ipath_qp *q
* @bth2p: pointer to the BTH PSN word
*
* Return 1 if constructed; otherwise, return 0.
- * Note the QP s_lock must be held.
+ * Note the QP s_lock must be held and interrupts disabled.
*/
int ipath_make_rc_req(struct ipath_qp *qp,
struct ipath_other_headers *ohdr,
@@ -356,6 +364,11 @@ int ipath_make_rc_req(struct ipath_qp *q
bth2 |= qp->s_psn++ & IPS_PSN_MASK;
if ((int)(qp->s_psn - qp->s_next_psn) > 0)
qp->s_next_psn = qp->s_psn;
+ /*
+ * Put the QP on the pending list so lost ACKs will cause
+ * a retry. More than one request can be pending so the
+ * QP may already be on the dev->pending list.
+ */
spin_lock(&dev->pending_lock);
if (list_empty(&qp->timerwait))
list_add_tail(&qp->timerwait,
@@ -365,8 +378,8 @@ int ipath_make_rc_req(struct ipath_qp *q
case OP(RDMA_READ_RESPONSE_FIRST):
/*
- * This case can only happen if a send is restarted. See
- * ipath_restart_rc().
+ * This case can only happen if a send is restarted.
+ * See ipath_restart_rc().
*/
ipath_init_restart(qp, wqe);
/* FALLTHROUGH */
@@ -526,11 +539,17 @@ static void send_rc_ack(struct ipath_qp
ohdr = &hdr.u.l.oth;
lrh0 = IPS_LRH_GRH;
}
+ /* read pkey_index w/o lock (its atomic) */
bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
- ohdr->u.aeth = ipath_compute_aeth(qp);
- if (qp->s_ack_state >= OP(COMPARE_SWAP)) {
- bth0 |= IB_OPCODE_ATOMIC_ACKNOWLEDGE << 24;
- ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->s_ack_atomic);
+ if (qp->r_nak_state)
+ ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPS_MSN_MASK) |
+ (qp->r_nak_state <<
+ IPS_AETH_CREDIT_SHIFT));
+ else
+ ohdr->u.aeth = ipath_compute_aeth(qp);
+ if (qp->r_ack_state >= OP(COMPARE_SWAP)) {
+ bth0 |= OP(ATOMIC_ACKNOWLEDGE) << 24;
+ ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->r_atomic_data);
hwords += sizeof(ohdr->u.at.atomic_ack_eth) / 4;
} else
bth0 |= OP(ACKNOWLEDGE) << 24;
@@ -541,15 +560,36 @@ static void send_rc_ack(struct ipath_qp
hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
ohdr->bth[0] = cpu_to_be32(bth0);
ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
- ohdr->bth[2] = cpu_to_be32(qp->s_ack_psn & IPS_PSN_MASK);
+ ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPS_PSN_MASK);
/*
* If we can send the ACK, clear the ACK state.
*/
if (ipath_verbs_send(dev->dd, hwords, (u32 *) &hdr, 0, NULL) == 0) {
- qp->s_ack_state = OP(ACKNOWLEDGE);
+ qp->r_ack_state = OP(ACKNOWLEDGE);
+ dev->n_unicast_xmit++;
+ } else {
+ /*
+ * We are out of PIO buffers at the moment.
+ * Pass responsibility for sending the ACK to the
+ * send tasklet so that when a PIO buffer becomes
+ * available, the ACK is sent ahead of other outgoing
+ * packets.
+ */
dev->n_rc_qacks++;
- dev->n_unicast_xmit++;
+ spin_lock_irq(&qp->s_lock);
+ /* Don't coalesce if a RDMA read or atomic is pending. */
+ if (qp->s_ack_state == OP(ACKNOWLEDGE) ||
+ qp->s_ack_state < OP(RDMA_READ_REQUEST)) {
+ qp->s_ack_state = qp->r_ack_state;
+ qp->s_nak_state = qp->r_nak_state;
+ qp->s_ack_psn = qp->r_ack_psn;
+ qp->r_ack_state = OP(ACKNOWLEDGE);
+ }
+ spin_unlock_irq(&qp->s_lock);
+
+ /* Call ipath_do_rc_send() in another thread. */
+ tasklet_hi_schedule(&qp->s_task);
}
}
@@ -641,7 +681,7 @@ done:
* @psn: packet sequence number for the request
* @wc: the work completion request
*
- * The QP s_lock should be held.
+ * The QP s_lock should be held and interrupts disabled.
*/
void ipath_restart_rc(struct ipath_qp *qp, u32 psn, struct ib_wc *wc)
{
@@ -705,7 +745,7 @@ bail:
*
* This is called from ipath_rc_rcv_resp() to process an incoming RC ACK
* for the given QP.
- * Called at interrupt level with the QP s_lock held.
+ * Called at interrupt level with the QP s_lock held and interrupts disabled.
* Returns 1 if OK, 0 if current operation should be aborted (NAK).
*/
static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
@@ -1126,18 +1166,16 @@ static inline int ipath_rc_rcv_error(str
* Don't queue the NAK if a RDMA read, atomic, or
* NAK is pending though.
*/
- spin_lock(&qp->s_lock);
- if ((qp->s_ack_state >= OP(RDMA_READ_REQUEST) &&
- qp->s_ack_state != IB_OPCODE_ACKNOWLEDGE) ||
- qp->s_nak_state != 0) {
- spin_unlock(&qp->s_lock);
+ if (qp->s_ack_state != OP(ACKNOWLEDGE) ||
+ qp->r_nak_state != 0)
goto done;
- }
- qp->s_ack_state = OP(SEND_ONLY);
- qp->s_nak_state = IB_NAK_PSN_ERROR;
- /* Use the expected PSN. */
- qp->s_ack_psn = qp->r_psn;
- goto resched;
+ if (qp->r_ack_state < OP(COMPARE_SWAP)) {
+ qp->r_ack_state = OP(SEND_ONLY);
+ qp->r_nak_state = IB_NAK_PSN_ERROR;
+ /* Use the expected PSN. */
+ qp->r_ack_psn = qp->r_psn;
+ }
+ goto send_ack;
}
/*
@@ -1151,33 +1189,29 @@ static inline int ipath_rc_rcv_error(str
* send the earliest so that RDMA reads can be restarted at
* the requester's expected PSN.
*/
- spin_lock(&qp->s_lock);
- if (qp->s_ack_state != IB_OPCODE_ACKNOWLEDGE &&
- ipath_cmp24(psn, qp->s_ack_psn) >= 0) {
- if (qp->s_ack_state < IB_OPCODE_RDMA_READ_REQUEST)
- qp->s_ack_psn = psn;
- spin_unlock(&qp->s_lock);
- goto done;
- }
- switch (opcode) {
- case OP(RDMA_READ_REQUEST):
- /*
- * We have to be careful to not change s_rdma_sge
- * while ipath_do_rc_send() is using it and not
- * holding the s_lock.
- */
- if (qp->s_ack_state != OP(ACKNOWLEDGE) &&
- qp->s_ack_state >= IB_OPCODE_RDMA_READ_REQUEST) {
- spin_unlock(&qp->s_lock);
- dev->n_rdma_dup_busy++;
- goto done;
- }
+ if (opcode == OP(RDMA_READ_REQUEST)) {
/* RETH comes after BTH */
if (!header_in_data)
reth = &ohdr->u.rc.reth;
else {
reth = (struct ib_reth *)data;
data += sizeof(*reth);
+ }
+ /*
+ * If we receive a duplicate RDMA request, it means the
+ * requester saw a sequence error and needs to restart
+ * from an earlier point. We can abort the current
+ * RDMA read send in that case.
+ */
+ spin_lock_irq(&qp->s_lock);
+ if (qp->s_ack_state != OP(ACKNOWLEDGE) &&
+ (qp->s_hdrwords || ipath_cmp24(psn, qp->s_ack_psn) >= 0)) {
+ /*
+ * We are already sending earlier requested data.
+ * Don't abort it to send later out of sequence data.
+ */
+ spin_unlock_irq(&qp->s_lock);
+ goto done;
}
qp->s_rdma_len = be32_to_cpu(reth->length);
if (qp->s_rdma_len != 0) {
@@ -1192,8 +1226,10 @@ static inline int ipath_rc_rcv_error(str
ok = ipath_rkey_ok(dev, &qp->s_rdma_sge,
qp->s_rdma_len, vaddr, rkey,
IB_ACCESS_REMOTE_READ);
- if (unlikely(!ok))
+ if (unlikely(!ok)) {
+ spin_unlock_irq(&qp->s_lock);
goto done;
+ }
} else {
qp->s_rdma_sge.sg_list = NULL;
qp->s_rdma_sge.num_sge = 0;
@@ -1202,25 +1238,44 @@ static inline int ipath_rc_rcv_error(str
qp->s_rdma_sge.sge.length = 0;
qp->s_rdma_sge.sge.sge_length = 0;
}
- break;
-
+ qp->s_ack_state = opcode;
+ qp->s_ack_psn = psn;
+ spin_unlock_irq(&qp->s_lock);
+ tasklet_hi_schedule(&qp->s_task);
+ goto send_ack;
+ }
+
+ /*
+ * A pending RDMA read will ACK anything before it so
+ * ignore earlier duplicate requests.
+ */
+ if (qp->s_ack_state != OP(ACKNOWLEDGE))
+ goto done;
+
+ /*
+ * If an ACK is pending, don't replace the pending ACK
+ * with an earlier one since the later one will ACK the earlier.
+ * Also, if we already have a pending atomic, send it.
+ */
+ if (qp->r_ack_state != OP(ACKNOWLEDGE) &&
+ (ipath_cmp24(psn, qp->r_ack_psn) <= 0 ||
+ qp->r_ack_state >= OP(COMPARE_SWAP)))
+ goto send_ack;
+ switch (opcode) {
case OP(COMPARE_SWAP):
case OP(FETCH_ADD):
/*
* Check for the PSN of the last atomic operation
* performed and resend the result if found.
*/
- if ((psn & IPS_PSN_MASK) != qp->r_atomic_psn) {
- spin_unlock(&qp->s_lock);
+ if ((psn & IPS_PSN_MASK) != qp->r_atomic_psn)
goto done;
- }
- qp->s_ack_atomic = qp->r_atomic_data;
break;
}
- qp->s_ack_state = opcode;
- qp->s_nak_state = 0;
- qp->s_ack_psn = psn;
-resched:
+ qp->r_ack_state = opcode;
+ qp->r_nak_state = 0;
+ qp->r_ack_psn = psn;
+send_ack:
return 0;
done:
@@ -1248,7 +1303,6 @@ void ipath_rc_rcv(struct ipath_ibdev *de
u32 hdrsize;
u32 psn;
u32 pad;
- unsigned long flags;
struct ib_wc wc;
u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
int diff;
@@ -1289,10 +1343,8 @@ void ipath_rc_rcv(struct ipath_ibdev *de
opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
ipath_rc_rcv_resp(dev, ohdr, data, tlen, qp, opcode, psn,
hdrsize, pmtu, header_in_data);
- goto bail;
- }
-
- spin_lock_irqsave(&qp->r_rq.lock, flags);
+ goto done;
+ }
/* Compute 24 bits worth of difference. */
diff = ipath_cmp24(psn, qp->r_psn);
@@ -1300,7 +1352,7 @@ void ipath_rc_rcv(struct ipath_ibdev *de
if (ipath_rc_rcv_error(dev, ohdr, data, qp, opcode,
psn, diff, header_in_data))
goto done;
- goto resched;
+ goto send_ack;
}
/* Check for opcode sequence errors. */
@@ -1312,22 +1364,19 @@ void ipath_rc_rcv(struct ipath_ibdev *de
opcode == OP(SEND_LAST_WITH_IMMEDIATE))
break;
nack_inv:
- /*
- * A NAK will ACK earlier sends and RDMA writes. Don't queue the
- * NAK if a RDMA read, atomic, or NAK is pending though.
- */
- spin_lock(&qp->s_lock);
- if (qp->s_ack_state >= OP(RDMA_READ_REQUEST) &&
- qp->s_ack_state != IB_OPCODE_ACKNOWLEDGE) {
- spin_unlock(&qp->s_lock);
- goto done;
- }
- /* XXX Flush WQEs */
- qp->state = IB_QPS_ERR;
- qp->s_ack_state = OP(SEND_ONLY);
- qp->s_nak_state = IB_NAK_INVALID_REQUEST;
- qp->s_ack_psn = qp->r_psn;
- goto resched;
+ /*
+ * A NAK will ACK earlier sends and RDMA writes.
+ * Don't queue the NAK if a RDMA read, atomic, or NAK
+ * is pending though.
+ */
+ if (qp->r_ack_state >= OP(COMPARE_SWAP))
+ goto send_ack;
+ /* XXX Flush WQEs */
+ qp->state = IB_QPS_ERR;
+ qp->r_ack_state = OP(SEND_ONLY);
+ qp->r_nak_state = IB_NAK_INVALID_REQUEST;
+ qp->r_ack_psn = qp->r_psn;
+ goto send_ack;
case OP(RDMA_WRITE_FIRST):
case OP(RDMA_WRITE_MIDDLE):
@@ -1336,20 +1385,6 @@ void ipath_rc_rcv(struct ipath_ibdev *de
opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
break;
goto nack_inv;
-
- case OP(RDMA_READ_REQUEST):
- case OP(COMPARE_SWAP):
- case OP(FETCH_ADD):
- /*
- * Drop all new requests until a response has been sent. A
- * new request then ACKs the RDMA response we sent. Relaxed
- * ordering would allow new requests to be processed but we
- * would need to keep a queue of rwqe's for all that are in
- * progress. Note that we can't RNR NAK this request since
- * the RDMA READ or atomic response is already queued to be
- * sent (unless we implement a response send queue).
- */
- goto done;
default:
if (opcode == OP(SEND_MIDDLE) ||
@@ -1359,6 +1394,11 @@ void ipath_rc_rcv(struct ipath_ibdev *de
opcode == OP(RDMA_WRITE_LAST) ||
opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
goto nack_inv;
+ /*
+ * Note that it is up to the requester to not send a new
+ * RDMA read or atomic operation before receiving an ACK
+ * for the previous operation.
+ */
break;
}
@@ -1375,17 +1415,12 @@ void ipath_rc_rcv(struct ipath_ibdev *de
* Don't queue the NAK if a RDMA read or atomic
* is pending though.
*/
- spin_lock(&qp->s_lock);
- if (qp->s_ack_state >=
- OP(RDMA_READ_REQUEST) &&
- qp->s_ack_state != IB_OPCODE_ACKNOWLEDGE) {
- spin_unlock(&qp->s_lock);
- goto done;
- }
- qp->s_ack_state = OP(SEND_ONLY);
- qp->s_nak_state = IB_RNR_NAK | qp->s_min_rnr_timer;
- qp->s_ack_psn = qp->r_psn;
- goto resched;
+ if (qp->r_ack_state >= OP(COMPARE_SWAP))
+ goto send_ack;
+ qp->r_ack_state = OP(SEND_ONLY);
+ qp->r_nak_state = IB_RNR_NAK | qp->r_min_rnr_timer;
+ qp->r_ack_psn = qp->r_psn;
+ goto send_ack;
}
qp->r_rcv_len = 0;
/* FALLTHROUGH */
@@ -1442,7 +1477,7 @@ void ipath_rc_rcv(struct ipath_ibdev *de
if (unlikely(wc.byte_len > qp->r_len))
goto nack_inv;
ipath_copy_sge(&qp->r_sge, data, tlen);
- atomic_inc(&qp->msn);
+ qp->r_msn++;
if (opcode == OP(RDMA_WRITE_LAST) ||
opcode == OP(RDMA_WRITE_ONLY))
break;
@@ -1486,29 +1521,8 @@ void ipath_rc_rcv(struct ipath_ibdev *de
ok = ipath_rkey_ok(dev, &qp->r_sge,
qp->r_len, vaddr, rkey,
IB_ACCESS_REMOTE_WRITE);
- if (unlikely(!ok)) {
- nack_acc:
- /*
- * A NAK will ACK earlier sends and RDMA
- * writes. Don't queue the NAK if a RDMA
- * read, atomic, or NAK is pending though.
- */
- spin_lock(&qp->s_lock);
- if (qp->s_ack_state >=
- OP(RDMA_READ_REQUEST) &&
- qp->s_ack_state !=
- IB_OPCODE_ACKNOWLEDGE) {
- spin_unlock(&qp->s_lock);
- goto done;
- }
- /* XXX Flush WQEs */
- qp->state = IB_QPS_ERR;
- qp->s_ack_state = OP(RDMA_WRITE_ONLY);
- qp->s_nak_state =
- IB_NAK_REMOTE_ACCESS_ERROR;
- qp->s_ack_psn = qp->r_psn;
- goto resched;
- }
+ if (unlikely(!ok))
+ goto nack_acc;
} else {
qp->r_sge.sg_list = NULL;
qp->r_sge.sge.mr = NULL;
@@ -1535,12 +1549,10 @@ void ipath_rc_rcv(struct ipath_ibdev *de
reth = (struct ib_reth *)data;
data += sizeof(*reth);
}
- spin_lock(&qp->s_lock);
- if (qp->s_ack_state != OP(ACKNOWLEDGE) &&
- qp->s_ack_state >= IB_OPCODE_RDMA_READ_REQUEST) {
- spin_unlock(&qp->s_lock);
- goto done;
- }
+ if (unlikely(!(qp->qp_access_flags &
+ IB_ACCESS_REMOTE_READ)))
+ goto nack_acc;
+ spin_lock_irq(&qp->s_lock);
qp->s_rdma_len = be32_to_cpu(reth->length);
if (qp->s_rdma_len != 0) {
u32 rkey = be32_to_cpu(reth->rkey);
@@ -1552,7 +1564,7 @@ void ipath_rc_rcv(struct ipath_ibdev *de
qp->s_rdma_len, vaddr, rkey,
IB_ACCESS_REMOTE_READ);
if (unlikely(!ok)) {
- spin_unlock(&qp->s_lock);
+ spin_unlock_irq(&qp->s_lock);
goto nack_acc;
}
/*
@@ -1569,21 +1581,25 @@ void ipath_rc_rcv(struct ipath_ibdev *de
qp->s_rdma_sge.sge.length = 0;
qp->s_rdma_sge.sge.sge_length = 0;
}
- if (unlikely(!(qp->qp_access_flags &
- IB_ACCESS_REMOTE_READ)))
- goto nack_acc;
/*
* We need to increment the MSN here instead of when we
* finish sending the result since a duplicate request would
* increment it more than once.
*/
- atomic_inc(&qp->msn);
+ qp->r_msn++;
+
qp->s_ack_state = opcode;
- qp->s_nak_state = 0;
qp->s_ack_psn = psn;
+ spin_unlock_irq(&qp->s_lock);
+
qp->r_psn++;
qp->r_state = opcode;
- goto rdmadone;
+ qp->r_nak_state = 0;
+
+ /* Call ipath_do_rc_send() in another thread. */
+ tasklet_hi_schedule(&qp->s_task);
+
+ goto done;
case OP(COMPARE_SWAP):
case OP(FETCH_ADD): {
@@ -1612,7 +1628,7 @@ void ipath_rc_rcv(struct ipath_ibdev *de
goto nack_acc;
/* Perform atomic OP and save result. */
sdata = be64_to_cpu(ateth->swap_data);
- spin_lock(&dev->pending_lock);
+ spin_lock_irq(&dev->pending_lock);
qp->r_atomic_data = *(u64 *) qp->r_sge.sge.vaddr;
if (opcode == OP(FETCH_ADD))
*(u64 *) qp->r_sge.sge.vaddr =
@@ -1620,8 +1636,8 @@ void ipath_rc_rcv(struct ipath_ibdev *de
else if (qp->r_atomic_data ==
be64_to_cpu(ateth->compare_data))
*(u64 *) qp->r_sge.sge.vaddr = sdata;
- spin_unlock(&dev->pending_lock);
- atomic_inc(&qp->msn);
+ spin_unlock_irq(&dev->pending_lock);
+ qp->r_msn++;
qp->r_atomic_psn = psn & IPS_PSN_MASK;
psn |= 1 << 31;
break;
@@ -1633,44 +1649,39 @@ void ipath_rc_rcv(struct ipath_ibdev *de
}
qp->r_psn++;
qp->r_state = opcode;
+ qp->r_nak_state = 0;
/* Send an ACK if requested or required. */
if (psn & (1 << 31)) {
/*
* Coalesce ACKs unless there is a RDMA READ or
* ATOMIC pending.
*/
- spin_lock(&qp->s_lock);
- if (qp->s_ack_state == OP(ACKNOWLEDGE) ||
- qp->s_ack_state < IB_OPCODE_RDMA_READ_REQUEST) {
- qp->s_ack_state = opcode;
- qp->s_nak_state = 0;
- qp->s_ack_psn = psn;
- qp->s_ack_atomic = qp->r_atomic_data;
- goto resched;
- }
- spin_unlock(&qp->s_lock);
- }
+ if (qp->r_ack_state < OP(COMPARE_SWAP)) {
+ qp->r_ack_state = opcode;
+ qp->r_ack_psn = psn;
+ }
+ goto send_ack;
+ }
+ goto done;
+
+nack_acc:
+ /*
+ * A NAK will ACK earlier sends and RDMA writes.
+ * Don't queue the NAK if a RDMA read, atomic, or NAK
+ * is pending though.
+ */
+ if (qp->r_ack_state < OP(COMPARE_SWAP)) {
+ /* XXX Flush WQEs */
+ qp->state = IB_QPS_ERR;
+ qp->r_ack_state = OP(RDMA_WRITE_ONLY);
+ qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
+ qp->r_ack_psn = qp->r_psn;
+ }
+send_ack:
+ /* Send ACK right away unless the send tasklet has a pending ACK. */
+ if (qp->s_ack_state == OP(ACKNOWLEDGE))
+ send_rc_ack(qp);
+
done:
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
- goto bail;
-
-resched:
- /*
- * Try to send ACK right away but not if ipath_do_rc_send() is
- * active.
- */
- if (qp->s_hdrwords == 0 &&
- (qp->s_ack_state < IB_OPCODE_RDMA_READ_REQUEST ||
- qp->s_ack_state >= IB_OPCODE_COMPARE_SWAP))
- send_rc_ack(qp);
-
-rdmadone:
- spin_unlock(&qp->s_lock);
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
-
- /* Call ipath_do_rc_send() in another thread. */
- tasklet_hi_schedule(&qp->s_task);
-
-bail:
return;
}
diff -r 5f3c0b2d446d -r 1bef8244297a drivers/infiniband/hw/ipath/ipath_ruc.c
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:26 2006 -0700
@@ -113,20 +113,23 @@ void ipath_insert_rnr_queue(struct ipath
*
* Return 0 if no RWQE is available, otherwise return 1.
*
- * Called at interrupt level with the QP r_rq.lock held.
+ * Can be called from interrupt level.
*/
int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
{
+ unsigned long flags;
struct ipath_rq *rq;
struct ipath_srq *srq;
struct ipath_rwqe *wqe;
- int ret;
+ int ret = 1;
if (!qp->ibqp.srq) {
rq = &qp->r_rq;
+ spin_lock_irqsave(&rq->lock, flags);
+
if (unlikely(rq->tail == rq->head)) {
ret = 0;
- goto bail;
+ goto done;
}
wqe = get_rwqe_ptr(rq, rq->tail);
qp->r_wr_id = wqe->wr_id;
@@ -138,17 +141,16 @@ int ipath_get_rwqe(struct ipath_qp *qp,
}
if (++rq->tail >= rq->size)
rq->tail = 0;
- ret = 1;
- goto bail;
+ goto done;
}
srq = to_isrq(qp->ibqp.srq);
rq = &srq->rq;
- spin_lock(&rq->lock);
+ spin_lock_irqsave(&rq->lock, flags);
+
if (unlikely(rq->tail == rq->head)) {
- spin_unlock(&rq->lock);
ret = 0;
- goto bail;
+ goto done;
}
wqe = get_rwqe_ptr(rq, rq->tail);
qp->r_wr_id = wqe->wr_id;
@@ -170,18 +172,18 @@ int ipath_get_rwqe(struct ipath_qp *qp,
n = rq->head - rq->tail;
if (n < srq->limit) {
srq->limit = 0;
- spin_unlock(&rq->lock);
+ spin_unlock_irqrestore(&rq->lock, flags);
ev.device = qp->ibqp.device;
ev.element.srq = qp->ibqp.srq;
ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
srq->ibsrq.event_handler(&ev,
srq->ibsrq.srq_context);
- } else
- spin_unlock(&rq->lock);
- } else
- spin_unlock(&rq->lock);
- ret = 1;
-
+ goto bail;
+ }
+ }
+
+done:
+ spin_unlock_irqrestore(&rq->lock, flags);
bail:
return ret;
}
@@ -248,10 +250,8 @@ again:
wc.imm_data = wqe->wr.imm_data;
/* FALLTHROUGH */
case IB_WR_SEND:
- spin_lock_irqsave(&qp->r_rq.lock, flags);
if (!ipath_get_rwqe(qp, 0)) {
rnr_nak:
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
/* Handle RNR NAK */
if (qp->ibqp.qp_type == IB_QPT_UC)
goto send_comp;
@@ -263,20 +263,17 @@ again:
sqp->s_rnr_retry--;
dev->n_rnr_naks++;
sqp->s_rnr_timeout =
- ib_ipath_rnr_table[sqp->s_min_rnr_timer];
+ ib_ipath_rnr_table[sqp->r_min_rnr_timer];
ipath_insert_rnr_queue(sqp);
goto done;
}
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
break;
case IB_WR_RDMA_WRITE_WITH_IMM:
wc.wc_flags = IB_WC_WITH_IMM;
wc.imm_data = wqe->wr.imm_data;
- spin_lock_irqsave(&qp->r_rq.lock, flags);
if (!ipath_get_rwqe(qp, 1))
goto rnr_nak;
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
/* FALLTHROUGH */
case IB_WR_RDMA_WRITE:
if (wqe->length == 0)
diff -r 5f3c0b2d446d -r 1bef8244297a drivers/infiniband/hw/ipath/ipath_uc.c
--- a/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:26 2006 -0700
@@ -241,7 +241,6 @@ void ipath_uc_rcv(struct ipath_ibdev *de
u32 hdrsize;
u32 psn;
u32 pad;
- unsigned long flags;
struct ib_wc wc;
u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
struct ib_reth *reth;
@@ -279,8 +278,6 @@ void ipath_uc_rcv(struct ipath_ibdev *de
wc.imm_data = 0;
wc.wc_flags = 0;
- spin_lock_irqsave(&qp->r_rq.lock, flags);
-
/* Compare the PSN verses the expected PSN. */
if (unlikely(ipath_cmp24(psn, qp->r_psn) != 0)) {
/*
@@ -537,15 +534,11 @@ void ipath_uc_rcv(struct ipath_ibdev *de
default:
/* Drop packet for unknown opcodes. */
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
dev->n_pkt_drops++;
- goto bail;
+ goto done;
}
qp->r_psn++;
qp->r_state = opcode;
done:
- spin_unlock_irqrestore(&qp->r_rq.lock, flags);
-
-bail:
return;
}
diff -r 5f3c0b2d446d -r 1bef8244297a drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:26 2006 -0700
@@ -307,32 +307,34 @@ struct ipath_qp {
u32 s_next_psn; /* PSN for next request */
u32 s_last_psn; /* last response PSN processed */
u32 s_psn; /* current packet sequence number */
+ u32 s_ack_psn; /* PSN for RDMA_READ */
u32 s_rnr_timeout; /* number of milliseconds for RNR timeout */
- u32 s_ack_psn; /* PSN for next ACK or RDMA_READ */
- u64 s_ack_atomic; /* data for atomic ACK */
+ u32 r_ack_psn; /* PSN for next ACK or atomic ACK */
u64 r_wr_id; /* ID for current receive WQE */
u64 r_atomic_data; /* data for last atomic op */
u32 r_atomic_psn; /* PSN of last atomic op */
u32 r_len; /* total length of r_sge */
u32 r_rcv_len; /* receive data len processed */
u32 r_psn; /* expected rcv packet sequence number */
+ u32 r_msn; /* message sequence number */
u8 state; /* QP state */
u8 s_state; /* opcode of last packet sent */
u8 s_ack_state; /* opcode of packet to ACK */
u8 s_nak_state; /* non-zero if NAK is pending */
u8 r_state; /* opcode of last packet received */
+ u8 r_ack_state; /* opcode of packet to ACK */
+ u8 r_nak_state; /* non-zero if NAK is pending */
+ u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */
u8 r_reuse_sge; /* for UC receive errors */
u8 r_sge_inx; /* current index into sg_list */
+ u8 qp_access_flags;
u8 s_max_sge; /* size of s_wq->sg_list */
- u8 qp_access_flags;
u8 s_retry_cnt; /* number of times to retry */
u8 s_rnr_retry_cnt;
- u8 s_min_rnr_timer;
u8 s_retry; /* requester retry counter */
u8 s_rnr_retry; /* requester RNR retry counter */
u8 s_pkey_index; /* PKEY index to use */
enum ib_mtu path_mtu;
- atomic_t msn; /* message sequence number */
u32 remote_qpn;
u32 qkey; /* QKEY for this QP (for UD or RD) */
u32 s_size; /* send work queue size */
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 30 of 39] IB/ipath - purge sps_lid and sps_mlid arrays
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (28 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 29 of 39] IB/ipath - RC receive interrupt performance changes Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 31 of 39] IB/ipath - drop the "stats" sysfs attribute group Bryan O'Sullivan
` (9 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
The two arrays only had space for 4 units.
Also changed from ipath_set_sps_lid() to ipath_set_lid(); the sps
was leftover.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 1bef8244297a -r 3ceb73f8bde0 drivers/infiniband/hw/ipath/ipath_common.h
--- a/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:26 2006 -0700
@@ -122,8 +122,7 @@ struct infinipath_stats {
__u64 sps_ports;
/* list of pkeys (other than default) accepted (0 means not set) */
__u16 sps_pkeys[4];
- /* lids for up to 4 infinipaths, indexed by infinipath # */
- __u16 sps_lid[4];
+ __u16 sps_unused16[4]; /* available; maintaining compatible layout */
/* number of user ports per chip (not IB ports) */
__u32 sps_nports;
/* not our interrupt, or already handled */
@@ -141,10 +140,8 @@ struct infinipath_stats {
* packets if ipath not configured, sma/mad, etc.)
*/
__u64 sps_krdrops;
- /* mlids for up to 4 infinipaths, indexed by infinipath # */
- __u16 sps_mlid[4];
/* pad for future growth */
- __u64 __sps_pad[45];
+ __u64 __sps_pad[46];
};
/*
diff -r 1bef8244297a -r 3ceb73f8bde0 drivers/infiniband/hw/ipath/ipath_init_chip.c
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:26 2006 -0700
@@ -811,8 +811,6 @@ int ipath_init_chip(struct ipath_devdata
/* clear any interrups up to this point (ints still not enabled) */
ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, -1LL);
- ipath_stats.sps_lid[dd->ipath_unit] = dd->ipath_lid;
-
/*
* Set up the port 0 (kernel) rcvhdr q and egr TIDs. If doing
* re-init, the simplest way to handle this is to free
diff -r 1bef8244297a -r 3ceb73f8bde0 drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:26 2006 -0700
@@ -300,9 +300,8 @@ bail:
EXPORT_SYMBOL_GPL(ipath_layer_set_mtu);
-int ipath_set_sps_lid(struct ipath_devdata *dd, u32 arg, u8 lmc)
-{
- ipath_stats.sps_lid[dd->ipath_unit] = arg;
+int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc)
+{
dd->ipath_lid = arg;
dd->ipath_lmc = lmc;
@@ -316,7 +315,7 @@ int ipath_set_sps_lid(struct ipath_devda
return 0;
}
-EXPORT_SYMBOL_GPL(ipath_set_sps_lid);
+EXPORT_SYMBOL_GPL(ipath_set_lid);
int ipath_layer_set_guid(struct ipath_devdata *dd, __be64 guid)
{
@@ -632,9 +631,9 @@ int ipath_layer_open(struct ipath_devdat
if (*dd->ipath_statusp & IPATH_STATUS_IB_READY)
intval |= IPATH_LAYER_INT_IF_UP;
- if (ipath_stats.sps_lid[dd->ipath_unit])
+ if (dd->ipath_lid)
intval |= IPATH_LAYER_INT_LID;
- if (ipath_stats.sps_mlid[dd->ipath_unit])
+ if (dd->ipath_mlid)
intval |= IPATH_LAYER_INT_BCAST;
/*
* do this on open, in case low level is already up and
diff -r 1bef8244297a -r 3ceb73f8bde0 drivers/infiniband/hw/ipath/ipath_layer.h
--- a/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.h Thu Jun 29 14:33:26 2006 -0700
@@ -129,7 +129,7 @@ u32 ipath_layer_get_cr_errpkey(struct ip
u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd);
int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 state);
int ipath_layer_set_mtu(struct ipath_devdata *, u16);
-int ipath_set_sps_lid(struct ipath_devdata *, u32, u8);
+int ipath_set_lid(struct ipath_devdata *, u32, u8);
int ipath_layer_send_hdr(struct ipath_devdata *dd,
struct ether_header *hdr);
int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords,
diff -r 1bef8244297a -r 3ceb73f8bde0 drivers/infiniband/hw/ipath/ipath_mad.c
--- a/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:26 2006 -0700
@@ -308,7 +308,7 @@ static int recv_subn_set_portinfo(struct
/* Must be a valid unicast LID address. */
if (lid == 0 || lid >= IPS_MULTICAST_LID_BASE)
goto err;
- ipath_set_sps_lid(dev->dd, lid, pip->mkeyprot_resv_lmc & 7);
+ ipath_set_lid(dev->dd, lid, pip->mkeyprot_resv_lmc & 7);
event.event = IB_EVENT_LID_CHANGE;
ib_dispatch_event(&event);
}
diff -r 1bef8244297a -r 3ceb73f8bde0 drivers/infiniband/hw/ipath/ipath_sysfs.c
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
@@ -115,11 +115,6 @@ DRIVER_STAT(pkey1, pkeys[1]);
DRIVER_STAT(pkey1, pkeys[1]);
DRIVER_STAT(pkey2, pkeys[2]);
DRIVER_STAT(pkey3, pkeys[3]);
-/* XXX fix the following when dynamic table of devices used */
-DRIVER_STAT(lid0, lid[0]);
-DRIVER_STAT(lid1, lid[1]);
-DRIVER_STAT(lid2, lid[2]);
-DRIVER_STAT(lid3, lid[3]);
DRIVER_STAT(nports, nports);
DRIVER_STAT(null_intr, nullintr);
@@ -128,11 +123,6 @@ DRIVER_STAT(page_locks, pagelocks);
DRIVER_STAT(page_locks, pagelocks);
DRIVER_STAT(page_unlocks, pageunlocks);
DRIVER_STAT(krdrops, krdrops);
-/* XXX fix the following when dynamic table of devices used */
-DRIVER_STAT(mlid0, mlid[0]);
-DRIVER_STAT(mlid1, mlid[1]);
-DRIVER_STAT(mlid2, mlid[2]);
-DRIVER_STAT(mlid3, mlid[3]);
static struct attribute *driver_stat_attributes[] = {
&driver_attr_intrs.attr,
@@ -155,10 +145,6 @@ static struct attribute *driver_stat_att
&driver_attr_pkey1.attr,
&driver_attr_pkey2.attr,
&driver_attr_pkey3.attr,
- &driver_attr_lid0.attr,
- &driver_attr_lid1.attr,
- &driver_attr_lid2.attr,
- &driver_attr_lid3.attr,
&driver_attr_nports.attr,
&driver_attr_null_intr.attr,
&driver_attr_max_pkts_call.attr,
@@ -166,10 +152,6 @@ static struct attribute *driver_stat_att
&driver_attr_page_locks.attr,
&driver_attr_page_unlocks.attr,
&driver_attr_krdrops.attr,
- &driver_attr_mlid0.attr,
- &driver_attr_mlid1.attr,
- &driver_attr_mlid2.attr,
- &driver_attr_mlid3.attr,
NULL
};
@@ -273,7 +255,7 @@ static ssize_t store_lid(struct device *
size_t count)
{
struct ipath_devdata *dd = dev_get_drvdata(dev);
- u16 lid;
+ u16 lid = 0;
int ret;
ret = ipath_parse_ushort(buf, &lid);
@@ -285,11 +267,11 @@ static ssize_t store_lid(struct device *
goto invalid;
}
- ipath_set_sps_lid(dd, lid, 0);
+ ipath_set_lid(dd, lid, 0);
goto bail;
invalid:
- ipath_dev_err(dd, "attempt to set invalid LID\n");
+ ipath_dev_err(dd, "attempt to set invalid LID 0x%x\n", lid);
bail:
return ret;
}
@@ -320,7 +302,6 @@ static ssize_t store_mlid(struct device
unit = dd->ipath_unit;
dd->ipath_mlid = mlid;
- ipath_stats.sps_mlid[unit] = mlid;
ipath_layer_intr(dd, IPATH_LAYER_INT_BCAST);
goto bail;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 31 of 39] IB/ipath - drop the "stats" sysfs attribute group
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (29 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 30 of 39] IB/ipath - purge sps_lid and sps_mlid arrays Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 32 of 39] IB/ipath - support more models of InfiniPath hardware Bryan O'Sullivan
` (8 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
This attribute group made it into the original driver, but should
not have.
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 3ceb73f8bde0 -r 21378f21e091 drivers/infiniband/hw/ipath/ipath_sysfs.c
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
@@ -84,81 +84,6 @@ static ssize_t show_num_units(struct dev
return scnprintf(buf, PAGE_SIZE, "%d\n",
ipath_count_units(NULL, NULL, NULL));
}
-
-#define DRIVER_STAT(name, attr) \
- static ssize_t show_stat_##name(struct device_driver *dev, \
- char *buf) \
- { \
- return scnprintf( \
- buf, PAGE_SIZE, "%llu\n", \
- (unsigned long long) ipath_stats.sps_ ##attr); \
- } \
- static DRIVER_ATTR(name, S_IRUGO, show_stat_##name, NULL)
-
-DRIVER_STAT(intrs, ints);
-DRIVER_STAT(err_intrs, errints);
-DRIVER_STAT(errs, errs);
-DRIVER_STAT(pkt_errs, pkterrs);
-DRIVER_STAT(crc_errs, crcerrs);
-DRIVER_STAT(hw_errs, hwerrs);
-DRIVER_STAT(ib_link, iblink);
-DRIVER_STAT(port0_pkts, port0pkts);
-DRIVER_STAT(ether_spkts, ether_spkts);
-DRIVER_STAT(ether_rpkts, ether_rpkts);
-DRIVER_STAT(sma_spkts, sma_spkts);
-DRIVER_STAT(sma_rpkts, sma_rpkts);
-DRIVER_STAT(hdrq_full, hdrqfull);
-DRIVER_STAT(etid_full, etidfull);
-DRIVER_STAT(no_piobufs, nopiobufs);
-DRIVER_STAT(ports, ports);
-DRIVER_STAT(pkey0, pkeys[0]);
-DRIVER_STAT(pkey1, pkeys[1]);
-DRIVER_STAT(pkey2, pkeys[2]);
-DRIVER_STAT(pkey3, pkeys[3]);
-
-DRIVER_STAT(nports, nports);
-DRIVER_STAT(null_intr, nullintr);
-DRIVER_STAT(max_pkts_call, maxpkts_call);
-DRIVER_STAT(avg_pkts_call, avgpkts_call);
-DRIVER_STAT(page_locks, pagelocks);
-DRIVER_STAT(page_unlocks, pageunlocks);
-DRIVER_STAT(krdrops, krdrops);
-
-static struct attribute *driver_stat_attributes[] = {
- &driver_attr_intrs.attr,
- &driver_attr_err_intrs.attr,
- &driver_attr_errs.attr,
- &driver_attr_pkt_errs.attr,
- &driver_attr_crc_errs.attr,
- &driver_attr_hw_errs.attr,
- &driver_attr_ib_link.attr,
- &driver_attr_port0_pkts.attr,
- &driver_attr_ether_spkts.attr,
- &driver_attr_ether_rpkts.attr,
- &driver_attr_sma_spkts.attr,
- &driver_attr_sma_rpkts.attr,
- &driver_attr_hdrq_full.attr,
- &driver_attr_etid_full.attr,
- &driver_attr_no_piobufs.attr,
- &driver_attr_ports.attr,
- &driver_attr_pkey0.attr,
- &driver_attr_pkey1.attr,
- &driver_attr_pkey2.attr,
- &driver_attr_pkey3.attr,
- &driver_attr_nports.attr,
- &driver_attr_null_intr.attr,
- &driver_attr_max_pkts_call.attr,
- &driver_attr_avg_pkts_call.attr,
- &driver_attr_page_locks.attr,
- &driver_attr_page_unlocks.attr,
- &driver_attr_krdrops.attr,
- NULL
-};
-
-static struct attribute_group driver_stat_attr_group = {
- .name = "stats",
- .attrs = driver_stat_attributes
-};
static ssize_t show_status(struct device *dev,
struct device_attribute *attr,
@@ -716,20 +641,12 @@ int ipath_driver_create_group(struct dev
int ret;
ret = sysfs_create_group(&drv->kobj, &driver_attr_group);
- if (ret)
- goto bail;
-
- ret = sysfs_create_group(&drv->kobj, &driver_stat_attr_group);
- if (ret)
- sysfs_remove_group(&drv->kobj, &driver_attr_group);
-
-bail:
+
return ret;
}
void ipath_driver_remove_group(struct device_driver *drv)
{
- sysfs_remove_group(&drv->kobj, &driver_stat_attr_group);
sysfs_remove_group(&drv->kobj, &driver_attr_group);
}
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 32 of 39] IB/ipath - support more models of InfiniPath hardware
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (30 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 31 of 39] IB/ipath - drop the "stats" sysfs attribute group Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 33 of 39] IB/ipath - read/write correct sizes through diag interface Bryan O'Sullivan
` (7 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
We do a few more explicit checks for specific models, and now also
support the old PathScale serial number style, or new QLogic style.
This is backwards compatible with previous versions of software and
hardware. That is, older software will see a plausible serial number
and correct GUID when used with a new board, while newer software will
correctly handle an older board.
Signed-off-by: Mike Albaugh <mike.albaugh@qlogic.com>
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 21378f21e091 -r 8fbb5d71823a drivers/infiniband/hw/ipath/ipath_common.h
--- a/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:26 2006 -0700
@@ -476,7 +476,7 @@ struct ipath_sma_pkt
* Data layout in I2C flash (for GUID, etc.)
* All fields are little-endian binary unless otherwise stated
*/
-#define IPATH_FLASH_VERSION 1
+#define IPATH_FLASH_VERSION 2
struct ipath_flash {
/* flash layout version (IPATH_FLASH_VERSION) */
__u8 if_fversion;
@@ -484,14 +484,14 @@ struct ipath_flash {
__u8 if_csum;
/*
* valid length (in use, protected by if_csum), including
- * if_fversion and if_sum themselves)
+ * if_fversion and if_csum themselves)
*/
__u8 if_length;
/* the GUID, in network order */
__u8 if_guid[8];
/* number of GUIDs to use, starting from if_guid */
__u8 if_numguid;
- /* the board serial number, in ASCII */
+ /* the (last 10 characters of) board serial number, in ASCII */
char if_serial[12];
/* board mfg date (YYYYMMDD ASCII) */
char if_mfgdate[8];
@@ -503,8 +503,10 @@ struct ipath_flash {
__u8 if_powerhour[2];
/* ASCII free-form comment field */
char if_comment[32];
- /* 78 bytes used, min flash size is 128 bytes */
- __u8 if_future[50];
+ /* Backwards compatible prefix for longer QLogic Serial Numbers */
+ char if_sprefix[4];
+ /* 82 bytes used, min flash size is 128 bytes */
+ __u8 if_future[46];
};
/*
diff -r 21378f21e091 -r 8fbb5d71823a drivers/infiniband/hw/ipath/ipath_eeprom.c
--- a/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c Thu Jun 29 14:33:26 2006 -0700
@@ -602,8 +602,31 @@ void ipath_get_eeprom_info(struct ipath_
guid = *(__be64 *) ifp->if_guid;
dd->ipath_guid = guid;
dd->ipath_nguid = ifp->if_numguid;
- memcpy(dd->ipath_serial, ifp->if_serial,
- sizeof(ifp->if_serial));
+ /*
+ * Things are slightly complicated by the desire to transparently
+ * support both the Pathscale 10-digit serial number and the QLogic
+ * 13-character version.
+ */
+ if ((ifp->if_fversion > 1) && ifp->if_sprefix[0]
+ && ((u8 *)ifp->if_sprefix)[0] != 0xFF) {
+ /* This board has a Serial-prefix, which is stored
+ * elsewhere for backward-compatibility.
+ */
+ char *snp = dd->ipath_serial;
+ int len;
+ memcpy(snp, ifp->if_sprefix, sizeof ifp->if_sprefix);
+ snp[sizeof ifp->if_sprefix] = '\0';
+ len = strlen(snp);
+ snp += len;
+ len = (sizeof dd->ipath_serial) - len;
+ if (len > sizeof ifp->if_serial) {
+ len = sizeof ifp->if_serial;
+ }
+ memcpy(snp, ifp->if_serial, len);
+ } else
+ memcpy(dd->ipath_serial, ifp->if_serial,
+ sizeof ifp->if_serial);
+
ipath_cdbg(VERBOSE, "Initted GUID to %llx from eeprom\n",
(unsigned long long) be64_to_cpu(dd->ipath_guid));
diff -r 21378f21e091 -r 8fbb5d71823a drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:26 2006 -0700
@@ -491,8 +491,11 @@ struct ipath_devdata {
u16 ipath_lid;
/* list of pkeys programmed; 0 if not set */
u16 ipath_pkeys[4];
- /* ASCII serial number, from flash */
- u8 ipath_serial[12];
+ /*
+ * ASCII serial number, from flash, large enough for original
+ * all digit strings, and longer QLogic serial number format
+ */
+ u8 ipath_serial[16];
/* human readable board version */
u8 ipath_boardversion[80];
/* chip major rev, from ipath_revision */
diff -r 21378f21e091 -r 8fbb5d71823a drivers/infiniband/hw/ipath/ipath_pe800.c
--- a/drivers/infiniband/hw/ipath/ipath_pe800.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_pe800.c Thu Jun 29 14:33:26 2006 -0700
@@ -533,7 +533,7 @@ static int ipath_pe_boardname(struct ipa
if (n)
snprintf(name, namelen, "%s", n);
- if (dd->ipath_majrev != 4 || dd->ipath_minrev != 1) {
+ if (dd->ipath_majrev != 4 || !dd->ipath_minrev || dd->ipath_minrev>2) {
ipath_dev_err(dd, "Unsupported PE-800 revision %u.%u!\n",
dd->ipath_majrev, dd->ipath_minrev);
ret = 1;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 33 of 39] IB/ipath - read/write correct sizes through diag interface
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (31 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 32 of 39] IB/ipath - support more models of InfiniPath hardware Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 34 of 39] IB/ipath - fix a bug that results in addresses near 0 being written via DMA Bryan O'Sullivan
` (6 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
We must increment uaddr by size we are reading or writing, since it's
passed as a char *, not a pointer to the appropriate size.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 8fbb5d71823a -r a7c1ad1e090b drivers/infiniband/hw/ipath/ipath_diag.c
--- a/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:26 2006 -0700
@@ -115,7 +115,7 @@ static int ipath_read_umem64(struct ipat
goto bail;
}
reg_addr++;
- uaddr++;
+ uaddr += sizeof(u64);
}
ret = 0;
bail:
@@ -154,7 +154,7 @@ static int ipath_write_umem64(struct ipa
writeq(data, reg_addr);
reg_addr++;
- uaddr++;
+ uaddr += sizeof(u64);
}
ret = 0;
bail:
@@ -192,7 +192,8 @@ static int ipath_read_umem32(struct ipat
}
reg_addr++;
- uaddr++;
+ uaddr += sizeof(u32);
+
}
ret = 0;
bail:
@@ -231,7 +232,7 @@ static int ipath_write_umem32(struct ipa
writel(data, reg_addr);
reg_addr++;
- uaddr++;
+ uaddr += sizeof(u32);
}
ret = 0;
bail:
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 34 of 39] IB/ipath - fix a bug that results in addresses near 0 being written via DMA
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (32 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 33 of 39] IB/ipath - read/write correct sizes through diag interface Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 35 of 39] IB/ipath - remove some #if 0 code related to lockable memory Bryan O'Sullivan
` (5 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
We can't tell for sure if any packets are in the infinipath receive buffer
when we shut down a chip port. Normally this is taken care of by orderly
shutdown, but when processes are terminated, or sending process has a bug,
we can continue to receive packets. So rather than writing zero to the
address registers for the closing port, we point it at a dummy memory.
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r a7c1ad1e090b -r b6ebaf2dd2fd drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -1824,6 +1824,12 @@ static void cleanup_device(struct ipath_
dd->ipath_pioavailregs_phys);
dd->ipath_pioavailregs_dma = NULL;
}
+ if (dd->ipath_dummy_hdrq) {
+ dma_free_coherent(&dd->pcidev->dev,
+ dd->ipath_pd[0]->port_rcvhdrq_size,
+ dd->ipath_dummy_hdrq, dd->ipath_dummy_hdrq_phys);
+ dd->ipath_dummy_hdrq = NULL;
+ }
if (dd->ipath_pageshadow) {
struct page **tmpp = dd->ipath_pageshadow;
diff -r a7c1ad1e090b -r b6ebaf2dd2fd drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:26 2006 -0700
@@ -1486,41 +1486,50 @@ static int ipath_close(struct inode *in,
}
if (dd->ipath_kregbase) {
- ipath_write_kreg_port(
- dd, dd->ipath_kregs->kr_rcvhdrtailaddr,
- port, 0ULL);
- ipath_write_kreg_port(
- dd, dd->ipath_kregs->kr_rcvhdraddr,
- pd->port_port, 0);
+ int i;
+ /* atomically clear receive enable port. */
+ clear_bit(INFINIPATH_R_PORTENABLE_SHIFT + port,
+ &dd->ipath_rcvctrl);
+ ipath_write_kreg( dd, dd->ipath_kregs->kr_rcvctrl,
+ dd->ipath_rcvctrl);
+ /* and read back from chip to be sure that nothing
+ * else is in flight when we do the rest */
+ (void)ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
/* clean up the pkeys for this port user */
ipath_clean_part_key(pd, dd);
- if (port < dd->ipath_cfgports) {
- int i = dd->ipath_pbufsport * (port - 1);
- ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport);
-
- /* atomically clear receive enable port. */
- clear_bit(INFINIPATH_R_PORTENABLE_SHIFT + port,
- &dd->ipath_rcvctrl);
- ipath_write_kreg(
- dd,
- dd->ipath_kregs->kr_rcvctrl,
- dd->ipath_rcvctrl);
-
- if (dd->ipath_pageshadow)
- unlock_expected_tids(pd);
- ipath_stats.sps_ports--;
- ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
- pd->port_comm, pd->port_pid,
- dd->ipath_unit, port);
- }
+
+ /*
+ * be paranoid, and never write 0's to these, just use an
+ * unused part of the port 0 tail page. Of course,
+ * rcvhdraddr points to a large chunk of memory, so this
+ * could still trash things, but at least it won't trash
+ * page 0, and by disabling the port, it should stop "soon",
+ * even if a packet or two is in already in flight after we
+ * disabled the port.
