[Intel-gfx] [PATCH] drm/i915/mtl: Print SSEU information of all GTs for debugfs

Tue Nov 7 02:40:41 UTC 2023

The reason to dump on GTs is to make consistent SSEU information in i915. It is fine that the information was moved to i915/gt*.

-----Original Message-----
From: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com> 
Sent: Monday, November 6, 2023 8:30 PM
To: Roper, Matthew D <matthew.d.roper at intel.com>
Cc: Yu, Gareth <gareth.yu at intel.com>; intel-gfx at lists.freedesktop.org; Shyti, Andi <andi.shyti at intel.com>
Subject: Re: [Intel-gfx] [PATCH] drm/i915/mtl: Print SSEU information of all GTs for debugfs

On 03/11/2023 17:24, Matt Roper wrote:
> On Fri, Nov 03, 2023 at 11:17:18AM +0000, Tvrtko Ursulin wrote:
>>
>> On 03/11/2023 05:29, Gareth Yu wrote:
>>> Print another SSEU information addition to the first one.
>>>
>>> Cc : Tejas Upadhyay <tejaskumarx.surendrakumar.upadhyay at intel.com>
>>> Cc : Matt Roper <matthew.d.roper at intel.com> Cc : Ville Syrjälä 
>>> <ville.syrjala at linux.intel.com>
>>> Signed-off-by: Gareth Yu <gareth.yu at intel.com>
>>> ---
>>>    drivers/gpu/drm/i915/i915_debugfs.c   | 13 ++++++++++---
>>>    drivers/gpu/drm/i915/i915_gpu_error.c |  6 +++++-
>>>    2 files changed, 15 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c 
>>> b/drivers/gpu/drm/i915/i915_debugfs.c
>>> index e9b79c2c37d8..b5914a2c0597 100644
>>> --- a/drivers/gpu/drm/i915/i915_debugfs.c
>>> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
>>> @@ -63,13 +63,16 @@ static int i915_capabilities(struct seq_file *m, void *data)
>>>    {
>>>    	struct drm_i915_private *i915 = node_to_i915(m->private);
>>>    	struct drm_printer p = drm_seq_file_printer(m);
>>> +	struct intel_gt *gt;
>>> +	unsigned int i;
>>>    	seq_printf(m, "pch: %d\n", INTEL_PCH_TYPE(i915));
>>>    	intel_device_info_print(INTEL_INFO(i915), RUNTIME_INFO(i915), &p);
>>>    	intel_display_device_info_print(DISPLAY_INFO(i915), DISPLAY_RUNTIME_INFO(i915), &p);
>>>    	i915_print_iommu_status(i915, &p);
>>> -	intel_gt_info_print(&to_gt(i915)->info, &p);
>>> +	for_each_gt(gt, i915, i)
>>> +		intel_gt_info_print(&gt->info, &p);
>>>    	intel_driver_caps_print(&i915->caps, &p);
>>>    	kernel_param_lock(THIS_MODULE);
>>> @@ -783,9 +786,13 @@ DEFINE_SIMPLE_ATTRIBUTE(i915_drop_caches_fops,
>>>    static int i915_sseu_status(struct seq_file *m, void *unused)
>>>    {
>>>    	struct drm_i915_private *i915 = node_to_i915(m->private);
>>> -	struct intel_gt *gt = to_gt(i915);
>>> +	struct intel_gt *gt;
>>> +	unsigned int i;
>>> +
>>> +	for_each_gt(gt, i915, i)
>>> +		intel_sseu_status(m, gt);
>>
>> Don't we have the per GT debugfs directories and files already!?
> 
> Yeah, we shouldn't be updating this.  Commit a00eda7d8996 ("drm/i915:
> Move sseu debugfs under gt/") notes:
> 
>          "The sseu_status debugfs has also been kept at the top level as
>          we do have tests that use it; it will be removed once we teach
>          the tests to look into the new path."
> 
> If there are still IGT tests that haven't been updated, dumping both 
> GTs here will probably break them since they aren't expecting it.  If 
> they have all been updated, then we should just move forward with 
> deleting this device-level SSEU instead.

Thanks for confirming the debugfs parts should be dropped.

> 
>>
>>> -	return intel_sseu_status(m, gt);
>>> +	return 0;
>>>    }
>>>    static int i915_forcewake_open(struct inode *inode, struct file 
>>> *file) diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c 
>>> b/drivers/gpu/drm/i915/i915_gpu_error.c
>>> index b4e31e59c799..2adc317af944 100644
>>> --- a/drivers/gpu/drm/i915/i915_gpu_error.c
>>> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
>>> @@ -722,9 +722,13 @@ static void err_print_gt_info(struct drm_i915_error_state_buf *m,
>>>    			      struct intel_gt_coredump *gt)
>>>    {
>>>    	struct drm_printer p = i915_error_printer(m);
>>> +	struct drm_i915_private *i915 = gt->_gt->i915;
>>> +	struct intel_gt *gt_n;
>>> +	unsigned int n;
>>>    	intel_gt_info_print(&gt->info, &p);
>>> -	intel_sseu_print_topology(gt->_gt->i915, &gt->info.sseu, &p);
>>> +	for_each_gt(gt_n, i915, n)
>>> +		intel_sseu_print_topology(gt_n->i915, &gt_n->info.sseu, &p);
>>
>> Do we need a consistent story across all error capture? Aka why is 
>> sseu special.
>>
>> Also the intel_gt_info_print() above calls intel_sseu_dump so we end 
>> up with root tile SSEU printed twice?
> 
> I'm guessing this call was supposed to be deleted by 0b6613c6b91e
> ("drm/i915/sseu: Move sseu_info under gt_info").  We should probably 
> go ahead and do that now do remove the redundancy.

Okay, just need a volunteer! :)

> err_print_gt_info() should be printing the GT information (including
> SSEU) for whichever GT had the error.  I don't see a reason why we'd 
> want to dump extra SSEU information for a different GT that wasn't 
> involved in the error.
> 
> Actually, SSEU is the _least_ useful thing to print for extra GTs 
> because once xehpsdv/pvc are gone from i915, the only platforms that 
> have multiple GTs are MTL/ARL and the SSEU information will always be 
> empty on the media GT (there's no DSS or EUs there).

Yes all good points. Perhaps Yu can shed more light on why the need to dump on GTs came about?

Then, on a second look, the correct GT info is already captured. What is missing is teaching, probably intel_gt_info_print, to skip dumping the SSEU info on media tiles. That coupled with removing one of the redundant calls should complete the picture. Although I don't know which one between intel_sseu_print_topology and intel_sseu_dump to keep. Or the data from intel_sseu_print_topology is a genuine value add on top of intel_sseu_dump?

Regards,

Tvrtko

> 
> 
> Matt
> 
>>
>> There possibly was a Jira years ago to do stuff about multi-tile 
>> error capture but maybe it got lost.
>>
>> Adding Andi if he has comments.
>>
>> Regards,
>>
>> Tvrtko
>>
>>>    }
>>>    static void err_print_gt_display(struct drm_i915_error_state_buf 
>>> *m,
> 