On Wed, Nov 20, 2013 at 12:39 PM, Keith Busch <keith.busch@xxxxxxxxx> wrote:
> On Wed, 20 Nov 2013, Alex Williamson wrote:
>>
>> On Wed, 2013-11-20 at 12:00 -0700, Bjorn Helgaas wrote:
>>>
>>> [+cc Alex]
>>>
>>> On Wed, Nov 20, 2013 at 10:26 AM, Keith Busch <keith.busch@xxxxxxxxx>
>>> wrote:
>>>>
>>>> A user can issue a pci function level reset to a device using sysfs
>>>> entry /sys/bus/pci/devices/.../reset. A kernel driver handling the
>>>> pci device probably would like to know that such a reset has occured,
>>>> so this patch adds a callback in to pci_driver's pci_error_handler.
>>
>>
>> Seems reasonable to me, but I'd like to see what useful thing you're
>> going to do in the driver when this is triggered too. Thanks,
>>
>> Alex
>
>
> Sure, a lot of devices require a driver reinitialize it after a function
> level reset. I specifically develop NVM-Express and the nvme driver
> currently sees the device as simply unresponsive after a user hits the
> reset sysfs file, and it doesn't know why its stopped completing commands,
> so the driver considers them all failed. If this callback is okay, I'd
> have the nvme driver take a short-cut to reinitialize the controller to
> a ready state rather than fail everything after long timeouts.
>
>
>>>> Signed-off-by: Keith Busch <keith.busch@xxxxxxxxx>
>>>> ---
>>>> Assuming this is a good idea, is this the right place to invoke the
>>>> reset callback, or should it be up the call stack closer to the sysfs
>>>> reset entry point? Or somewhere else entirely?
>>>>
>>>> drivers/pci/pci.c | 6 ++++++
>>>> include/linux/pci.h | 3 +++
>>>> 2 files changed, 9 insertions(+)
>>>>
>>>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>>>> index b127fbda..5ee6fc0 100644
>>>> --- a/drivers/pci/pci.c
>>>> +++ b/drivers/pci/pci.c
>>>> @@ -3435,6 +3435,12 @@ static int __pci_dev_reset(struct pci_dev *dev,
>>>> int probe)
>>>>
>>>> rc = pci_parent_bus_reset(dev, probe);
>>>> done:
>>>> + if (rc != -ENOTTY && !probe) {
>>>> + struct pci_error_handlers *err_handler =
>>>> + dev->driver ? dev->driver->err_handler :
>>>> NULL;
>>>> + if (err_handler && err_handler->function_reset)
>>>> + err_handler->function_reset(dev);
>>>> + }
>>>> return rc;
>>>> }
>>>>
>>>> diff --git a/include/linux/pci.h b/include/linux/pci.h
>>>> index 835ec7b..d1f100e 100644
>>>> --- a/include/linux/pci.h
>>>> +++ b/include/linux/pci.h
>>>> @@ -605,6 +605,9 @@ struct pci_error_handlers {
>>>> /* PCI slot has been reset */
>>>> pci_ers_result_t (*slot_reset)(struct pci_dev *dev);
>>>>
>>>> + /* PCI function has been reset */
>>>> + pci_ers_result_t (*function_reset)(struct pci_dev *dev);
Your intent makes perfect sense. Surely we should tell the driver
when we reset its device.
But I wonder whether we really need or want both "slot_reset" and
"function_reset"? I know some resets affect more than one device, but
it looks to me like "slot_reset" is already used to notify the driver
for each affected function.
It looks to me like "slot_reset" is really a function-oriented
interface just like most of the rest of PCI. Maybe you could just use
it rather than adding a new one?
When you sort this out, please post this patch along with the nvme
patch to take advantage of it. That way we can see the whole picture
and have some hope that the new hook has been tested and does what you
intend.
Bjorn
>>>> /* Device driver may resume normal operations */
>>>> void (*resume)(struct pci_dev *dev);
>>>> };
>>>> --
>>>> 1.7.10.4
>>>>
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
>>>> the body of a message to majordomo@xxxxxxxxxxxxxxx
>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
>>> the body of a message to majordomo@xxxxxxxxxxxxxxx
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>
>>
>>
>>
>>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