+ */
+ ipath_write_kreg_port(dd,
+ dd->ipath_kregs->kr_rcvhdrtailaddr, port,
+ dd->ipath_dummy_hdrq_phys);
+ ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdraddr,
+ pd->port_port, dd->ipath_dummy_hdrq_phys);
+
+ i = dd->ipath_pbufsport * (port - 1);
+ ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport);
+
+ if (dd->ipath_pageshadow)
+ unlock_expected_tids(pd);
+ ipath_stats.sps_ports--;
+ ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
+ pd->port_comm, pd->port_pid,
+ dd->ipath_unit, port);
+
+ dd->ipath_f_clear_tids(dd, pd->port_port);
}
pd->port_cnt = 0;
pd->port_pid = 0;
-
- dd->ipath_f_clear_tids(dd, pd->port_port);
dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */
mutex_unlock(&ipath_mutex);
diff -r a7c1ad1e090b -r b6ebaf2dd2fd drivers/infiniband/hw/ipath/ipath_init_chip.c
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:26 2006 -0700
@@ -647,6 +647,7 @@ int ipath_init_chip(struct ipath_devdata
u32 val32, kpiobufs;
u64 val;
struct ipath_portdata *pd = NULL; /* keep gcc4 happy */
+ gfp_t gfp_flags = GFP_USER | __GFP_COMP;
ret = init_housekeeping(dd, &pd, reinit);
if (ret)
@@ -833,6 +834,22 @@ int ipath_init_chip(struct ipath_devdata
"rcvhdrq and/or egr bufs\n");
else
enable_chip(dd, pd, reinit);
+
+
+ if (!ret && !reinit) {
+ /* used when we close a port, for DMA already in flight at close */
+ dd->ipath_dummy_hdrq = dma_alloc_coherent(
+ &dd->pcidev->dev, pd->port_rcvhdrq_size,
+ &dd->ipath_dummy_hdrq_phys,
+ gfp_flags);
+ if (!dd->ipath_dummy_hdrq ) {
+ dev_info(&dd->pcidev->dev,
+ "Couldn't allocate 0x%lx bytes for dummy hdrq\n",
+ pd->port_rcvhdrq_size);
+ /* fallback to just 0'ing */
+ dd->ipath_dummy_hdrq_phys = 0UL;
+ }
+ }
/*
* cause retrigger of pending interrupts ignored during init,
diff -r a7c1ad1e090b -r b6ebaf2dd2fd drivers/infiniband/hw/ipath/ipath_kernel.h
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h Thu Jun 29 14:33:26 2006 -0700
@@ -352,6 +352,8 @@ struct ipath_devdata {
/* check for stale messages in rcv queue */
/* only allow one intr at a time. */
unsigned long ipath_rcv_pending;
+ void *ipath_dummy_hdrq; /* used after port close */
+ dma_addr_t ipath_dummy_hdrq_phys;
/*
* Shadow copies of registers; size indicates read access size.
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 35 of 39] IB/ipath - remove some #if 0 code related to lockable memory
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (33 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 34 of 39] IB/ipath - fix a bug that results in addresses near 0 being written via DMA Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 36 of 39] IB/ipath - Ignore receive queue size if SRQ is specified Bryan O'Sullivan
` (4 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r b6ebaf2dd2fd -r 9b423c45af8b drivers/infiniband/hw/ipath/ipath_user_pages.c
--- a/drivers/infiniband/hw/ipath/ipath_user_pages.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_user_pages.c Thu Jun 29 14:33:26 2006 -0700
@@ -57,17 +57,6 @@ static int __get_user_pages(unsigned lon
unsigned long lock_limit;
size_t got;
int ret;
-
-#if 0
- /*
- * XXX - causes MPI programs to fail, haven't had time to check
- * yet
- */
- if (!capable(CAP_IPC_LOCK)) {
- ret = -EPERM;
- goto bail;
- }
-#endif
lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >>
PAGE_SHIFT;
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 36 of 39] IB/ipath - Ignore receive queue size if SRQ is specified
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (34 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 35 of 39] IB/ipath - remove some #if 0 code related to lockable memory Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 37 of 39] IB/ipath - namespace cleanup: replace ips with ipath Bryan O'Sullivan
` (3 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
The receive work queue size should be ignored if the QP is created
to use a shared receive queue according to the IB spec.
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 9b423c45af8b -r 31c382d8210a drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:26 2006 -0700
@@ -685,16 +685,22 @@ struct ib_qp *ipath_create_qp(struct ib_
ret = ERR_PTR(-ENOMEM);
goto bail;
}
- qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
- sz = sizeof(struct ipath_sge) *
- init_attr->cap.max_recv_sge +
- sizeof(struct ipath_rwqe);
- qp->r_rq.wq = vmalloc(qp->r_rq.size * sz);
- if (!qp->r_rq.wq) {
- kfree(qp);
- vfree(swq);
- ret = ERR_PTR(-ENOMEM);
- goto bail;
+ if (init_attr->srq) {
+ qp->r_rq.size = 0;
+ qp->r_rq.max_sge = 0;
+ qp->r_rq.wq = NULL;
+ } else {
+ qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
+ qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
+ sz = (sizeof(struct ipath_sge) * qp->r_rq.max_sge) +
+ sizeof(struct ipath_rwqe);
+ qp->r_rq.wq = vmalloc(qp->r_rq.size * sz);
+ if (!qp->r_rq.wq) {
+ kfree(qp);
+ vfree(swq);
+ ret = ERR_PTR(-ENOMEM);
+ goto bail;
+ }
}
/*
@@ -713,7 +719,6 @@ struct ib_qp *ipath_create_qp(struct ib_
qp->s_wq = swq;
qp->s_size = init_attr->cap.max_send_wr + 1;
qp->s_max_sge = init_attr->cap.max_send_sge;
- qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
qp->s_flags = init_attr->sq_sig_type == IB_SIGNAL_REQ_WR ?
1 << IPATH_S_SIGNAL_REQ_WR : 0;
dev = to_idev(ibpd->device);
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 37 of 39] IB/ipath - namespace cleanup: replace ips with ipath
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (35 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 36 of 39] IB/ipath - Ignore receive queue size if SRQ is specified Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:41 ` [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems Bryan O'Sullivan
` (2 subsequent siblings)
39 siblings, 0 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
Remove ips namespace from infinipath drivers. This renames ips_common.h
to ipath_common.h. Definitions, data structures, etc. that were not
used by kernel modules have moved to user-only headers. All names
including ips have been renamed to ipath. Some names have had an ipath
prefix added.
Signed-off-by: Christian Bell <christian.bell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_common.h
--- a/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_common.h Thu Jun 29 14:33:26 2006 -0700
@@ -39,7 +39,8 @@
* to communicate between kernel and user code.
*/
-/* This is the IEEE-assigned OUI for QLogic, Inc. InfiniPath */
+
+/* This is the IEEE-assigned OUI for QLogic Inc. InfiniPath */
#define IPATH_SRC_OUI_1 0x00
#define IPATH_SRC_OUI_2 0x11
#define IPATH_SRC_OUI_3 0x75
@@ -343,9 +344,9 @@ struct ipath_base_info {
/*
* Similarly, this is the kernel version going back to the user. It's
* slightly different, in that we want to tell if the driver was built as
- * part of a QLogic release, or from the driver from OpenIB, kernel.org,
- * or a standard distribution, for support reasons. The high bit is 0 for
- * non-QLogic, and 1 for QLogic-built/supplied.
+ * part of a QLogic release, or from the driver from openfabrics.org,
+ * kernel.org, or a standard distribution, for support reasons.
+ * The high bit is 0 for non-QLogic and 1 for QLogic-built/supplied.
*
* It's returned by the driver to the user code during initialization in the
* spi_sw_version field of ipath_base_info, so the user code can in turn
@@ -600,14 +601,118 @@ struct infinipath_counters {
#define INFINIPATH_KPF_INTR 0x1
/* SendPIO per-buffer control */
-#define INFINIPATH_SP_LENGTHP1_MASK 0x3FF
-#define INFINIPATH_SP_LENGTHP1_SHIFT 0
-#define INFINIPATH_SP_INTR 0x80000000
-#define INFINIPATH_SP_TEST 0x40000000
-#define INFINIPATH_SP_TESTEBP 0x20000000
+#define INFINIPATH_SP_TEST 0x40
+#define INFINIPATH_SP_TESTEBP 0x20
/* SendPIOAvail bits */
#define INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT 1
#define INFINIPATH_SENDPIOAVAIL_CHECK_SHIFT 0
+/* infinipath header format */
+struct ipath_header {
+ /*
+ * Version - 4 bits, Port - 4 bits, TID - 10 bits and Offset -
+ * 14 bits before ECO change ~28 Dec 03. After that, Vers 4,
+ * Port 3, TID 11, offset 14.
+ */
+ __le32 ver_port_tid_offset;
+ __le16 chksum;
+ __le16 pkt_flags;
+};
+
+/* infinipath user message header format.
+ * This structure contains the first 4 fields common to all protocols
+ * that employ infinipath.
+ */
+struct ipath_message_header {
+ __be16 lrh[4];
+ __be32 bth[3];
+ /* fields below this point are in host byte order */
+ struct ipath_header iph;
+ __u8 sub_opcode;
+};
+
+/* infinipath ethernet header format */
+struct ether_header {
+ __be16 lrh[4];
+ __be32 bth[3];
+ struct ipath_header iph;
+ __u8 sub_opcode;
+ __u8 cmd;
+ __be16 lid;
+ __u16 mac[3];
+ __u8 frag_num;
+ __u8 seq_num;
+ __le32 len;
+ /* MUST be of word size due to PIO write requirements */
+ __le32 csum;
+ __le16 csum_offset;
+ __le16 flags;
+ __u16 first_2_bytes;
+ __u8 unused[2]; /* currently unused */
+};
+
+
+/* IB - LRH header consts */
+#define IPATH_LRH_GRH 0x0003 /* 1. word of IB LRH - next header: GRH */
+#define IPATH_LRH_BTH 0x0002 /* 1. word of IB LRH - next header: BTH */
+
+/* misc. */
+#define SIZE_OF_CRC 1
+
+#define IPATH_DEFAULT_P_KEY 0xFFFF
+#define IPATH_PERMISSIVE_LID 0xFFFF
+#define IPATH_AETH_CREDIT_SHIFT 24
+#define IPATH_AETH_CREDIT_MASK 0x1F
+#define IPATH_AETH_CREDIT_INVAL 0x1F
+#define IPATH_PSN_MASK 0xFFFFFF
+#define IPATH_MSN_MASK 0xFFFFFF
+#define IPATH_QPN_MASK 0xFFFFFF
+#define IPATH_MULTICAST_LID_BASE 0xC000
+#define IPATH_MULTICAST_QPN 0xFFFFFF
+
+/* Receive Header Queue: receive type (from infinipath) */
+#define RCVHQ_RCV_TYPE_EXPECTED 0
+#define RCVHQ_RCV_TYPE_EAGER 1
+#define RCVHQ_RCV_TYPE_NON_KD 2
+#define RCVHQ_RCV_TYPE_ERROR 3
+
+
+/* sub OpCodes - ith4x */
+#define IPATH_ITH4X_OPCODE_ENCAP 0x81
+#define IPATH_ITH4X_OPCODE_LID_ARP 0x82
+
+#define IPATH_HEADER_QUEUE_WORDS 9
+
+/* functions for extracting fields from rcvhdrq entries for the driver.
+ */
+static inline __u32 ipath_hdrget_err_flags(const __le32 * rbuf)
+{
+ return __le32_to_cpu(rbuf[1]);
+}
+
+static inline __u32 ipath_hdrget_rcv_type(const __le32 * rbuf)
+{
+ return (__le32_to_cpu(rbuf[0]) >> INFINIPATH_RHF_RCVTYPE_SHIFT)
+ & INFINIPATH_RHF_RCVTYPE_MASK;
+}
+
+static inline __u32 ipath_hdrget_length_in_bytes(const __le32 * rbuf)
+{
+ return ((__le32_to_cpu(rbuf[0]) >> INFINIPATH_RHF_LENGTH_SHIFT)
+ & INFINIPATH_RHF_LENGTH_MASK) << 2;
+}
+
+static inline __u32 ipath_hdrget_index(const __le32 * rbuf)
+{
+ return (__le32_to_cpu(rbuf[0]) >> INFINIPATH_RHF_EGRINDEX_SHIFT)
+ & INFINIPATH_RHF_EGRINDEX_MASK;
+}
+
+static inline __u32 ipath_hdrget_ipath_ver(__le32 hdrword)
+{
+ return (__le32_to_cpu(hdrword) >> INFINIPATH_I_VERS_SHIFT)
+ & INFINIPATH_I_VERS_MASK;
+}
+
#endif /* _IPATH_COMMON_H */
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_diag.c
--- a/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_diag.c Thu Jun 29 14:33:26 2006 -0700
@@ -44,10 +44,9 @@
#include <linux/pci.h>
#include <asm/uaccess.h>
+#include "ipath_kernel.h"
+#include "ipath_layer.h"
#include "ipath_common.h"
-#include "ipath_kernel.h"
-#include "ips_common.h"
-#include "ipath_layer.h"
int ipath_diag_inuse;
static int diag_set_link;
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -39,8 +39,8 @@
#include <linux/vmalloc.h>
#include "ipath_kernel.h"
-#include "ips_common.h"
#include "ipath_layer.h"
+#include "ipath_common.h"
static void ipath_update_pio_bufs(struct ipath_devdata *);
@@ -823,7 +823,8 @@ static void ipath_rcv_layer(struct ipath
u8 pad, *bthbytes;
struct sk_buff *skb, *nskb;
- if (dd->ipath_port0_skbs && hdr->sub_opcode == OPCODE_ENCAP) {
+ if (dd->ipath_port0_skbs &&
+ hdr->sub_opcode == IPATH_ITH4X_OPCODE_ENCAP) {
/*
* Allocate a new sk_buff to replace the one we give
* to the network stack.
@@ -854,7 +855,7 @@ static void ipath_rcv_layer(struct ipath
/* another ether packet received */
ipath_stats.sps_ether_rpkts++;
}
- else if (hdr->sub_opcode == OPCODE_LID_ARP)
+ else if (hdr->sub_opcode == IPATH_ITH4X_OPCODE_LID_ARP)
__ipath_layer_rcv_lid(dd, hdr);
}
@@ -871,7 +872,7 @@ void ipath_kreceive(struct ipath_devdata
const u32 rsize = dd->ipath_rcvhdrentsize; /* words */
const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize; /* words */
u32 etail = -1, l, hdrqtail;
- struct ips_message_header *hdr;
+ struct ipath_message_header *hdr;
u32 eflags, i, etype, tlen, pkttot = 0, updegr=0, reloop=0;
static u64 totcalls; /* stats, may eventually remove */
char emsg[128];
@@ -897,7 +898,7 @@ reloop:
u8 *bthbytes;
rc = (u64 *) (dd->ipath_pd[0]->port_rcvhdrq + (l << 2));
- hdr = (struct ips_message_header *)&rc[1];
+ hdr = (struct ipath_message_header *)&rc[1];
/*
* could make a network order version of IPATH_KD_QP, and
* do the obvious shift before masking to speed this up.
