Export pcie_find_root_port() so we can use it outside of PCIe-AER error
injection.
Signed-off-by: Johannes Thumshirn <[email protected]>
---
drivers/pci/pcie/aer/aer_inject.c | 14 --------------
include/linux/pci.h | 14 ++++++++++++++
2 files changed, 14 insertions(+), 14 deletions(-)
diff --git a/drivers/pci/pcie/aer/aer_inject.c b/drivers/pci/pcie/aer/aer_inject.c
index db553dc..2b6a592 100644
--- a/drivers/pci/pcie/aer/aer_inject.c
+++ b/drivers/pci/pcie/aer/aer_inject.c
@@ -307,20 +307,6 @@ static int pci_bus_set_aer_ops(struct pci_bus *bus)
return 0;
}
-static struct pci_dev *pcie_find_root_port(struct pci_dev *dev)
-{
- while (1) {
- if (!pci_is_pcie(dev))
- break;
- if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
- return dev;
- if (!dev->bus->self)
- break;
- dev = dev->bus->self;
- }
- return NULL;
-}
-
static int find_aer_device_iter(struct device *device, void *data)
{
struct pcie_device **result = data;
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 0e49f70..a38772a 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -1928,6 +1928,20 @@ static inline int pci_pcie_type(const struct pci_dev *dev)
return (pcie_caps_reg(dev) & PCI_EXP_FLAGS_TYPE) >> 4;
}
+static inline struct pci_dev *pcie_find_root_port(struct pci_dev *dev)
+{
+ while (1) {
+ if (!pci_is_pcie(dev))
+ break;
+ if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
+ return dev;
+ if (!dev->bus->self)
+ break;
+ dev = dev->bus->self;
+ }
+ return NULL;
+}
+
void pci_request_acs(void);
bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags);
bool pci_acs_path_enabled(struct pci_dev *start,
--
2.10.0
The Read Completion Boundary (RCB) bit must only be set on a device or
endpoint if it is set on the root complex.
Certain BIOSes erroneously set the RCB Bit in their ACPI _HPX Tables
even if it is not set on the root port. This is a violation to the PCIe
Specification and is known to bring some Mellanox Connect-X 3 HCAs into
a state where they can't map their firmware and go into error recovery.
BIOS Information
Vendor: IBM
Version: -[A8E120CUS-1.30]-
Release Date: 08/22/2016
>From PCI Express Base Specification 1.1,
section 2.3.1.1. Data Return for Read Requests:
The Read Completion Boundary (RCB) parameter determines the naturally
aligned address boundaries on which a Read Request may be serviced with
multiple Completions
o For a Root Complex, RCB is 64 bytes or 128 bytes
o This value is reported through a configuration register
(see Section 7.8)
Note: Bridges and Endpoints may implement a corresponding command
bit which may be set by system software to indicate the RCB value
for the Root Complex, allowing the Bridge/Endpoint to optimize its
behavior when the Root Complex’s RCB is 128 bytes.
o For all other system elements, RCB is 128 bytes
Table 7-16: Link Control Register:
Configuration software must only Set this bit if the Root Port
Upstream from the Endpoint or Bridge reports an RCB value of
128 bytes (a value of 1b in the Read Completion Boundary bit).
Default value of this bit is 0b.
Functions that do not implement this feature must hardwire the
bit to 0b.
Before commit 7a1562d4f:
> 41:00.0 Ethernet controller: Mellanox Technologies MT27500 Family [ConnectX-3]
> LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
> ExtSynch+ ClockPM- AutWidDis- BWInt- AutBWInt-
>
> 40:02.0 PCI bridge: Intel Corporation Xeon E7 v2/Xeon E5 v2/Core i7 PCI Express Root Port 2a (rev 07) (prog-if 00 [Normal decode])
> LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
> ExtSynch+ ClockPM- AutWidDis- BWInt- AutBWInt-
After:
> 40:02.0 PCI bridge: Intel Corporation Xeon E7 v2/Xeon E5 v2/Core i7 PCI Express Root Port 2a (rev 07) (prog-if 00 [Normal decode])
> LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
> ExtSynch+ ClockPM- AutWidDis- BWInt- AutBWInt-
>
> 41:00.0 Ethernet controller: Mellanox Technologies MT27500 Family [ConnectX-3]
> LnkCtl: ASPM Disabled; RCB 128 bytes Disabled- CommClk+
> ExtSynch- ClockPM- AutWidDis- BWInt- AutBWInt-
Fixes: 7a1562d4f ("PCI: Apply _HPX Link Control settings to all devices with a link")
Signed-off-by: Johannes Thumshirn <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
---
drivers/pci/probe.c | 29 +++++++++++++++++++++++++++--
1 file changed, 27 insertions(+), 2 deletions(-)
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index ab00267..0a4ab9c 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -1439,6 +1439,19 @@ static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
dev_warn(&dev->dev, "PCI-X settings not supported\n");
}
+static bool pcie_get_root_rcb(struct pci_dev *dev)
+{
+ struct pci_dev *rp = pcie_find_root_port(dev);
+ u16 lnkctl;
+
+ if (!rp)
+ return false;
+
+ pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
+
+ return lnkctl & PCI_EXP_LNKCTL_RCB;
+}
+
static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
{
int pos;
@@ -1468,9 +1481,21 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
/* Initialize Link Control Register */
- if (pcie_cap_has_lnkctl(dev))
+ if (pcie_cap_has_lnkctl(dev)) {
+ bool rrcb;
+ u16 clear;
+ u16 set;
+
+ rrcb = pcie_get_root_rcb(dev);
+
+ clear = ~hpp->pci_exp_lnkctl_and;
+ set = hpp->pci_exp_lnkctl_or;
+ if (!rrcb)
+ set &= ~PCI_EXP_LNKCTL_RCB;
+
pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
- ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
+ clear, set);
+ }
/* Find Advanced Error Reporting Enhanced Capability */
pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
--
2.10.0
Hi Johannes,
On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> The Read Completion Boundary (RCB) bit must only be set on a device or
> endpoint if it is set on the root complex.
>
> Certain BIOSes erroneously set the RCB Bit in their ACPI _HPX Tables
> even if it is not set on the root port. This is a violation to the PCIe
> Specification and is known to bring some Mellanox Connect-X 3 HCAs into
> a state where they can't map their firmware and go into error recovery.
>
> BIOS Information
> Vendor: IBM
> Version: -[A8E120CUS-1.30]-
> Release Date: 08/22/2016
This seems like a pretty serious problem (sounds like maybe the HCA is
completely useless?)
Can you point us at a bugzilla or other problem report? It's nice to
have details of what this looks like to a user, so people who trip
over this problem have a little more chance of finding the solution.
7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
with a link") appeared in v3.18, so it's probably not a *new* problem,
so my guess is that this is v4.10 material.
> From PCI Express Base Specification 1.1,
> section 2.3.1.1. Data Return for Read Requests:
> The Read Completion Boundary (RCB) parameter determines the naturally
> aligned address boundaries on which a Read Request may be serviced with
> multiple Completions
> o For a Root Complex, RCB is 64 bytes or 128 bytes
> o This value is reported through a configuration register
> (see Section 7.8)
> Note: Bridges and Endpoints may implement a corresponding command
> bit which may be set by system software to indicate the RCB value
> for the Root Complex, allowing the Bridge/Endpoint to optimize its
> behavior when the Root Complex’s RCB is 128 bytes.
> o For all other system elements, RCB is 128 bytes
>
> Table 7-16: Link Control Register:
> Configuration software must only Set this bit if the Root Port
> Upstream from the Endpoint or Bridge reports an RCB value of
> 128 bytes (a value of 1b in the Read Completion Boundary bit).
> Default value of this bit is 0b.
>
> Functions that do not implement this feature must hardwire the
> bit to 0b.
>
> Before commit 7a1562d4f:
> > 41:00.0 Ethernet controller: Mellanox Technologies MT27500 Family [ConnectX-3]
> > LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
> > ExtSynch+ ClockPM- AutWidDis- BWInt- AutBWInt-
> >
> > 40:02.0 PCI bridge: Intel Corporation Xeon E7 v2/Xeon E5 v2/Core i7 PCI Express Root Port 2a (rev 07) (prog-if 00 [Normal decode])
> > LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
> > ExtSynch+ ClockPM- AutWidDis- BWInt- AutBWInt-
>
> After:
> > 40:02.0 PCI bridge: Intel Corporation Xeon E7 v2/Xeon E5 v2/Core i7 PCI Express Root Port 2a (rev 07) (prog-if 00 [Normal decode])
> > LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
> > ExtSynch+ ClockPM- AutWidDis- BWInt- AutBWInt-
> >
> > 41:00.0 Ethernet controller: Mellanox Technologies MT27500 Family [ConnectX-3]
> > LnkCtl: ASPM Disabled; RCB 128 bytes Disabled- CommClk+
> > ExtSynch- ClockPM- AutWidDis- BWInt- AutBWInt-
>
> Fixes: 7a1562d4f ("PCI: Apply _HPX Link Control settings to all devices with a link")
> Signed-off-by: Johannes Thumshirn <[email protected]>
> Reviewed-by: Hannes Reinecke <[email protected]>
> ---
> drivers/pci/probe.c | 29 +++++++++++++++++++++++++++--
> 1 file changed, 27 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index ab00267..0a4ab9c 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1439,6 +1439,19 @@ static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
> dev_warn(&dev->dev, "PCI-X settings not supported\n");
> }
>
> +static bool pcie_get_root_rcb(struct pci_dev *dev)
> +{
> + struct pci_dev *rp = pcie_find_root_port(dev);
> + u16 lnkctl;
> +
> + if (!rp)
> + return false;
> +
> + pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
> +
> + return lnkctl & PCI_EXP_LNKCTL_RCB;
> +}
> +
> static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
> {
> int pos;
> @@ -1468,9 +1481,21 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
> ~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
>
> /* Initialize Link Control Register */
> - if (pcie_cap_has_lnkctl(dev))
> + if (pcie_cap_has_lnkctl(dev)) {
> + bool rrcb;
> + u16 clear;
> + u16 set;
> +
> + rrcb = pcie_get_root_rcb(dev);
> +
> + clear = ~hpp->pci_exp_lnkctl_and;
> + set = hpp->pci_exp_lnkctl_or;
> + if (!rrcb)
> + set &= ~PCI_EXP_LNKCTL_RCB;
> +
> pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
> - ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
> + clear, set);
> + }
>
> /* Find Advanced Error Reporting Enhanced Capability */
> pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
> --
> 2.10.0
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Nov 09, 2016 at 11:11:40AM -0600, Bjorn Helgaas wrote:
> Hi Johannes,
>
> On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > The Read Completion Boundary (RCB) bit must only be set on a device or
> > endpoint if it is set on the root complex.
> >
> > Certain BIOSes erroneously set the RCB Bit in their ACPI _HPX Tables
> > even if it is not set on the root port. This is a violation to the PCIe
> > Specification and is known to bring some Mellanox Connect-X 3 HCAs into
> > a state where they can't map their firmware and go into error recovery.
> >
> > BIOS Information
> > Vendor: IBM
> > Version: -[A8E120CUS-1.30]-
> > Release Date: 08/22/2016
>
> This seems like a pretty serious problem (sounds like maybe the HCA is
> completely useless?)
Correct.
>
> Can you point us at a bugzilla or other problem report? It's nice to
> have details of what this looks like to a user, so people who trip
> over this problem have a little more chance of finding the solution.
As we already said, our bugzilla entry for this is not accessible from the
outside, but I know Red Hat does have a bugzilla entry for the same issue as
well. Maybe this is reachable from the outside (adding Don for this, as I know
he has worked on this problem as well).
>
> 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
> with a link") appeared in v3.18, so it's probably not a *new* problem,
> so my guess is that this is v4.10 material.
Yes 4.10 sounds good to me. I personally think, this problem hasn't
materialized yet, as this is the kind of hardware you run on a rather /stable/
kernel either you built on your own or get from an enterprise distribution and
until recently these kernels haven't been updated to something newer than
3.18.
Thanks,
Johannes
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N?rnberg
GF: Felix Imend?rffer, Jane Smithard, Graham Norton
HRB 21284 (AG N?rnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
On 11/14/2016 06:56 AM, Johannes Thumshirn wrote:
> On Wed, Nov 09, 2016 at 11:11:40AM -0600, Bjorn Helgaas wrote:
>> Hi Johannes,
>>
>> On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
>>> The Read Completion Boundary (RCB) bit must only be set on a device or
>>> endpoint if it is set on the root complex.
>>>
>>> Certain BIOSes erroneously set the RCB Bit in their ACPI _HPX Tables
>>> even if it is not set on the root port. This is a violation to the PCIe
>>> Specification and is known to bring some Mellanox Connect-X 3 HCAs into
>>> a state where they can't map their firmware and go into error recovery.
>>>
>>> BIOS Information
>>> Vendor: IBM
>>> Version: -[A8E120CUS-1.30]-
>>> Release Date: 08/22/2016
>>
>> This seems like a pretty serious problem (sounds like maybe the HCA is
>> completely useless?)
>
> Correct.
>
>>
>> Can you point us at a bugzilla or other problem report? It's nice to
>> have details of what this looks like to a user, so people who trip
>> over this problem have a little more chance of finding the solution.
>
> As we already said, our bugzilla entry for this is not accessible from the
> outside, but I know Red Hat does have a bugzilla entry for the same issue as
> well. Maybe this is reachable from the outside (adding Don for this, as I know
> he has worked on this problem as well).
>
RHEL bz's are not accessible from the outside.
I suggest capturing the content of the RH bz issue and creating a k.o. bz
with the information.
>>
>> 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
>> with a link") appeared in v3.18, so it's probably not a *new* problem,
>> so my guess is that this is v4.10 material.
>
> Yes 4.10 sounds good to me. I personally think, this problem hasn't
> materialized yet, as this is the kind of hardware you run on a rather /stable/
> kernel either you built on your own or get from an enterprise distribution and
> until recently these kernels haven't been updated to something newer than
> 3.18.
>
> Thanks,
> Johannes
>
On Mon, Nov 14, 2016 at 11:16:51AM -0500, Don Dutile wrote:
> On 11/14/2016 06:56 AM, Johannes Thumshirn wrote:
> > On Wed, Nov 09, 2016 at 11:11:40AM -0600, Bjorn Helgaas wrote:
> > > Hi Johannes,
> > >
> > > On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > > > The Read Completion Boundary (RCB) bit must only be set on a device or
> > > > endpoint if it is set on the root complex.
> > > >
> > > > Certain BIOSes erroneously set the RCB Bit in their ACPI _HPX Tables
> > > > even if it is not set on the root port. This is a violation to the PCIe
> > > > Specification and is known to bring some Mellanox Connect-X 3 HCAs into
> > > > a state where they can't map their firmware and go into error recovery.
> > > >
> > > > BIOS Information
> > > > Vendor: IBM
> > > > Version: -[A8E120CUS-1.30]-
> > > > Release Date: 08/22/2016
> > >
> > > This seems like a pretty serious problem (sounds like maybe the HCA is
> > > completely useless?)
> >
> > Correct.
> >
> > >
> > > Can you point us at a bugzilla or other problem report? It's nice to
> > > have details of what this looks like to a user, so people who trip
> > > over this problem have a little more chance of finding the solution.
> >
> > As we already said, our bugzilla entry for this is not accessible from the
> > outside, but I know Red Hat does have a bugzilla entry for the same issue as
> > well. Maybe this is reachable from the outside (adding Don for this, as I know
> > he has worked on this problem as well).
> >
> RHEL bz's are not accessible from the outside.
> I suggest capturing the content of the RH bz issue and creating a k.o. bz
> with the information.
I've created https://bugzilla.kernel.org/show_bug.cgi?id=187781 to track this
on b.k.o. Feel free to add any information you have.
@Bjorn anything else I can provide in order to get the fix applied?
Byte,
Johannes
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N?rnberg
GF: Felix Imend?rffer, Jane Smithard, Graham Norton
HRB 21284 (AG N?rnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
Hi Johannes,
On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> The Read Completion Boundary (RCB) bit must only be set on a device or
> endpoint if it is set on the root complex.
I propose the following slightly modified patch. The interesting
difference is that your patch only touches the _HPX "OR" mask, so it
refrains from *setting* RCB in some cases, but it never actually
*clears* it. The only time we clear RCB is when the _HPX "AND" mask
has RCB == 0.
My intent below is that we completely ignore the _HPX RCB bits, and we
set an Endpoint's RCB if and only if the Root Port's RCB is set.
I made an ugly ASCII table to think about the cases:
Root EP _HPX _HPX Final Endpoint RCB state
Port (init) AND OR (curr) (yours) (mine)
0) 0 0 0 0 0 0 0
1) 0 0 0 1 1 0 0
2) 0 0 1 0 0 0 0
3) 0 0 1 1 1 0 0
4) 0 1 0 0 0 0 0
5) 0 1 0 1 1 0 0
6) 0 1 1 0 1 1 0
7) 0 1 1 1 1 1 0
8) 1 0 0 0 0 0 1
9) 1 0 0 1 1 1 1
A) 1 0 1 0 0 0 1
B) 1 0 1 1 1 1 1
C) 1 1 0 0 0 0 1
D) 1 1 0 1 1 1 1
E) 1 1 1 0 1 1 1
F) 1 1 1 1 1 1 1
Cases 0-7 should all result in the Endpoint RCB being zero because the
Root Port RCB is zero. Case 1 is the bug you're fixing. Cases 3 & 5
are similar hypothetical bugs your patch also fixes.
Cases 6 & 7, where firmware left the Endpoint RCB set and _HPX didn't
tell us to clear it, are hypothetical firmware bugs that your patch
wouldn't fix.
In cases 8, A, and C, we currently leave the Endpoint RCB cleared,
either because firmware left it clear and _HPX didn't tell us to set
it (8 and A), or because firmware set it but _HPX told us to clear it
(C).
One could argue that 8, A, and C should stay as they currently are, as
a way for _HPX to work around hardware bugs, e.g., a Root Port that
advertises a 128-byte RCB but doesn't actually support it. I didn't
bother with that and set the Endpoint's RCB to 128 in all cases when
the Root Port claims to support it.
It'd be great if you could test this and comment.
If you get a chance, collect the /proc/iomem contents, too. That's
not for this bug; it's because I'm curious about the
ERST: Can not request [mem 0xb928b000-0xb928cbff] for ERST
problem in your dmesg log.
commit da567a6c67cb5c5a55db7b26b46e9b7e9af69299
Author: Johannes Thumshirn <[email protected]>
Date: Wed Nov 2 16:35:52 2016 -0600
PCI: Set Read Completion Boundary to 128 iff Root Port supports it
Per PCIe spec r3.0, sec 2.3.1.1, the Read Completion Boundary (RCB)
determines the naturally aligned address boundaries on which a Read Request
may be serviced with multiple Completions:
- For a Root Complex, RCB is 64 bytes or 128 bytes
This value is reported in the Link Control Register
Note: Bridges and Endpoints may implement a corresponding command bit
which may be set by system software to indicate the RCB value for the
Root Complex, allowing the Bridge/Endpoint to optimize its behavior
when the Root Complex’s RCB is 128 bytes.
- For all other system elements, RCB is 128 bytes
Per sec 7.8.7, if a Root Port only supports a 64-byte RCB, the RCB of all
downstream devices must be clear, indicating an RCB of 64 bytes. If the
Root Port supports a 128-byte RCB, we may optionally set the RCB of
downstream devices so they know they can generate larger Completions.
Some BIOSes supply an _HPX that tells us to set RCB, even though the Root
Port doesn't have RCB set, which may lead to Malformed TLP errors if the
Endpoint generates completions larger than the Root Port can handle.
The IBM x3850 X6 with BIOS version -[A8E120CUS-1.30]- 08/22/2016 supplies
such an _HPX and a Mellanox MT27500 ConnectX-3 device fails to initialize:
mlx4_core 0000:41:00.0: command 0xfff timed out (go bit not cleared)
mlx4_core 0000:41:00.0: device is going to be reset
mlx4_core 0000:41:00.0: Failed to obtain HW semaphore, aborting
mlx4_core 0000:41:00.0: Fail to reset HCA
------------[ cut here ]------------
kernel BUG at drivers/net/ethernet/mellanox/mlx4/catas.c:193!
After 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
and 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
with a link"), we apply _HPX settings to *all* devices, not just those
hot-added after boot.
Before 7a1562d4f2d0, we didn't touch the Mellanox RCB, and the device
worked. After 7a1562d4f2d0, we set its RCB to 128, and it failed.
Set the RCB to 128 iff the Root Port supports a 128-byte RCB. Otherwise,
set RCB to 64 bytes. This effectively ignores what _HPX tells us about
RCB.
[bhelgaas: changelog, clear RCB if not set for Root Port]
Fixes: 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
Fixes: 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices with a link")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=187781
Signed-off-by: Johannes Thumshirn <[email protected]>
Signed-off-by: Bjorn Helgaas <[email protected]>
CC: [email protected] # v3.18+
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index ab00267..8854161 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -1439,6 +1439,21 @@ static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
dev_warn(&dev->dev, "PCI-X settings not supported\n");
}
+static bool pcie_root_rcb_set(struct pci_dev *dev)
+{
+ struct pci_dev *rp = pcie_find_root_port(dev);
+ u16 lnkctl;
+
+ if (!rp)
+ return false;
+
+ pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
+ if (lnkctl & PCI_EXP_LNKCTL_RCB)
+ return true;
+
+ return false;
+}
+
static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
{
int pos;
@@ -1468,9 +1483,19 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
/* Initialize Link Control Register */
- if (pcie_cap_has_lnkctl(dev))
+ if (pcie_cap_has_lnkctl(dev)) {
+
+ /*
+ * If the Root Port supports Read Completion Boundary of
+ * 128, set RCB to 128. Otherwise, clear it.
+ */
+ hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB;
+ if (pcie_root_rcb_set(dev))
+ hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB;
+
pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
+ }
/* Find Advanced Error Reporting Enhanced Capability */
pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
> Hi Johannes,
>
> On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > The Read Completion Boundary (RCB) bit must only be set on a device or
> > endpoint if it is set on the root complex.
>
> I propose the following slightly modified patch. The interesting
> difference is that your patch only touches the _HPX "OR" mask, so it
> refrains from *setting* RCB in some cases, but it never actually
> *clears* it. The only time we clear RCB is when the _HPX "AND" mask
> has RCB == 0.
>
> My intent below is that we completely ignore the _HPX RCB bits, and we
> set an Endpoint's RCB if and only if the Root Port's RCB is set.
>
> I made an ugly ASCII table to think about the cases:
>
> Root EP _HPX _HPX Final Endpoint RCB state
> Port (init) AND OR (curr) (yours) (mine)
> 0) 0 0 0 0 0 0 0
> 1) 0 0 0 1 1 0 0
> 2) 0 0 1 0 0 0 0
> 3) 0 0 1 1 1 0 0
> 4) 0 1 0 0 0 0 0
> 5) 0 1 0 1 1 0 0
> 6) 0 1 1 0 1 1 0
> 7) 0 1 1 1 1 1 0
> 8) 1 0 0 0 0 0 1
> 9) 1 0 0 1 1 1 1
> A) 1 0 1 0 0 0 1
> B) 1 0 1 1 1 1 1
> C) 1 1 0 0 0 0 1
> D) 1 1 0 1 1 1 1
> E) 1 1 1 0 1 1 1
> F) 1 1 1 1 1 1 1
>
> Cases 0-7 should all result in the Endpoint RCB being zero because the
> Root Port RCB is zero. Case 1 is the bug you're fixing. Cases 3 & 5
> are similar hypothetical bugs your patch also fixes.
>
> Cases 6 & 7, where firmware left the Endpoint RCB set and _HPX didn't
> tell us to clear it, are hypothetical firmware bugs that your patch
> wouldn't fix.
>
> In cases 8, A, and C, we currently leave the Endpoint RCB cleared,
> either because firmware left it clear and _HPX didn't tell us to set
> it (8 and A), or because firmware set it but _HPX told us to clear it
> (C).
>
> One could argue that 8, A, and C should stay as they currently are, as
> a way for _HPX to work around hardware bugs, e.g., a Root Port that
> advertises a 128-byte RCB but doesn't actually support it. I didn't
> bother with that and set the Endpoint's RCB to 128 in all cases when
> the Root Port claims to support it.
>
> It'd be great if you could test this and comment.
I've lost access to the machines, but I'll try to delegate it to someone who
has access.
>
> If you get a chance, collect the /proc/iomem contents, too. That's
> not for this bug; it's because I'm curious about the
>
> ERST: Can not request [mem 0xb928b000-0xb928cbff] for ERST
>
> problem in your dmesg log.
I'll ask for this as well.
Byte,
Johannes
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N?rnberg
GF: Felix Imend?rffer, Jane Smithard, Graham Norton
HRB 21284 (AG N?rnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
> Hi Johannes,
>
> On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > The Read Completion Boundary (RCB) bit must only be set on a device or
> > endpoint if it is set on the root complex.
>
> I propose the following slightly modified patch. The interesting
> difference is that your patch only touches the _HPX "OR" mask, so it
> refrains from *setting* RCB in some cases, but it never actually
> *clears* it. The only time we clear RCB is when the _HPX "AND" mask
> has RCB == 0.
>
> My intent below is that we completely ignore the _HPX RCB bits, and we
> set an Endpoint's RCB if and only if the Root Port's RCB is set.
>
> I made an ugly ASCII table to think about the cases:
>
> Root EP _HPX _HPX Final Endpoint RCB state
> Port (init) AND OR (curr) (yours) (mine)
> 0) 0 0 0 0 0 0 0
> 1) 0 0 0 1 1 0 0
> 2) 0 0 1 0 0 0 0
> 3) 0 0 1 1 1 0 0
> 4) 0 1 0 0 0 0 0
> 5) 0 1 0 1 1 0 0
> 6) 0 1 1 0 1 1 0
> 7) 0 1 1 1 1 1 0
> 8) 1 0 0 0 0 0 1
> 9) 1 0 0 1 1 1 1
> A) 1 0 1 0 0 0 1
> B) 1 0 1 1 1 1 1
> C) 1 1 0 0 0 0 1
> D) 1 1 0 1 1 1 1
> E) 1 1 1 0 1 1 1
> F) 1 1 1 1 1 1 1
>
> Cases 0-7 should all result in the Endpoint RCB being zero because the
> Root Port RCB is zero. Case 1 is the bug you're fixing. Cases 3 & 5
> are similar hypothetical bugs your patch also fixes.
>
> Cases 6 & 7, where firmware left the Endpoint RCB set and _HPX didn't
> tell us to clear it, are hypothetical firmware bugs that your patch
> wouldn't fix.
>
> In cases 8, A, and C, we currently leave the Endpoint RCB cleared,
> either because firmware left it clear and _HPX didn't tell us to set
> it (8 and A), or because firmware set it but _HPX told us to clear it
> (C).
>
> One could argue that 8, A, and C should stay as they currently are, as
> a way for _HPX to work around hardware bugs, e.g., a Root Port that
> advertises a 128-byte RCB but doesn't actually support it. I didn't
> bother with that and set the Endpoint's RCB to 128 in all cases when
> the Root Port claims to support it.
>
> It'd be great if you could test this and comment.
>
> If you get a chance, collect the /proc/iomem contents, too. That's
> not for this bug; it's because I'm curious about the
>
> ERST: Can not request [mem 0xb928b000-0xb928cbff] for ERST
>
> problem in your dmesg log.
Oops, I goofed and forgot to clear RCB by default.
Here's the fixed one.
commit b7bff74c2e6babf12906291ee177f16444de81ad
Author: Johannes Thumshirn <[email protected]>
Date: Wed Nov 16 15:47:53 2016 -0600
PCI: Set Read Completion Boundary to 128 iff Root Port supports it
Per PCIe spec r3.0, sec 2.3.1.1, the Read Completion Boundary (RCB)
determines the naturally aligned address boundaries on which a Read Request
may be serviced with multiple Completions:
- For a Root Complex, RCB is 64 bytes or 128 bytes
This value is reported in the Link Control Register
Note: Bridges and Endpoints may implement a corresponding command bit
which may be set by system software to indicate the RCB value for the
Root Complex, allowing the Bridge/Endpoint to optimize its behavior
when the Root Complex’s RCB is 128 bytes.
- For all other system elements, RCB is 128 bytes
Per sec 7.8.7, if a Root Port only supports a 64-byte RCB, the RCB of all
downstream devices must be clear, indicating an RCB of 64 bytes. If the
Root Port supports a 128-byte RCB, we may optionally set the RCB of
downstream devices so they know they can generate larger Completions.
Some BIOSes supply an _HPX that tells us to set RCB, even though the Root
Port doesn't have RCB set, which may lead to Malformed TLP errors if the
Endpoint generates completions larger than the Root Port can handle.
The IBM x3850 X6 with BIOS version -[A8E120CUS-1.30]- 08/22/2016 supplies
such an _HPX and a Mellanox MT27500 ConnectX-3 device fails to initialize:
mlx4_core 0000:41:00.0: command 0xfff timed out (go bit not cleared)
mlx4_core 0000:41:00.0: device is going to be reset
mlx4_core 0000:41:00.0: Failed to obtain HW semaphore, aborting
mlx4_core 0000:41:00.0: Fail to reset HCA
------------[ cut here ]------------
kernel BUG at drivers/net/ethernet/mellanox/mlx4/catas.c:193!
After 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
and 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
with a link"), we apply _HPX settings to *all* devices, not just those
hot-added after boot.
Before 7a1562d4f2d0, we didn't touch the Mellanox RCB, and the device
worked. After 7a1562d4f2d0, we set its RCB to 128, and it failed.
