* [net-next-2.6 PATCH 06/23] igb: rework use of VMOLR in regards to PF and VFs
From: Jeff Kirsher @ 2009-10-28 9:47 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch updates the use of the VMOLR to include enabling multicast
promiscous for the VFs should they attempt to send over 30 multicast
addresses or if they use the new message type to enable multicast
promiscuous.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/e1000_mbx.h | 10 +-
drivers/net/igb/igb.h | 2
drivers/net/igb/igb_main.c | 173 +++++++++++++++++++++++++++++--------------
3 files changed, 125 insertions(+), 60 deletions(-)
diff --git a/drivers/net/igb/e1000_mbx.h b/drivers/net/igb/e1000_mbx.h
index ebc02ea..bb112fb 100644
--- a/drivers/net/igb/e1000_mbx.h
+++ b/drivers/net/igb/e1000_mbx.h
@@ -58,10 +58,12 @@
#define E1000_VT_MSGINFO_MASK (0xFF << E1000_VT_MSGINFO_SHIFT)
#define E1000_VF_RESET 0x01 /* VF requests reset */
-#define E1000_VF_SET_MAC_ADDR 0x02 /* VF requests PF to set MAC addr */
-#define E1000_VF_SET_MULTICAST 0x03 /* VF requests PF to set MC addr */
-#define E1000_VF_SET_VLAN 0x04 /* VF requests PF to set VLAN */
-#define E1000_VF_SET_LPE 0x05 /* VF requests PF to set VMOLR.LPE */
+#define E1000_VF_SET_MAC_ADDR 0x02 /* VF requests to set MAC addr */
+#define E1000_VF_SET_MULTICAST 0x03 /* VF requests to set MC addr */
+#define E1000_VF_SET_VLAN 0x04 /* VF requests to set VLAN */
+#define E1000_VF_SET_LPE 0x05 /* VF requests to set VMOLR.LPE */
+#define E1000_VF_SET_PROMISC 0x06 /*VF requests to clear VMOLR.ROPE/MPME*/
+#define E1000_VF_SET_PROMISC_MULTICAST (0x02 << E1000_VT_MSGINFO_SHIFT)
#define E1000_PF_CONTROL_MSG 0x0100 /* PF control message */
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index 7801d3f..c27dc1a 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -78,6 +78,8 @@ struct vf_data_storage {
};
#define IGB_VF_FLAG_CTS 0x00000001 /* VF is clear to send data */
+#define IGB_VF_FLAG_UNI_PROMISC 0x00000002 /* VF has unicast promisc */
+#define IGB_VF_FLAG_MULTI_PROMISC 0x00000004 /* VF has multicast promisc */
/* RX descriptor control thresholds.
* PTHRESH - MAC will consider prefetch if it has fewer than this number of
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index a9e8ba2..3ab2978 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -129,38 +129,6 @@ static void igb_vmm_control(struct igb_adapter *);
static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
-static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
-{
- u32 reg_data;
-
- reg_data = rd32(E1000_VMOLR(vfn));
- reg_data |= E1000_VMOLR_BAM | /* Accept broadcast */
- E1000_VMOLR_ROMPE | /* Accept packets matched in MTA */
- E1000_VMOLR_AUPE | /* Accept untagged packets */
- E1000_VMOLR_STRVLAN; /* Strip vlan tags */
- wr32(E1000_VMOLR(vfn), reg_data);
-}
-
-static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
- int vfn)
-{
- struct e1000_hw *hw = &adapter->hw;
- u32 vmolr;
-
- /* if it isn't the PF check to see if VFs are enabled and
- * increase the size to support vlan tags */
- if (vfn < adapter->vfs_allocated_count &&
- adapter->vf_data[vfn].vlans_enabled)
- size += VLAN_TAG_SIZE;
-
- vmolr = rd32(E1000_VMOLR(vfn));
- vmolr &= ~E1000_VMOLR_RLPML_MASK;
- vmolr |= size | E1000_VMOLR_LPE;
- wr32(E1000_VMOLR(vfn), vmolr);
-
- return 0;
-}
-
#ifdef CONFIG_PM
static int igb_suspend(struct pci_dev *, pm_message_t);
static int igb_resume(struct pci_dev *);
@@ -1115,8 +1083,6 @@ int igb_up(struct igb_adapter *adapter)
if (adapter->msix_entries)
igb_configure_msix(adapter);
- igb_set_vmolr(hw, adapter->vfs_allocated_count);
-
/* Clear any pending interrupts. */
rd32(E1000_ICR);
igb_irq_enable(adapter);
@@ -1892,8 +1858,6 @@ static int igb_open(struct net_device *netdev)
* clean_rx handler before we do so. */
igb_configure(adapter);
- igb_set_vmolr(hw, adapter->vfs_allocated_count);
-
err = igb_request_irq(adapter);
if (err)
goto err_req_irq;
@@ -2331,22 +2295,33 @@ void igb_setup_rctl(struct igb_adapter *adapter)
* if an un-trusted VF does not provide descriptors to hardware.
*/
if (adapter->vfs_allocated_count) {
- u32 vmolr;
-
/* set all queue drop enable bits */
wr32(E1000_QDE, ALL_QUEUES);
-
- vmolr = rd32(E1000_VMOLR(adapter->vfs_allocated_count));
- if (rctl & E1000_RCTL_LPE)
- vmolr |= E1000_VMOLR_LPE;
- if (adapter->num_rx_queues > 1)
- vmolr |= E1000_VMOLR_RSSE;
- wr32(E1000_VMOLR(adapter->vfs_allocated_count), vmolr);
}
wr32(E1000_RCTL, rctl);
}
+static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
+ int vfn)
+{
+ struct e1000_hw *hw = &adapter->hw;
+ u32 vmolr;
+
+ /* if it isn't the PF check to see if VFs are enabled and
+ * increase the size to support vlan tags */
+ if (vfn < adapter->vfs_allocated_count &&
+ adapter->vf_data[vfn].vlans_enabled)
+ size += VLAN_TAG_SIZE;
+
+ vmolr = rd32(E1000_VMOLR(vfn));
+ vmolr &= ~E1000_VMOLR_RLPML_MASK;
+ vmolr |= size | E1000_VMOLR_LPE;
+ wr32(E1000_VMOLR(vfn), vmolr);
+
+ return 0;
+}
+
/**
* igb_rlpml_set - set maximum receive packet size
* @adapter: board private structure
@@ -2366,12 +2341,43 @@ static void igb_rlpml_set(struct igb_adapter *adapter)
* size and set the VMOLR RLPML to the size we need */
if (pf_id) {
igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
- max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
+ max_frame_size = MAX_JUMBO_FRAME_SIZE;
}
wr32(E1000_RLPML, max_frame_size);
}
+static inline void igb_set_vmolr(struct igb_adapter *adapter, int vfn)
+{
+ struct e1000_hw *hw = &adapter->hw;
+ u32 vmolr;
+
+ /*
+ * This register exists only on 82576 and newer so if we are older then
+ * we should exit and do nothing
+ */
+ if (hw->mac.type < e1000_82576)
+ return;
+
+ vmolr = rd32(E1000_VMOLR(vfn));
+ vmolr |= E1000_VMOLR_AUPE | /* Accept untagged packets */
+ E1000_VMOLR_STRVLAN; /* Strip vlan tags */
+
+ /* clear all bits that might not be set */
+ vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
+
+ if (adapter->num_rx_queues > 1 && vfn == adapter->vfs_allocated_count)
+ vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
+ /*
+ * for VMDq only allow the VFs and pool 0 to accept broadcast and
+ * multicast packets
+ */
+ if (vfn <= adapter->vfs_allocated_count)
+ vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
+
+ wr32(E1000_VMOLR(vfn), vmolr);
+}
+
/**
* igb_configure_rx_ring - Configure a receive ring after Reset
* @adapter: board private structure
@@ -2425,6 +2431,9 @@ void igb_configure_rx_ring(struct igb_adapter *adapter,
wr32(E1000_SRRCTL(reg_idx), srrctl);
+ /* set filtering for VMDQ pools */
+ igb_set_vmolr(adapter, reg_idx & 0x7);
+
/* enable receive descriptor fetching */
rxdctl = rd32(E1000_RXDCTL(reg_idx));
rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
@@ -4101,6 +4110,45 @@ static void igb_ping_all_vfs(struct igb_adapter *adapter)
}
}
+static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
+{
+ struct e1000_hw *hw = &adapter->hw;
+ u32 vmolr = rd32(E1000_VMOLR(vf));
+ struct vf_data_storage *vf_data = &adapter->vf_data[vf];
+
+ vf_data->flags |= ~(IGB_VF_FLAG_UNI_PROMISC |
+ IGB_VF_FLAG_MULTI_PROMISC);
+ vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
+
+ if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
+ vmolr |= E1000_VMOLR_MPME;
+ *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
+ } else {
+ /*
+ * if we have hashes and we are clearing a multicast promisc
+ * flag we need to write the hashes to the MTA as this step
+ * was previously skipped
+ */
+ if (vf_data->num_vf_mc_hashes > 30) {
+ vmolr |= E1000_VMOLR_MPME;
+ } else if (vf_data->num_vf_mc_hashes) {
+ int j;
+ vmolr |= E1000_VMOLR_ROMPE;
+ for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
+ igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
+ }
+ }
+
+ wr32(E1000_VMOLR(vf), vmolr);
+
+ /* there are flags left unprocessed, likely not supported */
+ if (*msgbuf & E1000_VT_MSGINFO_MASK)
+ return -EINVAL;
+
+ return 0;
+
+}
+
static int igb_set_vf_multicasts(struct igb_adapter *adapter,
u32 *msgbuf, u32 vf)
{
@@ -4109,18 +4157,17 @@ static int igb_set_vf_multicasts(struct igb_adapter *adapter,
struct vf_data_storage *vf_data = &adapter->vf_data[vf];
int i;
- /* only up to 30 hash values supported */
- if (n > 30)
- n = 30;
-
- /* salt away the number of multi cast addresses assigned
+ /* salt away the number of multicast addresses assigned
* to this VF for later use to restore when the PF multi cast
* list changes
*/
vf_data->num_vf_mc_hashes = n;
- /* VFs are limited to using the MTA hash table for their multicast
- * addresses */
+ /* only up to 30 hash values supported */
+ if (n > 30)
+ n = 30;
+
+ /* store the hashes for later use */
for (i = 0; i < n; i++)
vf_data->vf_mc_hashes[i] = hash_list[i];
@@ -4137,9 +4184,20 @@ static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
int i, j;
for (i = 0; i < adapter->vfs_allocated_count; i++) {
+ u32 vmolr = rd32(E1000_VMOLR(i));
+ vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
+
vf_data = &adapter->vf_data[i];
- for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
- igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
+
+ if ((vf_data->num_vf_mc_hashes > 30) ||
+ (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
+ vmolr |= E1000_VMOLR_MPME;
+ } else if (vf_data->num_vf_mc_hashes) {
+ vmolr |= E1000_VMOLR_ROMPE;
+ for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
+ igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
+ }
+ wr32(E1000_VMOLR(i), vmolr);
}
}
@@ -4282,7 +4340,7 @@ static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
adapter->vf_data[vf].last_nack = jiffies;
/* reset offloads to defaults */
- igb_set_vmolr(&adapter->hw, vf);
+ igb_set_vmolr(adapter, vf);
/* reset vlans for device */
igb_clear_vf_vfta(adapter, vf);
@@ -4398,6 +4456,9 @@ static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
case E1000_VF_SET_MAC_ADDR:
retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
break;
+ case E1000_VF_SET_PROMISC:
+ retval = igb_set_vf_promisc(adapter, msgbuf, vf);
+ break;
case E1000_VF_SET_MULTICAST:
retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
break;
^ permalink raw reply related
* [net-next-2.6 PATCH 07/23] igb: rework handling of the vfta and vlvf registers in relation to mng_vlan
From: Jeff Kirsher @ 2009-10-28 9:47 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch corrects some errors in how vlans are being handled when vfs
start interacting with the management vlans.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 94 +++++++++++++++-----------------------------
1 files changed, 32 insertions(+), 62 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 3ab2978..9170927 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -963,24 +963,23 @@ static void igb_irq_enable(struct igb_adapter *adapter)
static void igb_update_mng_vlan(struct igb_adapter *adapter)
{
- struct net_device *netdev = adapter->netdev;
+ struct e1000_hw *hw = &adapter->hw;
u16 vid = adapter->hw.mng_cookie.vlan_id;
u16 old_vid = adapter->mng_vlan_id;
- if (adapter->vlgrp) {
- if (!vlan_group_get_device(adapter->vlgrp, vid)) {
- if (adapter->hw.mng_cookie.status &
- E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
- igb_vlan_rx_add_vid(netdev, vid);
- adapter->mng_vlan_id = vid;
- } else
- adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
- if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
- (vid != old_vid) &&
- !vlan_group_get_device(adapter->vlgrp, old_vid))
- igb_vlan_rx_kill_vid(netdev, old_vid);
- } else
- adapter->mng_vlan_id = vid;
+ if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
+ /* add VID to filter table */
+ igb_vfta_set(hw, vid, true);
+ adapter->mng_vlan_id = vid;
+ } else {
+ adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
+ }
+
+ if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
+ (vid != old_vid) &&
+ !vlan_group_get_device(adapter->vlgrp, old_vid)) {
+ /* remove VID from filter table */
+ igb_vfta_set(hw, old_vid, false);
}
}
@@ -1847,11 +1846,6 @@ static int igb_open(struct net_device *netdev)
/* e1000_power_up_phy(adapter); */
- adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
- if ((adapter->hw.mng_cookie.status &
- E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
- igb_update_mng_vlan(adapter);
-
/* before we allocate an interrupt, we must be ready to handle it.
* Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
* as soon as we call pci_request_irq, so we have to setup our
@@ -1924,14 +1918,6 @@ static int igb_close(struct net_device *netdev)
igb_free_all_tx_resources(adapter);
igb_free_all_rx_resources(adapter);
- /* kill manageability vlan ID if supported, but not if a vlan with
- * the same ID is registered on the host OS (let 8021q kill it) */
- if ((adapter->hw.mng_cookie.status &
- E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
- !(adapter->vlgrp &&
- vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
- igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
-
return 0;
}
@@ -4235,7 +4221,11 @@ static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
struct e1000_hw *hw = &adapter->hw;
u32 reg, i;
- /* It is an error to call this function when VFs are not enabled */
+ /* The vlvf table only exists on 82576 hardware and newer */
+ if (hw->mac.type < e1000_82576)
+ return -1;
+
+ /* we only need to do this if VMDq is enabled */
if (!adapter->vfs_allocated_count)
return -1;
@@ -4265,16 +4255,12 @@ static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
/* if !enabled we need to set this up in vfta */
if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
- /* add VID to filter table, if bit already set
- * PF must have added it outside of table */
- if (igb_vfta_set(hw, vid, true))
- reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
- adapter->vfs_allocated_count);
+ /* add VID to filter table */
+ igb_vfta_set(hw, vid, true);
reg |= E1000_VLVF_VLANID_ENABLE;
}
reg &= ~E1000_VLVF_VLANID_MASK;
reg |= vid;
-
wr32(E1000_VLVF(i), reg);
/* do not modify RLPML for PF devices */
@@ -4290,8 +4276,8 @@ static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
reg |= size;
wr32(E1000_VMOLR(vf), reg);
}
- adapter->vf_data[vf].vlans_enabled++;
+ adapter->vf_data[vf].vlans_enabled++;
return 0;
}
} else {
@@ -5393,21 +5379,15 @@ static void igb_vlan_rx_register(struct net_device *netdev,
ctrl |= E1000_CTRL_VME;
wr32(E1000_CTRL, ctrl);
- /* enable VLAN receive filtering */
+ /* Disable CFI check */
rctl = rd32(E1000_RCTL);
rctl &= ~E1000_RCTL_CFIEN;
wr32(E1000_RCTL, rctl);
- igb_update_mng_vlan(adapter);
} else {
/* disable VLAN tag insert/strip */
ctrl = rd32(E1000_CTRL);
ctrl &= ~E1000_CTRL_VME;
wr32(E1000_CTRL, ctrl);
-
- if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
- igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
- adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
- }
}
igb_rlpml_set(adapter);
@@ -5422,16 +5402,11 @@ static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
struct e1000_hw *hw = &adapter->hw;
int pf_id = adapter->vfs_allocated_count;
- if ((hw->mng_cookie.status &
- E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
- (vid == adapter->mng_vlan_id))
- return;
-
- /* add vid to vlvf if sr-iov is enabled,
- * if that fails add directly to filter table */
- if (igb_vlvf_set(adapter, vid, true, pf_id))
- igb_vfta_set(hw, vid, true);
+ /* attempt to add filter to vlvf array */
+ igb_vlvf_set(adapter, vid, true, pf_id);
+ /* add the filter since PF can receive vlans w/o entry in vlvf */
+ igb_vfta_set(hw, vid, true);
}
static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
@@ -5439,6 +5414,7 @@ static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
struct igb_adapter *adapter = netdev_priv(netdev);
struct e1000_hw *hw = &adapter->hw;
int pf_id = adapter->vfs_allocated_count;
+ s32 err;
igb_irq_disable(adapter);
vlan_group_set_device(adapter->vlgrp, vid, NULL);
@@ -5446,17 +5422,11 @@ static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
if (!test_bit(__IGB_DOWN, &adapter->state))
igb_irq_enable(adapter);
- if ((adapter->hw.mng_cookie.status &
- E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
- (vid == adapter->mng_vlan_id)) {
- /* release control to f/w */
- igb_release_hw_control(adapter);
- return;
- }
+ /* remove vlan from VLVF table array */
+ err = igb_vlvf_set(adapter, vid, false, pf_id);
- /* remove vid from vlvf if sr-iov is enabled,
- * if not in vlvf remove from vfta */
- if (igb_vlvf_set(adapter, vid, false, pf_id))
+ /* if vid was not present in VLVF just remove it from table */
+ if (err)
igb_vfta_set(hw, vid, false);
}
^ permalink raw reply related
* [net-next-2.6 PATCH 08/23] igb: move vf init into a seperate function
From: Jeff Kirsher @ 2009-10-28 9:47 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch moves VF initialization into a seperate function to help improve
the readability of igb_probe.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 98 +++++++++++++++++++++++++-------------------
1 files changed, 56 insertions(+), 42 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 9170927..54e8f02 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1409,46 +1409,6 @@ static int __devinit igb_probe(struct pci_dev *pdev,
if (err)
goto err_sw_init;
-#ifdef CONFIG_PCI_IOV
- /* since iov functionality isn't critical to base device function we
- * can accept failure. If it fails we don't allow iov to be enabled */
- if (hw->mac.type == e1000_82576) {
- /* 82576 supports a maximum of 7 VFs in addition to the PF */
- unsigned int num_vfs = (max_vfs > 7) ? 7 : max_vfs;
- int i;
- unsigned char mac_addr[ETH_ALEN];
-
- if (num_vfs) {
- adapter->vf_data = kcalloc(num_vfs,
- sizeof(struct vf_data_storage),
- GFP_KERNEL);
- if (!adapter->vf_data) {
- dev_err(&pdev->dev,
- "Could not allocate VF private data - "
- "IOV enable failed\n");
- } else {
- err = pci_enable_sriov(pdev, num_vfs);
- if (!err) {
- adapter->vfs_allocated_count = num_vfs;
- dev_info(&pdev->dev,
- "%d vfs allocated\n",
- num_vfs);
- for (i = 0;
- i < adapter->vfs_allocated_count;
- i++) {
- random_ether_addr(mac_addr);
- igb_set_vf_mac(adapter, i,
- mac_addr);
- }
- } else {
- kfree(adapter->vf_data);
- adapter->vf_data = NULL;
- }
- }
- }
- }
-
-#endif
/* setup the private structure */
err = igb_sw_init(adapter);
if (err)
@@ -1772,6 +1732,54 @@ static void __devexit igb_remove(struct pci_dev *pdev)
}
/**
+ * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
+ * @adapter: board private structure to initialize
+ *
+ * This function initializes the vf specific data storage and then attempts to
+ * allocate the VFs. The reason for ordering it this way is because it is much
+ * mor expensive time wise to disable SR-IOV than it is to allocate and free
+ * the memory for the VFs.
+ **/
+static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
+{
+#ifdef CONFIG_PCI_IOV
+ struct pci_dev *pdev = adapter->pdev;
+
+ if (adapter->vfs_allocated_count > 7)
+ adapter->vfs_allocated_count = 7;
+
+ if (adapter->vfs_allocated_count) {
+ adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
+ sizeof(struct vf_data_storage),
+ GFP_KERNEL);
+ /* if allocation failed then we do not support SR-IOV */
+ if (!adapter->vf_data) {
+ adapter->vfs_allocated_count = 0;
+ dev_err(&pdev->dev, "Unable to allocate memory for VF "
+ "Data Storage\n");
+ }
+ }
+
+ if (pci_enable_sriov(pdev, adapter->vfs_allocated_count)) {
+ kfree(adapter->vf_data);
+ adapter->vf_data = NULL;
+#endif /* CONFIG_PCI_IOV */
+ adapter->vfs_allocated_count = 0;
+#ifdef CONFIG_PCI_IOV
+ } else {
+ unsigned char mac_addr[ETH_ALEN];
+ int i;
+ dev_info(&pdev->dev, "%d vfs allocated\n",
+ adapter->vfs_allocated_count);
+ for (i = 0; i < adapter->vfs_allocated_count; i++) {
+ random_ether_addr(mac_addr);
+ igb_set_vf_mac(adapter, i, mac_addr);
+ }
+ }
+#endif /* CONFIG_PCI_IOV */
+}
+
+/**
* igb_sw_init - Initialize general software structures (struct igb_adapter)
* @adapter: board private structure to initialize
*
@@ -1795,13 +1803,19 @@ static int __devinit igb_sw_init(struct igb_adapter *adapter)
adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
- /* This call may decrease the number of queues depending on
- * interrupt mode. */
+#ifdef CONFIG_PCI_IOV
+ if (hw->mac.type == e1000_82576)
+ adapter->vfs_allocated_count = max_vfs;
+
+#endif /* CONFIG_PCI_IOV */
+ /* This call may decrease the number of queues */
if (igb_init_interrupt_scheme(adapter)) {
dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
return -ENOMEM;
}
+ igb_probe_vfs(adapter);
+
/* Explicitly disable IRQ since the NIC can be in any state. */
igb_irq_disable(adapter);
^ permalink raw reply related
* [net-next-2.6 PATCH 09/23] igb: only process global stats in igb_update_stats
From: Jeff Kirsher @ 2009-10-28 9:48 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch moves the update of adapter->net_stats.rx/tx values out of the
interrupt routine and into igb_update_stats by just adding together the
tx/rx byte/packet counts for the rings.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 61 +++++++++++++++++++++-----------------------
1 files changed, 29 insertions(+), 32 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 54e8f02..62b6876 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -3769,7 +3769,10 @@ void igb_update_stats(struct igb_adapter *adapter)
struct net_device *netdev = adapter->netdev;
struct e1000_hw *hw = &adapter->hw;
struct pci_dev *pdev = adapter->pdev;
+ u32 rnbc;
u16 phy_tmp;
+ int i;
+ u64 bytes, packets;
#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
@@ -3782,6 +3785,29 @@ void igb_update_stats(struct igb_adapter *adapter)
if (pci_channel_offline(pdev))
return;
+ bytes = 0;
+ packets = 0;
+ for (i = 0; i < adapter->num_rx_queues; i++) {
+ u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
+ adapter->rx_ring[i].rx_stats.drops += rqdpc_tmp;
+ netdev->stats.rx_fifo_errors += rqdpc_tmp;
+ bytes += adapter->rx_ring[i].rx_stats.bytes;
+ packets += adapter->rx_ring[i].rx_stats.packets;
+ }
+
+ netdev->stats.rx_bytes = bytes;
+ netdev->stats.rx_packets = packets;
+
+ bytes = 0;
+ packets = 0;
+ for (i = 0; i < adapter->num_tx_queues; i++) {
+ bytes += adapter->tx_ring[i].tx_stats.bytes;
+ packets += adapter->tx_ring[i].tx_stats.packets;
+ }
+ netdev->stats.tx_bytes = bytes;
+ netdev->stats.tx_packets = packets;
+
+ /* read stats registers */
adapter->stats.crcerrs += rd32(E1000_CRCERRS);
adapter->stats.gprc += rd32(E1000_GPRC);
adapter->stats.gorc += rd32(E1000_GORCL);
@@ -3814,7 +3840,9 @@ void igb_update_stats(struct igb_adapter *adapter)
adapter->stats.gptc += rd32(E1000_GPTC);
adapter->stats.gotc += rd32(E1000_GOTCL);
rd32(E1000_GOTCH); /* clear GOTCL */
- adapter->stats.rnbc += rd32(E1000_RNBC);
+ rnbc = rd32(E1000_RNBC);
+ adapter->stats.rnbc += rnbc;
+ netdev->stats.rx_fifo_errors += rnbc;
adapter->stats.ruc += rd32(E1000_RUC);
adapter->stats.rfc += rd32(E1000_RFC);
adapter->stats.rjc += rd32(E1000_RJC);
@@ -3861,33 +3889,6 @@ void igb_update_stats(struct igb_adapter *adapter)
/* Rx Errors */
- if (hw->mac.type != e1000_82575) {
- u32 rqdpc_tmp;
- u64 rqdpc_total = 0;
- int i;
- /* Read out drops stats per RX queue. Notice RQDPC (Receive
- * Queue Drop Packet Count) stats only gets incremented, if
- * the DROP_EN but it set (in the SRRCTL register for that
- * queue). If DROP_EN bit is NOT set, then the some what
- * equivalent count is stored in RNBC (not per queue basis).
- * Also note the drop count is due to lack of available
- * descriptors.
- */
- for (i = 0; i < adapter->num_rx_queues; i++) {
- rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0xFFF;
- adapter->rx_ring[i].rx_stats.drops += rqdpc_tmp;
- rqdpc_total += adapter->rx_ring[i].rx_stats.drops;
- }
- netdev->stats.rx_fifo_errors = rqdpc_total;
- }
-
- /* Note RNBC (Receive No Buffers Count) is an not an exact
- * drop count as the hardware FIFO might save the day. Thats
- * one of the reason for saving it in rx_fifo_errors, as its
- * potentially not a true drop.
- */
- netdev->stats.rx_fifo_errors += adapter->stats.rnbc;
-
/* RLEC on some newer hardware can be incorrect so build
* our own version based on RUC and ROC */
netdev->stats.rx_errors = adapter->stats.rxerrc +
@@ -4818,8 +4819,6 @@ static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
tx_ring->total_packets += total_packets;
tx_ring->tx_stats.bytes += total_bytes;
tx_ring->tx_stats.packets += total_packets;
- netdev->stats.tx_bytes += total_bytes;
- netdev->stats.tx_packets += total_packets;
return (count < tx_ring->count);
}
@@ -5043,8 +5042,6 @@ next_desc:
rx_ring->total_bytes += total_bytes;
rx_ring->rx_stats.packets += total_packets;
rx_ring->rx_stats.bytes += total_bytes;
- netdev->stats.rx_bytes += total_bytes;
- netdev->stats.rx_packets += total_packets;
return cleaned;
}
^ permalink raw reply related
* [net-next-2.6 PATCH 10/23] igb: move global_quad_port_a from global into local static define
From: Jeff Kirsher @ 2009-10-28 9:48 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This change moves global_quad_port_a into igb_probe as a static define
since it doesn't actually need to be global.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 8 ++------
1 files changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 62b6876..7f32211 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -179,8 +179,6 @@ static struct pci_driver igb_driver = {
.err_handler = &igb_err_handler
};
-static int global_quad_port_a; /* global quad port a indication */
-
MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
MODULE_LICENSE("GPL");
@@ -252,12 +250,9 @@ static int __init igb_init_module(void)
printk(KERN_INFO "%s\n", igb_copyright);
- global_quad_port_a = 0;
-
#ifdef CONFIG_IGB_DCA
dca_register_notify(&dca_notifier);
#endif
-
ret = pci_register_driver(&igb_driver);
return ret;
}
@@ -1318,10 +1313,11 @@ static int __devinit igb_probe(struct pci_dev *pdev,
struct net_device *netdev;
struct igb_adapter *adapter;
struct e1000_hw *hw;
+ u16 eeprom_data = 0;
+ static int global_quad_port_a; /* global quad port a indication */
const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
unsigned long mmio_start, mmio_len;
int err, pci_using_dac;
- u16 eeprom_data = 0;
u16 eeprom_apme_mask = IGB_EEPROM_APME;
u32 part_num;
^ permalink raw reply related
* [net-next-2.6 PATCH 11/23] igb: make tx hang check multiqueue, check eop descriptor
From: Jeff Kirsher @ 2009-10-28 9:48 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This change makes the tx hang check run over all tx queues instead of just
queue 0. Also have hang display info on EOP descriptor instead of the
descriptor at the start of the chain.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 10 ++++++----
1 files changed, 6 insertions(+), 4 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 7f32211..f75f90f 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -3002,6 +3002,10 @@ link_up:
}
}
+ /* Force detection of hung controller every watchdog period */
+ for (i = 0; i < adapter->num_tx_queues; i++)
+ adapter->tx_ring[i].detect_tx_hung = true;
+
/* Cause software interrupt to ensure rx ring is cleaned */
if (adapter->msix_entries) {
u32 eics = 0;
@@ -3014,9 +3018,6 @@ link_up:
wr32(E1000_ICS, E1000_ICS_RXDMT0);
}
- /* Force detection of hung controller every watchdog period */
- tx_ring->detect_tx_hung = true;
-
/* Reset the timer */
if (!test_bit(__IGB_DOWN, &adapter->state))
mod_timer(&adapter->watchdog_timer,
@@ -3667,6 +3668,7 @@ static void igb_tx_timeout(struct net_device *netdev)
/* Do the reset outside of interrupt context */
adapter->tx_timeout_count++;
+
schedule_work(&adapter->reset_task);
wr32(E1000_EICS,
(adapter->eims_enable_mask & ~adapter->eims_other));
@@ -4804,7 +4806,7 @@ static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
readl(tx_ring->tail),
tx_ring->next_to_use,
tx_ring->next_to_clean,
- tx_ring->buffer_info[i].time_stamp,
+ tx_ring->buffer_info[eop].time_stamp,
eop,
jiffies,
eop_desc->wb.status);
^ permalink raw reply related
* [net-next-2.6 PATCH 12/23] igb: cleanup code related to ring resource allocation and free
From: Jeff Kirsher @ 2009-10-28 9:49 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch cleans up some of the ring alloc and free code to better handle
exceptions such as attempting to free resources on an already freed ring.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 29 +++++++++++++++++------------
1 files changed, 17 insertions(+), 12 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index f75f90f..e67ff0e 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1952,7 +1952,8 @@ int igb_setup_tx_resources(struct igb_ring *tx_ring)
tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
tx_ring->size = ALIGN(tx_ring->size, 4096);
- tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
+ tx_ring->desc = pci_alloc_consistent(pdev,
+ tx_ring->size,
&tx_ring->dma);
if (!tx_ring->desc)
@@ -1978,13 +1979,13 @@ err:
**/
static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
{
+ struct pci_dev *pdev = adapter->pdev;
int i, err = 0;
- int r_idx;
for (i = 0; i < adapter->num_tx_queues; i++) {
err = igb_setup_tx_resources(&adapter->tx_ring[i]);
if (err) {
- dev_err(&adapter->pdev->dev,
+ dev_err(&pdev->dev,
"Allocation for Tx Queue %u failed\n", i);
for (i--; i >= 0; i--)
igb_free_tx_resources(&adapter->tx_ring[i]);
@@ -1993,7 +1994,7 @@ static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
}
for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
- r_idx = i % adapter->num_tx_queues;
+ int r_idx = i % adapter->num_tx_queues;
adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
}
return err;
@@ -2116,6 +2117,7 @@ int igb_setup_rx_resources(struct igb_ring *rx_ring)
err:
vfree(rx_ring->buffer_info);
+ rx_ring->buffer_info = NULL;
dev_err(&pdev->dev, "Unable to allocate memory for "
"the receive descriptor ring\n");
return -ENOMEM;
@@ -2130,12 +2132,13 @@ err:
**/
static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
{
+ struct pci_dev *pdev = adapter->pdev;
int i, err = 0;
for (i = 0; i < adapter->num_rx_queues; i++) {
err = igb_setup_rx_resources(&adapter->rx_ring[i]);
if (err) {
- dev_err(&adapter->pdev->dev,
+ dev_err(&pdev->dev,
"Allocation for Rx Queue %u failed\n", i);
for (i--; i >= 0; i--)
igb_free_rx_resources(&adapter->rx_ring[i]);
@@ -2476,6 +2479,10 @@ void igb_free_tx_resources(struct igb_ring *tx_ring)
vfree(tx_ring->buffer_info);
tx_ring->buffer_info = NULL;
+ /* if not set, then don't free */
+ if (!tx_ring->desc)
+ return;
+
pci_free_consistent(tx_ring->pdev, tx_ring->size,
tx_ring->desc, tx_ring->dma);
@@ -2534,14 +2541,10 @@ static void igb_clean_tx_ring(struct igb_ring *tx_ring)
memset(tx_ring->buffer_info, 0, size);
/* Zero out the descriptor ring */
-
memset(tx_ring->desc, 0, tx_ring->size);
tx_ring->next_to_use = 0;
tx_ring->next_to_clean = 0;
-
- writel(0, tx_ring->head);
- writel(0, tx_ring->tail);
}
/**
@@ -2569,6 +2572,10 @@ void igb_free_rx_resources(struct igb_ring *rx_ring)
vfree(rx_ring->buffer_info);
rx_ring->buffer_info = NULL;
+ /* if not set, then don't free */
+ if (!rx_ring->desc)
+ return;
+
pci_free_consistent(rx_ring->pdev, rx_ring->size,
rx_ring->desc, rx_ring->dma);
@@ -2601,6 +2608,7 @@ static void igb_clean_rx_ring(struct igb_ring *rx_ring)
if (!rx_ring->buffer_info)
return;
+
/* Free all the Rx ring sk_buffs */
for (i = 0; i < rx_ring->count; i++) {
buffer_info = &rx_ring->buffer_info[i];
@@ -2638,9 +2646,6 @@ static void igb_clean_rx_ring(struct igb_ring *rx_ring)
rx_ring->next_to_clean = 0;
rx_ring->next_to_use = 0;
-
- writel(0, rx_ring->head);
- writel(0, rx_ring->tail);
}
/**
^ permalink raw reply related
* [net-next-2.6 PATCH 13/23] igb: change queue ordering for 82576 based adapters
From: Jeff Kirsher @ 2009-10-28 9:49 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch changes the queue ordering for 82576 adapters so that if VFs are
enabled the queues will first be allocated out of the PF pool. Any
remaining queues will be allocated out of other VMDq pools.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 25 +++++++++++++------------
1 files changed, 13 insertions(+), 12 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index e67ff0e..52c4302 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -285,7 +285,7 @@ module_exit(igb_exit_module);
**/
static void igb_cache_ring_register(struct igb_adapter *adapter)
{
- int i;
+ int i = 0, j = 0;
u32 rbase_offset = adapter->vfs_allocated_count;
switch (adapter->hw.mac.type) {
@@ -295,19 +295,20 @@ static void igb_cache_ring_register(struct igb_adapter *adapter)
* In order to avoid collision we start at the first free queue
* and continue consuming queues in the same sequence
*/
- for (i = 0; i < adapter->num_rx_queues; i++)
- adapter->rx_ring[i].reg_idx = rbase_offset +
- Q_IDX_82576(i);
- for (i = 0; i < adapter->num_tx_queues; i++)
- adapter->tx_ring[i].reg_idx = rbase_offset +
- Q_IDX_82576(i);
- break;
+ if (adapter->vfs_allocated_count) {
+ for (; i < adapter->num_rx_queues; i++)
+ adapter->rx_ring[i].reg_idx = rbase_offset +
+ Q_IDX_82576(i);
+ for (; j < adapter->num_tx_queues; j++)
+ adapter->tx_ring[j].reg_idx = rbase_offset +
+ Q_IDX_82576(j);
+ }
case e1000_82575:
default:
- for (i = 0; i < adapter->num_rx_queues; i++)
- adapter->rx_ring[i].reg_idx = i;
- for (i = 0; i < adapter->num_tx_queues; i++)
- adapter->tx_ring[i].reg_idx = i;
+ for (; i < adapter->num_rx_queues; i++)
+ adapter->rx_ring[i].reg_idx = rbase_offset + i;
+ for (; j < adapter->num_tx_queues; j++)
+ adapter->tx_ring[j].reg_idx = rbase_offset + j;
break;
}
}
^ permalink raw reply related
* [net-next-2.6 PATCH 14/23] igb: cleanup interrupt enablement in regards to msix_other
From: Jeff Kirsher @ 2009-10-28 9:49 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch changes a few things so that instead of firing a link status
interrupt directly the get_link_status bit is set and the watchdog is
scheduled. In addition the mailbox bit is now only enabled if VFs are
enabled.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 31 +++++++++++++++++++++++--------
1 files changed, 23 insertions(+), 8 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 52c4302..5b0f939 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -919,6 +919,11 @@ static void igb_irq_disable(struct igb_adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
+ /*
+ * we need to be careful when disabling interrupts. The VFs are also
+ * mapped into these registers and so clearing the bits can cause
+ * issues on the VF drivers so we only need to clear what we set
+ */
if (adapter->msix_entries) {
u32 regval = rd32(E1000_EIAM);
wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
@@ -942,15 +947,17 @@ static void igb_irq_enable(struct igb_adapter *adapter)
struct e1000_hw *hw = &adapter->hw;
if (adapter->msix_entries) {
+ u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC;
u32 regval = rd32(E1000_EIAC);
wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
regval = rd32(E1000_EIAM);
wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
wr32(E1000_EIMS, adapter->eims_enable_mask);
- if (adapter->vfs_allocated_count)
+ if (adapter->vfs_allocated_count) {
wr32(E1000_MBVFIMR, 0xFF);
- wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
- E1000_IMS_DOUTSYNC));
+ ims |= E1000_IMS_VMMB;
+ }
+ wr32(E1000_IMS, ims);
} else {
wr32(E1000_IMS, IMS_ENABLE_MASK);
wr32(E1000_IAM, IMS_ENABLE_MASK);
@@ -1091,8 +1098,10 @@ int igb_up(struct igb_adapter *adapter)
netif_tx_start_all_queues(adapter->netdev);
- /* Fire a link change interrupt to start the watchdog. */
- wr32(E1000_ICS, E1000_ICS_LSC);
+ /* start the watchdog. */
+ hw->mac.get_link_status = 1;
+ schedule_work(&adapter->watchdog_task);
+
return 0;
}
@@ -1889,8 +1898,9 @@ static int igb_open(struct net_device *netdev)
netif_tx_start_all_queues(netdev);
- /* Fire a link status change interrupt to start the watchdog. */
- wr32(E1000_ICS, E1000_ICS_LSC);
+ /* start the watchdog. */
+ hw->mac.get_link_status = 1;
+ schedule_work(&adapter->watchdog_task);
return 0;
@@ -3952,7 +3962,12 @@ static irqreturn_t igb_msix_other(int irq, void *data)
mod_timer(&adapter->watchdog_timer, jiffies + 1);
}
- wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
+ if (adapter->vfs_allocated_count)
+ wr32(E1000_IMS, E1000_IMS_LSC |
+ E1000_IMS_VMMB |
+ E1000_IMS_DOUTSYNC);
+ else
+ wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC);
wr32(E1000_EIMS, adapter->eims_other);
return IRQ_HANDLED;
^ permalink raw reply related
* [net-next-2.6 PATCH 15/23] igb: Remove invalid stats counters
From: Jeff Kirsher @ 2009-10-28 9:50 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
There are several counters being used like they are static when in fact
they are clear on read. In order to prevent the values from being
incorrect I am removing the defunct counters.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb.h | 6 ------
drivers/net/igb/igb_main.c | 18 +-----------------
2 files changed, 1 insertions(+), 23 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index c27dc1a..b9fcfd3 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -269,10 +269,6 @@ struct igb_adapter {
/* TX */
struct igb_ring *tx_ring; /* One per active queue */
unsigned long tx_queue_len;
- u32 gotc;
- u64 gotc_old;
- u64 tpt_old;
- u64 colc_old;
u32 tx_timeout_count;
/* RX */
@@ -280,8 +276,6 @@ struct igb_adapter {
int num_tx_queues;
int num_rx_queues;
- u32 gorc;
- u64 gorc_old;
u32 max_frame_size;
u32 min_frame_size;
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 5b0f939..cb1acca 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -2925,9 +2925,6 @@ static void igb_watchdog_task(struct work_struct *work)
int i;
link = igb_has_link(adapter);
- if ((netif_carrier_ok(netdev)) && link)
- goto link_up;
-
if (link) {
if (!netif_carrier_ok(netdev)) {
u32 ctrl;
@@ -2990,20 +2987,8 @@ static void igb_watchdog_task(struct work_struct *work)
}
}
-link_up:
igb_update_stats(adapter);
-
- hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
- adapter->tpt_old = adapter->stats.tpt;
- hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
- adapter->colc_old = adapter->stats.colc;
-
- adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
- adapter->gorc_old = adapter->stats.gorc;
- adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
- adapter->gotc_old = adapter->stats.gotc;
-
- igb_update_adaptive(&adapter->hw);
+ igb_update_adaptive(hw);
if (!netif_carrier_ok(netdev)) {
if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
@@ -3875,7 +3860,6 @@ void igb_update_stats(struct igb_adapter *adapter)
adapter->stats.bptc += rd32(E1000_BPTC);
/* used for adaptive IFS */
-
hw->mac.tx_packet_delta = rd32(E1000_TPT);
adapter->stats.tpt += hw->mac.tx_packet_delta;
hw->mac.collision_delta = rd32(E1000_COLC);
^ permalink raw reply related
* [net-next-2.6 PATCH 16/23] igb: cleanup igb.h header whitespace and some structure formatting
From: Jeff Kirsher @ 2009-10-28 9:50 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch changes the layout of the ring and adapter structs to fill a few
holes in the structure. It also cleans up some whitespace and formatting
issues.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb.h | 12 ++++++------
1 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index b9fcfd3..3298f5a 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -185,16 +185,15 @@ struct igb_ring {
dma_addr_t dma; /* phys address of the ring */
void *desc; /* descriptor ring memory */
unsigned int size; /* length of desc. ring in bytes */
- unsigned int count; /* number of desc. in the ring */
+ u16 count; /* number of desc. in the ring */
u16 next_to_use;
u16 next_to_clean;
+ u8 queue_index;
+ u8 reg_idx;
void __iomem *head;
void __iomem *tail;
struct igb_buffer *buffer_info; /* array of buffer info structs */
- u8 queue_index;
- u8 reg_idx;
-
unsigned int total_bytes;
unsigned int total_packets;
@@ -249,6 +248,7 @@ struct igb_adapter {
u32 en_mng_pt;
u16 link_speed;
u16 link_duplex;
+
unsigned int total_tx_bytes;
unsigned int total_tx_packets;
unsigned int total_rx_bytes;
@@ -311,8 +311,8 @@ struct igb_adapter {
u32 eeprom_wol;
struct igb_ring *multi_tx_table[IGB_ABS_MAX_TX_QUEUES];
- unsigned int tx_ring_count;
- unsigned int rx_ring_count;
+ u16 tx_ring_count;
+ u16 rx_ring_count;
unsigned int vfs_allocated_count;
struct vf_data_storage *vf_data;
};
^ permalink raw reply related
* [net-next-2.6 PATCH 17/23] igb: cleanup igb xmit frame path
From: Jeff Kirsher @ 2009-10-28 9:50 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch cleans up the xmit frame path for igb to better handle xmit
frame errors and avoid null pointer exceptions. It also cleans up some
whitespace issues found in the xmit frame path.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 30 +++++++++++++++---------------
1 files changed, 15 insertions(+), 15 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index cb1acca..8f8b7cc 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -3245,9 +3245,9 @@ set_itr_now:
#define IGB_TX_FLAGS_VLAN 0x00000002
#define IGB_TX_FLAGS_TSO 0x00000004
#define IGB_TX_FLAGS_IPV4 0x00000008
-#define IGB_TX_FLAGS_TSTAMP 0x00000010
-#define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
-#define IGB_TX_FLAGS_VLAN_SHIFT 16
+#define IGB_TX_FLAGS_TSTAMP 0x00000010
+#define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
+#define IGB_TX_FLAGS_VLAN_SHIFT 16
static inline int igb_tso_adv(struct igb_ring *tx_ring,
struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
@@ -3346,6 +3346,7 @@ static inline bool igb_tx_csum_adv(struct igb_ring *tx_ring,
if (tx_flags & IGB_TX_FLAGS_VLAN)
info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
+
info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
if (skb->ip_summed == CHECKSUM_PARTIAL)
info |= skb_network_header_len(skb);
@@ -3462,17 +3463,17 @@ static inline int igb_tx_map_adv(struct igb_ring *tx_ring, struct sk_buff *skb,
tx_ring->buffer_info[i].skb = skb;
tx_ring->buffer_info[first].next_to_watch = i;
- return count + 1;
+ return ++count;
}
static inline void igb_tx_queue_adv(struct igb_ring *tx_ring,
int tx_flags, int count, u32 paylen,
u8 hdr_len)
{
- union e1000_adv_tx_desc *tx_desc = NULL;
+ union e1000_adv_tx_desc *tx_desc;
struct igb_buffer *buffer_info;
u32 olinfo_status = 0, cmd_type_len;
- unsigned int i;
+ unsigned int i = tx_ring->next_to_use;
cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
E1000_ADVTXD_DCMD_DEXT);
@@ -3505,18 +3506,18 @@ static inline void igb_tx_queue_adv(struct igb_ring *tx_ring,
olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
- i = tx_ring->next_to_use;
- while (count--) {
+ do {
buffer_info = &tx_ring->buffer_info[i];
tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
tx_desc->read.cmd_type_len =
cpu_to_le32(cmd_type_len | buffer_info->length);
tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
+ count--;
i++;
if (i == tx_ring->count)
i = 0;
- }
+ } while (count > 0);
tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_ADVTXD_DCMD);
/* Force memory writes to complete before letting h/w
@@ -3568,8 +3569,7 @@ netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
unsigned int first;
unsigned int tx_flags = 0;
u8 hdr_len = 0;
- int count = 0;
- int tso = 0;
+ int tso = 0, count;
union skb_shared_tx *shtx = skb_tx(skb);
/* need: 1 descriptor per page,
@@ -3587,7 +3587,7 @@ netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
tx_flags |= IGB_TX_FLAGS_TSTAMP;
}
- if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
+ if (vlan_tx_tag_present(skb) && adapter->vlgrp) {
tx_flags |= IGB_TX_FLAGS_VLAN;
tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
}
@@ -3598,6 +3598,7 @@ netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
first = tx_ring->next_to_use;
if (skb_is_gso(skb)) {
tso = igb_tso_adv(tx_ring, skb, tx_flags, &hdr_len);
+
if (tso < 0) {
dev_kfree_skb_any(skb);
return NETDEV_TX_OK;
@@ -3611,12 +3612,11 @@ netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
tx_flags |= IGB_TX_FLAGS_CSUM;
/*
- * count reflects descriptors mapped, if 0 then mapping error
+ * count reflects descriptors mapped, if 0 or less then mapping error
* has occured and we need to rewind the descriptor queue
*/
count = igb_tx_map_adv(tx_ring, skb, first);
-
- if (!count) {
+ if (count <= 0) {
dev_kfree_skb_any(skb);
tx_ring->buffer_info[first].time_stamp = 0;
tx_ring->next_to_use = first;
^ permalink raw reply related
* Re: [PATCH net-next] bnx2x: Do Tx handling in a separate tasklet.
From: Vladislav Zolotarov @ 2009-10-28 9:54 UTC (permalink / raw)
To: David Miller ; +Cc: Eilon Greenstein, netdev@vger.kernel.org
I'd like to start from your last remark: you r absolutely right, and this is the problem we have in the current net-next driver. More than that, this patch is fixing this problem: it moved liberation of Tx SKBs from hardIRQ context (ISR) to the softIRQ context (tasklet) thereby resolving the problem u've mentioned. So, total agreement with u on this one. I must have named the patch differently to emphasize it.
I'd like to summarize the patch I've sent:
- Take Tx SKB liberation out of hardIRQ.
- Instead schedule a DPC that handles Tx work.
- Optimize the access to status block indices: read only the index we are about to use in the current context.
So, could u, pls., apply the patch in order to fix the problem we currently have in bnx2x?
Bullet 1 is correct but not complete: what about SKB's needed for filling Rx ring, what about Tx-only scenarios where u'd prefer to liberate SKBs from start_xmit()? Generally, we'd like to do SKB liberation both from start_xmit and from NAPI. Making it straight forward would make us take a tx_lock from inside NAPI and this is what we surely don't wan't to do. We are working on this optimization at the moment and it will be the topic of one of the next patches.
Regarding the second bullet u wrote: saying "low CPU consumption" in regard of Tx work in my previous e-mail I meant that CPU per packet ratio for Tx is much lower than for Rx. Sorry for being unclear.
Best regards
vlad
-----Original Message-----
From: David Miller [mailto:davem@davemloft.net]
Sent: Tuesday, October 27, 2009 12:28 AM
To: Vladislav Zolotarov
Cc: Eilon Greenstein; netdev@vger.kernel.org
Subject: Re: [PATCH net-next] bnx2x: Do Tx handling in a separate tasklet.
From: "Vladislav Zolotarov" <vladz@broadcom.com>
Date: Mon, 26 Oct 2009 07:42:27 -0700
> The separation of Tx and Rx interrupt handling gives us the
> possibility to properly affinitize the Rx (heavy CPU consuming task)
> and Tx (low CPU consuming task) and to ensure that Tx work is done
> not long after the Tx interrupt without interference of Rx work thus
> letting the user benefit from Tx coalescing configuration in order
> to achieve the best performance in each specific scenario. This is
> most important in heavy load scenarios with mixed traffic (UDP + TCP
> for instance). If we didn't separate Tx and Rx interrupt handling Tx
> coalescing configuration was not worth much.
There are other issues:
1) Actually, it makes sense to do TX and RX work together, since TX
packet liberation makes fresh CPU local packets available for
responses generated by RX packet reception.
2) TX packet liberation is not low CPU consumption, it has to perform
many atomic instructions, reference socket state, enter the SLAB
allocator, potentially liberate netfilter state, etc.
Using NAPI also moves the TX freeing into softirq context.
If you do it from a hardirq you are making it more expensive. From
hardirq the free just puts the SKB on a list, schedules a softirq,
then does the real SKB free work from the softirq.
This needless SKB list management and softirq scheduling you'll
avoid if you do things from softirqs, and thus using NAPI makes
sense here.
^ permalink raw reply
* [net-next-2.6 PATCH 18/23] igb: cleanup clean_rx_irq_adv and alloc_rx_buffers_adv
From: Jeff Kirsher @ 2009-10-28 9:51 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch cleans up some whitespace issues in clean_rx_irq_adv. It also
adds NUMA aware page allocation and dma error handling to
alloc_rx_buffers_adv.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 24 +++++++++++++++++++-----
1 files changed, 19 insertions(+), 5 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 8f8b7cc..d3e8316 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -4952,6 +4952,7 @@ static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
i++;
if (i == rx_ring->count)
i = 0;
+
next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
prefetch(next_rxd);
next_buffer = &rx_ring->buffer_info[i];
@@ -4989,7 +4990,6 @@ static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
skb->len += length;
skb->data_len += length;
-
skb->truesize += length;
}
@@ -5071,7 +5071,7 @@ void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
if ((bufsz < IGB_RXBUFFER_1024) && !buffer_info->page_dma) {
if (!buffer_info->page) {
- buffer_info->page = alloc_page(GFP_ATOMIC);
+ buffer_info->page = netdev_alloc_page(netdev);
if (!buffer_info->page) {
rx_ring->rx_stats.alloc_failed++;
goto no_buffers;
@@ -5085,9 +5085,16 @@ void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
buffer_info->page_offset,
PAGE_SIZE / 2,
PCI_DMA_FROMDEVICE);
+ if (pci_dma_mapping_error(rx_ring->pdev,
+ buffer_info->page_dma)) {
+ buffer_info->page_dma = 0;
+ rx_ring->rx_stats.alloc_failed++;
+ goto no_buffers;
+ }
}
- if (!buffer_info->skb) {
+ skb = buffer_info->skb;
+ if (!skb) {
skb = netdev_alloc_skb_ip_align(netdev, bufsz);
if (!skb) {
rx_ring->rx_stats.alloc_failed++;
@@ -5095,10 +5102,18 @@ void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
}
buffer_info->skb = skb;
+ }
+ if (!buffer_info->dma) {
buffer_info->dma = pci_map_single(rx_ring->pdev,
skb->data,
bufsz,
PCI_DMA_FROMDEVICE);
+ if (pci_dma_mapping_error(rx_ring->pdev,
+ buffer_info->dma)) {
+ buffer_info->dma = 0;
+ rx_ring->rx_stats.alloc_failed++;
+ goto no_buffers;
+ }
}
/* Refresh the desc even if buffer_addrs didn't change because
* each write-back erases this info. */
@@ -5107,8 +5122,7 @@ void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
cpu_to_le64(buffer_info->page_dma);
rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
} else {
- rx_desc->read.pkt_addr =
- cpu_to_le64(buffer_info->dma);
+ rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
rx_desc->read.hdr_addr = 0;
}
^ permalink raw reply related
* [net-next-2.6 PATCH 19/23] igb: replace unecessary &adapter->hw with just hw where applicable
From: Jeff Kirsher @ 2009-10-28 9:51 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch just cleans up some unecessary references to the adapter->hw
member when it has already been placed in a local variable named hw.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 26 ++++++++++++--------------
1 files changed, 12 insertions(+), 14 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index d3e8316..b2c0c97 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1107,8 +1107,8 @@ int igb_up(struct igb_adapter *adapter)
void igb_down(struct igb_adapter *adapter)
{
- struct e1000_hw *hw = &adapter->hw;
struct net_device *netdev = adapter->netdev;
+ struct e1000_hw *hw = &adapter->hw;
u32 tctl, rctl;
int i;
@@ -1271,10 +1271,10 @@ void igb_reset(struct igb_adapter *adapter)
}
/* Allow time for pending master requests to run */
- adapter->hw.mac.ops.reset_hw(&adapter->hw);
+ hw->mac.ops.reset_hw(hw);
wr32(E1000_WUC, 0);
- if (adapter->hw.mac.ops.init_hw(&adapter->hw))
+ if (hw->mac.ops.init_hw(hw))
dev_err(&adapter->pdev->dev, "Hardware Error\n");
igb_update_mng_vlan(adapter);
@@ -1282,8 +1282,8 @@ void igb_reset(struct igb_adapter *adapter)
/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
- igb_reset_adaptive(&adapter->hw);
- igb_get_phy_info(&adapter->hw);
+ igb_reset_adaptive(hw);
+ igb_get_phy_info(hw);
}
static const struct net_device_ops igb_netdev_ops = {
@@ -1404,8 +1404,6 @@ static int __devinit igb_probe(struct pci_dev *pdev,
hw->subsystem_vendor_id = pdev->subsystem_vendor;
hw->subsystem_device_id = pdev->subsystem_device;
- /* setup the private structure */
- hw->back = adapter;
/* Copy the default MAC, PHY and NVM function pointers */
memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
@@ -1460,7 +1458,7 @@ static int __devinit igb_probe(struct pci_dev *pdev,
if (adapter->hw.mac.type == e1000_82576)
netdev->features |= NETIF_F_SCTP_CSUM;
- adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
+ adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
/* before reading the NVM, reset the controller to put the device in a
* known good starting state */
@@ -1705,8 +1703,8 @@ static void __devexit igb_remove(struct pci_dev *pdev)
unregister_netdev(netdev);
- if (!igb_check_reset_block(&adapter->hw))
- igb_reset_phy(&adapter->hw);
+ if (!igb_check_reset_block(hw))
+ igb_reset_phy(hw);
igb_clear_interrupt_scheme(adapter);
@@ -2928,9 +2926,9 @@ static void igb_watchdog_task(struct work_struct *work)
if (link) {
if (!netif_carrier_ok(netdev)) {
u32 ctrl;
- hw->mac.ops.get_speed_and_duplex(&adapter->hw,
- &adapter->link_speed,
- &adapter->link_duplex);
+ hw->mac.ops.get_speed_and_duplex(hw,
+ &adapter->link_speed,
+ &adapter->link_duplex);
ctrl = rd32(E1000_CTRL);
/* Links status message must follow this format */
@@ -5552,7 +5550,7 @@ static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
wr32(E1000_CTRL, ctrl);
/* Allow time for pending master requests to run */
- igb_disable_pcie_master(&adapter->hw);
+ igb_disable_pcie_master(hw);
wr32(E1000_WUC, E1000_WUC_PME_EN);
wr32(E1000_WUFC, wufc);
^ permalink raw reply related
* [net-next-2.6 PATCH 20/23] igb: add pci_dev in few spots to clean up use of dev_err/info/warn
From: Jeff Kirsher @ 2009-10-28 9:51 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch relpaces several references to adapter->pdev->dev with just
pdev->dev. This allows for cleanup of several multiline dev_err/info
calls.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 24 ++++++++++++------------
1 files changed, 12 insertions(+), 12 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index b2c0c97..264ff00 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1173,6 +1173,7 @@ void igb_reinit_locked(struct igb_adapter *adapter)
void igb_reset(struct igb_adapter *adapter)
{
+ struct pci_dev *pdev = adapter->pdev;
struct e1000_hw *hw = &adapter->hw;
struct e1000_mac_info *mac = &hw->mac;
struct e1000_fc_info *fc = &hw->fc;
@@ -1275,7 +1276,7 @@ void igb_reset(struct igb_adapter *adapter)
wr32(E1000_WUC, 0);
if (hw->mac.ops.init_hw(hw))
- dev_err(&adapter->pdev->dev, "Hardware Error\n");
+ dev_err(&pdev->dev, "Hardware Error\n");
igb_update_mng_vlan(adapter);
@@ -3704,17 +3705,18 @@ static struct net_device_stats *igb_get_stats(struct net_device *netdev)
static int igb_change_mtu(struct net_device *netdev, int new_mtu)
{
struct igb_adapter *adapter = netdev_priv(netdev);
+ struct pci_dev *pdev = adapter->pdev;
int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
u32 rx_buffer_len, i;
if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
(max_frame > MAX_JUMBO_FRAME_SIZE)) {
- dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
+ dev_err(&pdev->dev, "Invalid MTU setting\n");
return -EINVAL;
}
if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
- dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
+ dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
return -EINVAL;
}
@@ -3739,7 +3741,7 @@ static int igb_change_mtu(struct net_device *netdev, int new_mtu)
if (netif_running(netdev))
igb_down(adapter);
- dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
+ dev_info(&pdev->dev, "changing MTU from %d to %d\n",
netdev->mtu, new_mtu);
netdev->mtu = new_mtu;
@@ -4053,6 +4055,7 @@ static int __igb_notify_dca(struct device *dev, void *data)
{
struct net_device *netdev = dev_get_drvdata(dev);
struct igb_adapter *adapter = netdev_priv(netdev);
+ struct pci_dev *pdev = adapter->pdev;
struct e1000_hw *hw = &adapter->hw;
unsigned long event = *(unsigned long *)data;
@@ -4061,12 +4064,9 @@ static int __igb_notify_dca(struct device *dev, void *data)
/* if already enabled, don't do it again */
if (adapter->flags & IGB_FLAG_DCA_ENABLED)
break;
- /* Always use CB2 mode, difference is masked
- * in the CB driver. */
- wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
if (dca_add_requester(dev) == 0) {
adapter->flags |= IGB_FLAG_DCA_ENABLED;
- dev_info(&adapter->pdev->dev, "DCA enabled\n");
+ dev_info(&pdev->dev, "DCA enabled\n");
igb_setup_dca(adapter);
break;
}
@@ -4076,7 +4076,7 @@ static int __igb_notify_dca(struct device *dev, void *data)
/* without this a class_device is left
* hanging around in the sysfs model */
dca_remove_requester(dev);
- dev_info(&adapter->pdev->dev, "DCA disabled\n");
+ dev_info(&pdev->dev, "DCA disabled\n");
adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
}
@@ -4471,7 +4471,7 @@ static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
retval = igb_set_vf_vlan(adapter, msgbuf, vf);
break;
default:
- dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
+ dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
