Re: Intel graphics and kernel RT failure

Re: Intel graphics and kernel RT failure

[Jan Engelhardt]

Cc'ing linux-rt. Mabye someone has a flash of an idea.

On Thursday 2009-06-04 23:42, Hector Centeno wrote to Jan Engelhardt:
>
>I hope it is OK to contact you directly to this email. I've been
>trying to use your 2.6.29 RT kernel for OpenSuse 11.1 and came across
>some problems related to the Intel graphics driver (I have a laptop
>with intel GMA 950). The problem seems to be related to the new intel
>driver development that is going on and that seems to affect older
>hardware like mine. I get a constantly growing list of errors in
>/var/log/mesages like these:
>
>kernel: BUG: using smp_processor_id() in preemptible [00000000] code: X/3257
>kernel: caller is kmap_atomic_prot_pfn+0x1a/0x94
>kernel: Pid: 3257, comm: X Tainted: G   N  2.6.29.4-jen80-rt #1
>kernel: Call Trace:
>kernel:  [<b033c3f2>] ? printk+0xf/0x15
>kernel:  [<b022712e>] debug_smp_processor_id+0xa6/0xbc
>kernel:  [<b011d617>] kmap_atomic_prot_pfn+0x1a/0x94
>kernel:  [<f8a23c4c>] i915_gem_object_flush_cpu_write_domain+0x1c/0x27 [i915]
>kernel:  [<b011d6b7>] iomap_atomic_prot_pfn+0x26/0x2b
>kernel:  [<f8a270be>] i915_gem_gtt_pwrite+0x102/0x1b8 [i915]
>kernel:  [<f8a27290>] i915_gem_pwrite_ioctl+0x11c/0x1a8 [i915]
>kernel:  [<b0223e73>] ? copy_from_user+0x34/0x11b
>kernel:  [<f89d2776>] drm_ioctl+0x1b7/0x22f [drm]
>kernel:  [<f8a27174>] ? i915_gem_pwrite_ioctl+0x0/0x1a8 [i915]
>kernel:  [<b01ae63f>] vfs_ioctl+0x4e/0x67
>kernel:  [<b01aec30>] do_vfs_ioctl+0x22a/0x242
>kernel:  [<b01aec88>] sys_ioctl+0x40/0x5a
>kernel:  [<b01032fc>] sysenter_do_call+0x12/0x28
>kernel: BUG: using smp_processor_id() in preemptible [00000000] code: X/3257
>kernel: caller is iounmap_atomic+0x18/0x9b
>kernel: Pid: 3257, comm: X Tainted: G   N  2.6.29.4-jen80-rt #1
>kernel: Call Trace:
>kernel:  [<b033c3f2>] ? printk+0xf/0x15
>kernel:  [<b022712e>] debug_smp_processor_id+0xa6/0xbc
>kernel:  [<b011d57a>] iounmap_atomic+0x18/0x9b
>kernel:  [<f8a270df>] i915_gem_gtt_pwrite+0x123/0x1b8 [i915]
>kernel:  [<f8a27290>] i915_gem_pwrite_ioctl+0x11c/0x1a8 [i915]
>kernel:  [<b0223e73>] ? copy_from_user+0x34/0x11b
>kernel:  [<f89d2776>] drm_ioctl+0x1b7/0x22f [drm]
>kernel:  [<f8a27174>] ? i915_gem_pwrite_ioctl+0x0/0x1a8 [i915]
>kernel:  [<b01ae63f>] vfs_ioctl+0x4e/0x67
>kernel:  [<b01aec30>] do_vfs_ioctl+0x22a/0x242
>kernel:  [<b01aec88>] sys_ioctl+0x40/0x5a
>kernel:  [<b01032fc>] sysenter_do_call+0x12/0x28
>
>I'm assuming it is related to the Intel graphics driver because there
>is mention to i915 but I might be wrong. If the Intel driver is
>causing this I wonder if this has been fixed in upstream [...]

Re: Intel graphics and kernel RT failure

[Sven-Thorsten Dietrich]

On Fri, 2009-06-05 at 00:21 +0200, Jan Engelhardt wrote:
> 
> 
> Cc'ing linux-rt. Mabye someone has a flash of an idea.
> 
> On Thursday 2009-06-04 23:42, Hector Centeno wrote to Jan Engelhardt:
> >
> >I hope it is OK to contact you directly to this email. I've been
> >trying to use your 2.6.29 RT kernel for OpenSuse 11.1 and came across
> >some problems related to the Intel graphics driver (I have a laptop
> >with intel GMA 950). The problem seems to be related to the new intel
> >driver development that is going on and that seems to affect older
> >hardware like mine. I get a constantly growing list of errors in
> >/var/log/mesages like these:
> >

Try this patch:

