ixgbe: fix 82598 SFP initialization after driver load.
Subsystems (2)
PATCH 1/4 ixgbe: fix 82598 SFP initialization after driver load.
From: Don Skidmore <[email protected]>
If we loaded the driver with out a SFP module plugged in it would leave it in a state that make it later unable to link when a module was plugged in. This patch corrects that by:
ixgbe_probe() - moving the check for IXGBE_ERR_SFP_NOT_PRESENT from after get_invariants() to after reset_hw() as now reset_hw() is where this condition will be indentified.
ixgbe_reset_hw_82598() - Enable this function to now return IXGBE_ERR_SFP_NOT_PRESENT.
ixgbe_identify_sfp_module_generic() - This where the lack of SFP module is detected. Modifications are added to allow a different return value for modules that just haven't been plugged in yet.
Other functions were updated to allow correct logging.
---
drivers/net/ixgbe/ixgbe_82598.c | 12 ++++++++++-- drivers/net/ixgbe/ixgbe_main.c | 33 +++++++++++++++++---------------- drivers/net/ixgbe/ixgbe_phy.c | 13 ++++++++++++- 3 files changed, 39 insertions(+), 19 deletions(-)
drivers/net/ixgbe/ixgbe_82598.c +10 -2
@@ -698,6 +698,7 @@ static s32 ixgbe_setup_copper_link_speed_82598(struct ixgbe_hw *hw,static s32 ixgbe_reset_hw_82598 (struct ixgbe_hw *hw) {s32 status =0 ; +s32 phy_status =0 ;u32 ctrl;u32 gheccr;u32 i;
@@ -745,13 +746,17 @@ static s32 ixgbe_reset_hw_82598(struct ixgbe_hw *hw)/* PHY ops must be identified and initialized prior to reset */ /* Init PHY and function pointers, perform SFP setup */ - status = hw->phy.ops.init (hw); -if (status ==IXGBE_ERR_SFP_NOT_SUPPORTED ) + phy_status = hw->phy.ops.init (hw); +if (phy_status ==IXGBE_ERR_SFP_NOT_SUPPORTED ) goto reset_hw_out; +else if (phy_status ==IXGBE_ERR_SFP_NOT_PRESENT ) + goto no_phy_reset; + hw->phy.ops.reset (hw); } +no_phy_reset:/* * Prevent the PCI-E bus from from hanging by disabling PCI-E master * access and verify no pending requests before reset
@@ -811,6 +816,9 @@ static s32 ixgbe_reset_hw_82598(struct ixgbe_hw *hw) hw->mac.ops.get_mac_addr (hw, hw->mac.perm_addr); reset_hw_out: +if (phy_status) + status = phy_status; +return status; }
drivers/net/ixgbe/ixgbe_main.c +17 -16
@@ -2599,7 +2599,10 @@ int ixgbe_up(struct ixgbe_adapter *adapter)void ixgbe_reset (struct ixgbe_adapter *adapter) {struct ixgbe_hw *hw = &adapter->hw; -if (hw->mac.ops.init_hw (hw)) +int err; + + err = hw->mac.ops.init_hw (hw); +if (err && (err !=IXGBE_ERR_SFP_NOT_PRESENT ))dev_err (&adapter->pdev->dev,"Hardware Error\n" );/* reprogram the RAR[0] in case user changed it. */