Set the RCB to 128 iff the Root Port supports a 128-byte RCB. Otherwise,
set RCB to 64 bytes. This effectively ignores what _HPX tells us about
RCB.
[bhelgaas: changelog, clear RCB if not set for Root Port]
Fixes: 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
Fixes: 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices with a link")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=187781
Signed-off-by: Johannes Thumshirn <[email protected]>
Signed-off-by: Bjorn Helgaas <[email protected]>
CC: [email protected] # v3.18+
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index ab00267..104c46d 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -1439,6 +1439,21 @@ static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
dev_warn(&dev->dev, "PCI-X settings not supported\n");
}
+static bool pcie_root_rcb_set(struct pci_dev *dev)
+{
+ struct pci_dev *rp = pcie_find_root_port(dev);
+ u16 lnkctl;
+
+ if (!rp)
+ return false;
+
+ pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
+ if (lnkctl & PCI_EXP_LNKCTL_RCB)
+ return true;
+
+ return false;
+}
+
static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
{
int pos;
@@ -1468,9 +1483,20 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
/* Initialize Link Control Register */
- if (pcie_cap_has_lnkctl(dev))
+ if (pcie_cap_has_lnkctl(dev)) {
+
+ /*
+ * If the Root Port supports Read Completion Boundary of
+ * 128, set RCB to 128. Otherwise, clear it.
+ */
+ hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB;
+ hpp->pci_exp_lnkctl_or &= ~PCI_EXP_LNKCTL_RCB;
+ if (pcie_root_rcb_set(dev))
+ hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB;
+
pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
+ }
/* Find Advanced Error Reporting Enhanced Capability */
pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
On Mon, Nov 21, 2016 at 10:53:52AM -0600, Bjorn Helgaas wrote:
> On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
> > Hi Johannes,
> >
> > On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > > The Read Completion Boundary (RCB) bit must only be set on a device or
> > > endpoint if it is set on the root complex.
> >
> > I propose the following slightly modified patch. The interesting
> > difference is that your patch only touches the _HPX "OR" mask, so it
> > refrains from *setting* RCB in some cases, but it never actually
> > *clears* it. The only time we clear RCB is when the _HPX "AND" mask
> > has RCB == 0.
> >
> > My intent below is that we completely ignore the _HPX RCB bits, and we
> > set an Endpoint's RCB if and only if the Root Port's RCB is set.
> >
> > I made an ugly ASCII table to think about the cases:
> >
> > Root EP _HPX _HPX Final Endpoint RCB state
> > Port (init) AND OR (curr) (yours) (mine)
> > 0) 0 0 0 0 0 0 0
> > 1) 0 0 0 1 1 0 0
> > 2) 0 0 1 0 0 0 0
> > 3) 0 0 1 1 1 0 0
> > 4) 0 1 0 0 0 0 0
> > 5) 0 1 0 1 1 0 0
> > 6) 0 1 1 0 1 1 0
> > 7) 0 1 1 1 1 1 0
> > 8) 1 0 0 0 0 0 1
> > 9) 1 0 0 1 1 1 1
> > A) 1 0 1 0 0 0 1
> > B) 1 0 1 1 1 1 1
> > C) 1 1 0 0 0 0 1
> > D) 1 1 0 1 1 1 1
> > E) 1 1 1 0 1 1 1
> > F) 1 1 1 1 1 1 1
> >
> > Cases 0-7 should all result in the Endpoint RCB being zero because the
> > Root Port RCB is zero. Case 1 is the bug you're fixing. Cases 3 & 5
> > are similar hypothetical bugs your patch also fixes.
> >
> > Cases 6 & 7, where firmware left the Endpoint RCB set and _HPX didn't
> > tell us to clear it, are hypothetical firmware bugs that your patch
> > wouldn't fix.
> >
> > In cases 8, A, and C, we currently leave the Endpoint RCB cleared,
> > either because firmware left it clear and _HPX didn't tell us to set
> > it (8 and A), or because firmware set it but _HPX told us to clear it
> > (C).
> >
> > One could argue that 8, A, and C should stay as they currently are, as
> > a way for _HPX to work around hardware bugs, e.g., a Root Port that
> > advertises a 128-byte RCB but doesn't actually support it. I didn't
> > bother with that and set the Endpoint's RCB to 128 in all cases when
> > the Root Port claims to support it.
> >
> > It'd be great if you could test this and comment.
> >
> > If you get a chance, collect the /proc/iomem contents, too. That's
> > not for this bug; it's because I'm curious about the
> >
> > ERST: Can not request [mem 0xb928b000-0xb928cbff] for ERST
> >
> > problem in your dmesg log.
>
> Oops, I goofed and forgot to clear RCB by default.
> Here's the fixed one.
Yep, my contact already noticed. I have heard rumors that the first two
patches worked on RHEL and the 3rd one didn't (but that's just rumors) so I
try to persuade our field engineer to spend another day testing the patches.
But please be aware this is a bit cumbersome as I don't have access to the
machine and our field engineer only has remote access as well.
Byte,
Johannes
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N?rnberg
GF: Felix Imend?rffer, Jane Smithard, Graham Norton
HRB 21284 (AG N?rnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
On Mon, Nov 21, 2016 at 10:53:52AM -0600, Bjorn Helgaas wrote:
> On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
> > Hi Johannes,
[...]
> >
> > If you get a chance, collect the /proc/iomem contents, too. That's
> > not for this bug; it's because I'm curious about the
> >
> > ERST: Can not request [mem 0xb928b000-0xb928cbff] for ERST
> >
> > problem in your dmesg log.
It's pasted at the end (not sure whether attaching works).
>
> Oops, I goofed and forgot to clear RCB by default.
> Here's the fixed one.
So this version works :-).
Do you plan to merge it for 4.9-rc7 or is it more 4.10 material?
Byte,
Johannes
/proc/iomem:
00000000-00000fff : reserved
00001000-0009bfff : System RAM
0009c000-0009ffff : reserved
000a0000-000bffff : PCI Bus 0000:00
000c0000-000fffff : PCI Bus 0000:00
000c0000-000c7fff : Video ROM
000c8000-000c8fff : Adapter ROM
000c9000-000c9fff : Adapter ROM
000ca000-000cafff : Adapter ROM
000cb000-000cbfff : Adapter ROM
000cc000-000d3bff : Adapter ROM
000e0000-000fffff : reserved
000f0000-000fffff : System ROM
00100000-b6a32fff : System RAM
01000000-015e8742 : Kernel code
015e8743-01d549ff : Kernel data
01f57000-021cefff : Kernel bss
b6a33000-b6a41fff : reserved
b6a42000-b7ffefff : System RAM
b7fff000-b83fefff : reserved
b83ff000-ba3fefff : ACPI Non-volatile Storage
ba3ff000-ba7a7fff : ACPI Tables
ba7a8000-ba7fffff : System RAM
ba800000-bb7fffff : RAM buffer
bb800000-cfffffff : reserved
c0000000-cfffffff : PCI MMCONFIG 0000 [bus 00-ff]
d0004000-d0004fff : dmar3
d0007000-d3ffffff : PCI Bus 0000:00
d0100000-d02fffff : PCI Bus 0000:11
d0300000-d04fffff : PCI Bus 0000:17
d20fd000-d20fdfff : 0000:00:05.4
d20fe000-d20fe3ff : 0000:00:1a.0
d20fe000-d20fe3ff : ehci_hcd
d20ff000-d20ff3ff : 0000:00:1d.0
d20ff000-d20ff3ff : ehci_hcd
d2100000-d22fffff : PCI Bus 0000:01
d2100000-d213ffff : 0000:01:00.0
d2140000-d217ffff : 0000:01:00.1
d2180000-d21bffff : 0000:01:00.2
d21f0000-d21f3fff : 0000:01:00.0
d21f0000-d21f3fff : igb
d21f4000-d21f7fff : 0000:01:00.1
d21f4000-d21f7fff : igb
d21f8000-d21fbfff : 0000:01:00.2
d21f8000-d21fbfff : igb
d21fc000-d21fffff : 0000:01:00.3
d21fc000-d21fffff : igb
d2200000-d223ffff : 0000:01:00.0
d2200000-d223ffff : igb
d2240000-d227ffff : 0000:01:00.1
d2240000-d227ffff : igb
d2280000-d22bffff : 0000:01:00.2
d2280000-d22bffff : igb
d22c0000-d22fffff : 0000:01:00.3
d22c0000-d22fffff : igb
d2300000-d24fffff : PCI Bus 0000:07
d23f0000-d23fffff : 0000:07:00.0
d23f0000-d23fffff : megasas: LSI
d2400000-d24fffff : 0000:07:00.0
d2500000-d25fffff : PCI Bus 0000:1d
d25fe000-d25fffff : 0000:1d:00.0
d25fe000-d25fffff : xhci-hcd
d2600000-d2ffffff : PCI Bus 0000:18
d2600000-d2ffffff : PCI Bus 0000:19
d2600000-d26fffff : PCI Bus 0000:1c
d2700000-d2ffffff : PCI Bus 0000:1a
d2700000-d2ffffff : PCI Bus 0000:1b
d27fc000-d27fffff : 0000:1b:00.0
d27fc000-d27fffff : mgadrmfb_mmio
d2800000-d2ffffff : 0000:1b:00.0
d3000000-d3ffffff : PCI Bus 0000:18
d3000000-d3ffffff : PCI Bus 0000:19
d3000000-d3ffffff : PCI Bus 0000:1a
d3000000-d3ffffff : PCI Bus 0000:1b
d3000000-d3ffffff : 0000:1b:00.0
d3000000-d3ffffff : mgadrmfb_vram
d4000000-d4000fff : dmar0
d4003000-dbffffff : PCI Bus 0000:40
d4100000-d42fffff : PCI Bus 0000:51
d76ff000-d76fffff : 0000:40:05.4
d7700000-d77fffff : PCI Bus 0000:41
d7700000-d77fffff : 0000:41:00.0
d7800000-db7fffff : PCI Bus 0000:47
db800000-dbffffff : PCI Bus 0000:41
db800000-dbffffff : 0000:41:00.0
dc000000-dc000fff : dmar1
dc003000-ebffffff : PCI Bus 0000:80
dffff000-dfffffff : 0000:80:05.4
e0000000-e3ffffff : PCI Bus 0000:81
e4000000-e7ffffff : PCI Bus 0000:8b
e8000000-ebffffff : PCI Bus 0000:95
ec000000-ec000fff : dmar2
ec003000-fbffffff : PCI Bus 0000:c0
effff000-efffffff : 0000:c0:05.4
f0000000-f3ffffff : PCI Bus 0000:c1
f4000000-f7ffffff : PCI Bus 0000:cb
f8000000-fbffffff : PCI Bus 0000:d5
fec00000-fecfffff : PNP0003:00
fec00000-fec003ff : IOAPIC 0
fec01000-fec013ff : IOAPIC 1
fec21000-fec213ff : IOAPIC 2
fec41000-fec413ff : IOAPIC 3
fec61000-fec613ff : IOAPIC 4
fed00000-fed003ff : HPET 0
fed00000-fed003ff : PNP0103:00
fed12000-fed1200f : pnp 00:01
fed12010-fed1201f : pnp 00:01
fed1b000-fed1bfff : pnp 00:01
fed1c000-fed1ffff : reserved
fed1f410-fed1f414 : iTCO_wdt.0.auto
fed45000-fed8bfff : pnp 00:01
fee00000-feefffff : pnp 00:01
fee00000-fee00fff : Local APIC
ff000000-ffffffff : reserved
ff000000-ffffffff : pnp 00:01
100000000-803fffffff : System RAM
20000000000-23fffffffff : PCI Bus 0000:00
20000000000-200001fffff : PCI Bus 0000:07
20000200000-200003fffff : PCI Bus 0000:11
20000400000-200005fffff : PCI Bus 0000:17
20000600000-200007fffff : PCI Bus 0000:1d
23fffee0000-23fffee3fff : 0000:00:04.0
23fffee0000-23fffee3fff : ioatdma
23fffee4000-23fffee7fff : 0000:00:04.1
23fffee4000-23fffee7fff : ioatdma
23fffee8000-23fffeebfff : 0000:00:04.2
23fffee8000-23fffeebfff : ioatdma
23fffeec000-23fffeeffff : 0000:00:04.3
23fffeec000-23fffeeffff : ioatdma
23fffef0000-23fffef3fff : 0000:00:04.4
23fffef0000-23fffef3fff : ioatdma
23fffef4000-23fffef7fff : 0000:00:04.5
23fffef4000-23fffef7fff : ioatdma
23fffef8000-23fffefbfff : 0000:00:04.6
23fffef8000-23fffefbfff : ioatdma
23fffefc000-23fffefffff : 0000:00:04.7
23fffefc000-23fffefffff : ioatdma
23ffff00000-23fffffffff : PCI Bus 0000:01
23ffff00000-23ffff1ffff : 0000:01:00.0
23ffff20000-23ffff3ffff : 0000:01:00.0
23ffff40000-23ffff5ffff : 0000:01:00.1
23ffff60000-23ffff7ffff : 0000:01:00.1
23ffff80000-23ffff9ffff : 0000:01:00.2
23ffffa0000-23ffffbffff : 0000:01:00.2
23ffffc0000-23ffffdffff : 0000:01:00.3
23ffffe0000-23fffffffff : 0000:01:00.3
24000000000-27fffffffff : PCI Bus 0000:40
24000000000-240001fffff : PCI Bus 0000:51
26ffffe0000-26ffffe3fff : 0000:40:04.0
26ffffe0000-26ffffe3fff : ioatdma
26ffffe4000-26ffffe7fff : 0000:40:04.1
26ffffe4000-26ffffe7fff : ioatdma
26ffffe8000-26ffffebfff : 0000:40:04.2
26ffffe8000-26ffffebfff : ioatdma
26ffffec000-26ffffeffff : 0000:40:04.3
26ffffec000-26ffffeffff : ioatdma
26fffff0000-26fffff3fff : 0000:40:04.4
26fffff0000-26fffff3fff : ioatdma
26fffff4000-26fffff7fff : 0000:40:04.5
26fffff4000-26fffff7fff : ioatdma
26fffff8000-26fffffbfff : 0000:40:04.6
26fffff8000-26fffffbfff : ioatdma
26fffffc000-26fffffffff : 0000:40:04.7
26fffffc000-26fffffffff : ioatdma
27000000000-27fffffffff : PCI Bus 0000:47
28000000000-2bfffffffff : PCI Bus 0000:80
28ffffe0000-28ffffe3fff : 0000:80:04.0
28ffffe0000-28ffffe3fff : ioatdma
28ffffe4000-28ffffe7fff : 0000:80:04.1
28ffffe4000-28ffffe7fff : ioatdma
28ffffe8000-28ffffebfff : 0000:80:04.2
28ffffe8000-28ffffebfff : ioatdma
28ffffec000-28ffffeffff : 0000:80:04.3
28ffffec000-28ffffeffff : ioatdma
28fffff0000-28fffff3fff : 0000:80:04.4
28fffff0000-28fffff3fff : ioatdma
28fffff4000-28fffff7fff : 0000:80:04.5
28fffff4000-28fffff7fff : ioatdma
28fffff8000-28fffffbfff : 0000:80:04.6
28fffff8000-28fffffbfff : ioatdma
28fffffc000-28fffffffff : 0000:80:04.7
28fffffc000-28fffffffff : ioatdma
29000000000-29fffffffff : PCI Bus 0000:95
2a000000000-2afffffffff : PCI Bus 0000:8b
2b000000000-2bfffffffff : PCI Bus 0000:81
2c000000000-2ffffffffff : PCI Bus 0000:c0
2cffffe0000-2cffffe3fff : 0000:c0:04.0
2cffffe0000-2cffffe3fff : ioatdma
2cffffe4000-2cffffe7fff : 0000:c0:04.1
2cffffe4000-2cffffe7fff : ioatdma
2cffffe8000-2cffffebfff : 0000:c0:04.2
2cffffe8000-2cffffebfff : ioatdma
2cffffec000-2cffffeffff : 0000:c0:04.3
2cffffec000-2cffffeffff : ioatdma
2cfffff0000-2cfffff3fff : 0000:c0:04.4
2cfffff0000-2cfffff3fff : ioatdma
2cfffff4000-2cfffff7fff : 0000:c0:04.5
2cfffff4000-2cfffff7fff : ioatdma
2cfffff8000-2cfffffbfff : 0000:c0:04.6
2cfffff8000-2cfffffbfff : ioatdma
2cfffffc000-2cfffffffff : 0000:c0:04.7
2cfffffc000-2cfffffffff : ioatdma
2d000000000-2dfffffffff : PCI Bus 0000:d5
2e000000000-2efffffffff : PCI Bus 0000:cb
2f000000000-2ffffffffff : PCI Bus 0000:c1
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N?rnberg
GF: Felix Imend?rffer, Jane Smithard, Graham Norton
HRB 21284 (AG N?rnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
On Mon, Nov 21, 2016 at 9:53 AM, Bjorn Helgaas <[email protected]> wrote:
> On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
>> Hi Johannes,
>>
>> On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
>> > The Read Completion Boundary (RCB) bit must only be set on a device or
>> > endpoint if it is set on the root complex.
>>
>> I propose the following slightly modified patch. The interesting
>> difference is that your patch only touches the _HPX "OR" mask, so it
>> refrains from *setting* RCB in some cases, but it never actually
>> *clears* it. The only time we clear RCB is when the _HPX "AND" mask
>> has RCB == 0.
>>
>> My intent below is that we completely ignore the _HPX RCB bits, and we
>> set an Endpoint's RCB if and only if the Root Port's RCB is set.
>>
>> I made an ugly ASCII table to think about the cases:
>>
>> Root EP _HPX _HPX Final Endpoint RCB state
>> Port (init) AND OR (curr) (yours) (mine)
>> 0) 0 0 0 0 0 0 0
>> 1) 0 0 0 1 1 0 0
>> 2) 0 0 1 0 0 0 0
>> 3) 0 0 1 1 1 0 0
>> 4) 0 1 0 0 0 0 0
>> 5) 0 1 0 1 1 0 0
>> 6) 0 1 1 0 1 1 0
>> 7) 0 1 1 1 1 1 0
>> 8) 1 0 0 0 0 0 1
>> 9) 1 0 0 1 1 1 1
>> A) 1 0 1 0 0 0 1
>> B) 1 0 1 1 1 1 1
>> C) 1 1 0 0 0 0 1
>> D) 1 1 0 1 1 1 1
>> E) 1 1 1 0 1 1 1
>> F) 1 1 1 1 1 1 1
>>
>> Cases 0-7 should all result in the Endpoint RCB being zero because the
>> Root Port RCB is zero. Case 1 is the bug you're fixing. Cases 3 & 5
>> are similar hypothetical bugs your patch also fixes.
>>
>> Cases 6 & 7, where firmware left the Endpoint RCB set and _HPX didn't
>> tell us to clear it, are hypothetical firmware bugs that your patch
>> wouldn't fix.
>>
>> In cases 8, A, and C, we currently leave the Endpoint RCB cleared,
>> either because firmware left it clear and _HPX didn't tell us to set
>> it (8 and A), or because firmware set it but _HPX told us to clear it
>> (C).
>>
>> One could argue that 8, A, and C should stay as they currently are, as
>> a way for _HPX to work around hardware bugs, e.g., a Root Port that
>> advertises a 128-byte RCB but doesn't actually support it. I didn't
>> bother with that and set the Endpoint's RCB to 128 in all cases when
>> the Root Port claims to support it.
>>
>> It'd be great if you could test this and comment.
>>
>> If you get a chance, collect the /proc/iomem contents, too. That's
>> not for this bug; it's because I'm curious about the
>>
>> ERST: Can not request [mem 0xb928b000-0xb928cbff] for ERST
>>
>> problem in your dmesg log.
>
> Oops, I goofed and forgot to clear RCB by default.
> Here's the fixed one.
>
Testing on our end was successful again with this fix ;)
Tested-by: Frank Danapfel <[email protected]>
Acked-by: Myron Stowe <[email protected]>
>
> commit b7bff74c2e6babf12906291ee177f16444de81ad
> Author: Johannes Thumshirn <[email protected]>
> Date: Wed Nov 16 15:47:53 2016 -0600
>
> PCI: Set Read Completion Boundary to 128 iff Root Port supports it
>
> Per PCIe spec r3.0, sec 2.3.1.1, the Read Completion Boundary (RCB)
> determines the naturally aligned address boundaries on which a Read Request
> may be serviced with multiple Completions:
>
> - For a Root Complex, RCB is 64 bytes or 128 bytes
> This value is reported in the Link Control Register
>
> Note: Bridges and Endpoints may implement a corresponding command bit
> which may be set by system software to indicate the RCB value for the
> Root Complex, allowing the Bridge/Endpoint to optimize its behavior
> when the Root Complex’s RCB is 128 bytes.
>
> - For all other system elements, RCB is 128 bytes
>
> Per sec 7.8.7, if a Root Port only supports a 64-byte RCB, the RCB of all
> downstream devices must be clear, indicating an RCB of 64 bytes. If the
> Root Port supports a 128-byte RCB, we may optionally set the RCB of
> downstream devices so they know they can generate larger Completions.
>
> Some BIOSes supply an _HPX that tells us to set RCB, even though the Root
> Port doesn't have RCB set, which may lead to Malformed TLP errors if the
> Endpoint generates completions larger than the Root Port can handle.
>
> The IBM x3850 X6 with BIOS version -[A8E120CUS-1.30]- 08/22/2016 supplies
> such an _HPX and a Mellanox MT27500 ConnectX-3 device fails to initialize:
>
> mlx4_core 0000:41:00.0: command 0xfff timed out (go bit not cleared)
> mlx4_core 0000:41:00.0: device is going to be reset
> mlx4_core 0000:41:00.0: Failed to obtain HW semaphore, aborting
> mlx4_core 0000:41:00.0: Fail to reset HCA
> ------------[ cut here ]------------
> kernel BUG at drivers/net/ethernet/mellanox/mlx4/catas.c:193!
>
> After 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
> and 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
> with a link"), we apply _HPX settings to *all* devices, not just those
> hot-added after boot.
>
> Before 7a1562d4f2d0, we didn't touch the Mellanox RCB, and the device
> worked. After 7a1562d4f2d0, we set its RCB to 128, and it failed.
>
> Set the RCB to 128 iff the Root Port supports a 128-byte RCB. Otherwise,
> set RCB to 64 bytes. This effectively ignores what _HPX tells us about
> RCB.
>
> [bhelgaas: changelog, clear RCB if not set for Root Port]
> Fixes: 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
> Fixes: 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices with a link")
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=187781
> Signed-off-by: Johannes Thumshirn <[email protected]>
> Signed-off-by: Bjorn Helgaas <[email protected]>
> CC: [email protected] # v3.18+
>
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index ab00267..104c46d 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1439,6 +1439,21 @@ static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
> dev_warn(&dev->dev, "PCI-X settings not supported\n");
> }
>
> +static bool pcie_root_rcb_set(struct pci_dev *dev)
> +{
> + struct pci_dev *rp = pcie_find_root_port(dev);
> + u16 lnkctl;
> +
> + if (!rp)
> + return false;
> +
> + pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
> + if (lnkctl & PCI_EXP_LNKCTL_RCB)
> + return true;
> +
> + return false;
> +}
> +
> static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
> {
> int pos;
> @@ -1468,9 +1483,20 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
> ~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
>
> /* Initialize Link Control Register */
> - if (pcie_cap_has_lnkctl(dev))
> + if (pcie_cap_has_lnkctl(dev)) {
> +
> + /*
> + * If the Root Port supports Read Completion Boundary of
> + * 128, set RCB to 128. Otherwise, clear it.
> + */
> + hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB;
> + hpp->pci_exp_lnkctl_or &= ~PCI_EXP_LNKCTL_RCB;
> + if (pcie_root_rcb_set(dev))
> + hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB;
> +
> pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
> ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
> + }
>
> /* Find Advanced Error Reporting Enhanced Capability */
> pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Nov 21, 2016 at 10:53:52AM -0600, Bjorn Helgaas wrote:
> On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
> > Hi Johannes,
> >
> > On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > > The Read Completion Boundary (RCB) bit must only be set on a device or
> > > endpoint if it is set on the root complex.
> ...
> Here's the fixed one.
I applied the one below to for-linus for v4.9.
I did tweak the preceding patch to only share pcie_find_root_port()
inside drivers/pci -- moved the declaration from include/linux/pci.h
to drivers/pci/pci.h and moved the definition to drivers/pci/search.c.
I don't think it needs to be visible to the whole kernel.
> commit b7bff74c2e6babf12906291ee177f16444de81ad
> Author: Johannes Thumshirn <[email protected]>
> Date: Wed Nov 16 15:47:53 2016 -0600
>
> PCI: Set Read Completion Boundary to 128 iff Root Port supports it
>
> Per PCIe spec r3.0, sec 2.3.1.1, the Read Completion Boundary (RCB)
> determines the naturally aligned address boundaries on which a Read Request
> may be serviced with multiple Completions:
>
> - For a Root Complex, RCB is 64 bytes or 128 bytes
> This value is reported in the Link Control Register
>
> Note: Bridges and Endpoints may implement a corresponding command bit
> which may be set by system software to indicate the RCB value for the
> Root Complex, allowing the Bridge/Endpoint to optimize its behavior
> when the Root Complex’s RCB is 128 bytes.
>
> - For all other system elements, RCB is 128 bytes
>
> Per sec 7.8.7, if a Root Port only supports a 64-byte RCB, the RCB of all
> downstream devices must be clear, indicating an RCB of 64 bytes. If the
> Root Port supports a 128-byte RCB, we may optionally set the RCB of
> downstream devices so they know they can generate larger Completions.
>
> Some BIOSes supply an _HPX that tells us to set RCB, even though the Root
> Port doesn't have RCB set, which may lead to Malformed TLP errors if the
> Endpoint generates completions larger than the Root Port can handle.
>
> The IBM x3850 X6 with BIOS version -[A8E120CUS-1.30]- 08/22/2016 supplies
> such an _HPX and a Mellanox MT27500 ConnectX-3 device fails to initialize:
>
> mlx4_core 0000:41:00.0: command 0xfff timed out (go bit not cleared)
> mlx4_core 0000:41:00.0: device is going to be reset
> mlx4_core 0000:41:00.0: Failed to obtain HW semaphore, aborting
> mlx4_core 0000:41:00.0: Fail to reset HCA
> ------------[ cut here ]------------
> kernel BUG at drivers/net/ethernet/mellanox/mlx4/catas.c:193!
>
> After 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
> and 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
> with a link"), we apply _HPX settings to *all* devices, not just those
> hot-added after boot.
>
> Before 7a1562d4f2d0, we didn't touch the Mellanox RCB, and the device
> worked. After 7a1562d4f2d0, we set its RCB to 128, and it failed.
>
> Set the RCB to 128 iff the Root Port supports a 128-byte RCB. Otherwise,
> set RCB to 64 bytes. This effectively ignores what _HPX tells us about
> RCB.
>
> [bhelgaas: changelog, clear RCB if not set for Root Port]
> Fixes: 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
> Fixes: 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices with a link")
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=187781
> Signed-off-by: Johannes Thumshirn <[email protected]>
> Signed-off-by: Bjorn Helgaas <[email protected]>
> CC: [email protected] # v3.18+
>
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index ab00267..104c46d 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1439,6 +1439,21 @@ static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
> dev_warn(&dev->dev, "PCI-X settings not supported\n");
> }
>
> +static bool pcie_root_rcb_set(struct pci_dev *dev)
> +{
> + struct pci_dev *rp = pcie_find_root_port(dev);
> + u16 lnkctl;
> +
> + if (!rp)
> + return false;
> +
> + pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
> + if (lnkctl & PCI_EXP_LNKCTL_RCB)
> + return true;
> +
> + return false;
> +}
> +
> static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
> {
> int pos;
> @@ -1468,9 +1483,20 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
> ~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
>
> /* Initialize Link Control Register */
> - if (pcie_cap_has_lnkctl(dev))
> + if (pcie_cap_has_lnkctl(dev)) {
> +
> + /*
> + * If the Root Port supports Read Completion Boundary of
> + * 128, set RCB to 128. Otherwise, clear it.
> + */
> + hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB;
> + hpp->pci_exp_lnkctl_or &= ~PCI_EXP_LNKCTL_RCB;
> + if (pcie_root_rcb_set(dev))
> + hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB;
> +
> pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
> ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
> + }
>
> /* Find Advanced Error Reporting Enhanced Capability */
> pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Nov 23, 2016 at 11:31:52AM -0600, Bjorn Helgaas wrote:
> On Mon, Nov 21, 2016 at 10:53:52AM -0600, Bjorn Helgaas wrote:
> > On Wed, Nov 16, 2016 at 12:11:58PM -0600, Bjorn Helgaas wrote:
> > > Hi Johannes,
> > >
> > > On Wed, Nov 02, 2016 at 04:35:52PM -0600, Johannes Thumshirn wrote:
> > > > The Read Completion Boundary (RCB) bit must only be set on a device or
> > > > endpoint if it is set on the root complex.
> > ...
>
> > Here's the fixed one.
>
> I applied the one below to for-linus for v4.9.
>
> I did tweak the preceding patch to only share pcie_find_root_port()
> inside drivers/pci -- moved the declaration from include/linux/pci.h
> to drivers/pci/pci.h and moved the definition to drivers/pci/search.c.
>
> I don't think it needs to be visible to the whole kernel.
Sigh, I guess it *does*, because the aer_inject user can be a module.
I'll fix that.
Bjorn