retval = -1;
break;
}
@@ -5472,6 +5472,7 @@ static void igb_restore_vlan(struct igb_adapter *adapter)
int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
{
+ struct pci_dev *pdev = adapter->pdev;
struct e1000_mac_info *mac = &adapter->hw.mac;
mac->autoneg = 0;
@@ -5495,8 +5496,7 @@ int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
break;
case SPEED_1000 + DUPLEX_HALF: /* not supported */
default:
- dev_err(&adapter->pdev->dev,
- "Unsupported Speed/Duplex configuration\n");
+ dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
return -EINVAL;
}
return 0;
^ permalink raw reply related
* [net-next-2.6 PATCH 21/23] igb: limit minimum mtu to 68 to keep ip bound to interface
From: Jeff Kirsher @ 2009-10-28 9:52 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
Limit the minimum mtu to 68 in order to prevent ip from being unbound from
the interface.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 3 +--
1 files changed, 1 insertions(+), 2 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 264ff00..846e64f 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -3709,8 +3709,7 @@ static int igb_change_mtu(struct net_device *netdev, int new_mtu)
int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
u32 rx_buffer_len, i;
- if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
- (max_frame > MAX_JUMBO_FRAME_SIZE)) {
+ if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
dev_err(&pdev->dev, "Invalid MTU setting\n");
return -EINVAL;
}
^ permalink raw reply related
* [net-next-2.6 PATCH 22/23] igb: open up SCTP checksum offloads to all MACs 82576 and newer
From: Jeff Kirsher @ 2009-10-28 9:52 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
Going forward the plan is to have the MACs support SCTP checksum offloads
so change the check from == to >=.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 846e64f..1a6c074 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1456,7 +1456,7 @@ static int __devinit igb_probe(struct pci_dev *pdev,
if (pci_using_dac)
netdev->features |= NETIF_F_HIGHDMA;
- if (adapter->hw.mac.type == e1000_82576)
+ if (hw->mac.type >= e1000_82576)
netdev->features |= NETIF_F_SCTP_CSUM;
adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
^ permalink raw reply related
* [net-next-2.6 PATCH 23/23] igb: cleanup whitespace issues in igb_main.c
From: Jeff Kirsher @ 2009-10-28 9:52 UTC (permalink / raw)
To: davem; +Cc: netdev, gospo, Alexander Duyck, Jeff Kirsher
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch goes through and cleans up whitespace issues in igb_main.c
to help improve readability.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 41 +++++++++++++++++++----------------------
1 files changed, 19 insertions(+), 22 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 1a6c074..b044c98 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1006,7 +1006,6 @@ static void igb_release_hw_control(struct igb_adapter *adapter)
ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
}
-
/**
* igb_get_hw_control - get control of the h/w from f/w
* @adapter: address of board private structure
@@ -1067,7 +1066,6 @@ static void igb_configure(struct igb_adapter *adapter)
* igb_up - Open the interface and prepare it to handle traffic
* @adapter: board private structure
**/
-
int igb_up(struct igb_adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
@@ -1288,7 +1286,7 @@ void igb_reset(struct igb_adapter *adapter)
}
static const struct net_device_ops igb_netdev_ops = {
- .ndo_open = igb_open,
+ .ndo_open = igb_open,
.ndo_stop = igb_close,
.ndo_start_xmit = igb_xmit_frame_adv,
.ndo_get_stats = igb_get_stats,
@@ -1444,7 +1442,6 @@ static int __devinit igb_probe(struct pci_dev *pdev,
netdev->features |= NETIF_F_IPV6_CSUM;
netdev->features |= NETIF_F_TSO;
netdev->features |= NETIF_F_TSO6;
-
netdev->features |= NETIF_F_GRO;
netdev->vlan_features |= NETIF_F_TSO;
@@ -1569,7 +1566,6 @@ static int __devinit igb_probe(struct pci_dev *pdev,
}
#endif
-
switch (hw->mac.type) {
case e1000_82576:
/*
@@ -1624,8 +1620,8 @@ static int __devinit igb_probe(struct pci_dev *pdev,
/* print bus type/speed/width info */
dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
netdev->name,
- ((hw->bus.speed == e1000_bus_speed_2500)
- ? "2.5Gb/s" : "unknown"),
+ ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
+ "unknown"),
((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
(hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
(hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
@@ -1658,8 +1654,8 @@ err_sw_init:
err_ioremap:
free_netdev(netdev);
err_alloc_etherdev:
- pci_release_selected_regions(pdev, pci_select_bars(pdev,
- IORESOURCE_MEM));
+ pci_release_selected_regions(pdev,
+ pci_select_bars(pdev, IORESOURCE_MEM));
err_pci_reg:
err_dma:
pci_disable_device(pdev);
@@ -1723,11 +1719,12 @@ static void __devexit igb_remove(struct pci_dev *pdev)
dev_info(&pdev->dev, "IOV Disabled\n");
}
#endif
+
iounmap(hw->hw_addr);
if (hw->flash_address)
iounmap(hw->flash_address);
- pci_release_selected_regions(pdev, pci_select_bars(pdev,
- IORESOURCE_MEM));
+ pci_release_selected_regions(pdev,
+ pci_select_bars(pdev, IORESOURCE_MEM));
free_netdev(netdev);
@@ -2288,9 +2285,7 @@ void igb_setup_rctl(struct igb_adapter *adapter)
*/
rctl |= E1000_RCTL_SECRC;
- /*
- * disable store bad packets and clear size bits.
- */
+ /* disable store bad packets and clear size bits. */
rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
/* enable LPE to prevent packets larger than max_frame_size */
@@ -2916,7 +2911,8 @@ static void igb_watchdog(unsigned long data)
static void igb_watchdog_task(struct work_struct *work)
{
struct igb_adapter *adapter = container_of(work,
- struct igb_adapter, watchdog_task);
+ struct igb_adapter,
+ watchdog_task);
struct e1000_hw *hw = &adapter->hw;
struct net_device *netdev = adapter->netdev;
struct igb_ring *tx_ring = adapter->tx_ring;
@@ -2935,14 +2931,14 @@ static void igb_watchdog_task(struct work_struct *work)
/* Links status message must follow this format */
printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
"Flow Control: %s\n",
- netdev->name,
- adapter->link_speed,
- adapter->link_duplex == FULL_DUPLEX ?
+ netdev->name,
+ adapter->link_speed,
+ adapter->link_duplex == FULL_DUPLEX ?
"Full Duplex" : "Half Duplex",
- ((ctrl & E1000_CTRL_TFCE) && (ctrl &
- E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
- E1000_CTRL_RFCE) ? "RX" : ((ctrl &
- E1000_CTRL_TFCE) ? "TX" : "None")));
+ ((ctrl & E1000_CTRL_TFCE) &&
+ (ctrl & E1000_CTRL_RFCE)) ? "RX/TX" :
+ ((ctrl & E1000_CTRL_RFCE) ? "RX" :
+ ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None")));
/* tweak tx_queue_len according to speed/duplex and
* adjust the timeout factor */
@@ -3724,6 +3720,7 @@ static int igb_change_mtu(struct net_device *netdev, int new_mtu)
/* igb_down has a dependency on max_frame_size */
adapter->max_frame_size = max_frame;
+
/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
* means we reserve 2 more, this pushes us to allocate from the next
* larger slab size.
^ permalink raw reply related
* Re: [PATCH net-next] bnx2x: Do Tx handling in a separate tasklet.
From: David Miller @ 2009-10-28 9:57 UTC (permalink / raw)
To: vladz; +Cc: IMCEAMAILTO-davem+40davemloft+2Enet, eilong, netdev
In-Reply-To: <8628FE4E7912BF47A96AE7DD7BAC0AADCB2CFF2028@SJEXCHCCR02.corp.ad.broadcom.com>
From: "Vladislav Zolotarov" <vladz@broadcom.com>
Date: Wed, 28 Oct 2009 02:54:37 -0700
> I'd like to start from your last remark: you r absolutely right, and this is the problem we have in the current net-next driver. More than that, this patch is fixing this problem: it moved liberation of Tx SKBs from hardIRQ context (ISR) to the softIRQ context (tasklet) thereby resolving the problem u've mentioned. So, total agreement with u on this one. I must have named the patch differently to emphasize it.
>
> I'd like to summarize the patch I've sent:
> - Take Tx SKB liberation out of hardIRQ.
> - Instead schedule a DPC that handles Tx work.
> - Optimize the access to status block indices: read only the index we are about to use in the current context.
>
> So, could u, pls., apply the patch in order to fix the problem we currently have in bnx2x?
There is no reason not to use NAPI to achieve this objective and that's
the main objection I have to your patch.
Using NAPI will not only allow you to move the SKB freeing to softirq
context but it will also provide fairness between multiple NAPI
contexts active at the same time on the same cpu.
Furthermore, if you combine RX and TX NAPI work for a specific queue
into the same NAPI context, TX liberation can run first and provide
fresh CPU local SKBs for RX packet input processing created replies
to allocate.
You haven't addressed any of that, and I am not going to apply your
patch becuase I don't want your driver to set a precedence here.
^ permalink raw reply
* Re: [PATCH v3 1/7] Only parse time stamp TCP option in time wait sock
From: Gilad Ben-Yossef @ 2009-10-28 10:14 UTC (permalink / raw)
To: William Allen Simpson; +Cc: netdev, ori, Yony Amit
In-Reply-To: <4AE5D4AE.2080108@gmail.com>
Hi William,
William Allen Simpson wrote:
> Gilad Ben-Yossef wrote:
>> Since we only use tcp_parse_options here to check for the exietence
>> of TCP timestamp option in the header, it is better to call with
>> the "established" flag on.
>>
> Please explain how this patch is required for the other patches?
Gladly (and suggestions to do it differently are welcome) :
For the purpose of the patch tcp_parse_options was changed to consult
dst_entry options when parsing non established packets.
This means that for any place that we call tcp_parse_options with the
established parameter set to false, we need to supply it with a dst_entry.
In all other locations in kernel code when tcp_parse_options is called
such a dst_entry is easily available already.
The time wait mini socket exists so that we would not waste resource
keeping around the full socket state of a "real socket". As such, it
does not cache the dst_entry. Adding it to the TIME_WAIT mini sockets
jsut for this purpose defeats the purpose of having a mini socket in the
first place.
One other possible way to go about it is to re-compute the dst_entry at
this location, but this seemed an expensive operation to perform for
what should be a light weight operation. I asked myself if there might
be another way?
So I took a good look at the code and discovered that there is no need
to call tcp_parse_options there in "non established" mode at all.
>
> And more importantly, why it is better to call with established on?
Sure. This is kind of long written down, although it's really simple. I
will try to describe it as best I can.
Take a look at what tcp_parse_options() does as a function -
It has only one output: changing the fields of the tcp_options_received
struct which it gets a pointer to as a parameter. It also has a single
side effect: it updates the SKB TCP control block sacked field, if a
SACK option is detected in the packet header.
Its behavior is dictated by the established parameter. If false, it will
try to parse all supported TCP options, if found in the packet header.
If true it will only try to parse the time stamp and SACK options.
Now take a look what happens at tcp_timewait_state_process() when we
call tcp_parse_options() -
We allocate (on stack) a temporary tcp_options_received struct, and if
our TIME_WAIT mini socket had recent timestamp data
(tcptw->tw_ts_recent_stamp), we call tcp_parse_options() with our
temporay tcp_options_received struct.
Here is the important bit: we never ever look at anything in the
tcp_options_received struct after the call returns, except for the time
stamp data if it is available!
So, passing established as false here makes us try to parse, if found in
the packet, a bunch of options that we never ever look at the result of.
(The fact that time wait minisocket code also zeros the saw_tstamp
before the call to tcp_parse_options although the same field is being
zeroed again inside the function is just icing on the cake...)
I have one more issue to explain, and this is regarding the single side
effect tcp_parse_option has - if the SACK option is found,
tcp_parse_options updates the skb control block sacked field. However,
not that it does this regardless of whether established is true or
false, so it doesn't matter which we pass. (I will leave out the fact
that whether or not the SACK option is parsed depends on a non
initialized field of the tcp_options_received struct now as an obscure
detail... nothing obviously looks at that later).
So bottom line: passing a true value in established does the exact same
thing, result wise, as current code, except it does so in fewer cycles.
I do confess to having goofed here in one regard: the patch I posted did
not set the tstamp_ok field of the tcp_options_received struct, which
can lead to randomly not parsing the time stamp option even when you
need to.
Perhaps this is what masks my intent. This is a bug of course and I'm
grateful for you for helping me catch it :-)
I will send an updated patch set with this fixed ASAP.
> And most importantly, what end cases you considered, and how this
> interacts with the proposed rfc1323bis changes, especially on reset?
>
My whole point was that this "change" does not change the behavior of
the code in any way. In fact, it is no different then a compile time
optimization (don't execute code paths nothing later uses the result
thereof) and if the compiler was smart enough, it would have done the
same. So corner cases and RFC compliance stay exactly as before.
I hope I managed to explain myself better this time around and thanks
again for taking the time to review this. ;-)
Gilad
--
Gilad Ben-Yossef
Chief Coffee Drinker & CTO
Codefidence Ltd.
Web: http://codefidence.com
Cell: +972-52-8260388
Skype: gilad_codefidence
Tel: +972-8-9316883 ext. 201
Fax: +972-8-9316884
Email: gilad@codefidence.com
Check out our Open Source technology and training blog - http://tuxology.net
"The biggest risk you can take it is to take no risk."
-- Mark Zuckerberg and probably others
^ permalink raw reply
* Re: [PATCH v3 4/7] Add the no SACK route option feature
From: Gilad Ben-Yossef @ 2009-10-28 10:18 UTC (permalink / raw)
To: William Allen Simpson; +Cc: netdev, ori
In-Reply-To: <4AE5D089.2050606@gmail.com>
William Allen Simpson wrote:
> Gilad Ben-Yossef wrote:
>> Implement querying and acting upon the no sack bit in the features
>> field.
>>
>> Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
>> Sigend-off-by: Ori Finkelman <ori@comsleep.com>
>> Sigend-off-by: Yony Amit <yony@comsleep.com>
>>
> Please explain how this code turns SACK on when it is off globally?
>
> As both Eric and I asked?
It doesn't. Please see my discussion with Eric for the why.
In short, doing so introduce a very subtle change to what the existing
interface do today, which will break
backwards compatibility by changing the meaning of writing zero to the
relevant sysctl. I don't want to be hunt down by angry sys admins :-)
Thanks,
Gilad
--
Gilad Ben-Yossef
Chief Coffee Drinker & CTO
Codefidence Ltd.
Web: http://codefidence.com
Cell: +972-52-8260388
Skype: gilad_codefidence
Tel: +972-8-9316883 ext. 201
Fax: +972-8-9316884
Email: gilad@codefidence.com
Check out our Open Source technology and training blog - http://tuxology.net
"The biggest risk you can take it is to take no risk."
-- Mark Zuckerberg and probably others
^ permalink raw reply
* Re: [PATCH] Multicast packet reassembly can fail
From: Eric Dumazet @ 2009-10-28 10:18 UTC (permalink / raw)
To: Steve Chen; +Cc: netdev
In-Reply-To: <1256683583.3153.389.camel@linux-1lbu>
Steve Chen a écrit :
> Multicast packet reassembly can fail
>
> When multicast connections with multiple fragments are received by the same
> node from more than one Ethernet ports, race condition between fragments
> from each Ethernet port can cause fragment reassembly to fail leading to
> packet drop. This is because packets from each Ethernet port appears identical
> to the the code that reassembles the Ethernet packet.
>
> The solution is evaluate the Ethernet interface number in addition to all other
> parameters so that every packet can be uniquely identified. The existing
> iif field in struct ipq is now used to generate the hash key, and iif is also
> used for comparison in case of hash collision.
>
> Please note that q->saddr ^ (q->iif << 5) is now being passed into
> ipqhashfn to generate the hash key. This is borrowed from the routing
> code.
>
> Signed-off-by: Steve Chen <schen@mvista.com>
> Signed-off-by: Mark Huth <mhuth@mvista.com>
>
This makes no sense to me, but I need to check the code.
How interface could matter in IP defragmentation ?
And why multicast is part of the equation ?
If defrag fails, this must be for other reason,
and probably needs another fix.
Check line 219 of net/ipv4/inet_fragment.c
#ifdef CONFIG_SMP
/* With SMP race we have to recheck hash table, because
* such entry could be created on other cpu, while we
* promoted read lock to write lock.
*/
hlist_for_each_entry(qp, n, &f->hash[hash], list) {
if (qp->net == nf && f->match(qp, arg)) {
atomic_inc(&qp->refcnt);
write_unlock(&f->lock);
qp_in->last_in |= INET_FRAG_COMPLETE; <<< HERE >>>
inet_frag_put(qp_in, f);
return qp;
}
}
#endif
I really wonder why we set INET_FRAG_COMPLETE here
^ permalink raw reply
* Re: [net-next-2.6 PATCH 01/23] igb: add support for seperate tx-usecs setting in ethtool
From: David Miller @ 2009-10-28 10:39 UTC (permalink / raw)
To: jeffrey.t.kirsher; +Cc: netdev, gospo, alexander.h.duyck
In-Reply-To: <20091028094540.13156.2637.stgit@localhost.localdomain>
All applied to net-next-2.6, but then I had to add the following
patch to kill a warning:
igb: Fix warnings in igb_set_ringparam()
drivers/net/igb/igb_ethtool.c: In function ‘igb_set_ringparam’:
drivers/net/igb/igb_ethtool.c:744: warning: comparison of distinct pointer types lacks a cast
drivers/net/igb/igb_ethtool.c:748: warning: comparison of distinct pointer types lacks a cast
Casts were to u16 on the constant, but the type of new_{r,t}x_count is
u32. Cast to u32 instead.
Signed-off-by: David S. Miller <davem@davemloft.net>
---
drivers/net/igb/igb_ethtool.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/igb/igb_ethtool.c b/drivers/net/igb/igb_ethtool.c
index d24b902..90b89a8 100644
--- a/drivers/net/igb/igb_ethtool.c
+++ b/drivers/net/igb/igb_ethtool.c
@@ -741,11 +741,11 @@ static int igb_set_ringparam(struct net_device *netdev,
return -EINVAL;
new_rx_count = min(ring->rx_pending, (u32)IGB_MAX_RXD);
- new_rx_count = max(new_rx_count, (u16)IGB_MIN_RXD);
+ new_rx_count = max(new_rx_count, (u32)IGB_MIN_RXD);
new_rx_count = ALIGN(new_rx_count, REQ_RX_DESCRIPTOR_MULTIPLE);
new_tx_count = min(ring->tx_pending, (u32)IGB_MAX_TXD);
- new_tx_count = max(new_tx_count, (u16)IGB_MIN_TXD);
+ new_tx_count = max(new_tx_count, (u32)IGB_MIN_TXD);
new_tx_count = ALIGN(new_tx_count, REQ_TX_DESCRIPTOR_MULTIPLE);
if ((new_tx_count == adapter->tx_ring_count) &&
--
1.6.5.1
^ permalink raw reply related
* Re: [PATCH] cnic: Fix L2CTX_STATUSB_NUM offset in context memory.
From: David Miller @ 2009-10-28 10:42 UTC (permalink / raw)
To: mchan; +Cc: davem, netdev, benli
In-Reply-To: <1256662728-21864-1-git-send-email-mchan@broadcom.com>
From: "Michael Chan" <mchan@broadcom.com>
Date: Tue, 27 Oct 2009 08:58:48 -0800
> The BNX2_L2CTX_STATUSB_NUM definition needs to be changed to match
> the recent firmware update:
>
> commit 078b0735881c7969aaf21469f3577831cddd9f8c
> bnx2: Update firmware to 5.0.0.j3.
>
> Without the fix, bnx2 can crash intermittently in bnx2_rx_int() when
> iSCSI is enabled.
>
> Signed-off-by: Michael Chan <mchan@broadcom.com>
> Signed-off-by: Benjamin Li <benli@broadcom.com>
Applied to net-2.6, but please be explicit about what tree you
want me to apply this to in the future.
Sure I could deduce this by running "git describe" on that
commit ID mentioned in the commit message, but why not be
explicit? :-)
^ 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