[Intel-gfx] interference on display, Intel 945GM, 32bit system, kernel next

[Krzysztof Kolasa]

On 22.07.2015 10:27, Daniel Vetter wrote:
> On Tue, Jul 21, 2015 at 08:07:47PM +0200, Krzysztof Kolasa wrote:
>> On 21.07.2015 16:03, Daniel Vetter wrote:
>>> On Tue, Jul 21, 2015 at 02:48:21PM +0200, Krzysztof Kolasa wrote:
>>>> On 21.07.2015 11:43, Chris Wilson wrote:
>>>>> On Tue, Jul 21, 2015 at 11:07:20AM +0200, Daniel Vetter wrote:
>>>>>> On Tue, Jul 21, 2015 at 10:58:50AM +0200, Krzysztof Kolasa wrote:
>>>>>>> On 21.07.2015 10:41, Daniel Vetter wrote:
>>>>>>>> I meant whether you can reset the bad commit and it's immediate parent
>>>>>>>> extensively to make sure the bisect is really correct. gpu's occasionally
>>>>>>>> take a while to hang themselves, so could be that the bisect was
>>>>>>>> mislead somewhere.
>>>>>>> Again I will bisect, more testing good sections.
>>>>>>>
>>>>>>> At the moment on my old laptop, the latest official, properly working kernel is 4.1.2
>>>>>> You don't need to redo the full bisect, only retest the bad and the parent
>>>>>> of the bad extensively. That's enough to confirm the bisect result for
>>>>>> sure.
>>>>> The bisection is misleading. It's a missing write-barrier.
>>>>> -Chris
>>>>>
>>>> tested again:
>>>>
>>>> # good: [cd102a687beed1042824d5fa81c6ba8bfe78e6a4] drm/i915: Remove misleading comment around bind_to_vm
>>>> git bisect good cd102a687beed1042824d5fa81c6ba8bfe78e6a4
>>>> # bad: [0875546c5318c85c13d07014af5350e9000bc9e9] drm/i915: Fix up the vma aliasing ppgtt binding
>>>> git bisect bad 0875546c5318c85c13d07014af5350e9000bc9e9
>>>> # first bad commit: [0875546c5318c85c13d07014af5350e9000bc9e9] drm/i915: Fix up the vma aliasing ppgtt binding
>>>>
>>>>
>>>> Screen does not lie :) first bad is BAD, good is GOOD
>>>>
>>>> on commit cd102a687beed10, computer worked steadily (with a large number of windows, applications) about an hour and stopped the tests, everything was OK
>>>>
>>>> on commit 0875546c5318c85, after about one minute, the screen gone mad
>>> Suprising. Can you try out the below patch, it should disable the active
>>> ingredient of the offending. On older kernels just remove the if
>>> (bind_flags) before the call to ->bind_vma - the point is to call
>>> ->bind_vma unconditionally.
>>> -Daniel
>>>
>>> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
>>> index cc133c700686..1227cd69c624 100644
>>> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
>>> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
>>> @@ -2908,9 +2908,6 @@ int i915_vma_bind(struct i915_vma *vma, enum i915_cache_level cache_level,
>>>  	else
>>>  		bind_flags &= ~vma->bound;
>>>  
>>> -	if (bind_flags == 0)
>>> -		return 0;
>>> -
>>>  	if (vma->bound == 0 && vma->vm->allocate_va_range) {
>>>  		trace_i915_va_alloc(vma->vm,
>>>  				    vma->node.start,
>>
>> I tested the patch and did not help ...
> Does this one below help?
> -Daniel
>
> diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
> index d2df321ba634..38dce61e028d 100644
> --- a/drivers/gpu/drm/i915/i915_dma.c
> +++ b/drivers/gpu/drm/i915/i915_dma.c
> @@ -150,9 +150,6 @@ static int i915_getparam(struct drm_device *dev, void *data,
>  	case I915_PARAM_HAS_COHERENT_PHYS_GTT:
>  		value = 1;
>  		break;
> -	case I915_PARAM_MMAP_VERSION:
> -		value = 1;
> -		break;
>  	case I915_PARAM_SUBSLICE_TOTAL:
>  		value = INTEL_INFO(dev)->subslice_total;
>  		if (!value)
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 52b446b27b4d..c71525cefdf9 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1676,7 +1676,7 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data,
>  	struct drm_gem_object *obj;
>  	unsigned long addr;
>  
> -	if (args->flags & ~(I915_MMAP_WC))
> +	if (args->flags)
>  		return -EINVAL;
>  
>  	if (args->flags & I915_MMAP_WC && !cpu_has_pat)
This commit solved the problem:

http://cgit.freedesktop.org/drm-intel/commit/?id=d0e30adc42d979e4adc36b6c112b57337423b70c

drm/i915: Mark PIN_USER binding as GLOBAL_BIND without the aliasing ppgtt

Krzysztof