* [PATCH 21 of 39] IB/ipath - fixed bug 9776 for real. The problem was that I was updating
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 22 of 39] IB/ipath - fix lost interrupts on HT-400
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 23 of 39] IB/ipath - disallow send of invalid packet sizes over UD
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 24 of 39] IB/ipath - don't confuse the max message size with the MTU
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 25 of 39] IB/ipath - removed redundant statements
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 26 of 39] IB/ipath - check for valid LID and multicast LIDs
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 36 of 39] IB/ipath - Ignore receive queue size if SRQ is specified
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* Re: 2.6.17-mm4
From: Michal Piotrowski @ 2006-06-29 21:41 UTC (permalink / raw)
To: Andrew Morton; +Cc: davej, linux-kernel, netdev
In-Reply-To: <20060629135826.20328067.akpm@osdl.org>
On 29/06/06, Andrew Morton <akpm@osdl.org> wrote:
[snip]
> OK. Perhaps try setting 8k stacks?
>
> Disable lockdep?
>
I don't see that now. I will try to reproduce this with a previous
config + Ingo's "Ignore loglevel on printks" patch.
Regards,
Michal
--
Michal K. K. Piotrowski
LTG - Linux Testers Group
(http://www.stardust.webpages.pl/ltg/wiki/)
^ permalink raw reply
* [PATCH 8 of 39] IB/ipath - remove some duplicate code
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 17 of 39] IB/ipath - use more appropriate gfp flags
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 6 of 39] IB/ipath - Allow diags on any unit
From: Bryan O'Sullivan @ 2006-06-29 21:40 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 29 of 39] IB/ipath - RC receive interrupt performance changes
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 34 of 39] IB/ipath - fix a bug that results in addresses near 0 being written via DMA
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 37 of 39] IB/ipath - namespace cleanup: replace ips with ipath
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 28 of 39] IB/ipath - Fixes a bug where our delay for EEPROM no longer works due to compiler reordering
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 33 of 39] IB/ipath - read/write correct sizes through diag interface
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 35 of 39] IB/ipath - remove some #if 0 code related to lockable memory
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 32 of 39] IB/ipath - support more models of InfiniPath hardware
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 31 of 39] IB/ipath - drop the "stats" sysfs attribute group
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 27 of 39] IB/ipath - fixes to performance get counters for IB compliance
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* [PATCH 30 of 39] IB/ipath - purge sps_lid and sps_mlid arrays
From: Bryan O'Sullivan @ 2006-06-29 21:41 UTC (permalink / raw)
To: akpm, rdreier, mst; +Cc: openib-general, linux-kernel, netdev
In-Reply-To: <patchbomb.1151617251@eng-12.pathscale.com>
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
* Re: [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss
From: David Miller @ 2006-06-29 21:50 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
In-Reply-To: <1b00209ef20a0e7893d8.1151617290@eng-12.pathscale.com>
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
* Re: [PATCH 38 of 39] IB/ipath - More changes to support InfiniPath on PowerPC 970 systems
From: David Miller @ 2006-06-29 21:53 UTC (permalink / raw)
To: bos; +Cc: akpm, rdreier, mst, openib-general, linux-kernel, netdev
In-Reply-To: <c22b6c244d5db77f7b1d.1151617289@eng-12.pathscale.com>
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
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox