Skip to contentSkip to conversation

Revert: "PCI: Add ACS quirk for all Cavium devices"

PATCH In review Alex Williamson 2 participants 4 messages Feb 14, 2017 – Feb 15, 2017 PCI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert: "PCI: Add ACS quirk for all Cavium devices"

This reverts commit b404bcfbf035413dcce539c8ba2c9986d220d8ed.

The reverted commit makes no attempt to selectively consider devices, current or future. Instead, it whitelists the entire PCI vendor ID. This is a reckless approach as we clearly cannot know whether future device IDs within this vendor ID are susceptible to peer-to-peer. Additionally, the comment suggests this quirk is only relevant to ThunderX, which raises further doubt whether it is appropriate to apply to the entire vendor ID.

Cc Manish Jaggi <[email protected]>
Cc Tirumalesh Chalamarla <[email protected]>

---

Please submit a new quirk targeting specific devices

 drivers/pci/quirks.c |   15 ---------------
 1 file changed, 15 deletions(-)
drivers/pci/quirks.c +0 -15
index 1800befa8b8b..449eabb438e2 100644
@@ -4060,19 +4060,6 @@ static int pci_quirk_amd_sb_acs(struct pci_dev *dev, u16 acs_flags)
 #endif
 }
 
-static int pci_quirk_cavium_acs(struct pci_dev *dev, u16 acs_flags)
-{
-	/*
-	 * Cavium devices matching this quirk do not perform peer-to-peer
-	 * with other functions, allowing masking out these bits as if they
-	 * were unimplemented in the ACS capability.
-	 */
-	acs_flags &= ~(PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
-		       PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT);
-
-	return acs_flags ? 0 : 1;
-}
-
 /*
  * Many Intel PCH root ports do provide ACS-like features to disable peer
  * transactions and validate bus numbers in requests, but do not provide an
@@ -4276,8 +4263,6 @@ static int pci_quirk_mf_endpoint_acs(struct pci_dev *dev, u16 acs_flags)
 	{ PCI_VENDOR_ID_INTEL, PCI_ANY_ID, pci_quirk_intel_spt_pch_acs },
 	{ 0x19a2, 0x710, pci_quirk_mf_endpoint_acs }, /* Emulex BE3-R */
 	{ 0x10df, 0x720, pci_quirk_mf_endpoint_acs }, /* Emulex Skyhawk-R */
-	/* Cavium ThunderX */
-	{ PCI_VENDOR_ID_CAVIUM, PCI_ANY_ID, pci_quirk_cavium_acs },
 	{ 0 }
 };
 
#1 lore raw
Quoted 10 lines · On Tue, 14 Feb 2017 09:18:18 -0700 Alex Williamson…

This reverts commit b404bcfbf035413dcce539c8ba2c9986d220d8ed.

The reverted commit makes no attempt to selectively consider devices, current or future. Instead, it whitelists the entire PCI vendor ID. This is a reckless approach as we clearly cannot know whether future device IDs within this vendor ID are susceptible to peer-to-peer. Additionally, the comment suggests this quirk is only relevant to ThunderX, which raises further doubt whether it is appropriate to apply to the entire vendor ID.

Quoted 6 lines
Cc Manish Jaggi <[email protected]>
Cc Tirumalesh Chalamarla <[email protected]>

---

Please submit a new quirk targeting specific devices

Sorry, I saw the date on the original commit as January 30, 2016 and forgot we're now in 2017, so I figured that this had only recently gone in. Since it's actually been around since 4.6, I'll take a different approach. Instead of reverting, I'm going to assume that this applies to ThunderX devices as per the comment. I'll follow-up with a new patch that only applies this to device IDs matching ThunderX as found here http://pci-ids.ucw.cz/read/PC/177d Perhaps in the meantime, someone from Cavium will comment on which subset of those devices this is actually relevant to. Thanks,

Alex

Quoted 38 lines
 drivers/pci/quirks.c |   15 ---------------
 1 file changed, 15 deletions(-)
drivers/pci/quirks.c +0 -15
index 1800befa8b8b..449eabb438e2 100644
@@ -4060,19 +4060,6 @@ static int pci_quirk_amd_sb_acs(struct pci_dev *dev, u16 acs_flags)
 #endif
 }
 
-static int pci_quirk_cavium_acs(struct pci_dev *dev, u16 acs_flags)
-{
-	/*
-	 * Cavium devices matching this quirk do not perform peer-to-peer
-	 * with other functions, allowing masking out these bits as if they
-	 * were unimplemented in the ACS capability.
-	 */
-	acs_flags &= ~(PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
-		       PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT);
-
-	return acs_flags ? 0 : 1;
-}
-
 /*
  * Many Intel PCH root ports do provide ACS-like features to disable peer
  * transactions and validate bus numbers in requests, but do not provide an
@@ -4276,8 +4263,6 @@ static int pci_quirk_mf_endpoint_acs(struct pci_dev *dev, u16 acs_flags)
 	{ PCI_VENDOR_ID_INTEL, PCI_ANY_ID, pci_quirk_intel_spt_pch_acs },
 	{ 0x19a2, 0x710, pci_quirk_mf_endpoint_acs }, /* Emulex BE3-R */
 	{ 0x10df, 0x720, pci_quirk_mf_endpoint_acs }, /* Emulex Skyhawk-R */
-	/* Cavium ThunderX */
-	{ PCI_VENDOR_ID_CAVIUM, PCI_ANY_ID, pci_quirk_cavium_acs },
 	{ 0 }
 };
 
#2 lore raw

Hi Alex,

Quoted 19 lines · Alex Williamson
Quoted 10 lines · On Tue, 14 Feb 2017 09:18:18 -0700 Alex Williamson…

This reverts commit b404bcfbf035413dcce539c8ba2c9986d220d8ed.

The reverted commit makes no attempt to selectively consider devices, current or future. Instead, it whitelists the entire PCI vendor ID. This is a reckless approach as we clearly cannot know whether future device IDs within this vendor ID are susceptible to peer-to-peer. Additionally, the comment suggests this quirk is only relevant to ThunderX, which raises further doubt whether it is appropriate to apply to the entire vendor ID.

Quoted 6 lines
Cc Manish Jaggi <[email protected]>
Cc Tirumalesh Chalamarla <[email protected]>

---

Please submit a new quirk targeting specific devices

Quoted 9 lines

Sorry, I saw the date on the original commit as January 30, 2016 and forgot we're now in 2017, so I figured that this had only recently gone in. Since it's actually been around since 4.6, I'll take a different approach. Instead of reverting, I'm going to assume that this applies to ThunderX devices as per the comment. I'll follow-up with a new patch that only applies this to device IDs matching ThunderX as found here http://pci-ids.ucw.cz/read/PC/177d Perhaps in the meantime, someone from Cavium will comment on which subset of those devices this is actually relevant to.

I will provide another patch with the specific device Ids.

Quoted 41 lines

Thanks,

Alex

Quoted 37 lines
  drivers/pci/quirks.c |   15 ---------------
  1 file changed, 15 deletions(-)
drivers/pci/quirks.c +0 -15
index 1800befa8b8b..449eabb438e2 100644
@@ -4060,19 +4060,6 @@ static int pci_quirk_amd_sb_acs(struct pci_dev *dev, u16 acs_flags)
  #endif
  }
 
