Patchwork [1/2] PCI: ASPM exit link state code could skip devices

login
register
mail settings
Submitter Myron Stowe
Date Jan. 31, 2013, 11:29 p.m.
Message ID <1359674965.2426.13.camel@zim.stowe>
Download mbox | patch
Permalink /patch/217306/
State Accepted
Headers show

Comments

Myron Stowe - Jan. 31, 2013, 11:29 p.m.
On Fri, 2013-01-18 at 13:23 -0500, Joe Lawrence wrote:
> From 3a51bbad5555ba6c6e144aa5176c8112eb449325 Mon Sep 17 00:00:00 2001
> From: Joe Lawrence <joe.lawrence@stratus.com>
> Date: Tue, 15 Jan 2013 14:51:57 -0500
> Subject: [PATCH 1/2] PCI: ASPM exit link state code could skip devices
> 
> On PCI bus hotplug removal, pcie_aspm_exit_link_state could potentially skip
> parent devices that have link_state allocated.  Instead of exiting early if
> a given device is not PCIe, check that the device's parent is PCIe.  This
> enables pcie_aspm_exit_link_state to properly clean up parent link_state when
> the last function in a slot might not be PCIe.
> 
> Reviewed-by: David Bulkow <david.bulkow@stratus.com>
> Signed-off-by: Joe Lawrence <joe.lawrence@stratus.com>
> ---
>  drivers/pci/pcie/aspm.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> index b52630b..6122447 100644
> --- a/drivers/pci/pcie/aspm.c
> +++ b/drivers/pci/pcie/aspm.c
> @@ -634,7 +634,7 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
>  	struct pci_dev *parent = pdev->bus->self;
>  	struct pcie_link_state *link, *root, *parent_link;
>  
> -	if (!pci_is_pcie(pdev) || !parent || !parent->link_state)
> +	if (!parent || !pci_is_pcie(parent) || !parent->link_state)
>  		return;
>  	if ((pci_pcie_type(parent) != PCI_EXP_TYPE_ROOT_PORT) &&
>  	    (pci_pcie_type(parent) != PCI_EXP_TYPE_DOWNSTREAM))

Joe:

Bjorn and I looked at this again today and think the checks that occur
early in 'pcie_aspm_exit_link_state()' could be simplified (as shown
below) to avoid issues such as you are encountering.

I think part of the confusion concerning the asymmetry between
'pcie_aspm_init_link_state()' and 'pcie_aspm_exit_link_state()' is that
the former is passed in a pointer to a bridge device whereas the latter
is passed in a pointer to a end-point device.  Frankly, I find the
entire driver to be confusing and very hard to understand.

Could you please test with the following patch and let us know the
results?


PCI: ASPM exit link state code is skipping devices

From: Myron Stowe <myron.stowe@redhat.com>

On PCI bus hotplug removal, 'pcie_aspm_exit_link_state' can potentially
skip parent devices that have link_state allocated.  Instead of exiting
early if a given device in not PCIe, check whether or not the device's
parent has link_state allocated.  This enables 'pcie_aspm_exit_link_state'
to properly clean up parent link_state when the last function in a slot
might not be PCIe.

Reported-by: Joe Lawrence <joe.lawrence@stratus.com>
Signed-off-by: Myron Stowe <myron.stowe@redhat.com>
---

 drivers/pci/pcie/aspm.c |    5 +----
 1 files changed, 1 insertions(+), 4 deletions(-)




--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Joe Lawrence - Feb. 1, 2013, 7:55 p.m.
On Thu, 31 Jan 2013, Myron Stowe wrote:

> On Fri, 2013-01-18 at 13:23 -0500, Joe Lawrence wrote:
> > From 3a51bbad5555ba6c6e144aa5176c8112eb449325 Mon Sep 17 00:00:00 2001
> > From: Joe Lawrence <joe.lawrence@stratus.com>
> > Date: Tue, 15 Jan 2013 14:51:57 -0500
> > Subject: [PATCH 1/2] PCI: ASPM exit link state code could skip devices
> > 
> > On PCI bus hotplug removal, pcie_aspm_exit_link_state could potentially skip
> > parent devices that have link_state allocated.  Instead of exiting early if
> > a given device is not PCIe, check that the device's parent is PCIe.  This
> > enables pcie_aspm_exit_link_state to properly clean up parent link_state when
> > the last function in a slot might not be PCIe.
> > 
> > Reviewed-by: David Bulkow <david.bulkow@stratus.com>
> > Signed-off-by: Joe Lawrence <joe.lawrence@stratus.com>
> > ---
> >  drivers/pci/pcie/aspm.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> > index b52630b..6122447 100644
> > --- a/drivers/pci/pcie/aspm.c
> > +++ b/drivers/pci/pcie/aspm.c
> > @@ -634,7 +634,7 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
> >  	struct pci_dev *parent = pdev->bus->self;
> >  	struct pcie_link_state *link, *root, *parent_link;
> >  
> > -	if (!pci_is_pcie(pdev) || !parent || !parent->link_state)
> > +	if (!parent || !pci_is_pcie(parent) || !parent->link_state)
> >  		return;
> >  	if ((pci_pcie_type(parent) != PCI_EXP_TYPE_ROOT_PORT) &&
> >  	    (pci_pcie_type(parent) != PCI_EXP_TYPE_DOWNSTREAM))
> 
> Joe:
> 
> Bjorn and I looked at this again today and think the checks that occur
> early in 'pcie_aspm_exit_link_state()' could be simplified (as shown
> below) to avoid issues such as you are encountering.
> 
> I think part of the confusion concerning the asymmetry between
> 'pcie_aspm_init_link_state()' and 'pcie_aspm_exit_link_state()' is that
> the former is passed in a pointer to a bridge device whereas the latter
> is passed in a pointer to a end-point device.  Frankly, I find the
> entire driver to be confusing and very hard to understand.
> 
> Could you please test with the following patch and let us know the
> results?
> 
> 
> PCI: ASPM exit link state code is skipping devices
> 
> From: Myron Stowe <myron.stowe@redhat.com>
> 
> On PCI bus hotplug removal, 'pcie_aspm_exit_link_state' can potentially
> skip parent devices that have link_state allocated.  Instead of exiting
> early if a given device in not PCIe, check whether or not the device's
> parent has link_state allocated.  This enables 'pcie_aspm_exit_link_state'
> to properly clean up parent link_state when the last function in a slot
> might not be PCIe.
> 
> Reported-by: Joe Lawrence <joe.lawrence@stratus.com>
> Signed-off-by: Myron Stowe <myron.stowe@redhat.com>
> ---
> 
>  drivers/pci/pcie/aspm.c |    5 +----
>  1 files changed, 1 insertions(+), 4 deletions(-)
> 
> 
> diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> index 8474b6a..aa52727 100644
> --- a/drivers/pci/pcie/aspm.c
> +++ b/drivers/pci/pcie/aspm.c
> @@ -634,10 +634,7 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
>  	struct pci_dev *parent = pdev->bus->self;
>  	struct pcie_link_state *link, *root, *parent_link;
>  
> -	if (!pci_is_pcie(pdev) || !parent || !parent->link_state)
> -		return;
> -	if ((pci_pcie_type(parent) != PCI_EXP_TYPE_ROOT_PORT) &&
> -	    (pci_pcie_type(parent) != PCI_EXP_TYPE_DOWNSTREAM))
> +	if (!parent || !parent->link_state)
>  		return;
>  
>  	down_read(&pci_bus_sem);
> 
> 
> 

Hi Myron,

The patch tests eliminates the reported crash when hotplugging our 
problem PCI bus.

The additional checks removed from pcie_aspm_exit_link_state (ie, the 
simplification) do appear to be unnecessary as pcie_aspm_init_link_state 
is already preventing such devices from allocating link state.  

Thanks,

-- Joe
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Bjorn Helgaas - Feb. 1, 2013, 10:31 p.m.
On Fri, Feb 1, 2013 at 12:55 PM, Joe Lawrence <Joe.Lawrence@stratus.com> wrote:
> On Thu, 31 Jan 2013, Myron Stowe wrote:

>> PCI: ASPM exit link state code is skipping devices
>>
>> From: Myron Stowe <myron.stowe@redhat.com>
>>
>> On PCI bus hotplug removal, 'pcie_aspm_exit_link_state' can potentially
>> skip parent devices that have link_state allocated.  Instead of exiting
>> early if a given device in not PCIe, check whether or not the device's
>> parent has link_state allocated.  This enables 'pcie_aspm_exit_link_state'
>> to properly clean up parent link_state when the last function in a slot
>> might not be PCIe.
>>
>> Reported-by: Joe Lawrence <joe.lawrence@stratus.com>
>> Signed-off-by: Myron Stowe <myron.stowe@redhat.com>
>> ---
>>
>>  drivers/pci/pcie/aspm.c |    5 +----
>>  1 files changed, 1 insertions(+), 4 deletions(-)
>>
>>
>> diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
>> index 8474b6a..aa52727 100644
>> --- a/drivers/pci/pcie/aspm.c
>> +++ b/drivers/pci/pcie/aspm.c
>> @@ -634,10 +634,7 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
>>       struct pci_dev *parent = pdev->bus->self;
>>       struct pcie_link_state *link, *root, *parent_link;
>>
>> -     if (!pci_is_pcie(pdev) || !parent || !parent->link_state)
>> -             return;
>> -     if ((pci_pcie_type(parent) != PCI_EXP_TYPE_ROOT_PORT) &&
>> -         (pci_pcie_type(parent) != PCI_EXP_TYPE_DOWNSTREAM))
>> +     if (!parent || !parent->link_state)
>>               return;
>>
>>       down_read(&pci_bus_sem);
>>
>>
>>
>
> Hi Myron,
>
> The patch tests eliminates the reported crash when hotplugging our
> problem PCI bus.
>
> The additional checks removed from pcie_aspm_exit_link_state (ie, the
> simplification) do appear to be unnecessary as pcie_aspm_init_link_state
> is already preventing such devices from allocating link state.

I added a Tested-by for you, Joe, and applied this to my pci/joe-aspm
branch for v3.9.  Thanks!
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 8474b6a..aa52727 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -634,10 +634,7 @@  void pcie_aspm_exit_link_state(struct pci_dev *pdev)
 	struct pci_dev *parent = pdev->bus->self;
 	struct pcie_link_state *link, *root, *parent_link;
 
-	if (!pci_is_pcie(pdev) || !parent || !parent->link_state)
-		return;
-	if ((pci_pcie_type(parent) != PCI_EXP_TYPE_ROOT_PORT) &&
-	    (pci_pcie_type(parent) != PCI_EXP_TYPE_DOWNSTREAM))
+	if (!parent || !parent->link_state)
 		return;
 
 	down_read(&pci_bus_sem);