@@ -905,10 +906,10 @@ reloop:
qp = ntohl(hdr->bth[1]) & 0xffffff;
bthbytes = (u8 *) hdr->bth;
- eflags = ips_get_hdr_err_flags((__le32 *) rc);
- etype = ips_get_rcv_type((__le32 *) rc);
+ eflags = ipath_hdrget_err_flags((__le32 *) rc);
+ etype = ipath_hdrget_rcv_type((__le32 *) rc);
/* total length */
- tlen = ips_get_length_in_bytes((__le32 *) rc);
+ tlen = ipath_hdrget_length_in_bytes((__le32 *) rc);
ebuf = NULL;
if (etype != RCVHQ_RCV_TYPE_EXPECTED) {
/*
@@ -918,7 +919,7 @@ reloop:
* set ebuf (so we try to copy data) unless the
* length requires it.
*/
- etail = ips_get_index((__le32 *) rc);
+ etail = ipath_hdrget_index((__le32 *) rc);
if (tlen > sizeof(*hdr) ||
etype == RCVHQ_RCV_TYPE_NON_KD)
ebuf = ipath_get_egrbuf(dd, etail, 0);
@@ -930,7 +931,7 @@ reloop:
*/
if (etype != RCVHQ_RCV_TYPE_NON_KD && etype !=
- RCVHQ_RCV_TYPE_ERROR && ips_get_ipath_ver(
+ RCVHQ_RCV_TYPE_ERROR && ipath_hdrget_ipath_ver(
hdr->iph.ver_port_tid_offset) !=
IPS_PROTO_VERSION) {
ipath_cdbg(PKT, "Bad InfiniPath protocol version "
@@ -943,7 +944,7 @@ reloop:
ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u "
"tlen=%x opcode=%x egridx=%x: %s\n",
eflags, l, etype, tlen, bthbytes[0],
- ips_get_index((__le32 *) rc), emsg);
+ ipath_hdrget_index((__le32 *) rc), emsg);
/* Count local link integrity errors. */
if (eflags & (INFINIPATH_RHF_H_ICRCERR |
INFINIPATH_RHF_H_VCRCERR)) {
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:26 2006 -0700
@@ -39,8 +39,8 @@
#include <asm/pgtable.h>
#include "ipath_kernel.h"
-#include "ips_common.h"
#include "ipath_layer.h"
+#include "ipath_common.h"
static int ipath_open(struct inode *, struct file *);
static int ipath_close(struct inode *, struct file *);
@@ -458,7 +458,7 @@ static int ipath_set_part_key(struct ipa
u16 lkey = key & 0x7FFF;
int ret;
- if (lkey == (IPS_DEFAULT_P_KEY & 0x7FFF)) {
+ if (lkey == (IPATH_DEFAULT_P_KEY & 0x7FFF)) {
/* nothing to do; this key always valid */
ret = 0;
goto bail;
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_init_chip.c
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c Thu Jun 29 14:33:26 2006 -0700
@@ -36,7 +36,7 @@
#include <linux/vmalloc.h>
#include "ipath_kernel.h"
-#include "ips_common.h"
+#include "ipath_common.h"
/*
* min buffers we want to have per port, after driver
@@ -277,7 +277,7 @@ static int init_chip_first(struct ipath_
pd->port_port = 0;
pd->port_cnt = 1;
/* The port 0 pkey table is used by the layer interface. */
- pd->port_pkeys[0] = IPS_DEFAULT_P_KEY;
+ pd->port_pkeys[0] = IPATH_DEFAULT_P_KEY;
dd->ipath_rcvtidcnt =
ipath_read_kreg32(dd, dd->ipath_kregs->kr_rcvtidcnt);
dd->ipath_rcvtidbase =
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_intr.c
--- a/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c Thu Jun 29 14:33:26 2006 -0700
@@ -34,8 +34,8 @@
#include <linux/pci.h>
#include "ipath_kernel.h"
-#include "ips_common.h"
#include "ipath_layer.h"
+#include "ipath_common.h"
/* These are all rcv-related errors which we want to count for stats */
#define E_SUM_PKTERRS \
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_layer.c
--- a/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c Thu Jun 29 14:33:26 2006 -0700
@@ -41,8 +41,8 @@
#include <asm/byteorder.h>
#include "ipath_kernel.h"
-#include "ips_common.h"
#include "ipath_layer.h"
+#include "ipath_common.h"
/* Acquire before ipath_devs_lock. */
static DEFINE_MUTEX(ipath_layer_mutex);
@@ -622,7 +622,7 @@ int ipath_layer_open(struct ipath_devdat
goto bail;
}
- ret = ipath_setrcvhdrsize(dd, NUM_OF_EXTRA_WORDS_IN_HEADER_QUEUE);
+ ret = ipath_setrcvhdrsize(dd, IPATH_HEADER_QUEUE_WORDS);
if (ret < 0)
goto bail;
@@ -1106,10 +1106,10 @@ int ipath_layer_send_hdr(struct ipath_de
}
vlsllnh = *((__be16 *) hdr);
- if (vlsllnh != htons(IPS_LRH_BTH)) {
+ if (vlsllnh != htons(IPATH_LRH_BTH)) {
ipath_dbg("Warning: lrh[0] wrong (%x, not %x); "
"not sending\n", be16_to_cpu(vlsllnh),
- IPS_LRH_BTH);
+ IPATH_LRH_BTH);
ret = -EINVAL;
}
if (ret)
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_mad.c
--- a/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c Thu Jun 29 14:33:26 2006 -0700
@@ -35,7 +35,7 @@
#include "ipath_kernel.h"
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
#define IB_SMP_UNSUP_VERSION __constant_htons(0x0004)
#define IB_SMP_UNSUP_METHOD __constant_htons(0x0008)
@@ -306,7 +306,7 @@ static int recv_subn_set_portinfo(struct
lid = be16_to_cpu(pip->lid);
if (lid != ipath_layer_get_lid(dev->dd)) {
/* Must be a valid unicast LID address. */
- if (lid == 0 || lid >= IPS_MULTICAST_LID_BASE)
+ if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
goto err;
ipath_set_lid(dev->dd, lid, pip->mkeyprot_resv_lmc & 7);
event.event = IB_EVENT_LID_CHANGE;
@@ -316,7 +316,7 @@ static int recv_subn_set_portinfo(struct
smlid = be16_to_cpu(pip->sm_lid);
if (smlid != dev->sm_lid) {
/* Must be a valid unicast LID address. */
- if (smlid == 0 || smlid >= IPS_MULTICAST_LID_BASE)
+ if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
goto err;
dev->sm_lid = smlid;
event.event = IB_EVENT_SM_CHANGE;
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_qp.c
--- a/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c Thu Jun 29 14:33:26 2006 -0700
@@ -35,7 +35,7 @@
#include <linux/vmalloc.h>
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
#define BITS_PER_PAGE (PAGE_SIZE*BITS_PER_BYTE)
#define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
@@ -450,7 +450,7 @@ int ipath_modify_qp(struct ib_qp *ibqp,
if (attr_mask & IB_QP_AV)
if (attr->ah_attr.dlid == 0 ||
- attr->ah_attr.dlid >= IPS_MULTICAST_LID_BASE)
+ attr->ah_attr.dlid >= IPATH_MULTICAST_LID_BASE)
goto inval;
if (attr_mask & IB_QP_PKEY_INDEX)
@@ -585,14 +585,14 @@ int ipath_query_qp(struct ib_qp *ibqp, s
*/
__be32 ipath_compute_aeth(struct ipath_qp *qp)
{
- u32 aeth = qp->r_msn & IPS_MSN_MASK;
+ u32 aeth = qp->r_msn & IPATH_MSN_MASK;
if (qp->ibqp.srq) {
/*
* Shared receive queues don't generate credits.
* Set the credit field to the invalid value.
*/
- aeth |= IPS_AETH_CREDIT_INVAL << IPS_AETH_CREDIT_SHIFT;
+ aeth |= IPATH_AETH_CREDIT_INVAL << IPATH_AETH_CREDIT_SHIFT;
} else {
u32 min, max, x;
u32 credits;
@@ -622,7 +622,7 @@ __be32 ipath_compute_aeth(struct ipath_q
else
min = x;
}
- aeth |= x << IPS_AETH_CREDIT_SHIFT;
+ aeth |= x << IPATH_AETH_CREDIT_SHIFT;
}
return cpu_to_be32(aeth);
}
@@ -888,18 +888,18 @@ void ipath_sqerror_qp(struct ipath_qp *q
*/
void ipath_get_credit(struct ipath_qp *qp, u32 aeth)
{
- u32 credit = (aeth >> IPS_AETH_CREDIT_SHIFT) & IPS_AETH_CREDIT_MASK;
+ u32 credit = (aeth >> IPATH_AETH_CREDIT_SHIFT) & IPATH_AETH_CREDIT_MASK;
/*
* If the credit is invalid, we can send
* as many packets as we like. Otherwise, we have to
* honor the credit field.
*/
- if (credit == IPS_AETH_CREDIT_INVAL)
+ if (credit == IPATH_AETH_CREDIT_INVAL)
qp->s_lsn = (u32) -1;
else if (qp->s_lsn != (u32) -1) {
/* Compute new LSN (i.e., MSN + credit) */
- credit = (aeth + credit_table[credit]) & IPS_MSN_MASK;
+ credit = (aeth + credit_table[credit]) & IPATH_MSN_MASK;
if (ipath_cmp24(credit, qp->s_lsn) > 0)
qp->s_lsn = credit;
}
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_rc.c
--- a/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c Thu Jun 29 14:33:26 2006 -0700
@@ -32,7 +32,7 @@
*/
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
/* cut down ridiculously long IB macro names */
#define OP(x) IB_OPCODE_RC_##x
@@ -49,7 +49,7 @@ static void ipath_init_restart(struct ip
struct ipath_ibdev *dev;
u32 len;
- len = ((qp->s_psn - wqe->psn) & IPS_PSN_MASK) *
+ len = ((qp->s_psn - wqe->psn) & IPATH_PSN_MASK) *
ib_mtu_enum_to_int(qp->path_mtu);
qp->s_sge.sge = wqe->sg_list[0];
qp->s_sge.sg_list = wqe->sg_list + 1;
@@ -159,9 +159,9 @@ u32 ipath_make_rc_ack(struct ipath_qp *q
qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
bth0 = OP(ACKNOWLEDGE) << 24;
if (qp->s_nak_state)
- ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPS_MSN_MASK) |
+ ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
(qp->s_nak_state <<
- IPS_AETH_CREDIT_SHIFT));
+ IPATH_AETH_CREDIT_SHIFT));
else
ohdr->u.aeth = ipath_compute_aeth(qp);
hwords++;
@@ -361,7 +361,7 @@ int ipath_make_rc_req(struct ipath_qp *q
if (qp->s_tail >= qp->s_size)
qp->s_tail = 0;
}
- bth2 |= qp->s_psn++ & IPS_PSN_MASK;
+ bth2 |= qp->s_psn++ & IPATH_PSN_MASK;
if ((int)(qp->s_psn - qp->s_next_psn) > 0)
qp->s_next_psn = qp->s_psn;
/*
@@ -387,7 +387,7 @@ int ipath_make_rc_req(struct ipath_qp *q
qp->s_state = OP(SEND_MIDDLE);
/* FALLTHROUGH */
case OP(SEND_MIDDLE):
- bth2 = qp->s_psn++ & IPS_PSN_MASK;
+ bth2 = qp->s_psn++ & IPATH_PSN_MASK;
if ((int)(qp->s_psn - qp->s_next_psn) > 0)
qp->s_next_psn = qp->s_psn;
ss = &qp->s_sge;
@@ -429,7 +429,7 @@ int ipath_make_rc_req(struct ipath_qp *q
qp->s_state = OP(RDMA_WRITE_MIDDLE);
/* FALLTHROUGH */
case OP(RDMA_WRITE_MIDDLE):
- bth2 = qp->s_psn++ & IPS_PSN_MASK;
+ bth2 = qp->s_psn++ & IPATH_PSN_MASK;
if ((int)(qp->s_psn - qp->s_next_psn) > 0)
qp->s_next_psn = qp->s_psn;
ss = &qp->s_sge;
@@ -466,7 +466,7 @@ int ipath_make_rc_req(struct ipath_qp *q
* See ipath_restart_rc().
*/
ipath_init_restart(qp, wqe);
- len = ((qp->s_psn - wqe->psn) & IPS_PSN_MASK) * pmtu;
+ len = ((qp->s_psn - wqe->psn) & IPATH_PSN_MASK) * pmtu;
ohdr->u.rc.reth.vaddr =
cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
ohdr->u.rc.reth.rkey =
@@ -474,7 +474,7 @@ int ipath_make_rc_req(struct ipath_qp *q
ohdr->u.rc.reth.length = cpu_to_be32(qp->s_len);
qp->s_state = OP(RDMA_READ_REQUEST);
hwords += sizeof(ohdr->u.rc.reth) / 4;
- bth2 = qp->s_psn++ & IPS_PSN_MASK;
+ bth2 = qp->s_psn++ & IPATH_PSN_MASK;
if ((int)(qp->s_psn - qp->s_next_psn) > 0)
qp->s_next_psn = qp->s_psn;
ss = NULL;
@@ -529,7 +529,7 @@ static void send_rc_ack(struct ipath_qp
/* Construct the header. */
ohdr = &hdr.u.oth;
- lrh0 = IPS_LRH_BTH;
+ lrh0 = IPATH_LRH_BTH;
/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4. */
hwords = 6;
if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
@@ -537,14 +537,14 @@ static void send_rc_ack(struct ipath_qp
&qp->remote_ah_attr.grh,
hwords, 0);
ohdr = &hdr.u.l.oth;
- lrh0 = IPS_LRH_GRH;
+ lrh0 = IPATH_LRH_GRH;
}
/* read pkey_index w/o lock (its atomic) */
bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
if (qp->r_nak_state)
- ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPS_MSN_MASK) |
+ ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
(qp->r_nak_state <<
- IPS_AETH_CREDIT_SHIFT));
+ IPATH_AETH_CREDIT_SHIFT));
else
ohdr->u.aeth = ipath_compute_aeth(qp);
if (qp->r_ack_state >= OP(COMPARE_SWAP)) {
@@ -560,7 +560,7 @@ static void send_rc_ack(struct ipath_qp
hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
ohdr->bth[0] = cpu_to_be32(bth0);
ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
- ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPS_PSN_MASK);
+ ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK);
/*
* If we can send the ACK, clear the ACK state.
@@ -890,8 +890,8 @@ static int do_rc_ack(struct ipath_qp *qp
reset_psn(qp, psn);
qp->s_rnr_timeout =
- ib_ipath_rnr_table[(aeth >> IPS_AETH_CREDIT_SHIFT) &
- IPS_AETH_CREDIT_MASK];
+ ib_ipath_rnr_table[(aeth >> IPATH_AETH_CREDIT_SHIFT) &
+ IPATH_AETH_CREDIT_MASK];
ipath_insert_rnr_queue(qp);
goto bail;
@@ -899,8 +899,8 @@ static int do_rc_ack(struct ipath_qp *qp
/* The last valid PSN seen is the previous request's. */
if (qp->s_last != qp->s_tail)
qp->s_last_psn = wqe->psn - 1;
- switch ((aeth >> IPS_AETH_CREDIT_SHIFT) &
- IPS_AETH_CREDIT_MASK) {
+ switch ((aeth >> IPATH_AETH_CREDIT_SHIFT) &
+ IPATH_AETH_CREDIT_MASK) {
case 0: /* PSN sequence error */
dev->n_seq_naks++;
/*
@@ -1268,7 +1268,7 @@ static inline int ipath_rc_rcv_error(str
* Check for the PSN of the last atomic operation
* performed and resend the result if found.
*/
- if ((psn & IPS_PSN_MASK) != qp->r_atomic_psn)
+ if ((psn & IPATH_PSN_MASK) != qp->r_atomic_psn)
goto done;
break;
}
@@ -1638,7 +1638,7 @@ void ipath_rc_rcv(struct ipath_ibdev *de
*(u64 *) qp->r_sge.sge.vaddr = sdata;
spin_unlock_irq(&dev->pending_lock);
qp->r_msn++;
- qp->r_atomic_psn = psn & IPS_PSN_MASK;
+ qp->r_atomic_psn = psn & IPATH_PSN_MASK;
psn |= 1 << 31;
break;
}
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_ruc.c
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c Thu Jun 29 14:33:26 2006 -0700
@@ -32,7 +32,7 @@
*/
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
/*
* Convert the AETH RNR timeout code into the number of milliseconds.
