* RE: [BUG]: Intel uncore boot warning introduced in 4.1
@ 2015-08-06 18:10 ` Liang, Kan
0 siblings, 0 replies; 13+ messages in thread
From: Liang, Kan @ 2015-08-06 18:10 UTC (permalink / raw)
To: Matthew Leach, Ingo Molnar
Cc: linux-kernel@vger.kernel.org, linux-ia64@vger.kernel.org,
eranian@google.com, 'Andi Kleen', 'Bjorn Helgaas',
'Vince Weaver', 'Peter Zijlstra',
'Sonny Rao'
>
> Hi Ingo,
>
> Matthew Leach <matthew@mattleach.net> writes:
>
> [...]
>
> > I've bisected the kernel and found that commit [2] seems to introduce
> > the warning message. I have checked on a v4.2-rc5 kernel and the
> > warning message is still there.
>
> [...]
>
> > [2]: 8cf1a3de97804b047973dd44cfacdc1930da8403
>
> Apologies, I got it wrong. The commit that is causing the issue is [1].
> If I revert it, the warning goes away. I'm also testing to see if this is the
> cause of the random freezing that occurs (which I can confirm is also
> happening with v4.2-rc5).
>
> [1]: 15c1247953e8a45232ed5a5540f291d2d0a77665
>
The issue may be caused by uncore box initialization.
For preventing the potential issues of uncore box initialization, I once
moved the uncore_box_init() out of driver initialization in commit
c05199e5a57a579fea1e8fa65e2b511ceb524ffc.
However, it cause some desktop crash, because the box initialization
codes were moved in IPI context.
For fixing the crash issue, we had two choice at that time.
- Simply revert the codes. That's where is
15c1247953e8a45232ed5a5540f291d2d0a77665 from.
- Move uncore_box_init out of IPI context to uncore event
init. I provided a patch for it. https://lkml.org/lkml/2015/4/28/21
Stephane Eranian also verified it on his platform
At that time, we chose first option. But it looks there is some
issue now. I guess we may try the second option this time.
Matthew,
Could you please revert
15c1247953e8a45232ed5a5540f291d2d0a77665
and apply the patch https://lkml.org/lkml/2015/4/26/294?
See if it works?
Thanks,
Kan
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [BUG]: Intel uncore boot warning introduced in 4.1
2015-08-06 18:10 ` Liang, Kan
@ 2015-08-06 18:44 ` Matthew Leach
-1 siblings, 0 replies; 13+ messages in thread
From: Matthew Leach @ 2015-08-06 18:44 UTC (permalink / raw)
To: Liang, Kan
Cc: Ingo Molnar, linux-kernel@vger.kernel.org,
linux-ia64@vger.kernel.org, eranian@google.com,
'Andi Kleen', 'Bjorn Helgaas',
'Vince Weaver', 'Peter Zijlstra',
'Sonny Rao'
Hi Kan,
"Liang, Kan" <kan.liang@intel.com> writes:
[...]
> Matthew,
>
> Could you please revert
> 15c1247953e8a45232ed5a5540f291d2d0a77665
> and apply the patch https://lkml.org/lkml/2015/4/26/294?
> See if it works?
That works for me. I no longer get the warning in my kernel boot log.
Thanks,
Matt
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [BUG]: Intel uncore boot warning introduced in 4.1
@ 2015-08-06 18:44 ` Matthew Leach
0 siblings, 0 replies; 13+ messages in thread
From: Matthew Leach @ 2015-08-06 18:44 UTC (permalink / raw)
To: Liang, Kan
Cc: Ingo Molnar, linux-kernel@vger.kernel.org,
linux-ia64@vger.kernel.org, eranian@google.com,
'Andi Kleen', 'Bjorn Helgaas',
'Vince Weaver', 'Peter Zijlstra',
'Sonny Rao'
Hi Kan,
"Liang, Kan" <kan.liang@intel.com> writes:
[...]
> Matthew,
>
> Could you please revert
> 15c1247953e8a45232ed5a5540f291d2d0a77665
> and apply the patch https://lkml.org/lkml/2015/4/26/294?
> See if it works?
That works for me. I no longer get the warning in my kernel boot log.
Thanks,
Matt
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [BUG]: Intel uncore boot warning introduced in 4.1
2015-08-06 18:10 ` Liang, Kan
@ 2015-08-07 9:05 ` Peter Zijlstra
-1 siblings, 0 replies; 13+ messages in thread
From: Peter Zijlstra @ 2015-08-07 9:05 UTC (permalink / raw)
To: Liang, Kan
Cc: Matthew Leach, Ingo Molnar, linux-kernel@vger.kernel.org,
linux-ia64@vger.kernel.org, eranian@google.com,
'Andi Kleen', 'Bjorn Helgaas',
'Vince Weaver', 'Sonny Rao'
On Thu, Aug 06, 2015 at 06:10:40PM +0000, Liang, Kan wrote:
> The issue may be caused by uncore box initialization.