@@ -5167,20 +5170,7 @@ static int __devinit ixgbe_probe(struct pci_dev *pdev,INIT_WORK (&adapter->sfp_config_module_task, ixgbe_sfp_config_module_task); - err = ii->get_invariants (hw); -if (err ==IXGBE_ERR_SFP_NOT_PRESENT ) { -/* start a kernel thread to watch for a module to arrive */ -set_bit (__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state); -mod_timer (&adapter->sfp_timer, -round_jiffies (jiffies + (2 *HZ ))); - err =0 ; - }else if (err ==IXGBE_ERR_SFP_NOT_SUPPORTED ) { -DPRINTK (PROBE ,ERR ,"failed to load because an " -"unsupported SFP+ module type was detected.\n" ); - goto err_hw_init; - }else if (err) { - goto err_hw_init; - } + ii->get_invariants (hw);/* setup the private structure */ err =ixgbe_sw_init (adapter);
@@ -5200,7 +5190,18 @@ static int __devinit ixgbe_probe(struct pci_dev *pdev,/* reset_hw fills in the perm_addr as well */ err = hw->mac.ops.reset_hw (hw); -if (err ==IXGBE_ERR_SFP_NOT_SUPPORTED ) { +if (err ==IXGBE_ERR_SFP_NOT_PRESENT && + hw->mac.type == ixgbe_mac_82598EB) { +/* +* Start a kernel thread to watch for a module to arrive. +* Only do this for 82598, since 82599 will generate +* interrupts on module arrival. +*/ +set_bit (__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state); +mod_timer (&adapter->sfp_timer, +round_jiffies (jiffies + (2 *HZ ))); + err =0 ; + }else if (err ==IXGBE_ERR_SFP_NOT_SUPPORTED ) {dev_err (&adapter->pdev->dev,"failed to load because an " "unsupported SFP+ module type was detected.\n" ); goto err_sw_init;
drivers/net/ixgbe/ixgbe_phy.c +12 -1
@@ -530,11 +530,22 @@ s32 ixgbe_identify_sfp_module_generic(struct ixgbe_hw *hw)u8 cable_tech =0 ;u16 enforce_sfp =0 ; +if (hw->mac.ops.get_media_type (hw) != ixgbe_media_type_fiber) { + hw->phy.sfp_type = ixgbe_sfp_type_not_present; + status =IXGBE_ERR_SFP_NOT_PRESENT ; + goto out; + } + status = hw->phy.ops.read_i2c_eeprom (hw,IXGBE_SFF_IDENTIFIER , &identifier); -if (status ==IXGBE_ERR_SFP_NOT_PRESENT ) { +if (status ==IXGBE_ERR_SFP_NOT_PRESENT || status ==IXGBE_ERR_I2C ) { + status =IXGBE_ERR_SFP_NOT_PRESENT ; hw->phy.sfp_type = ixgbe_sfp_type_not_present; +if (hw->phy.type != ixgbe_phy_nl) { + hw->phy.id =0 ; + hw->phy.type = ixgbe_phy_unknown; + } goto out; }
PATCH 2/4 ixgbe: fix driver loading with unsupported module on 82598
From: Don Skidmore <[email protected]>
Driver loads even when an unsupported module inserted. This was caused by mdio45_probe not returning all the information we needed. The fix is to replace that call with the new ixgbe_validate_phy_addr().
We will now need to store the phy address which we do in the hw->phy struct. Likewise we no longer need the mdio struct in hw->phy along with the mdio functions ixgbe_mdio_write(read) which have been removed. ixgbe_ioctl was no longer used and referenced hw->phy.mdio so it was also removed.
