Skip to contentSkip to conversation

ixgbe: Implement PCIe AER support

PATCH Superseded Jeff Kirsher 5 participants 11 messages Dec 2, 2008 – Dec 9, 2008

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ixgbe: Implement PCIe AER support

From: PJ Waskiewicz <[email protected]>

This patch implements the PCIe Advanced Error Reporting callbacks in ixgbe. The 82598 hardware supports AER, so we enable it.

---

 drivers/net/ixgbe/ixgbe.h      |    1 +
 drivers/net/ixgbe/ixgbe_main.c |   26 ++++++++++++++++++--------
 2 files changed, 19 insertions(+), 8 deletions(-)
drivers/net/ixgbe/ixgbe.h +1 -0
index 6cbf26e..e112008 100644
@@ -32,6 +32,7 @@
 #include <linux/pci.h>
 #include <linux/netdevice.h>
 #include <linux/inet_lro.h>
+#include <linux/aer.h>
 
 #include "ixgbe_type.h"
 #include "ixgbe_common.h"
drivers/net/ixgbe/ixgbe_main.c +18 -8
index e014a73..e9cb430 100644
@@ -4021,6 +4021,8 @@ static int __devinit ixgbe_probe(struct pci_dev *pdev,
 		goto err_pci_reg;
 	}
 
+	pci_enable_pcie_error_reporting(pdev);
+
 	pci_set_master(pdev);
 	pci_save_state(pdev);
 
@@ -4296,6 +4298,8 @@ static void __devexit ixgbe_remove(struct pci_dev *pdev)
 
 	free_netdev(netdev);
 
+	pci_disable_pcie_error_reporting(pdev);
+
 	pci_disable_device(pdev);
 }
 
@@ -4333,21 +4337,27 @@ static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
 {
 	struct net_device *netdev = pci_get_drvdata(pdev);
 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
+	pci_ers_result_t result;
 
 	if (pci_enable_device(pdev)) {
 		DPRINTK(PROBE, ERR,
 		        "Cannot re-enable PCI device after reset.\n");
-		return PCI_ERS_RESULT_DISCONNECT;
-	}
-	pci_set_master(pdev);
-	pci_restore_state(pdev);
+		result = PCI_ERS_RESULT_DISCONNECT;
+	} else {
+		pci_set_master(pdev);
+		pci_restore_state(pdev);
 
-	pci_enable_wake(pdev, PCI_D3hot, 0);
-	pci_enable_wake(pdev, PCI_D3cold, 0);
+		pci_enable_wake(pdev, PCI_D3hot, 0);
+		pci_enable_wake(pdev, PCI_D3cold, 0);
 
-	ixgbe_reset(adapter);
+		ixgbe_reset(adapter);
+
+		result = PCI_ERS_RESULT_RECOVERED;
+	}
+
+	pci_cleanup_aer_uncorrect_error_status(pdev);
 
-	return PCI_ERS_RESULT_RECOVERED;
+	return result;
 }
 
 /**
#1 lore raw
Quoted 8 lines · On Mon, 01 Dec 2008 16:41:15 -0800 Jeff Kirsher <[email protected]>…

From: PJ Waskiewicz <[email protected]>

This patch implements the PCIe Advanced Error Reporting callbacks in ixgbe. The 82598 hardware supports AER, so we enable it.

---

Once again: pcie-aer is not checking for mmconfig working! If you turn it on, you may get screaming errors.

pci_enable_pcie_error_reporting should be checking for pci_read/pci_write config errors and does not.

Your driver should be checking for pci_enable_pcie_error_reporting failing.

maintainer
#2 lore raw

From: Stephen Hemminger <[email protected]>
Date: Mon, 1 Dec 2008 17:47:54 -0800

Quoted 4 lines

pci_enable_pcie_error_reporting should be checking for pci_read/pci_write config errors and does not.

Your driver should be checking for pci_enable_pcie_error_reporting failing.

Agreed.