>
> For preventing the potential issues of uncore box initialization, I once
> moved the uncore_box_init() out of driver initialization in commit
> c05199e5a57a579fea1e8fa65e2b511ceb524ffc.
>
> However, it cause some desktop crash, because the box initialization
> codes were moved in IPI context.
>
> For fixing the crash issue, we had two choice at that time.
> - Simply revert the codes. That's where is
> 15c1247953e8a45232ed5a5540f291d2d0a77665 from.
> - Move uncore_box_init out of IPI context to uncore event
> init. I provided a patch for it. https://lkml.org/lkml/2015/4/28/21
> Stephane Eranian also verified it on his platform
>
> At that time, we chose first option. But it looks there is some
> issue now. I guess we may try the second option this time.
>
> Matthew,
>
> Could you please revert
> 15c1247953e8a45232ed5a5540f291d2d0a77665
> and apply the patch https://lkml.org/lkml/2015/4/26/294?
> See if it works?
That patch is wrong though; how can even publish a PMU which is not
initialized?
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [BUG]: Intel uncore boot warning introduced in 4.1
@ 2015-08-07 9:05 ` Peter Zijlstra
0 siblings, 0 replies; 13+ messages in thread
From: Peter Zijlstra @ 2015-08-07 9:05 UTC (permalink / raw)
To: Liang, Kan
Cc: Matthew Leach, Ingo Molnar, linux-kernel@vger.kernel.org,
linux-ia64@vger.kernel.org, eranian@google.com,
'Andi Kleen', 'Bjorn Helgaas',
'Vince Weaver', 'Sonny Rao'
On Thu, Aug 06, 2015 at 06:10:40PM +0000, Liang, Kan wrote:
> The issue may be caused by uncore box initialization.
>
> For preventing the potential issues of uncore box initialization, I once
> moved the uncore_box_init() out of driver initialization in commit
> c05199e5a57a579fea1e8fa65e2b511ceb524ffc.
>
> However, it cause some desktop crash, because the box initialization
> codes were moved in IPI context.
>
> For fixing the crash issue, we had two choice at that time.
> - Simply revert the codes. That's where is
> 15c1247953e8a45232ed5a5540f291d2d0a77665 from.
> - Move uncore_box_init out of IPI context to uncore event
> init. I provided a patch for it. https://lkml.org/lkml/2015/4/28/21
> Stephane Eranian also verified it on his platform
>
> At that time, we chose first option. But it looks there is some
> issue now. I guess we may try the second option this time.
>
> Matthew,
>
> Could you please revert
> 15c1247953e8a45232ed5a5540f291d2d0a77665
> and apply the patch https://lkml.org/lkml/2015/4/26/294?
> See if it works?
That patch is wrong though; how can even publish a PMU which is not
initialized?
^ permalink raw reply [flat|nested] 13+ messages in thread
* RE: [BUG]: Intel uncore boot warning introduced in 4.1
2015-08-07 9:05 ` Peter Zijlstra
(?)
@ 2015-08-10 13:23 ` Liang, Kan
2015-09-15 13:35 ` Josh Boyer
-1 siblings, 1 reply; 13+ messages in thread
From: Liang, Kan @ 2015-08-10 13:23 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Matthew Leach, Ingo Molnar, linux-kernel@vger.kernel.org,
linux-ia64@vger.kernel.org, eranian@google.com,
'Andi Kleen', 'Bjorn Helgaas',
'Vince Weaver', 'Sonny Rao'
> On Thu, Aug 06, 2015 at 06:10:40PM +0000, Liang, Kan wrote:
> > The issue may be caused by uncore box initialization.
> >
> > For preventing the potential issues of uncore box initialization, I
> > once moved the uncore_box_init() out of driver initialization in
> > commit c05199e5a57a579fea1e8fa65e2b511ceb524ffc.
> >
> > However, it cause some desktop crash, because the box initialization
> > codes were moved in IPI context.
> >
> > For fixing the crash issue, we had two choice at that time.
> > - Simply revert the codes. That's where is
> > 15c1247953e8a45232ed5a5540f291d2d0a77665 from.
> > - Move uncore_box_init out of IPI context to uncore event
> > init. I provided a patch for it. https://lkml.org/lkml/2015/4/28/21
> > Stephane Eranian also verified it on his platform
> >
> > At that time, we chose first option. But it looks there is some issue
> > now. I guess we may try the second option this time.
> >
> > Matthew,
> >
> > Could you please revert
> > 15c1247953e8a45232ed5a5540f291d2d0a77665
> > and apply the patch https://lkml.org/lkml/2015/4/26/294?
> > See if it works?
>
> That patch is wrong though; how can even publish a PMU which is not
> initialized?
It's initialized but not in the driver initialization.
We once encountered boot crashes which caused by uncore
driver who trying to access non-existing boxes. Also this uncore
boot warning.
So I think it's better to move the box init code out of driver
initialization to prevent such potential boot failures.
Uncore event init should be a good place to do box init.
Only when the box is not initialized and user tries to use
uncore event, we do box initialization.