@@ -632,7 +632,7 @@ again:
/* Sending responses has higher priority over sending requests. */
if (qp->s_ack_state != IB_OPCODE_RC_ACKNOWLEDGE &&
(bth0 = ipath_make_rc_ack(qp, ohdr, pmtu)) != 0)
- bth2 = qp->s_ack_psn++ & IPS_PSN_MASK;
+ bth2 = qp->s_ack_psn++ & IPATH_PSN_MASK;
else if (!((qp->ibqp.qp_type == IB_QPT_RC) ?
ipath_make_rc_req(qp, ohdr, pmtu, &bth0, &bth2) :
ipath_make_uc_req(qp, ohdr, pmtu, &bth0, &bth2))) {
@@ -651,12 +651,12 @@ again:
/* Construct the header. */
extra_bytes = (4 - qp->s_cur_size) & 3;
nwords = (qp->s_cur_size + extra_bytes) >> 2;
- lrh0 = IPS_LRH_BTH;
+ lrh0 = IPATH_LRH_BTH;
if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
qp->s_hdrwords += ipath_make_grh(dev, &qp->s_hdr.u.l.grh,
&qp->remote_ah_attr.grh,
qp->s_hdrwords, nwords);
- lrh0 = IPS_LRH_GRH;
+ lrh0 = IPATH_LRH_GRH;
}
lrh0 |= qp->remote_ah_attr.sl << 4;
qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_sysfs.c
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c Thu Jun 29 14:33:26 2006 -0700
@@ -35,8 +35,8 @@
#include <linux/pci.h>
#include "ipath_kernel.h"
-#include "ips_common.h"
#include "ipath_layer.h"
+#include "ipath_common.h"
/**
* ipath_parse_ushort - parse an unsigned short value in an arbitrary base
@@ -187,7 +187,7 @@ static ssize_t store_lid(struct device *
if (ret < 0)
goto invalid;
- if (lid == 0 || lid >= IPS_MULTICAST_LID_BASE) {
+ if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE) {
ret = -EINVAL;
goto invalid;
}
@@ -221,7 +221,7 @@ static ssize_t store_mlid(struct device
int ret;
ret = ipath_parse_ushort(buf, &mlid);
- if (ret < 0 || mlid < IPS_MULTICAST_LID_BASE)
+ if (ret < 0 || mlid < IPATH_MULTICAST_LID_BASE)
goto invalid;
unit = dd->ipath_unit;
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_uc.c
--- a/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c Thu Jun 29 14:33:26 2006 -0700
@@ -32,7 +32,7 @@
*/
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
/* cut down ridiculously long IB macro names */
#define OP(x) IB_OPCODE_UC_##x
@@ -213,7 +213,7 @@ int ipath_make_uc_req(struct ipath_qp *q
qp->s_cur_sge = &qp->s_sge;
qp->s_cur_size = len;
*bth0p = bth0 | (qp->s_state << 24);
- *bth2p = qp->s_next_psn++ & IPS_PSN_MASK;
+ *bth2p = qp->s_next_psn++ & IPATH_PSN_MASK;
return 1;
done:
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_ud.c
--- a/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c Thu Jun 29 14:33:26 2006 -0700
@@ -34,7 +34,7 @@
#include <rdma/ib_smi.h>
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
/**
* ipath_ud_loopback - handle send on loopback QPs
@@ -289,8 +289,8 @@ int ipath_post_ud_send(struct ipath_qp *
ret = -EINVAL;
goto bail;
}
- if (ah_attr->dlid >= IPS_MULTICAST_LID_BASE) {
- if (ah_attr->dlid != IPS_PERMISSIVE_LID)
+ if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE) {
+ if (ah_attr->dlid != IPATH_PERMISSIVE_LID)
dev->n_multicast_xmit++;
else
dev->n_unicast_xmit++;
@@ -310,7 +310,7 @@ int ipath_post_ud_send(struct ipath_qp *
if (ah_attr->ah_flags & IB_AH_GRH) {
/* Header size in 32-bit words. */
hwords = 17;
- lrh0 = IPS_LRH_GRH;
+ lrh0 = IPATH_LRH_GRH;
ohdr = &qp->s_hdr.u.l.oth;
qp->s_hdr.u.l.grh.version_tclass_flow =
cpu_to_be32((6 << 28) |
@@ -336,7 +336,7 @@ int ipath_post_ud_send(struct ipath_qp *
} else {
/* Header size in 32-bit words. */
hwords = 7;
- lrh0 = IPS_LRH_BTH;
+ lrh0 = IPATH_LRH_BTH;
ohdr = &qp->s_hdr.u.oth;
}
if (wr->opcode == IB_WR_SEND_WITH_IMM) {
@@ -367,18 +367,18 @@ int ipath_post_ud_send(struct ipath_qp *
if (wr->send_flags & IB_SEND_SOLICITED)
bth0 |= 1 << 23;
bth0 |= extra_bytes << 20;
- bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? IPS_DEFAULT_P_KEY :
+ bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? IPATH_DEFAULT_P_KEY :
ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
ohdr->bth[0] = cpu_to_be32(bth0);
/*
* Use the multicast QP if the destination LID is a multicast LID.
*/
- ohdr->bth[1] = ah_attr->dlid >= IPS_MULTICAST_LID_BASE &&
- ah_attr->dlid != IPS_PERMISSIVE_LID ?
- __constant_cpu_to_be32(IPS_MULTICAST_QPN) :
+ ohdr->bth[1] = ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
+ ah_attr->dlid != IPATH_PERMISSIVE_LID ?
+ __constant_cpu_to_be32(IPATH_MULTICAST_QPN) :
cpu_to_be32(wr->wr.ud.remote_qpn);
/* XXX Could lose a PSN count but not worth locking */
- ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & IPS_PSN_MASK);
+ ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & IPATH_PSN_MASK);
/*
* Qkeys with the high order bit set mean use the
* qkey from the QP context instead of the WR (see 10.2.5).
@@ -469,7 +469,7 @@ void ipath_ud_rcv(struct ipath_ibdev *de
src_qp = be32_to_cpu(ohdr->u.ud.deth[1]);
}
}
- src_qp &= IPS_QPN_MASK;
+ src_qp &= IPATH_QPN_MASK;
/*
* Check that the permissive LID is only used on QP0
@@ -627,7 +627,7 @@ void ipath_ud_rcv(struct ipath_ibdev *de
/*
* Save the LMC lower bits if the destination LID is a unicast LID.
*/
- wc.dlid_path_bits = dlid >= IPS_MULTICAST_LID_BASE ? 0 :
+ wc.dlid_path_bits = dlid >= IPATH_MULTICAST_LID_BASE ? 0 :
dlid & ((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
/* Signal completion event if the solicited bit is set. */
ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
@@ -37,7 +37,7 @@
#include "ipath_kernel.h"
#include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
/* Not static, because we don't want the compiler removing it */
const char ipath_verbs_version[] = "ipath_verbs " IPATH_IDSTR;
@@ -429,7 +429,7 @@ static void ipath_ib_rcv(void *arg, void
/* Check for a valid destination LID (see ch. 7.11.1). */
lid = be16_to_cpu(hdr->lrh[1]);
- if (lid < IPS_MULTICAST_LID_BASE) {
+ if (lid < IPATH_MULTICAST_LID_BASE) {
lid &= ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
if (unlikely(lid != ipath_layer_get_lid(dev->dd))) {
dev->rcv_errors++;
@@ -439,9 +439,9 @@ static void ipath_ib_rcv(void *arg, void
/* Check for GRH */
lnh = be16_to_cpu(hdr->lrh[0]) & 3;
- if (lnh == IPS_LRH_BTH)
+ if (lnh == IPATH_LRH_BTH)
ohdr = &hdr->u.oth;
- else if (lnh == IPS_LRH_GRH)
+ else if (lnh == IPATH_LRH_GRH)
ohdr = &hdr->u.l.oth;
else {
dev->rcv_errors++;
@@ -453,8 +453,8 @@ static void ipath_ib_rcv(void *arg, void
dev->opstats[opcode].n_packets++;
/* Get the destination QP number. */
- qp_num = be32_to_cpu(ohdr->bth[1]) & IPS_QPN_MASK;
- if (qp_num == IPS_MULTICAST_QPN) {
+ qp_num = be32_to_cpu(ohdr->bth[1]) & IPATH_QPN_MASK;
+ if (qp_num == IPATH_MULTICAST_QPN) {
struct ipath_mcast *mcast;
struct ipath_mcast_qp *p;
@@ -465,7 +465,7 @@ static void ipath_ib_rcv(void *arg, void
}
dev->n_multicast_rcv++;
list_for_each_entry_rcu(p, &mcast->qp_list, list)
- ipath_qp_rcv(dev, hdr, lnh == IPS_LRH_GRH, data,
+ ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
tlen, p->qp);
/*
* Notify ipath_multicast_detach() if it is waiting for us
@@ -477,7 +477,7 @@ static void ipath_ib_rcv(void *arg, void
qp = ipath_lookup_qpn(&dev->qp_table, qp_num);
if (qp) {
dev->n_unicast_rcv++;
- ipath_qp_rcv(dev, hdr, lnh == IPS_LRH_GRH, data,
+ ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
tlen, qp);
/*
* Notify ipath_destroy_qp() if it is waiting
@@ -860,8 +860,8 @@ static struct ib_ah *ipath_create_ah(str
}
/* A multicast address requires a GRH (see ch. 8.4.1). */
- if (ah_attr->dlid >= IPS_MULTICAST_LID_BASE &&
- ah_attr->dlid != IPS_PERMISSIVE_LID &&
+ if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
+ ah_attr->dlid != IPATH_PERMISSIVE_LID &&
!(ah_attr->ah_flags & IB_AH_GRH)) {
ret = ERR_PTR(-EINVAL);
goto bail;
diff -r 31c382d8210a -r 2a721e1f490b drivers/infiniband/hw/ipath/ips_common.h
--- a/drivers/infiniband/hw/ipath/ips_common.h Thu Jun 29 14:33:26 2006 -0700
+++ /dev/null Thu Jan 01 00:00:00 1970 +0000
@@ -1,264 +0,0 @@
-#ifndef IPS_COMMON_H
-#define IPS_COMMON_H
-/*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
- * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
- *
- * This software is available to you under a choice of one of two
- * licenses. You may choose to be licensed under the terms of the GNU
- * General Public License (GPL) Version 2, available from the file
- * COPYING in the main directory of this source tree, or the
- * OpenIB.org BSD license below:
- *
- * Redistribution and use in source and binary forms, with or
- * without modification, are permitted provided that the following
- * conditions are met:
- *
- * - Redistributions of source code must retain the above
- * copyright notice, this list of conditions and the following
- * disclaimer.
- *
- * - Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following
- * disclaimer in the documentation and/or other materials
- * provided with the distribution.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
- * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#include "ipath_common.h"
-
-struct ipath_header {
- /*
- * Version - 4 bits, Port - 4 bits, TID - 10 bits and Offset -
- * 14 bits before ECO change ~28 Dec 03. After that, Vers 4,
- * Port 3, TID 11, offset 14.
- */
- __le32 ver_port_tid_offset;
- __le16 chksum;
- __le16 pkt_flags;
-};
-
-struct ips_message_header {
- __be16 lrh[4];
- __be32 bth[3];
- /* fields below this point are in host byte order */
- struct ipath_header iph;
- __u8 sub_opcode;
- __u8 flags;
- __u16 src_rank;
- /* 24 bits. The upper 8 bit is available for other use */
- union {
- struct {
- unsigned ack_seq_num:24;
- unsigned port:4;
- unsigned unused:4;
- };
- __u32 ack_seq_num_org;
- };
- __u8 expected_tid_session_id;
- __u8 tinylen; /* to aid MPI */
- union {
- __u16 tag; /* to aid MPI */
- __u16 mqhdr; /* for PSM MQ */
- };
- union {
- __u32 mpi[4]; /* to aid MPI */
- __u32 data[4];
- __u64 mq[2]; /* for PSM MQ */
- struct {
- __u16 mtu;
- __u8 major_ver;
- __u8 minor_ver;
- __u32 not_used; //free
- __u32 run_id;
- __u32 client_ver;
- };
- };
-};
-
-struct ether_header {
- __be16 lrh[4];
- __be32 bth[3];
- struct ipath_header iph;
- __u8 sub_opcode;
- __u8 cmd;
- __be16 lid;
- __u16 mac[3];
- __u8 frag_num;
- __u8 seq_num;
- __le32 len;
- /* MUST be of word size due to PIO write requirements */
- __le32 csum;
- __le16 csum_offset;
- __le16 flags;
- __u16 first_2_bytes;
- __u8 unused[2]; /* currently unused */
-};
-
-/*
- * The PIO buffer used for sending infinipath messages must only be written
- * in 32-bit words, all the data must be written, and no writes can occur
- * after the last word is written (which transfers "ownership" of the buffer
- * to the chip and triggers the message to be sent).
- * Since the Linux sk_buff structure can be recursive, non-aligned, and
- * any number of bytes in each segment, we use the following structure
- * to keep information about the overall state of the copy operation.
- * This is used to save the information needed to store the checksum
- * in the right place before sending the last word to the hardware and
- * to buffer the last 0-3 bytes of non-word sized segments.
- */
-struct copy_data_s {
- struct ether_header *hdr;
- /* addr of PIO buf to write csum to */
- __u32 __iomem *csum_pio;
- __u32 __iomem *to; /* addr of PIO buf to write data to */
- __u32 device; /* which device to allocate PIO bufs from */
- __s32 error; /* set if there is an error. */
- __s32 extra; /* amount of data saved in u.buf below */
- __u32 len; /* total length to send in bytes */
- __u32 flen; /* frament length in words */
- __u32 csum; /* partial IP checksum */
- __u32 pos; /* position for partial checksum */
- __u32 offset; /* offset to where data currently starts */
- __s32 checksum_calc; /* set to 1 when csum has been calculated */
- struct sk_buff *skb;
- union {
- __u32 w;
- __u8 buf[4];
- } u;
-};
-
-/* IB - LRH header consts */
-#define IPS_LRH_GRH 0x0003 /* 1. word of IB LRH - next header: GRH */
-#define IPS_LRH_BTH 0x0002 /* 1. word of IB LRH - next header: BTH */
-
-#define IPS_OFFSET 0
-
-/*
- * defines the cut-off point between the header queue and eager/expected
- * TID queue
- */
-#define NUM_OF_EXTRA_WORDS_IN_HEADER_QUEUE \
- ((sizeof(struct ips_message_header) - \
- offsetof(struct ips_message_header, iph)) >> 2)
-
-/* OpCodes */
-#define OPCODE_IPS 0xC0
-#define OPCODE_ITH4X 0xC1
-
-/* OpCode 30 is use by stand-alone test programs */
-#define OPCODE_RAW_DATA 0xDE
-/* last OpCode (31) is reserved for test */
-#define OPCODE_TEST 0xDF
-
-/* sub OpCodes - ips */
-#define OPCODE_SEQ_DATA 0x01
-#define OPCODE_SEQ_CTRL 0x02
-
-#define OPCODE_SEQ_MQ_DATA 0x03
-#define OPCODE_SEQ_MQ_CTRL 0x04
-
-#define OPCODE_ACK 0x10
-#define OPCODE_NAK 0x11
-
-#define OPCODE_ERR_CHK 0x20
-#define OPCODE_ERR_CHK_PLS 0x21
-
-#define OPCODE_STARTUP 0x30
-#define OPCODE_STARTUP_ACK 0x31
-#define OPCODE_STARTUP_NAK 0x32
-
-#define OPCODE_STARTUP_EXT 0x34
-#define OPCODE_STARTUP_ACK_EXT 0x35
-#define OPCODE_STARTUP_NAK_EXT 0x36
-
-#define OPCODE_TIDS_RELEASE 0x40
-#define OPCODE_TIDS_RELEASE_CONFIRM 0x41
-
-#define OPCODE_CLOSE 0x50
-#define OPCODE_CLOSE_ACK 0x51
-/*
- * like OPCODE_CLOSE, but no complaint if other side has already closed.
- * Used when doing abort(), MPI_Abort(), etc.
- */
-#define OPCODE_ABORT 0x52
-
-/* sub OpCodes - ith4x */
-#define OPCODE_ENCAP 0x81
-#define OPCODE_LID_ARP 0x82
-
-/* Receive Header Queue: receive type (from infinipath) */
-#define RCVHQ_RCV_TYPE_EXPECTED 0
-#define RCVHQ_RCV_TYPE_EAGER 1
-#define RCVHQ_RCV_TYPE_NON_KD 2
-#define RCVHQ_RCV_TYPE_ERROR 3
-
-/* misc. */
-#define SIZE_OF_CRC 1
-
-#define EAGER_TID_ID INFINIPATH_I_TID_MASK
-
-#define IPS_DEFAULT_P_KEY 0xFFFF
-
-#define IPS_PERMISSIVE_LID 0xFFFF
-#define IPS_MULTICAST_LID_BASE 0xC000
-
-#define IPS_AETH_CREDIT_SHIFT 24
-#define IPS_AETH_CREDIT_MASK 0x1F
-#define IPS_AETH_CREDIT_INVAL 0x1F
-
-#define IPS_PSN_MASK 0xFFFFFF
-#define IPS_MSN_MASK 0xFFFFFF
-#define IPS_QPN_MASK 0xFFFFFF
-#define IPS_MULTICAST_QPN 0xFFFFFF
-
-/* functions for extracting fields from rcvhdrq entries */
-static inline __u32 ips_get_hdr_err_flags(const __le32 * rbuf)
-{
- return __le32_to_cpu(rbuf[1]);
-}
-
-static inline __u32 ips_get_index(const __le32 * rbuf)
-{
- return (__le32_to_cpu(rbuf[0]) >> INFINIPATH_RHF_EGRINDEX_SHIFT)
- & INFINIPATH_RHF_EGRINDEX_MASK;
-}
-
-static inline __u32 ips_get_rcv_type(const __le32 * rbuf)
-{
- return (__le32_to_cpu(rbuf[0]) >> INFINIPATH_RHF_RCVTYPE_SHIFT)
- & INFINIPATH_RHF_RCVTYPE_MASK;
-}
-
-static inline __u32 ips_get_length_in_bytes(const __le32 * rbuf)
-{
- return ((__le32_to_cpu(rbuf[0]) >> INFINIPATH_RHF_LENGTH_SHIFT)
- & INFINIPATH_RHF_LENGTH_MASK) << 2;
-}
-
-static inline void *ips_get_first_protocol_header(const __u32 * rbuf)
-{
- return (void *)&rbuf[2];
-}
-
-static inline struct ips_message_header *ips_get_ips_header(const __u32 *
- rbuf)
-{
- return (struct ips_message_header *)&rbuf[2];
-}
-
-static inline __u32 ips_get_ipath_ver(__le32 hdrword)
-{
- return (__le32_to_cpu(hdrword) >> INFINIPATH_I_VERS_SHIFT)
- & INFINIPATH_I_VERS_MASK;
-}
-
-#endif /* IPS_COMMON_H */
^ permalink raw reply [flat|nested] 65+ messages in thread* [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (36 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 37 of 39] IB/ipath - namespace cleanup: replace ips with ipath Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:53 ` David Miller
2006-07-06 22:37 ` Benjamin Herrenschmidt
2006-06-29 21:41 ` [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss Bryan O'Sullivan
2006-06-30 16:31 ` [PATCH 0 of 39] ipath - bug fixes, performance enhancements,and portability improvements Michael S. Tsirkin
39 siblings, 2 replies; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
Ordering of writethrough store buffers needs to be forced, and we need
to use ifdef to get writethrough behavior to InfiniPath buffers, because
there is no generic way to specify that at this time (similar to code
in char/drm/drm_vm.c and block/z2ram.c).
Signed-off-by: John Gregor <john.gregor@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r 2a721e1f490b -r c22b6c244d5d drivers/infiniband/hw/ipath/Makefile
--- a/drivers/infiniband/hw/ipath/Makefile Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/Makefile Thu Jun 29 14:33:26 2006 -0700
@@ -20,6 +20,7 @@ ipath_core-y := \
ipath_user_pages.o
ipath_core-$(CONFIG_X86_64) += ipath_wc_x86_64.o
+ipath_core-$(CONFIG_PPC64) += ipath_wc_ppc64.o
ib_ipath-y := \
ipath_cq.o \
diff -r 2a721e1f490b -r c22b6c244d5d drivers/infiniband/hw/ipath/ipath_driver.c
--- a/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c Thu Jun 29 14:33:26 2006 -0700
@@ -440,7 +440,13 @@ static int __devinit ipath_init_one(stru
}
dd->ipath_pcirev = rev;
+#if defined(__powerpc__)
+ /* There isn't a generic way to specify writethrough mappings */
+ dd->ipath_kregbase = __ioremap(addr, len,
+ (_PAGE_NO_CACHE|_PAGE_WRITETHRU));
+#else
dd->ipath_kregbase = ioremap_nocache(addr, len);
+#endif
if (!dd->ipath_kregbase) {
ipath_dbg("Unable to map io addr %llx to kvirt, failing\n",
diff -r 2a721e1f490b -r c22b6c244d5d drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c Thu Jun 29 14:33:26 2006 -0700
@@ -985,6 +985,13 @@ static int mmap_piobufs(struct vm_area_s
* write combining behavior we want on the PIO buffers!
*/
+#if defined(__powerpc__)
+ /* There isn't a generic way to specify writethrough mappings */
+ pgprot_val(vma->vm_page_prot) |= _PAGE_NO_CACHE;
+ pgprot_val(vma->vm_page_prot) |= _PAGE_WRITETHRU;
+ pgprot_val(vma->vm_page_prot) &= ~_PAGE_GUARDED;
+#endif
+
if (vma->vm_flags & VM_READ) {
dev_info(&dd->pcidev->dev,
"Can't map piobufs as readable (flags=%lx)\n",
diff -r 2a721e1f490b -r c22b6c244d5d drivers/infiniband/hw/ipath/ipath_wc_ppc64.c
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/infiniband/hw/ipath/ipath_wc_ppc64.c Thu Jun 29 14:33:26 2006 -0700
@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file is conditionally built on PowerPC only. Otherwise weak symbol
+ * versions of the functions exported from here are used.
+ */
+
+#include "ipath_kernel.h"
+
+/**
+ * ipath_unordered_wc - indicate whether write combining is ordered
+ *
+ * PowerPC systems (at least those in the 970 processor family)
+ * write partially filled store buffers in address order, but will write
+ * completely filled store buffers in "random" order, and therefore must
+ * have serialization for correctness with current InfiniPath chips.
+ *
+ */
+int ipath_unordered_wc(void)
+{
+ return 1;
+}
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-06-29 21:41 ` [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems Bryan O'Sullivan
@ 2006-06-29 21:53 ` David Miller
2006-06-29 22:01 ` Bryan O'Sullivan
2006-07-06 22:37 ` Benjamin Herrenschmidt
1 sibling, 1 reply; 65+ messages in thread
From: David Miller @ 2006-06-29 21:53 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Bryan O'Sullivan <bos@pathscale.com>
Date: Thu, 29 Jun 2006 14:41:29 -0700
> ipath_core-$(CONFIG_X86_64) += ipath_wc_x86_64.o
> +ipath_core-$(CONFIG_PPC64) += ipath_wc_ppc64.o
Again, don't put these kinds of cpu specific functions
into the infiniband driver. They are potentially globally
useful, not something only Infiniband might want to do.
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-06-29 21:53 ` David Miller
@ 2006-06-29 22:01 ` Bryan O'Sullivan
2006-06-29 22:04 ` David Miller
0 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 22:01 UTC (permalink / raw)
To: David Miller; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
On Thu, 2006-06-29 at 14:53 -0700, David Miller wrote:
> From: Bryan O'Sullivan <bos@pathscale.com>
> Date: Thu, 29 Jun 2006 14:41:29 -0700
>
> > ipath_core-$(CONFIG_X86_64) += ipath_wc_x86_64.o
> > +ipath_core-$(CONFIG_PPC64) += ipath_wc_ppc64.o
>
> Again, don't put these kinds of cpu specific functions
> into the infiniband driver. They are potentially globally
> useful, not something only Infiniband might want to do.
The support for write combining in the kernel is not in a state where
that makes any sense at the moment. Also, this is a single-statement
function.
<b
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-06-29 22:01 ` Bryan O'Sullivan
@ 2006-06-29 22:04 ` David Miller
2006-07-03 22:25 ` Anton Blanchard
0 siblings, 1 reply; 65+ messages in thread
From: David Miller @ 2006-06-29 22:04 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Bryan O'Sullivan <bos@pathscale.com>
Date: Thu, 29 Jun 2006 15:01:39 -0700
> The support for write combining in the kernel is not in a state where
> that makes any sense at the moment.
Please fix the generic code if it doesn't provide the facility
you need at the moment. Don't shoe horn it into your driver
just to make up for that.
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-06-29 22:04 ` David Miller
@ 2006-07-03 22:25 ` Anton Blanchard
2006-07-04 4:55 ` Michael S. Tsirkin
0 siblings, 1 reply; 65+ messages in thread
From: Anton Blanchard @ 2006-07-03 22:25 UTC (permalink / raw)
To: David Miller
Cc: bos, akpm, rdreier, mst, openib-general, linux-kernel, netdev,
matthew
Hi,
> Please fix the generic code if it doesn't provide the facility
> you need at the moment. Don't shoe horn it into your driver
> just to make up for that.
Ive had 3 drivers asking for write combining recently so I agree this is
a good idea. How about ioremap_wc as suggested by Willy:
http://marc.theaimsgroup.com/?l=linux-kernel&m=114374741828040&w=2
Anton
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-07-03 22:25 ` Anton Blanchard
@ 2006-07-04 4:55 ` Michael S. Tsirkin
0 siblings, 0 replies; 65+ messages in thread
From: Michael S. Tsirkin @ 2006-07-04 4:55 UTC (permalink / raw)
To: Anton Blanchard
Cc: eli, David Miller, bos, akpm, Roland Dreier, openib-general,
linux-kernel, netdev, matthew, ak, ak, vojtech
Quoting r. Anton Blanchard <anton@samba.org>:
> Subject: Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
>
>
> Hi,
>
> > Please fix the generic code if it doesn't provide the facility
> > you need at the moment. Don't shoe horn it into your driver
> > just to make up for that.
>
> Ive had 3 drivers asking for write combining recently so I agree this is
> a good idea. How about ioremap_wc as suggested by Willy:
>
> http://marc.theaimsgroup.com/?l=linux-kernel&m=114374741828040&w=2
Hmm ... I think ioremap_wc isn't sufficient by itself to make drivers using
write-combined memory, portable. Here's another thread on a related subject:
http://lkml.org/lkml/2006/2/24/347
--
MST
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-06-29 21:41 ` [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems Bryan O'Sullivan
2006-06-29 21:53 ` David Miller
@ 2006-07-06 22:37 ` Benjamin Herrenschmidt
2006-07-06 23:34 ` Bryan O'Sullivan
1 sibling, 1 reply; 65+ messages in thread
From: Benjamin Herrenschmidt @ 2006-07-06 22:37 UTC (permalink / raw)
To: Bryan O'Sullivan
Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
> +#if defined(__powerpc__)
> + /* There isn't a generic way to specify writethrough mappings */
> + pgprot_val(vma->vm_page_prot) |= _PAGE_NO_CACHE;
> + pgprot_val(vma->vm_page_prot) |= _PAGE_WRITETHRU;
> + pgprot_val(vma->vm_page_prot) &= ~_PAGE_GUARDED;
> +#endif
I don't see any case where having both NO_CACHE and WRITE_THRU can be
legal... It's one or the other.
> +/**
> + * ipath_unordered_wc - indicate whether write combining is ordered
> + *
> + * PowerPC systems (at least those in the 970 processor family)
> + * write partially filled store buffers in address order, but will write
> + * completely filled store buffers in "random" order, and therefore must
> + * have serialization for correctness with current InfiniPath chips.
> + *
> + */
> +int ipath_unordered_wc(void)
> +{
> + return 1;
> +}
How is the above providing any kind of serialisation ?
Ben.
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-07-06 22:37 ` Benjamin Herrenschmidt
@ 2006-07-06 23:34 ` Bryan O'Sullivan
2006-07-06 23:39 ` Benjamin Herrenschmidt
0 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-07-06 23:34 UTC (permalink / raw)
To: Benjamin Herrenschmidt
Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
On Fri, 2006-07-07 at 08:37 +1000, Benjamin Herrenschmidt wrote:
> > +int ipath_unordered_wc(void)
> > +{
> > + return 1;
> > +}
>
> How is the above providing any kind of serialisation ?
It's not intended to; it tells the *caller* whether to do it.
<b
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
2006-07-06 23:34 ` Bryan O'Sullivan
@ 2006-07-06 23:39 ` Benjamin Herrenschmidt
0 siblings, 0 replies; 65+ messages in thread
From: Benjamin Herrenschmidt @ 2006-07-06 23:39 UTC (permalink / raw)
To: Bryan O'Sullivan
Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
On Thu, 2006-07-06 at 16:34 -0700, Bryan O'Sullivan wrote:
> On Fri, 2006-07-07 at 08:37 +1000, Benjamin Herrenschmidt wrote:
>
> > > +int ipath_unordered_wc(void)
> > > +{
> > > + return 1;
> > > +}
> >
> > How is the above providing any kind of serialisation ?
>
> It's not intended to; it tells the *caller* whether to do it.
Ah ok. What barrier do you use for that ?
Ben.
^ permalink raw reply [flat|nested] 65+ messages in thread
* [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (37 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems Bryan O'Sullivan
@ 2006-06-29 21:41 ` Bryan O'Sullivan
2006-06-29 21:50 ` David Miller
2006-06-30 16:31 ` [PATCH 0 of 39] ipath - bug fixes, performance enhancements,and portability improvements Michael S. Tsirkin
39 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In cases where a large incoming RDMA is being received, we have to
copy data inside the interrupt handler before we can ACK each packet.
The source is DMAed to by the hardware, which means that the CPU won't
have it cached. We only read the source this one time; using normal load
instructions pollutes the dcache with useless data, reducing performance
to the point where we can lose a significant number of packets.
Using a (memcpy-compatible) copy routine that loads with streaming
instructions, we try to not fill the dcache with useless data. Avoiding
the cache refill penalty lets us keep up better with the sender, resulting
in many fewer dropped packets. We use normal stores to the destination,
because the copied-to data will be used soon after the interrupt handler
completes.
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
diff -r c22b6c244d5d -r 1b00209ef20a drivers/infiniband/hw/ipath/Makefile
--- a/drivers/infiniband/hw/ipath/Makefile Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/Makefile Thu Jun 29 14:33:26 2006 -0700
@@ -35,3 +35,5 @@ ib_ipath-y := \
ipath_ud.o \
ipath_verbs.o \
ipath_verbs_mcast.o
+
+ib_ipath-$(CONFIG_X86_64) += ipath_memcpy_x86_64.o
diff -r c22b6c244d5d -r 1b00209ef20a drivers/infiniband/hw/ipath/ipath_verbs.c
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c Thu Jun 29 14:33:26 2006 -0700
@@ -159,7 +159,7 @@ void ipath_copy_sge(struct ipath_sge_sta
BUG_ON(len == 0);
if (len > length)
len = length;
- memcpy(sge->vaddr, data, len);
+ ipath_memcpy_nc(sge->vaddr, data, len);
sge->vaddr += len;
sge->length -= len;
sge->sge_length -= len;
diff -r c22b6c244d5d -r 1b00209ef20a drivers/infiniband/hw/ipath/ipath_verbs.h
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:26 2006 -0700
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h Thu Jun 29 14:33:26 2006 -0700
@@ -728,4 +728,14 @@ extern unsigned int ib_ipath_max_srq_wrs
extern const u32 ib_ipath_rnr_table[];
+/*
+ * Copy data. Try not to pollute the dcache with the source data,
+ * because we won't be reading it again.
+ */
+#if defined(CONFIG_X86_64)
+void *ipath_memcpy_nc(void *dest, const void *src, size_t n);
+#else
+#define ipath_memcpy_nc(dest, src, n) memcpy(dest, src, n)
+#endif
+
#endif /* IPATH_VERBS_H */
diff -r c22b6c244d5d -r 1b00209ef20a drivers/infiniband/hw/ipath/ipath_memcpy_x86_64.S
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/infiniband/hw/ipath/ipath_memcpy_x86_64.S Thu Jun 29 14:33:26 2006 -0700
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * ipath_memcpy_nc - memcpy-compatible copy routine, using streaming loads
+ * @dest: destination address
+ * @src: source address
+ * @count: number of bytes to copy
+ *
+ * Use streaming loads and normal stores for a special-case copy where
+ * we know we won't be reading the source again, but will be reading the
+ * destination again soon.