Signed-off-by: Sven-Thorsten Dietrich <sdietrich@suse.de>

Index: b/arch/x86/mm/iomap_32.c
===================================================================
--- a/arch/x86/mm/iomap_32.c
+++ b/arch/x86/mm/iomap_32.c
@@ -38,7 +38,7 @@ void *kmap_atomic_prot_pfn(unsigned long
 
 	pagefault_disable();
 
-	idx = type + KM_TYPE_NR * smp_processor_id();
+	idx = type + KM_TYPE_NR * raw_smp_processor_id();
 	vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
 	set_pte(kmap_pte - idx, pfn_pte(pfn, prot));
 	arch_flush_lazy_mmu_mode();
@@ -69,7 +69,7 @@ void
 iounmap_atomic(void *kvaddr, enum km_type type)
 {
 	unsigned long vaddr = (unsigned long) kvaddr & PAGE_MASK;
-	enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();
+	enum fixed_addresses idx = type + KM_TYPE_NR*raw_smp_processor_id();
 
 	/*
 	 * Force other mappings to Oops if they'll try to access this pte

Re: Intel graphics and kernel RT failure

[Jan Engelhardt]

On Friday 2009-06-05 03:14, Sven-Thorsten Dietrich wrote:
>On Fri, 2009-06-05 at 00:21 +0200, Jan Engelhardt wrote:
>> 
>> Cc'ing linux-rt. Mabye someone has a flash of an idea.
>> 
>> On Thursday 2009-06-04 23:42, Hector Centeno wrote to Jan Engelhardt:
>> >
>> >I hope it is OK to contact you directly to this email. I've been
>> >trying to use your 2.6.29 RT kernel for OpenSuse 11.1 and came across
>> >some problems related to the Intel graphics driver (I have a laptop
>> >with intel GMA 950). The problem seems to be related to the new intel
>> >driver development that is going on and that seems to affect older
>> >hardware like mine. I get a constantly growing list of errors in
>> >/var/log/mesages like these:
>> >
>
>Try this patch:
>
>Signed-off-by: Sven-Thorsten Dietrich <sdietrich@suse.de>
>
>Index: b/arch/x86/mm/iomap_32.c
>===================================================================
>--- a/arch/x86/mm/iomap_32.c
>+++ b/arch/x86/mm/iomap_32.c
>@@ -38,7 +38,7 @@ void *kmap_atomic_prot_pfn(unsigned long
>-	idx = type + KM_TYPE_NR * smp_processor_id();
>+	idx = type + KM_TYPE_NR * raw_smp_processor_id();
>@@ -69,7 +69,7 @@ void
>-	enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();
>+	enum fixed_addresses idx = type + KM_TYPE_NR*raw_smp_processor_id();

Yes, it does fix the problem. Enabling PAE and not having applied
this patch also produces a working system for the original poster.

My thought was that this function kmap_atomic_prot_pfn is not used,
or used differently, under PAE. Could this be?

Re: Intel graphics and kernel RT failure

[Mark Knecht]

On Fri, Jun 5, 2009 at 8:07 AM, Jan Engelhardt<jengelh@medozas.de> wrote:
>
> On Friday 2009-06-05 03:14, Sven-Thorsten Dietrich wrote:
>>On Fri, 2009-06-05 at 00:21 +0200, Jan Engelhardt wrote:
>>>
>>> Cc'ing linux-rt. Mabye someone has a flash of an idea.
>>>
>>> On Thursday 2009-06-04 23:42, Hector Centeno wrote to Jan Engelhardt:
>>> >
>>> >I hope it is OK to contact you directly to this email. I've been
>>> >trying to use your 2.6.29 RT kernel for OpenSuse 11.1 and came across
>>> >some problems related to the Intel graphics driver (I have a laptop
>>> >with intel GMA 950). The problem seems to be related to the new intel
>>> >driver development that is going on and that seems to affect older
>>> >hardware like mine. I get a constantly growing list of errors in
>>> >/var/log/mesages like these:
>>> >
>>
>>Try this patch:
>>
>>Signed-off-by: Sven-Thorsten Dietrich <sdietrich@suse.de>
>>
>>Index: b/arch/x86/mm/iomap_32.c
>>===================================================================
>>--- a/arch/x86/mm/iomap_32.c
>>+++ b/arch/x86/mm/iomap_32.c
>>@@ -38,7 +38,7 @@ void *kmap_atomic_prot_pfn(unsigned long
>>-      idx = type + KM_TYPE_NR * smp_processor_id();
>>+      idx = type + KM_TYPE_NR * raw_smp_processor_id();
>>@@ -69,7 +69,7 @@ void
>>-      enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();
>>+      enum fixed_addresses idx = type + KM_TYPE_NR*raw_smp_processor_id();
>
> Yes, it does fix the problem. Enabling PAE and not having applied
> this patch also produces a working system for the original poster.
>
> My thought was that this function kmap_atomic_prot_pfn is not used,
> or used differently, under PAE. Could this be?

I have a system that's got an Intel chip on-board. It started failing
in strange ways when Gentoo decided that xorg-server-1.5 should go
stable and a newer version of the intel-video driver came along with
it. Searching around I found people saying turn on PAE, turn off PAE,
that it had nothing to do with PAE. I got the same failures on both
the standard and rt kernels. There are a number of bugs reported at
the Xorg site about this recent driver.

After 3 weeks of no MythTV on that machine I gave up and bought a
newer NVidia card. No problems since.

Just info.

cheers,
Mark
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Intel graphics and kernel RT failure

[Jan Engelhardt]

On Friday 2009-06-05 17:07, Jan Engelhardt wrote:
>On Friday 2009-06-05 03:14, Sven-Thorsten Dietrich wrote:
>>On Fri, 2009-06-05 at 00:21 +0200, Jan Engelhardt wrote:
>>> 
>>> Cc'ing linux-rt. Mabye someone has a flash of an idea.
>>> 
>>> On Thursday 2009-06-04 23:42, Hector Centeno wrote to Jan Engelhardt:
>>> >
>>> >I hope it is OK to contact you directly to this email. I've been
>>> >trying to use your 2.6.29 RT kernel for OpenSuse 11.1 and came across
>>> >some problems related to the Intel graphics driver (I have a laptop
>>> >with intel GMA 950). The problem seems to be related to the new intel
>>> >driver development that is going on and that seems to affect older
>>> >hardware like mine. I get a constantly growing list of errors in
>>> >/var/log/mesages like these:
>>> >
>>
>>Try this patch:
>>
>>Signed-off-by: Sven-Thorsten Dietrich <sdietrich@suse.de>
>>
>>Index: b/arch/x86/mm/iomap_32.c
>>===================================================================
>>--- a/arch/x86/mm/iomap_32.c
>>+++ b/arch/x86/mm/iomap_32.c
>>@@ -38,7 +38,7 @@ void *kmap_atomic_prot_pfn(unsigned long
>>-	idx = type + KM_TYPE_NR * smp_processor_id();
>>+	idx = type + KM_TYPE_NR * raw_smp_processor_id();
>>@@ -69,7 +69,7 @@ void
>>-	enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();
>>+	enum fixed_addresses idx = type + KM_TYPE_NR*raw_smp_processor_id();
>
>Yes, it does fix the problem.[...]

This change did not seem to have made it into 2.6.29.5-rt20.
Can it be added?

Re: Intel graphics and kernel RT failurery

[Thomas Gleixner]

On Thu, 4 Jun 2009, Sven-Thorsten Dietrich wrote:
> On Fri, 2009-06-05 at 00:21 +0200, Jan Engelhardt wrote:
> > 
> > 
> > Cc'ing linux-rt. Mabye someone has a flash of an idea.
> > 
> > On Thursday 2009-06-04 23:42, Hector Centeno wrote to Jan Engelhardt:
> > >
> > >I hope it is OK to contact you directly to this email. I've been
> > >trying to use your 2.6.29 RT kernel for OpenSuse 11.1 and came across
> > >some problems related to the Intel graphics driver (I have a laptop
> > >with intel GMA 950). The problem seems to be related to the new intel
> > >driver development that is going on and that seems to affect older
> > >hardware like mine. I get a constantly growing list of errors in
> > >/var/log/mesages like these:
> > >
> 
> Try this patch:
> 
> Signed-off-by: Sven-Thorsten Dietrich <sdietrich@suse.de>
> 
> Index: b/arch/x86/mm/iomap_32.c
> ===================================================================
> --- a/arch/x86/mm/iomap_32.c
> +++ b/arch/x86/mm/iomap_32.c
> @@ -38,7 +38,7 @@ void *kmap_atomic_prot_pfn(unsigned long
>  
>  	pagefault_disable();
>  
> -	idx = type + KM_TYPE_NR * smp_processor_id();
> +	idx = type + KM_TYPE_NR * raw_smp_processor_id();
>  	vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
>  	set_pte(kmap_pte - idx, pfn_pte(pfn, prot));
>  	arch_flush_lazy_mmu_mode();
> @@ -69,7 +69,7 @@ void
>  iounmap_atomic(void *kvaddr, enum km_type type)
>  {
>  	unsigned long vaddr = (unsigned long) kvaddr & PAGE_MASK;
> -	enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();
> +	enum fixed_addresses idx = type + KM_TYPE_NR*raw_smp_processor_id();
>  
>  	/*
>  	 * Force other mappings to Oops if they'll try to access this pte

No, that's wrong. We need to disable preemption across that
pagefault_disabled region. We might get preempted or migrated
otherwise.

Actually the correct fix for preempt-rt is to map this atomic mapping
magic to standard mappings, but that's more than a 5 minutes job.

Thanks,

	tglx