-static int pci_quirk_cavium_acs(struct pci_dev *dev, u16 acs_flags)
-{
-	/*
-	 * Cavium devices matching this quirk do not perform peer-to-peer
-	 * with other functions, allowing masking out these bits as if they
-	 * were unimplemented in the ACS capability.
-	 */
-	acs_flags &= ~(PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
-		       PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT);
-
-	return acs_flags ? 0 : 1;
-}
-
  /*
   * Many Intel PCH root ports do provide ACS-like features to disable peer
   * transactions and validate bus numbers in requests, but do not provide an
@@ -4276,8 +4263,6 @@ static int pci_quirk_mf_endpoint_acs(struct pci_dev *dev, u16 acs_flags)
  	{ PCI_VENDOR_ID_INTEL, PCI_ANY_ID, pci_quirk_intel_spt_pch_acs },
  	{ 0x19a2, 0x710, pci_quirk_mf_endpoint_acs }, /* Emulex BE3-R */
  	{ 0x10df, 0x720, pci_quirk_mf_endpoint_acs }, /* Emulex Skyhawk-R */
-	/* Cavium ThunderX */
-	{ PCI_VENDOR_ID_CAVIUM, PCI_ANY_ID, pci_quirk_cavium_acs },
  	{ 0 }
  };
 
#3 lore raw
Quoted 31 lines · On Wed, 15 Feb 2017 22:12:54 +0530 Manish Jaggi <[email protected]>…

Hi Alex,

Quoted 19 lines · Alex Williamson
Quoted 10 lines · On Tue, 14 Feb 2017 09:18:18 -0700 Alex Williamson…

This reverts commit b404bcfbf035413dcce539c8ba2c9986d220d8ed.

The reverted commit makes no attempt to selectively consider devices, current or future. Instead, it whitelists the entire PCI vendor ID. This is a reckless approach as we clearly cannot know whether future device IDs within this vendor ID are susceptible to peer-to-peer. Additionally, the comment suggests this quirk is only relevant to ThunderX, which raises further doubt whether it is appropriate to apply to the entire vendor ID.

Quoted 6 lines
Cc Manish Jaggi <[email protected]>
Cc Tirumalesh Chalamarla <[email protected]>

---

Please submit a new quirk targeting specific devices

Quoted 9 lines

Sorry, I saw the date on the original commit as January 30, 2016 and forgot we're now in 2017, so I figured that this had only recently gone in. Since it's actually been around since 4.6, I'll take a different approach. Instead of reverting, I'm going to assume that this applies to ThunderX devices as per the comment. I'll follow-up with a new patch that only applies this to device IDs matching ThunderX as found here http://pci-ids.ucw.cz/read/PC/177d Perhaps in the meantime, someone from Cavium will comment on which subset of those devices this is actually relevant to.

Quoted 1 line

I will provide another patch with the specific device Ids.

Perfect, thank you Manish.

Alex

Quoted 38 lines
Quoted 37 lines
Quoted 37 lines
  drivers/pci/quirks.c |   15 ---------------
  1 file changed, 15 deletions(-)
drivers/pci/quirks.c +0 -15
index 1800befa8b8b..449eabb438e2 100644
@@ -4060,19 +4060,6 @@ static int pci_quirk_amd_sb_acs(struct pci_dev *dev, u16 acs_flags)
  #endif
  }
 
-static int pci_quirk_cavium_acs(struct pci_dev *dev, u16 acs_flags)
-{
-	/*
-	 * Cavium devices matching this quirk do not perform peer-to-peer
-	 * with other functions, allowing masking out these bits as if they
-	 * were unimplemented in the ACS capability.
-	 */
-	acs_flags &= ~(PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
-		       PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT);
-
-	return acs_flags ? 0 : 1;
-}
-
  /*
   * Many Intel PCH root ports do provide ACS-like features to disable peer
   * transactions and validate bus numbers in requests, but do not provide an
@@ -4276,8 +4263,6 @@ static int pci_quirk_mf_endpoint_acs(struct pci_dev *dev, u16 acs_flags)
  	{ PCI_VENDOR_ID_INTEL, PCI_ANY_ID, pci_quirk_intel_spt_pch_acs },
  	{ 0x19a2, 0x710, pci_quirk_mf_endpoint_acs }, /* Emulex BE3-R */
  	{ 0x10df, 0x720, pci_quirk_mf_endpoint_acs }, /* Emulex Skyhawk-R */
-	/* Cavium ThunderX */
-	{ PCI_VENDOR_ID_CAVIUM, PCI_ANY_ID, pci_quirk_cavium_acs },
  	{ 0 }
  };