#3 lore raw

PATCH 1/1 pciaer: report config read/write errors

This patch does more error checking in the Advanced Error Reporting code. Since AER needs to access PCI registers > 255, it won't work without MMCONFIG and other quirks may stop it as well. The code must check this by looking at return values from pci_read/write_config_XXX calls.

I don't have any hardware that uses AER routines but discovered this in earlier versions of the sky2 driver that tried to use
pci AER routines. Ended up just giving up and using other ways to access PCI config space on sky2 since there were too many platform glitches.

drivers/pci/pcie/aer/aerdrv_core.c +32 -20
@@ -31,80 +31,92 @@ module_param(forceload, bool, 0);
 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
 {
 	u16 reg16 = 0;
-	int pos;
+	int pos, err;
+	u32 status;
 
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+	if (err)
+		return err;
+
 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	err = pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	if (err)
+		return err;
+
 	reg16 = reg16 |
 		PCI_EXP_DEVCTL_CERE |
 		PCI_EXP_DEVCTL_NFERE |
 		PCI_EXP_DEVCTL_FERE |
 		PCI_EXP_DEVCTL_URRE;
-	pci_write_config_word(dev, pos+PCI_EXP_DEVCTL,
-			reg16);
-	return 0;
+	return pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
 }
 
 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
 {
 	u16 reg16 = 0;
-	int pos;
+	int pos, err;
 
 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	err = pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	if (err)
+		return err;
+
 	reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
 			PCI_EXP_DEVCTL_NFERE |
 			PCI_EXP_DEVCTL_FERE |
 			PCI_EXP_DEVCTL_URRE);
-	pci_write_config_word(dev, pos+PCI_EXP_DEVCTL,
-			reg16);
-	return 0;
+	return pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
 }
 
 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
 {
-	int pos;
+	int pos, err;
 	u32 status, mask;
 
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
-	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+	if (err)
+		return err;
+
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
+	if (err)
+		return err;
+
 	if (dev->error_state == pci_channel_io_normal)
 		status &= ~mask; /* Clear corresponding nonfatal bits */
 	else
 		status &= mask; /* Clear corresponding fatal bits */
-	pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
-
-	return 0;
+	return pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
 }
 
 #if 0
 int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
 {
-	int pos;
+	int pos, err;
 	u32 status;
 
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
-	pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
+	err = pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
+	if (err)
+		return err;
 
-	return 0;
+	return pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
 }
 #endif  /*  0  */
 
#4 lore raw

Re: [PATCH] pciaer: report config read/write errors

Quoted 118 lines · Stephen Hemminger

This patch does more error checking in the Advanced Error Reporting code. Since AER needs to access PCI registers > 255, it won't work without MMCONFIG and other quirks may stop it as well. The code must check this by looking at return values from pci_read/write_config_XXX calls.

I don't have any hardware that uses AER routines but discovered this in earlier versions of the sky2 driver that tried to use
pci AER routines. Ended up just giving up and using other ways to access PCI config space on sky2 since there were too many platform glitches.

drivers/pci/pcie/aer/aerdrv_core.c +28 -17
@@ -31,80 +31,92 @@ module_param(forceload, bool, 0);
 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
 {
 	u16 reg16 = 0;
-	int pos;
+	int pos, err;
+	u32 status;
 
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+	if (err)
+		return err;
+
 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	err = pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	if (err)
+		return err;
+
 	reg16 = reg16 |
 		PCI_EXP_DEVCTL_CERE |
 		PCI_EXP_DEVCTL_NFERE |
 		PCI_EXP_DEVCTL_FERE |
 		PCI_EXP_DEVCTL_URRE;
-	pci_write_config_word(dev, pos+PCI_EXP_DEVCTL,
-			reg16);
-	return 0;
+	return pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
 }
 
 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
 {
 	u16 reg16 = 0;
-	int pos;
+	int pos, err;
 
 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	err = pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
+	if (err)
+		return err;
+
 	reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
 			PCI_EXP_DEVCTL_NFERE |
 			PCI_EXP_DEVCTL_FERE |
 			PCI_EXP_DEVCTL_URRE);
-	pci_write_config_word(dev, pos+PCI_EXP_DEVCTL,
-			reg16);
-	return 0;
+	return pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
 }
 
 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
 {
-	int pos;
+	int pos, err;
 	u32 status, mask;
 
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 
-	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
-	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+	if (err)
+		return err;
+
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
+	if (err)
+		return err;
+
 	if (dev->error_state == pci_channel_io_normal)
 		status &= ~mask; /* Clear corresponding nonfatal bits */
 	else
 		status &= mask; /* Clear corresponding fatal bits */
-	pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
-
-	return 0;
+	return pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
 }
 
 #if 0
 int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
 {
-	int pos;
+	int pos, err;
 	u32 status;
 
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 

drivers/pci/pcie/aer/aerdrv_core.c module_param(forceload, bool, 0); · line 121

Quoted 10 lines
drivers/pci/pcie/aer/aerdrv_core.c +4 -3
-	pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
-	pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
+	err = pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
+	if (err)
+		return err;
 
-	return 0;
+	return pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
 }
 #endif  /*  0  */

This looks fine to me. Thanks very much Stephen.

-PJ Waskiewicz

#5 lore raw

Re: [PATCH] pciaer: report config read/write errors

In reply to Stephen Hemminger
Quoted 9 lines · Stephen Hemminger

This patch does more error checking in the Advanced Error Reporting code. Since AER needs to access PCI registers > 255, it won't work without MMCONFIG and other quirks may stop it as well. The code must check this by looking at return values from pci_read/write_config_XXX calls.

I don't have any hardware that uses AER routines but discovered this in earlier versions of the sky2 driver that tried to use pci AER routines. Ended up just giving up and using other ways to access PCI
config space on sky2 since there were too many platform glitches.

When experimenting with sky2 driver, was pci_find_ext_capability() returning non-zero although further ext-space accesses were failing?

Quoted 14 lines
drivers/pci/pcie/aer/aerdrv_core.c +2 -1
@@ -31,80 +31,92 @@ module_param(forceload, bool, 0);
 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
 {
 	u16 reg16 = 0;
-	int pos;
+	int pos, err;
+	u32 status;
 

drivers/pci/pcie/aer/aerdrv_core.c module_param(forceload, bool, 0); · line 46

Quoted 10 lines
drivers/pci/pcie/aer/aerdrv_core.c +4 -0
 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
 	if (!pos)
 		return -EIO;
 
+	err = pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+	if (err)
+		return err;
+
 
 

For legacy-conf-space, most kernel code assumes success without checking. For ext-conf-space, wouldn't it be convenient to be able to make the same assumption when pci_find_ext_capability() returns a valid offset?

The patch looks good to me, but I am just asking whether there is a known case where pcie_find_ext_capability() returns a valid offset, although that offset might turn out unusable (it might be worth investigating pci_find_ext_capability() then).

Loic

#6 lore raw

Re: [PATCH] pciaer: report config read/write errors

Quoted 15 lines · On Tue, 02 Dec 2008 14:04:15 -0500 Loic Prylli <[email protected]> wrote:
Quoted 10 lines · Stephen Hemminger

This patch does more error checking in the Advanced Error Reporting code. Since AER needs to access PCI registers > 255, it won't work without MMCONFIG and other quirks may stop it as well. The code must check this by looking at return values from pci_read/write_config_XXX calls.

I don't have any hardware that uses AER routines but discovered this in earlier versions of the sky2 driver that tried to use
pci AER routines. Ended up just giving up and using other ways to access PCI config space on sky2 since there were too many platform glitches.

Quoted 3 lines

When experimenting with sky2 driver, was pci_find_ext_capability() returning non-zero although further ext-space accesses were failing?

No pci_find_ext_capability would succeed but all access to registers >= 256 would fail if MMCONFIG failed. Since the device often asserts stray errors on boot, if the error could not be cleared, the driver would get stuck when IRQ was enabled.

P.s: you can look back into LKML for Linus discussion about why MMCONFIG is broken anyway.

#7 lore raw

Re: [PATCH] pciaer: report config read/write errors

Quoted 2 lines · Stephen Hemminger

pci_find_ext_capability would succeed but all access to registers >= 256 would fail if MMCONFIG failed.

Your sentence seems self-contradicting to me. pci_find_ext_capability() (!= pci_find_capability()) only tries to access registers >= 256, so I don't see how it would succeed if *all* those accesses are failing.

Quoted 3 lines

P.s: you can look back into LKML for Linus discussion about why MMCONFIG is broken anyway.

I was an active participant in one of those discussions (although it is possible I missed a later one), so I am quite aware about the limitations of MMCONFIG. But pci_find_ext_capability() looked to me a good filter to check about those limitations().

This is not just about pciaer, for instance, it might be that a lot of the SR-IOV code submitted recently assumes that if the corresponding extended-capability can be detected, little checking needs to be done afterwards.

Loic

#8 lore raw

Re: [PATCH] pciaer: report config read/write errors

Quoted 22 lines · On Tue, 02 Dec 2008 15:14:35 -0500 Loic Prylli <[email protected]> wrote:
Quoted 2 lines · Stephen Hemminger

pci_find_ext_capability would succeed but all access to registers >= 256 would fail if MMCONFIG failed.

Your sentence seems self-contradicting to me. pci_find_ext_capability() (!= pci_find_capability()) only tries to access registers >= 256, so I don't see how it would succeed if *all* those accesses are failing.

Quoted 3 lines

P.s: you can look back into LKML for Linus discussion about why MMCONFIG is broken anyway.

I was an active participant in one of those discussions (although it is possible I missed a later one), so I am quite aware about the limitations of MMCONFIG. But pci_find_ext_capability() looked to me a good filter to check about those limitations().

Quoted 4 lines

This is not just about pciaer, for instance, it might be that a lot of the SR-IOV code submitted recently assumes that if the corresponding extended-capability can be detected, little checking needs to be done afterwards.

I misremembered the issue.
The problem in sky2 was related but not the same. pci_find_ext_capability would return 0, but the device driver needed to be able to clear errors (or IRQ would scream). So since the existing extended capability code wouldn't work on all platforms, another alternative had to be found.

#9 lore raw
In reply to David Miller
Quoted 9 lines · David Miller

From: Stephen Hemminger <[email protected]>
Date: Mon, 1 Dec 2008 17:47:54 -0800

Quoted 4 lines

pci_enable_pcie_error_reporting should be checking for pci_read/pci_write config errors and does not.

Your driver should be checking for pci_enable_pcie_error_reporting failing.

Agreed.

I've finally been able to get back to this. I've looked at any other prior use of the PCIe AER routines, and the only driver using it to date is the QLogic QLA2xxx driver in SCSI-land. I've also looked at the return codes, with and without Stephen's recent AER patch, and there's not much the driver can do. If it's catching the errors, the best I can do is print an error message, then continue along.

I've put that patch together, and will have it sent along shortly. If there's something additional I'm missing here, please let me know.

Cheers,
-PJ

#10 lore raw
In reply to Stephen Hemminger
Quoted 12 lines · Stephen Hemminger
Quoted 8 lines · On Mon, 01 Dec 2008 16:41:15 -0800 Jeff Kirsher <[email protected]>…

From: PJ Waskiewicz <[email protected]>

This patch implements the PCIe Advanced Error Reporting callbacks in ixgbe. The 82598 hardware supports AER, so we enable it.

---

Quoted 7 lines

Once again: pcie-aer is not checking for mmconfig working! If you turn it on, you may get screaming errors.

pci_enable_pcie_error_reporting should be checking for pci_read/pci_write config errors and does not.

Your driver should be checking for pci_enable_pcie_error_reporting failing.

I'll respin and have a new patch sent for this. I just finally saw the igb thread, so my apologies for missing this earlier.

Cheers,
-PJ