---
drivers/net/ixgbe/ixgbe.h | 1 + drivers/net/ixgbe/ixgbe_main.c | 42 ---------------------------------------- drivers/net/ixgbe/ixgbe_phy.c | 32 ++++++++++++++++++++++++------ drivers/net/ixgbe/ixgbe_phy.h | 1 + drivers/net/ixgbe/ixgbe_type.h | 5 ++++- 5 files changed, 31 insertions(+), 50 deletions(-)
drivers/net/ixgbe/ixgbe.h +1 -0
@@ -394,6 +394,7 @@ extern void ixgbe_update_stats(struct ixgbe_adapter *adapter);extern int ixgbe_init_interrupt_scheme (struct ixgbe_adapter *adapter);extern void ixgbe_clear_interrupt_scheme (struct ixgbe_adapter *adapter);extern void ixgbe_write_eitr (struct ixgbe_adapter *,int ,u32 ); +extern int ethtool_ioctl (struct ifreq *ifr);#ifdef IXGBE_FCOE extern void ixgbe_configure_fcoe (struct ixgbe_adapter *adapter); extern int ixgbe_fso(struct ixgbe_adapter *adapter,
drivers/net/ixgbe/ixgbe_main.c +0 -42
60 lines not on this page · raw message
drivers/net/ixgbe/ixgbe_phy.c +25 -7
@@ -60,7 +60,7 @@ s32 ixgbe_identify_phy_generic(struct ixgbe_hw *hw)if (hw->phy.type == ixgbe_phy_unknown) {for (phy_addr =0 ; phy_addr <IXGBE_MAX_PHY_ADDR ; phy_addr++) { -if (mdio45_probe (&hw->phy.mdio, phy_addr) ==0 ) { +if (ixgbe_validate_phy_addr (hw, phy_addr)) {ixgbe_get_phy_id (hw); hw->phy.type =ixgbe_get_phy_type_from_id (hw->phy.id);
@@ -76,6 +76,26 @@ s32 ixgbe_identify_phy_generic(struct ixgbe_hw *hw) }/** +* ixgbe_validate_phy_addr - Determines phy address is valid +* @hw: pointer to hardware structure +* +**/ +bool ixgbe_validate_phy_addr (struct ixgbe_hw *hw,u32 phy_addr) +{ +u16 phy_id =0 ; +bool valid = false; + + hw->phy.addr = phy_addr; + hw->phy.ops.read_reg (hw,IXGBE_MDIO_PHY_ID_HIGH , +IXGBE_MDIO_PMA_PMD_DEV_TYPE , &phy_id); + +if (phy_id !=0xFFFF && phy_id !=0x0 ) + valid = true; + +return valid; +} + +/** * ixgbe_get_phy_id - Get the phy type * @hw: pointer to hardware structure *
@@ -167,7 +187,7 @@ s32 ixgbe_read_phy_reg_generic(struct ixgbe_hw *hw, u32 reg_addr,/* Setup and write the address cycle command */ command = ((reg_addr <<IXGBE_MSCA_NP_ADDR_SHIFT ) | (device_type <<IXGBE_MSCA_DEV_TYPE_SHIFT ) | - (hw->phy.mdio.prtad <<IXGBE_MSCA_PHY_ADDR_SHIFT ) | + (hw->phy.addr <<IXGBE_MSCA_PHY_ADDR_SHIFT ) | (IXGBE_MSCA_ADDR_CYCLE |IXGBE_MSCA_MDI_COMMAND ));IXGBE_WRITE_REG (hw,IXGBE_MSCA , command);
@@ -198,8 +218,7 @@ s32 ixgbe_read_phy_reg_generic(struct ixgbe_hw *hw, u32 reg_addr,*/ command = ((reg_addr <<IXGBE_MSCA_NP_ADDR_SHIFT ) | (device_type <<IXGBE_MSCA_DEV_TYPE_SHIFT ) | - (hw->phy.mdio.prtad << -IXGBE_MSCA_PHY_ADDR_SHIFT ) | + (hw->phy.addr <<IXGBE_MSCA_PHY_ADDR_SHIFT ) | (IXGBE_MSCA_READ |IXGBE_MSCA_MDI_COMMAND ));IXGBE_WRITE_REG (hw,IXGBE_MSCA , command);
@@ -268,7 +287,7 @@ s32 ixgbe_write_phy_reg_generic(struct ixgbe_hw *hw, u32 reg_addr,/* Setup and write the address cycle command */ command = ((reg_addr <<IXGBE_MSCA_NP_ADDR_SHIFT ) | (device_type <<IXGBE_MSCA_DEV_TYPE_SHIFT ) | - (hw->phy.mdio.prtad <<IXGBE_MSCA_PHY_ADDR_SHIFT ) | + (hw->phy.addr <<IXGBE_MSCA_PHY_ADDR_SHIFT ) | (IXGBE_MSCA_ADDR_CYCLE |IXGBE_MSCA_MDI_COMMAND ));IXGBE_WRITE_REG (hw,IXGBE_MSCA , command);
@@ -299,8 +318,7 @@ s32 ixgbe_write_phy_reg_generic(struct ixgbe_hw *hw, u32 reg_addr,*/ command = ((reg_addr <<IXGBE_MSCA_NP_ADDR_SHIFT ) | (device_type <<IXGBE_MSCA_DEV_TYPE_SHIFT ) | - (hw->phy.mdio.prtad << -IXGBE_MSCA_PHY_ADDR_SHIFT ) | + (hw->phy.addr <<IXGBE_MSCA_PHY_ADDR_SHIFT ) | (IXGBE_MSCA_WRITE |IXGBE_MSCA_MDI_COMMAND ));IXGBE_WRITE_REG (hw,IXGBE_MSCA , command);
drivers/net/ixgbe/ixgbe_phy.h +1 -0
@@ -79,6 +79,7 @@s32 ixgbe_init_phy_ops_generic (struct ixgbe_hw *hw); +bool ixgbe_validate_phy_addr (struct ixgbe_hw *hw,u32 phy_addr);s32 ixgbe_identify_phy_generic (struct ixgbe_hw *hw);s32 ixgbe_reset_phy_generic (struct ixgbe_hw *hw); s32 ixgbe_read_phy_reg_generic(struct ixgbe_hw *hw, u32 reg_addr,
drivers/net/ixgbe/ixgbe_type.h +4 -1
@@ -915,6 +915,7 @@#define IXGBE_CORECTL_WRITE_CMD 0x00010000/* MDIO definitions */ +#define IXGBE_MDIO_PMA_PMD_DEV_TYPE 0x1#define IXGBE_MDIO_COMMAND_TIMEOUT 100/* PHY Timeout for 1 GB mode */
@@ -925,6 +926,8 @@#define IXGBE_MDIO_VENDOR_SPECIFIC_1_10G_SPEED 0x0018#define IXGBE_MDIO_VENDOR_SPECIFIC_1_1G_SPEED 0x0010 +#define IXGBE_MDIO_PHY_ID_HIGH 0x2/* PHY ID High Reg */ +#define IXGBE_MDIO_PMA_PMD_SDA_SCL_ADDR 0xC30A/* PHY_XS SDA/SCL Addr Reg */ #define IXGBE_MDIO_PMA_PMD_SDA_SCL_DATA 0xC30B/* PHY_XS SDA/SCL Data Reg */ #define IXGBE_MDIO_PMA_PMD_SDA_SCL_STAT 0xC30C/* PHY_XS SDA/SCL Status Reg */
@@ -2286,8 +2289,8 @@ struct ixgbe_mac_info {struct ixgbe_phy_info {struct ixgbe_phy_operations ops; -struct mdio_if_info mdio;enum ixgbe_phy_type type; +u32 addr;u32 id;enum ixgbe_sfp_type sfp_type;bool sfp_setup_needed;
PATCH 3/4 ixgbe: remove rx_hdr_split debug counter for non-debug configurations
In reply to Jeff KirsherFrom: Alexander Duyck <[email protected]>
The counter rx_hdr_split was meant to be a debug counter. As such it should only be enabled when debugging.
---
drivers/net/ixgbe/ixgbe.h | 2 ++ drivers/net/ixgbe/ixgbe_ethtool.c | 2 ++ drivers/net/ixgbe/ixgbe_main.c | 2 ++ 3 files changed, 6 insertions(+), 0 deletions(-)
drivers/net/ixgbe/ixgbe.h +2 -0
@@ -284,7 +284,9 @@ struct ixgbe_adapter {struct ixgbe_ring_feature ring_feature[RING_F_ARRAY_SIZE ];struct msix_entry *msix_entries; +#ifdef DEBUG u64 rx_hdr_split; +#endif u32 alloc_rx_page_failed;u32 alloc_rx_buff_failed;
drivers/net/ixgbe/ixgbe_ethtool.c +2 -0
@@ -87,7 +87,9 @@ static struct ixgbe_stats ixgbe_gstrings_stats[] = { {"rx_csum_offload_good", IXGBE_STAT(hw_csum_rx_good)}, {"rx_csum_offload_errors", IXGBE_STAT(hw_csum_rx_error)}, {"tx_csum_offload_ctxt", IXGBE_STAT(hw_csum_tx_good)}, +#ifdef DEBUG {"rx_header_split", IXGBE_STAT(rx_hdr_split)}, +#endif {"alloc_rx_page_failed", IXGBE_STAT(alloc_rx_page_failed)}, {"alloc_rx_buff_failed", IXGBE_STAT(alloc_rx_buff_failed)}, {"rx_no_dma_resources", IXGBE_STAT(hw_rx_no_dma_resources)},
drivers/net/ixgbe/ixgbe_main.c +2 -0
@@ -694,8 +694,10 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector, hdr_info =le16_to_cpu (ixgbe_get_hdr_info (rx_desc)); len = (hdr_info &IXGBE_RXDADV_HDRBUFLEN_MASK ) >>IXGBE_RXDADV_HDRBUFLEN_SHIFT ; +#ifdef DEBUG if (hdr_info &IXGBE_RXDADV_SPH ) adapter->rx_hdr_split++; +#endif if (len >IXGBE_RX_HDR_SIZE ) len =IXGBE_RX_HDR_SIZE ; upper_len =le16_to_cpu (rx_desc->wb.upper.length);
PATCH 4/4 ixgbe: Remove device ID 0x10d8
In reply to Jeff KirsherFrom: Peter P Waskiewicz Jr <[email protected]>
Device ID 0x10d8 is the default silicon device ID for 82599. However, the device will not be functional without an EEPROM, so we want to prevent the driver from loading on the device. Otherwise, the driver will load, but no PHY setup or PCIe setup will occur, causing the device to be unusable. To prevent users from encountering this, just remove the device ID.
---
drivers/net/ixgbe/ixgbe_82599.c | 1 - drivers/net/ixgbe/ixgbe_type.h | 1 - 2 files changed, 0 insertions(+), 2 deletions(-)
drivers/net/ixgbe/ixgbe_82599.c +0 -1
@@ -357,7 +357,6 @@ enum ixgbe_media_type ixgbe_get_media_type_82599(struct ixgbe_hw *hw) }switch (hw->device_id) { -case IXGBE_DEV_ID_82599 :case IXGBE_DEV_ID_82599_KX4 :case IXGBE_DEV_ID_82599_XAUI_LOM :/* Default device ID is mezzanine card KX/KX4 */
drivers/net/ixgbe/ixgbe_type.h +0 -1
@@ -46,7 +46,6 @@#define IXGBE_DEV_ID_82598_DA_DUAL_PORT 0x10F1#define IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM 0x10E1#define IXGBE_DEV_ID_82598EB_XF_LR 0x10F4 -#define IXGBE_DEV_ID_82599 0x10D8#define IXGBE_DEV_ID_82599_KX4 0x10F7#define IXGBE_DEV_ID_82599_SFP 0x10FB#define IXGBE_DEV_ID_82599_XAUI_LOM 0x10FC
Re: [net-next-2.6 PATCH 2/4] ixgbe: fix driver loading with unsupported module on 82598
In reply to Jeff KirsherQuoted 3 lines · Jeff Kirsher
From: Don Skidmore <[email protected]>
Driver loads even when an unsupported module inserted.
Why not; they're hot-pluggable.
Quoted 4 lines
This was
caused bymdio45_probenot returning all the information we needed. The fix is to replace that call with the newixgbe_validate_phy_addr().
You could improve mdio45_probe()...
Quoted 5 lines
We will now need to store the phy address which we do in the
hw->phystruct. Likewise we no longer need the mdio struct inhw->phyalong with the mdio functionsixgbe_mdio_write(read)which have been removed.ixgbe_ioctlwas no longer used and referencedhw->phy.mdioso it was also removed.
The ioctl is for the benefit of user-space!
[...]
drivers/net/ixgbe/ixgbe_type.h line 918
Quoted 9 lines
drivers/net/ixgbe/ixgbe_type.h +1
-0@@ -915,6 +915,7 @@#define IXGBE_CORECTL_WRITE_CMD 0x00010000/* MDIO definitions */ +#define IXGBE_MDIO_PMA_PMD_DEV_TYPE 0x1
This is called MDIO_MMD_PMAPMD.
drivers/net/ixgbe/ixgbe_type.h line 930
Quoted 8 lines
drivers/net/ixgbe/ixgbe_type.h +2
-0#define IXGBE_MDIO_COMMAND_TIMEOUT 100/* PHY Timeout for 1 GB mode */ @@ -925,6 +926,8 @@#define IXGBE_MDIO_VENDOR_SPECIFIC_1_10G_SPEED 0x0018#define IXGBE_MDIO_VENDOR_SPECIFIC_1_1G_SPEED 0x0010 +#define IXGBE_MDIO_PHY_ID_HIGH 0x2/* PHY ID High Reg */ +
This is called MDIO_DEVID1.
Ben.
Signature · 3 lines
Ben Hutchings, Senior Software Engineer, Solarflare Communications Not speaking for my employer; that's the marketing department's job. They asked us to note that Solarflare product names are trademarked.
David Miller
maintainerRe: [net-next-2.6 PATCH 3/4] ixgbe: remove rx_hdr_split debug counter for non-debug configurations
In reply to Jeff KirsherFrom: Jeff Kirsher <[email protected]>
Date: Tue, 26 May 2009 17:54:45 -0700
Quoted 5 lines
The counter
rx_hdr_splitwas meant to be a debug counter. As such it should only be enabled when debugging.Signed-off-by Alexander Duyck <[email protected]>Signed-off-by Jeff Kirsher <[email protected]>
I disagree with this change.
Any statistic is useful for diagnosing problems by users, and if you hide it behind DEBUG then users never see the facility.
Either it's unconditionally in the driver or it's unconditionally out. I say keep it in :-)
Re: [net-next-2.6 PATCH 4/4] ixgbe: Remove device ID 0x10d8
In reply to Jeff KirsherFrom: Jeff Kirsher <[email protected]>
Date: Tue, 26 May 2009 17:55:02 -0700
Quoted 8 lines
Device ID
0x10d8is the default silicon device ID for 82599. However, the device will not be functional without an EEPROM, so we want to prevent the driver from loading on the device. Otherwise, the driver will load, but no PHY setup or PCIe setup will occur, causing the device to be unusable. To prevent users from encountering this, just remove the device ID.Signed-off-by Peter P Waskiewicz Jr <[email protected]>Signed-off-by Jeff Kirsher <[email protected]>
Applied.
From: Jeff Kirsher <[email protected]>
Date: Tue, 26 May 2009 17:54:04 -0700
Quoted 15 lines
If we loaded the driver with out a SFP module plugged in it would leave it in a state that make it later unable to link when a module was plugged in. This patch corrects that by:
ixgbe_probe()- moving the check forIXGBE_ERR_SFP_NOT_PRESENTfrom afterget_invariants()to afterreset_hw()as nowreset_hw()is where this condition will be indentified.
ixgbe_reset_hw_82598()- Enable this function to now returnIXGBE_ERR_SFP_NOT_PRESENT.
ixgbe_identify_sfp_module_generic()- This where the lack of SFP module is detected. Modifications are added to allow a different return value for modules that just haven't been plugged in yet.
Quoted 5 lines
Other functions were updated to allow correct logging.
Signed-off-by Don Skidmore <[email protected]>Signed-off-by Peter P Waskiewicz Jr <[email protected]>Signed-off-by Jeff Kirsher <[email protected]>
Applied.
David Miller
maintainerRe: [net-next-2.6 PATCH 2/4] ixgbe: fix driver loading with unsupported module on 82598
In reply to Ben HutchingsFrom: Ben Hutchings <[email protected]>
Date: Wed, 27 May 2009 02:28:57 +0100
Quoted 10 lines · Jeff Kirsher
Quoted 7 lines
From: Don Skidmore <[email protected]>
We will now need to store the phy address which we do in the
hw->phystruct. Likewise we no longer need the mdio struct inhw->phyalong with the mdio functionsixgbe_mdio_write(read)which have been removed.ixgbe_ioctlwas no longer used and referencedhw->phy.mdioso it was also removed.The ioctl is for the benefit of user-space!
Indeed, this is completely bogus. I'm not applying this.
Re: [net-next-2.6 PATCH 3/4] ixgbe: remove rx_hdr_split debug counter for non-debug configurations
In reply to David MillerQuoted 9 lines · David Miller
From: Jeff Kirsher <[email protected]>
Date: Tue, 26 May 2009 17:54:45 -0700Quoted 5 lines
The counter
rx_hdr_splitwas meant to be a debug counter. As such it should only be enabled when debugging.Signed-off-by Alexander Duyck <[email protected]>Signed-off-by Jeff Kirsher <[email protected]>
Quoted 7 lines
I disagree with this change.
Any statistic is useful for diagnosing problems by users, and if you hide it behind DEBUG then users never see the facility.
Either it's unconditionally in the driver or it's unconditionally out. I say keep it in :-)
If rx_hdr_split means the same thing at Intel as it does here, it's a hardware feature Microsoft wants for use in Hyper-V and it won't be enabled under Linux.
Ben.
Signature · 3 lines
Ben Hutchings, Senior Software Engineer, Solarflare Communications Not speaking for my employer; that's the marketing department's job. They asked us to note that Solarflare product names are trademarked.
David Miller
maintainerRe: [net-next-2.6 PATCH 3/4] ixgbe: remove rx_hdr_split debug counter for non-debug configurations
From: Ben Hutchings <[email protected]>
Date: Wed, 27 May 2009 12:16:57 +0100
Quoted 3 lines
If
rx_hdr_splitmeans the same thing at Intel as it does here, it's a hardware feature Microsoft wants for use in Hyper-V and it won't be enabled under Linux.
Nobody explains anything, so it could or it could not be, who knows. :-)
RE: [net-next-2.6 PATCH 3/4] ixgbe: remove rx_hdr_split debug counter for non-debug configurations
Quoted 9 lines · David Miller
From: Ben Hutchings <[email protected]>
Date: Wed, 27 May 2009 12:16:57 +0100Quoted 3 lines
If
rx_hdr_splitmeans the same thing at Intel as it does here, it's a hardware feature Microsoft wants for use in Hyper-V and it won't be enabled under Linux.Nobody explains anything, so it could or it could not be, who knows. :-)
The feature mentioned here is header-data split. When it is enabled the igb driver puts the header into skb data, and the remainder into the page frags list. It is used when we enable jumbo frames on the igb driver. The counter I had removed in the patch was keeping track of the split header bit which is 1 when the header size reflected in the descriptor is correct.
The counter didn't really have much use other than in our early validation work to make certain the silicon was working as documented. If it stays it isn't too big a concern I just figured it might be nice to remove it since igb is a multiqueue driver and having a global counter can cause some cache thrash on multicore systems.
Thanks,
Alex
David Miller
maintainerRe: [net-next-2.6 PATCH 3/4] ixgbe: remove rx_hdr_split debug counter for non-debug configurations
From: "Duyck, Alexander H" <[email protected]>
Date: Thu, 28 May 2009 17:01:17 -0700
Quoted 6 lines
The counter didn't really have much use other than in our early validation work to make certain the silicon was working as documented. If it stays it isn't too big a concern I just figured it might be nice to remove it since igb is a multiqueue driver and having a global counter can cause some cache thrash on multicore systems.
If it really is just a silicon validation thing, then given the multiqueue concerns it's probably best to just remove it unconditionaly.