+ */
+ .text
+ .p2align 4,,15
+ /* rdi destination, rsi source, rdx count */
+ .globl ipath_memcpy_nc
+ .type ipath_memcpy_nc, @function
+ipath_memcpy_nc:
+ movq %rdi, %rax
+.L5:
+ cmpq $15, %rdx
+ ja .L34
+.L3:
+ cmpl $8, %edx /* rdx is 0..15 */
+ jbe .L9
+.L6:
+ testb $8, %dxl /* rdx is 3,5,6,7,9..15 */
+ je .L13
+ movq (%rsi), %rcx
+ addq $8, %rsi
+ movq %rcx, (%rdi)
+ addq $8, %rdi
+.L13:
+ testb $4, %dxl
+ je .L15
+ movl (%rsi), %ecx
+ addq $4, %rsi
+ movl %ecx, (%rdi)
+ addq $4, %rdi
+.L15:
+ testb $2, %dxl
+ je .L17
+ movzwl (%rsi), %ecx
+ addq $2, %rsi
+ movw %cx, (%rdi)
+ addq $2, %rdi
+.L17:
+ testb $1, %dxl
+ je .L33
+.L1:
+ movzbl (%rsi), %ecx
+ movb %cl, (%rdi)
+.L33:
+ ret
+.L34:
+ cmpq $63, %rdx /* rdx is > 15 */
+ ja .L64
+ movl $16, %ecx /* rdx is 16..63 */
+.L25:
+ movq 8(%rsi), %r8
+ movq (%rsi), %r9
+ addq %rcx, %rsi
+ movq %r8, 8(%rdi)
+ movq %r9, (%rdi)
+ addq %rcx, %rdi
+ subq %rcx, %rdx
+ cmpl %edx, %ecx /* is rdx >= 16? */
+ jbe .L25
+ jmp .L3 /* rdx is 0..15 */
+ .p2align 4,,7
+.L64:
+ movl $64, %ecx
+.L42:
+ prefetchnta 128(%rsi)
+ movq (%rsi), %r8
+ movq 8(%rsi), %r9
+ movq 16(%rsi), %r10
+ movq 24(%rsi), %r11
+ subq %rcx, %rdx
+ movq %r8, (%rdi)
+ movq 32(%rsi), %r8
+ movq %r9, 8(%rdi)
+ movq 40(%rsi), %r9
+ movq %r10, 16(%rdi)
+ movq 48(%rsi), %r10
+ movq %r11, 24(%rdi)
+ movq 56(%rsi), %r11
+ addq %rcx, %rsi
+ movq %r8, 32(%rdi)
+ movq %r9, 40(%rdi)
+ movq %r10, 48(%rdi)
+ movq %r11, 56(%rdi)
+ addq %rcx, %rdi
+ cmpq %rdx, %rcx /* is rdx >= 64? */
+ jbe .L42
+ sfence
+ orl %edx, %edx
+ je .L33
+ jmp .L5
+.L9:
+ jmp *.L12(,%rdx,8) /* rdx is 0..8 */
+ .section .rodata
+ .align 8
+ .align 4
+.L12:
+ .quad .L33
+ .quad .L1
+ .quad .L2
+ .quad .L6
+ .quad .L4
+ .quad .L6
+ .quad .L6
+ .quad .L6
+ .quad .L8
+ .text
+.L2:
+ movzwl (%rsi), %ecx
+ movw %cx, (%rdi)
+ ret
+.L4:
+ movl (%rsi), %ecx
+ movl %ecx, (%rdi)
+ ret
+.L8:
+ movq (%rsi), %rcx
+ movq %rcx, (%rdi)
+ ret
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 21:41 ` [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss Bryan O'Sullivan
@ 2006-06-29 21:50 ` David Miller
2006-06-29 21:59 ` Bryan O'Sullivan
0 siblings, 1 reply; 65+ messages in thread
From: David Miller @ 2006-06-29 21:50 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Bryan O'Sullivan <bos@pathscale.com>
Date: Thu, 29 Jun 2006 14:41:30 -0700
> +/*
> + * Copy data. Try not to pollute the dcache with the source data,
> + * because we won't be reading it again.
> + */
> +#if defined(CONFIG_X86_64)
> +void *ipath_memcpy_nc(void *dest, const void *src, size_t n);
> +#else
> +#define ipath_memcpy_nc(dest, src, n) memcpy(dest, src, n)
> +#endif
A facility like this doesn't belong in some arbitrary driver layer.
It belongs as a generic facility the whole kernel could make use
of.
Please stop polluting the infiniband drivers with Opteron crap.
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 21:50 ` David Miller
@ 2006-06-29 21:59 ` Bryan O'Sullivan
2006-06-29 22:03 ` David Miller
0 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 21:59 UTC (permalink / raw)
To: David Miller; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
On Thu, 2006-06-29 at 14:50 -0700, David Miller wrote:
> A facility like this doesn't belong in some arbitrary driver layer.
> It belongs as a generic facility the whole kernel could make use
> of.
It could, indeed. In fact, we had that discussion here before I sent
this patch in. It presumably wants to live in lib/, and acquire a more
generic name. What name will capture the uncached-read-but-cached-write
semantics in a useful fashion? memcpy_nc?
<b
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 21:59 ` Bryan O'Sullivan
@ 2006-06-29 22:03 ` David Miller
2006-06-29 23:34 ` Bryan O'Sullivan
0 siblings, 1 reply; 65+ messages in thread
From: David Miller @ 2006-06-29 22:03 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Bryan O'Sullivan <bos@pathscale.com>
Date: Thu, 29 Jun 2006 14:59:37 -0700
> It could, indeed. In fact, we had that discussion here before I sent
> this patch in. It presumably wants to live in lib/, and acquire a more
> generic name. What name will capture the uncached-read-but-cached-write
> semantics in a useful fashion? memcpy_nc?
I'm not good with names :-)
Note that there also might be cases where using such a memcpy
variant might be the wrong thing to do. For example, for a very
tightly coupled CMT cpu implementation which has the memory controller,
L2 cache, PCI controller, etc. all on the same die and the PCI controller
makes use of the L2 cache just like the cpu threads do, using this
kind of memcpy would always be the wrong thing to do.
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 22:03 ` David Miller
@ 2006-06-29 23:34 ` Bryan O'Sullivan
2006-06-29 23:46 ` David Miller
0 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-29 23:34 UTC (permalink / raw)
To: David Miller; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
On Thu, 2006-06-29 at 15:03 -0700, David Miller wrote:
> I'm not good with names :-)
Heh. I'll call it memcpy_nc for now, then, and people can retch all
over the name as they please when I submit a more suitably generic
patch.
> Note that there also might be cases where using such a memcpy
> variant might be the wrong thing to do. For example, for a very
> tightly coupled CMT cpu implementation which has the memory controller,
> L2 cache, PCI controller, etc. all on the same die and the PCI controller
> makes use of the L2 cache just like the cpu threads do, using this
> kind of memcpy would always be the wrong thing to do.
I'm not quite following you, though I assume you're referring to Niagara
or Rock :-) Are you saying a memcpy_nc would do worse than plain
memcpy, or worse than some other memcpy-like routine?
<b
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 23:34 ` Bryan O'Sullivan
@ 2006-06-29 23:46 ` David Miller
2006-06-29 23:55 ` [openib-general] " Ralph Campbell
2006-06-30 0:28 ` Rick Jones
0 siblings, 2 replies; 65+ messages in thread
From: David Miller @ 2006-06-29 23:46 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Bryan O'Sullivan <bos@pathscale.com>
Date: Thu, 29 Jun 2006 16:34:23 -0700
> I'm not quite following you, though I assume you're referring to Niagara
> or Rock :-) Are you saying a memcpy_nc would do worse than plain
> memcpy, or worse than some other memcpy-like routine?
It would do worse than memcpy.
If you bypass the L2 cache, it's pointless because the next
agent (PCI controller, CPU thread, etc.) is going to need the
data in the L2 cache.
It's better in that kind of setup to eat the L2 cache miss overhead in
memcpy since memcpy can usually prefetch and store buffer in order to
absorb some of the L2 miss costs.
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [openib-general] [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 23:46 ` David Miller
@ 2006-06-29 23:55 ` Ralph Campbell
2006-06-30 0:28 ` Rick Jones
1 sibling, 0 replies; 65+ messages in thread
From: Ralph Campbell @ 2006-06-29 23:55 UTC (permalink / raw)
To: David Miller; +Cc: bos, akpm, netdev, rdreier, linux-kernel, openib-general
This is intended to be an architecture specific function
so if the CPU does support HW dma to the CPU's L2 cache, the
architecture specific version of memcpy_nc() would not replace
the default definition which maps memcpy_nc() to memcpy().
For CPUs like the vast majority currently available, there
is a performance benefit by not reading data into the cache
that won't be read a second time.
On Thu, 2006-06-29 at 16:46 -0700, David Miller wrote:
> From: Bryan O'Sullivan <bos@pathscale.com>
> Date: Thu, 29 Jun 2006 16:34:23 -0700
>
> > I'm not quite following you, though I assume you're referring to Niagara
> > or Rock :-) Are you saying a memcpy_nc would do worse than plain
> > memcpy, or worse than some other memcpy-like routine?
>
> It would do worse than memcpy.
>
> If you bypass the L2 cache, it's pointless because the next
> agent (PCI controller, CPU thread, etc.) is going to need the
> data in the L2 cache.
>
> It's better in that kind of setup to eat the L2 cache miss overhead in
> memcpy since memcpy can usually prefetch and store buffer in order to
> absorb some of the L2 miss costs.
>
> _______________________________________________
> openib-general mailing list
> openib-general@openib.org
> http://openib.org/mailman/listinfo/openib-general
>
> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
>
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-29 23:46 ` David Miller
2006-06-29 23:55 ` [openib-general] " Ralph Campbell
@ 2006-06-30 0:28 ` Rick Jones
2006-06-30 0:32 ` David Miller
1 sibling, 1 reply; 65+ messages in thread
From: Rick Jones @ 2006-06-30 0:28 UTC (permalink / raw)
To: David Miller
Cc: bos, akpm, rdreier, mst, openib-general, linux-kernel, netdev
> If you bypass the L2 cache, it's pointless because the next
> agent (PCI controller, CPU thread, etc.) is going to need the
> data in the L2 cache.
>
> It's better in that kind of setup to eat the L2 cache miss overhead in
> memcpy since memcpy can usually prefetch and store buffer in order to
> absorb some of the L2 miss costs.
I thought that most PCI controllers (that is to say the things bridging
PCI to the rest of the system) could do prefetching and/or that PCI-X
(if not PCI, no idea about PCI-e) cards could issue multiple
transactions anyway?
rick jones
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-30 0:28 ` Rick Jones
@ 2006-06-30 0:32 ` David Miller
2006-06-30 0:44 ` Rick Jones
0 siblings, 1 reply; 65+ messages in thread
From: David Miller @ 2006-06-30 0:32 UTC (permalink / raw)
To: rick.jones2; +Cc: bos, akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Rick Jones <rick.jones2@hp.com>
Date: Thu, 29 Jun 2006 17:28:50 -0700
> I thought that most PCI controllers (that is to say the things bridging
> PCI to the rest of the system) could do prefetching and/or that PCI-X
> (if not PCI, no idea about PCI-e) cards could issue multiple
> transactions anyway?
People doing deep CMT chips have found out that all of that
prefetching and store buffering is unnecessary when everything is so
tightly integrated.
All of the previous UltraSPARC boxes before Niagara had a
streaming cache sitting on the PCI controller. It basically
prefetched for reads and collected writes from PCI devices
into cacheline sized chunks.
The PCI controller in the current Niagara systems has none of that
stuff.
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-30 0:32 ` David Miller
@ 2006-06-30 0:44 ` Rick Jones
2006-06-30 0:47 ` David Miller
0 siblings, 1 reply; 65+ messages in thread
From: Rick Jones @ 2006-06-30 0:44 UTC (permalink / raw)
To: David Miller
Cc: bos, akpm, rdreier, mst, openib-general, linux-kernel, netdev
David Miller wrote:
> From: Rick Jones <rick.jones2@hp.com>
> Date: Thu, 29 Jun 2006 17:28:50 -0700
>
>
>>I thought that most PCI controllers (that is to say the things bridging
>>PCI to the rest of the system) could do prefetching and/or that PCI-X
>>(if not PCI, no idea about PCI-e) cards could issue multiple
>>transactions anyway?
>
>
> People doing deep CMT chips have found out that all of that
> prefetching and store buffering is unnecessary when everything is so
> tightly integrated.
Then is prefetching in memcpy really that important to them (BTW besides
Sun/Niagra who are doing "deep CMT"?)
> All of the previous UltraSPARC boxes before Niagara had a
> streaming cache sitting on the PCI controller. It basically
> prefetched for reads and collected writes from PCI devices
> into cacheline sized chunks.
>
> The PCI controller in the current Niagara systems has none of that
> stuff.
Relying on PCI-X devices to issue multiple requests then?
rick jones
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
2006-06-30 0:44 ` Rick Jones
@ 2006-06-30 0:47 ` David Miller
0 siblings, 0 replies; 65+ messages in thread
From: David Miller @ 2006-06-30 0:47 UTC (permalink / raw)
To: rick.jones2; +Cc: bos, akpm, rdreier, mst, openib-general, linux-kernel, netdev
From: Rick Jones <rick.jones2@hp.com>
Date: Thu, 29 Jun 2006 17:44:05 -0700
> Then is prefetching in memcpy really that important to them.
Not really, the thread just blocks while waiting for memory.
On stores they do a cacheline fill optimization similar to
the powerpc.
> Relying on PCI-X devices to issue multiple requests then?
Perhaps :)
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 0 of 39] ipath - bug fixes, performance enhancements,and portability improvements
2006-06-29 21:40 [PATCH 0 of 39] ipath - bug fixes, performance enhancements, and portability improvements Bryan O'Sullivan
` (38 preceding siblings ...)
2006-06-29 21:41 ` [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss Bryan O'Sullivan
@ 2006-06-30 16:31 ` Michael S. Tsirkin
2006-06-30 17:00 ` Bryan O'Sullivan
39 siblings, 1 reply; 65+ messages in thread
From: Michael S. Tsirkin @ 2006-06-30 16:31 UTC (permalink / raw)
To: Bryan O'Sullivan
Cc: akpm, Roland Dreier, openib-general, linux-kernel, netdev
Quoting r. Bryan O'Sullivan <bos@pathscale.com>:
> Subject: [PATCH 0 of 39] ipath - bug fixes, performance enhancements,and portability improvements
>
> Hi, Andrew -
>
> These patches bring the ipath driver up to date with a number of bug fixes,
> performance improvements, and better PowerPC support. There are a few
> whitespace and formatting patches in the series, but they're all self-
> contained. The patches have been tested internally, and shouldn't contain
> anything controversial.
>
> My hope is that they'll sit in -mm for a little bit, and make it into
> an early 2.6.18 -rc kernel.
OK, next week I'll put these into my tree, too.
Bryan, as far as I can see there were some comments with regard to patches 38
and 39 in the series. Will you be sending updated revisions of these?
--
MST
^ permalink raw reply [flat|nested] 65+ messages in thread* Re: [PATCH 0 of 39] ipath - bug fixes, performance enhancements,and portability improvements
2006-06-30 16:31 ` [PATCH 0 of 39] ipath - bug fixes, performance enhancements,and portability improvements Michael S. Tsirkin
@ 2006-06-30 17:00 ` Bryan O'Sullivan
2006-07-01 19:43 ` [PATCH 0 of 39] ipath - bug fixes, performanceenhancements,and " Michael S. Tsirkin
0 siblings, 1 reply; 65+ messages in thread
From: Bryan O'Sullivan @ 2006-06-30 17:00 UTC (permalink / raw)
To: Michael S. Tsirkin
Cc: akpm, Roland Dreier, openib-general, linux-kernel, netdev
On Fri, 2006-06-30 at 19:31 +0300, Michael S. Tsirkin wrote:
> OK, next week I'll put these into my tree, too.
Thanks. The first 37 are in -mm; the last two you can drop until I sort
them out.
<b
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 0 of 39] ipath - bug fixes, performanceenhancements,and portability improvements
2006-06-30 17:00 ` Bryan O'Sullivan
@ 2006-07-01 19:43 ` Michael S. Tsirkin
2006-07-01 22:35 ` Andrew Morton
0 siblings, 1 reply; 65+ messages in thread
From: Michael S. Tsirkin @ 2006-07-01 19:43 UTC (permalink / raw)
To: Bryan O'Sullivan
Cc: akpm, Roland Dreier, openib-general, linux-kernel, netdev
Quoting r. Bryan O'Sullivan <bos@pathscale.com>:
> Subject: Re: [PATCH 0 of 39] ipath - bug fixes, performanceenhancements,and portability improvements
>
> On Fri, 2006-06-30 at 19:31 +0300, Michael S. Tsirkin wrote:
>
> > OK, next week I'll put these into my tree, too.
>
> Thanks. The first 37 are in -mm; the last two you can drop until I sort
> them out.
What about patch 28?
--
MST
^ permalink raw reply [flat|nested] 65+ messages in thread
* Re: [PATCH 0 of 39] ipath - bug fixes, performanceenhancements,and portability improvements
2006-07-01 19:43 ` [PATCH 0 of 39] ipath - bug fixes, performanceenhancements,and " Michael S. Tsirkin
@ 2006-07-01 22:35 ` Andrew Morton
0 siblings, 0 replies; 65+ messages in thread
From: Andrew Morton @ 2006-07-01 22:35 UTC (permalink / raw)
To: Michael S. Tsirkin; +Cc: bos, rdreier, openib-general, linux-kernel, netdev
On Sat, 1 Jul 2006 22:43:23 +0300
"Michael S. Tsirkin" <mst@mellanox.co.il> wrote:
> Quoting r. Bryan O'Sullivan <bos@pathscale.com>:
> > Subject: Re: [PATCH 0 of 39] ipath - bug fixes, performanceenhancements,and portability improvements
> >
> > On Fri, 2006-06-30 at 19:31 +0300, Michael S. Tsirkin wrote:
> >
> > > OK, next week I'll put these into my tree, too.
> >
> > Thanks. The first 37 are in -mm; the last two you can drop until I sort
> > them out.
>
> What about patch 28?
>
I sent 1-27,29-37 to Linus and he has merged them. But I held #28 back based on
Roland's concerns.
^ permalink raw reply [flat|nested] 65+ messages in thread