Re: [PATCH v4 30/32] drm/xe/gt: Report 'Fault response' pagefault error using SIGID

On 8/14/2026 12:12 AM, Daniele Ceraolo Spurio wrote:
> 
> 
> On 8/13/2026 12:46 PM, Michal Wajdeczko wrote:
>>
>> On 8/13/2026 8:37 PM, Rodrigo Vivi wrote:
>>> On Wed, Aug 12, 2026 at 09:14:46PM +0200, Michal Wajdeczko wrote:
>>>> Report 'Unsuccessful response' error using xe_log_err() helper.
>>>>
>>>> Signed-off-by: Michal Wajdeczko <[email protected]>
>>>> Cc: Rodrigo Vivi <[email protected]>
>>>> Cc: Matthew Brost <[email protected]>
>>>> Cc: Daniele Ceraolo Spurio <[email protected]>
>>>> Cc: Aravind Iddamsetty <[email protected]>
>>>> Cc: Mallesh Koujalagi <[email protected]>
>>>> ---
>>>> v2: rebased (Michal) reword message (Mallesh)
>>>> ---
>>>>   drivers/gpu/drm/xe/xe_pagefault.c | 4 ++--
>>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
>>>> index 7e829c61387f..eb600c466187 100644
>>>> --- a/drivers/gpu/drm/xe/xe_pagefault.c
>>>> +++ b/drivers/gpu/drm/xe/xe_pagefault.c
>>>> @@ -14,6 +14,7 @@
>>>>   #include "xe_gt_types.h"
>>>>   #include "xe_gt_stats.h"
>>>>   #include "xe_hw_engine.h"
>>>> +#include "xe_log.h"
>>>>   #include "xe_pagefault.h"
>>>>   #include "xe_pagefault_types.h"
>>>>   #include "xe_svm.h"
>>>> @@ -625,8 +626,7 @@ static void xe_pagefault_queue_work(struct work_struct *w)
>>>>                   xe_pagefault_save_to_vm(gt_to_xe(gt), pf);
>>>>                   xe_pagefault_cache_start_invalidate(cache_start);
>>>>                   xe_pagefault_print(pf);
>>>> -                xe_gt_info(pf->gt, "Fault response: Unsuccessful %pe\n",
>>>> -                       ERR_PTR(err));
>>>> +                xe_log_err(pf->gt, PAGEFAULT, err, "Unsuccessful response\n");
>>> I'm wondering if we should really go from info to err here...
>> my question was different:
>>     "why this err is only reported as info level?"
>> as once we hit this condition we also print large dump
>> (see xe_pagefault_print) likely for debug/diagnostics
>> so it seems important
> 
> drive-by comment, but AFAIU this error can be intentionally triggered by UMD by simply accessing an invalid PPGTT address (see e.g. xe_exec_fault_mode@invalid-va), and we usually avoid using error-level logs for something that UMD can cause on demand.

thanks for explanation! 

so we have two choices here:

1) drop this patch completely, or
2) switch to SIGID reporting with INFO severity

but currently xe_log_info() does not take an 'err' parameter
so we will not print the actual error code as it is today

so we have two more choices:

a) update xe_log_info() macro to accept 'err', or
b) add new xe_log_err_info() macro that takes 'err'
   but it will still use CPER_SEV_INFO severity

I'm in favor for 2) + b)

> 
> Daniele
> 
>>
>> @Matthew, comments?
>>
>>> Why not staying with xe_log_info or now?
>> I can change that for xe_log_info, but then (by design)
>> we will not print an error
>>
>>> But also, why err and not ERR_PTR(err)?
>> patch 10/32 [1] introduced helpers that take just errno
>> value and convert that into data { ERR_PTR(err), 0 } pair
>>
>> [1] https://patchwork.freedesktop.org/patch/745993/?series=171022&rev=4
>>
>>>>               } else {
>>>>                   xe_gt_stats_incr(pf->gt, XE_GT_STATS_ID_INVALID_PREFETCH_PAGEFAULT_COUNT, 1);
>>>>                   xe_gt_dbg(pf->gt, "Prefetch Fault response: Unsuccessful %pe\n",
>>>> -- 
>>>> 2.47.1
>>>>
>