testing of Xen PCIbackend support for slot and bus reset

testing of Xen PCIbackend support for slot and bus reset

[Richie]

[-- Attachment #1.1: Type: text/plain, Size: 421 bytes --]

@Konrad

I'm interesting in testing your latest iteration of bus reset support.  
Actually I cloned your devel/xen-pciback.slot_and_bus.v2.1 
<http://git.kernel.org/cgit/linux/kernel/git/konrad/xen.git/log/?h=devel/xen-pciback.slot_and_bus.v2.1> 
branch and tested that but did not seem to reset the vga card.  I did 
capture dmesg output which was nice and clean regarding resetting the 
device but don't have it handy.

[-- Attachment #1.2: Type: text/html, Size: 642 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: testing of Xen PCIbackend support for slot and bus reset

[Martin Öhrling]

[-- Attachment #1.1: Type: text/plain, Size: 1053 bytes --]

Any progress in testing the branch?

I can see one thing that could explain if the function isn't called. The
reset file is only created if it doesn't exist. I know it's unlikely for a
gpu, but have you verified that your driver didn't export a "reset"-file
without these patches? As far as I can tell, the drivers reset function
will be used instead of pcistub_reset_pci_dev() in the presence of a
"reset"-file.

/Martin

2014-08-17 3:52 GMT+02:00 Richie <listmail@triad.rr.com>:

>  @Konrad
>
> I'm interesting in testing your latest iteration of bus reset support.
> Actually I cloned your devel/xen-pciback.slot_and_bus.v2.1
> <http://git.kernel.org/cgit/linux/kernel/git/konrad/xen.git/log/?h=devel/xen-pciback.slot_and_bus.v2.1>
> branch and tested that but did not seem to reset the vga card.  I did
> capture dmesg output which was nice and clean regarding resetting the
> device but don't have it handy.
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>
>

[-- Attachment #1.2: Type: text/html, Size: 1696 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: testing of Xen PCIbackend support for slot and bus reset

[Konrad Rzeszutek Wilk]

On Tue, Oct 28, 2014 at 11:58:41AM +0100, Martin Öhrling wrote:
> Any progress in testing the branch?
> 
> I can see one thing that could explain if the function isn't called. The
> reset file is only created if it doesn't exist. I know it's unlikely for a
> gpu, but have you verified that your driver didn't export a "reset"-file
> without these patches? As far as I can tell, the drivers reset function
> will be used instead of pcistub_reset_pci_dev() in the presence of a
> "reset"-file.

Use the 'do_flr' SysFS instead of the 'reset'. Once an PCI device
is 'owned' by PCIback the 'reset' is no longer routed to the GPU driver.

And the 'reset' does not do a bus-reset (whcih is what you need for
most GPUs).


> 
> /Martin
> 
> 2014-08-17 3:52 GMT+02:00 Richie <listmail@triad.rr.com>:
> 
> >  @Konrad
> >
> > I'm interesting in testing your latest iteration of bus reset support.
> > Actually I cloned your devel/xen-pciback.slot_and_bus.v2.1
> > <http://git.kernel.org/cgit/linux/kernel/git/konrad/xen.git/log/?h=devel/xen-pciback.slot_and_bus.v2.1>
> > branch and tested that but did not seem to reset the vga card.  I did
> > capture dmesg output which was nice and clean regarding resetting the
> > device but don't have it handy.
> >
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@lists.xen.org
> > http://lists.xen.org/xen-devel
> >
> >

> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

Re: testing of Xen PCIbackend support for slot and bus reset

[Martin Öhrling]

[-- Attachment #1.1: Type: text/plain, Size: 2190 bytes --]

> Use the 'do_flr' SysFS instead of the 'reset'. Once an PCI device
> is 'owned' by PCIback the 'reset' is no longer routed to the GPU driver.
>
>
Got a bit confused by the conditional check for the presence of reset file
in pcistub_try_create_reset_file(). You are absolutely correct in that
these devices are owned by pciback and not another driver. Can an alternate
reset file (with another reset function) be created at an earlier stage or
is this a guard to prevent the reset file (calling pcistub_reset_store())
from being created multiple times?

Still a bit confused. I may be missing something, but the sysfs file name
used for pcistub_reset_pci_dev() seems to be "reset". The function shall do
FLR and, if possible, D3/bus/slot reset as fallback but only if it has no
side effects for other functions on the device (equivalent to a real FLR).
Shouldn't this function be registered as en emulated "do_flr" file instead
of "reset"?