Thanks,
Kan
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [BUG]: Intel uncore boot warning introduced in 4.1
2015-08-10 13:23 ` Liang, Kan
@ 2015-09-15 13:35 ` Josh Boyer
0 siblings, 0 replies; 13+ messages in thread
From: Josh Boyer @ 2015-09-15 13:35 UTC (permalink / raw)
To: Liang, Kan
Cc: Peter Zijlstra, Matthew Leach, Ingo Molnar,
linux-kernel@vger.kernel.org, linux-ia64@vger.kernel.org,
eranian@google.com, Andi Kleen, Bjorn Helgaas, Vince Weaver,
Sonny Rao
On Mon, Aug 10, 2015 at 9:23 AM, Liang, Kan <kan.liang@intel.com> wrote:
>
>> On Thu, Aug 06, 2015 at 06:10:40PM +0000, Liang, Kan wrote:
>> > The issue may be caused by uncore box initialization.
>> >
>> > For preventing the potential issues of uncore box initialization, I
>> > once moved the uncore_box_init() out of driver initialization in
>> > commit c05199e5a57a579fea1e8fa65e2b511ceb524ffc.
>> >
>> > However, it cause some desktop crash, because the box initialization
>> > codes were moved in IPI context.
>> >
>> > For fixing the crash issue, we had two choice at that time.
>> > - Simply revert the codes. That's where is
>> > 15c1247953e8a45232ed5a5540f291d2d0a77665 from.
>> > - Move uncore_box_init out of IPI context to uncore event
>> > init. I provided a patch for it. https://lkml.org/lkml/2015/4/28/21
>> > Stephane Eranian also verified it on his platform
>> >
>> > At that time, we chose first option. But it looks there is some issue
>> > now. I guess we may try the second option this time.
>> >
>> > Matthew,
>> >
>> > Could you please revert
>> > 15c1247953e8a45232ed5a5540f291d2d0a77665
>> > and apply the patch https://lkml.org/lkml/2015/4/26/294?
>> > See if it works?
>>
>> That patch is wrong though; how can even publish a PMU which is not
>> initialized?
>
> It's initialized but not in the driver initialization.
> We once encountered boot crashes which caused by uncore
> driver who trying to access non-existing boxes. Also this uncore
> boot warning.
> So I think it's better to move the box init code out of driver
> initialization to prevent such potential boot failures.
> Uncore event init should be a good place to do box init.
> Only when the box is not initialized and user tries to use
> uncore event, we do box initialization.
We're still getting reports of this in Fedora with 4.1.y kernels. Was
there any resolution to this?
josh
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [BUG]: Intel uncore boot warning introduced in 4.1
@ 2015-09-15 13:35 ` Josh Boyer
0 siblings, 0 replies; 13+ messages in thread
From: Josh Boyer @ 2015-09-15 13:35 UTC (permalink / raw)
To: Liang, Kan
Cc: Peter Zijlstra, Matthew Leach, Ingo Molnar,
linux-kernel@vger.kernel.org, linux-ia64@vger.kernel.org,
eranian@google.com, Andi Kleen, Bjorn Helgaas, Vince Weaver,
Sonny Rao
On Mon, Aug 10, 2015 at 9:23 AM, Liang, Kan <kan.liang@intel.com> wrote:
>
>> On Thu, Aug 06, 2015 at 06:10:40PM +0000, Liang, Kan wrote:
>> > The issue may be caused by uncore box initialization.
>> >
>> > For preventing the potential issues of uncore box initialization, I
>> > once moved the uncore_box_init() out of driver initialization in
>> > commit c05199e5a57a579fea1e8fa65e2b511ceb524ffc.
>> >
>> > However, it cause some desktop crash, because the box initialization
>> > codes were moved in IPI context.
>> >
>> > For fixing the crash issue, we had two choice at that time.
>> > - Simply revert the codes. That's where is
>> > 15c1247953e8a45232ed5a5540f291d2d0a77665 from.
>> > - Move uncore_box_init out of IPI context to uncore event
>> > init. I provided a patch for it. https://lkml.org/lkml/2015/4/28/21
>> > Stephane Eranian also verified it on his platform
>> >
>> > At that time, we chose first option. But it looks there is some issue
>> > now. I guess we may try the second option this time.
>> >
>> > Matthew,
>> >
>> > Could you please revert
>> > 15c1247953e8a45232ed5a5540f291d2d0a77665
>> > and apply the patch https://lkml.org/lkml/2015/4/26/294?
>> > See if it works?
>>
>> That patch is wrong though; how can even publish a PMU which is not
>> initialized?
>
> It's initialized but not in the driver initialization.
> We once encountered boot crashes which caused by uncore
> driver who trying to access non-existing boxes. Also this uncore
> boot warning.
> So I think it's better to move the box init code out of driver
> initialization to prevent such potential boot failures.
> Uncore event init should be a good place to do box init.
> Only when the box is not initialized and user tries to use
> uncore event, we do box initialization.
We're still getting reports of this in Fedora with 4.1.y kernels. Was
there any resolution to this?
josh
^ permalink raw reply [flat|nested] 13+ messages in thread