And the 'reset' does not do a bus-reset (whcih is what you need for
> most GPUs).
>

The reset function in this branch looks similar to the one in
devel/xen-pciback.slot_and_bus.v0. I have tested the latter and I can
verify that it works well. I had to do some additional changes to avoid
infinite wait after a couple of reboots but the reset function worked well.

/Martin


>
> >
> > /Martin
> >
> > 2014-08-17 3:52 GMT+02:00 Richie <listmail@triad.rr.com>:
> >
> > >  @Konrad
> > >
> > > I'm interesting in testing your latest iteration of bus reset support.
> > > Actually I cloned your devel/xen-pciback.slot_and_bus.v2.1
> > > <
> http://git.kernel.org/cgit/linux/kernel/git/konrad/xen.git/log/?h=devel/xen-pciback.slot_and_bus.v2.1
> >
> > > branch and tested that but did not seem to reset the vga card.  I did
> > > capture dmesg output which was nice and clean regarding resetting the
> > > device but don't have it handy.
> > >
> > > _______________________________________________
> > > Xen-devel mailing list
> > > Xen-devel@lists.xen.org
> > > http://lists.xen.org/xen-devel
> > >
> > >
>
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@lists.xen.org
> > http://lists.xen.org/xen-devel
>
>

[-- Attachment #1.2: Type: text/html, Size: 3593 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: testing of Xen PCIbackend support for slot and bus reset

[Martin Öhrling]

[-- Attachment #1.1: Type: text/plain, Size: 2524 bytes --]

My mistake. Just realised that the "reset" file is the name of what I
though should be "do_flr" in sysfs.

/Martin


2014-10-28 16:06 GMT+01:00 Martin Öhrling <martin@öhrling.se>:

>
> Use the 'do_flr' SysFS instead of the 'reset'. Once an PCI device
>> is 'owned' by PCIback the 'reset' is no longer routed to the GPU driver.
>>
>>
> Got a bit confused by the conditional check for the presence of reset file
> in pcistub_try_create_reset_file(). You are absolutely correct in that
> these devices are owned by pciback and not another driver. Can an alternate
> reset file (with another reset function) be created at an earlier stage or
> is this a guard to prevent the reset file (calling pcistub_reset_store())
> from being created multiple times?
>
> Still a bit confused. I may be missing something, but the sysfs file name
> used for pcistub_reset_pci_dev() seems to be "reset". The function shall do
> FLR and, if possible, D3/bus/slot reset as fallback but only if it has no
> side effects for other functions on the device (equivalent to a real FLR).
> Shouldn't this function be registered as en emulated "do_flr" file instead
> of "reset"?
>
> And the 'reset' does not do a bus-reset (whcih is what you need for
>> most GPUs).
>>
>
> The reset function in this branch looks similar to the one in
> devel/xen-pciback.slot_and_bus.v0. I have tested the latter and I can
> verify that it works well. I had to do some additional changes to avoid
> infinite wait after a couple of reboots but the reset function worked well.
>
> /Martin
>
>
>>
>> >
>> > /Martin
>> >
>> > 2014-08-17 3:52 GMT+02:00 Richie <listmail@triad.rr.com>:
>> >
>> > >  @Konrad
>> > >
>> > > I'm interesting in testing your latest iteration of bus reset support.
>> > > Actually I cloned your devel/xen-pciback.slot_and_bus.v2.1
>> > > <
>> http://git.kernel.org/cgit/linux/kernel/git/konrad/xen.git/log/?h=devel/xen-pciback.slot_and_bus.v2.1
>> >
>> > > branch and tested that but did not seem to reset the vga card.  I did
>> > > capture dmesg output which was nice and clean regarding resetting the
>> > > device but don't have it handy.
>> > >
>> > > _______________________________________________
>> > > Xen-devel mailing list
>> > > Xen-devel@lists.xen.org
>> > > http://lists.xen.org/xen-devel
>> > >
>> > >
>>
>> > _______________________________________________
>> > Xen-devel mailing list
>> > Xen-devel@lists.xen.org
>> > http://lists.xen.org/xen-devel
>>
>>
>

[-- Attachment #1.2: Type: text/html, Size: 4294 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel