[PATCH v8 2/6] drm/xe/xe_gt_pagefault: Move pagefault struct to header

Tue Mar 18 08:44:49 UTC 2025


On 17.03.2025 21:55, Cavitt, Jonathan wrote:
> -----Original Message-----
> From: Wajdeczko, Michal <Michal.Wajdeczko at intel.com> 
> Sent: Saturday, March 15, 2025 7:45 AM
> To: Cavitt, Jonathan <jonathan.cavitt at intel.com>; intel-xe at lists.freedesktop.org
> Cc: Gupta, saurabhg <saurabhg.gupta at intel.com>; Zuo, Alex <alex.zuo at intel.com>; joonas.lahtinen at linux.intel.com; Brost, Matthew <matthew.brost at intel.com>; Zhang, Jianxun <jianxun.zhang at intel.com>; Lin, Shuicheng <shuicheng.lin at intel.com>; dri-devel at lists.freedesktop.org; michal.mzorek at intel.com
> Subject: Re: [PATCH v8 2/6] drm/xe/xe_gt_pagefault: Move pagefault struct to header
>>
>> On 14.03.2025 23:06, Cavitt, Jonathan wrote:
>>> -----Original Message-----
>>> From: Wajdeczko, Michal <Michal.Wajdeczko at intel.com> 
>>> Sent: Friday, March 14, 2025 10:02 AM
>>> To: Cavitt, Jonathan <jonathan.cavitt at intel.com>; intel-xe at lists.freedesktop.org
>>> Cc: Gupta, saurabhg <saurabhg.gupta at intel.com>; Zuo, Alex <alex.zuo at intel.com>; joonas.lahtinen at linux.intel.com; Brost, Matthew <matthew.brost at intel.com>; Zhang, Jianxun <jianxun.zhang at intel.com>; Lin, Shuicheng <shuicheng.lin at intel.com>; dri-devel at lists.freedesktop.org; michal.mzorek at intel.com
>>> Subject: Re: [PATCH v8 2/6] drm/xe/xe_gt_pagefault: Move pagefault struct to header
>>>>
>>>> On 13.03.2025 19:34, Jonathan Cavitt wrote:
>>>>> Move the pagefault struct from xe_gt_pagefault.c to the
>>>>> xe_gt_pagefault_types.h header file, along with the associated enum values.
>>>>>
>>>>> v2:
>>>>> - Normalize names for common header (Matt Brost)
>>>>>
>>>>> v3:
>>>>> - s/Migrate/Move (Michal W)
>>>>> - s/xe_pagefault/xe_gt_pagefault (Michal W)
>>>>> - Create new header file, xe_gt_pagefault_types.h (Michal W)
>>>>> - Add kernel docs (Michal W)
>>>>>
>>>>> Signed-off-by: Jonathan Cavitt <jonathan.cavitt at intel.com>
>>>>> Cc: Michal Wajdeczko <Michal.Wajdeczko at intel.com>
>>>>> ---
>>>>
>>>> ...
>>>>
>>>>> diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.h b/drivers/gpu/drm/xe/xe_gt_pagefault.h
>>>>> index 839c065a5e4c..69b700c4915a 100644
>>>>> --- a/drivers/gpu/drm/xe/xe_gt_pagefault.h
>>>>> +++ b/drivers/gpu/drm/xe/xe_gt_pagefault.h
>>>>> @@ -8,6 +8,8 @@
>>>>>  
>>>>>  #include <linux/types.h>
>>>>>  
>>>>> +#include "xe_gt_pagefault_types.h"
>>>>
>>>> it's not needed here, move it to .c
>>>>
>>>>> +
>>>>>  struct xe_gt;
>>>>>  struct xe_guc;
>>>>>  
>>>>> diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault_types.h b/drivers/gpu/drm/xe/xe_gt_pagefault_types.h
>>>>> new file mode 100644
>>>>> index 000000000000..90b7085d4b8e
>>>>> --- /dev/null
>>>>> +++ b/drivers/gpu/drm/xe/xe_gt_pagefault_types.h
>>>>> @@ -0,0 +1,67 @@
>>>>> +/* SPDX-License-Identifier: MIT */
>>>>> +/*
>>>>> + * Copyright (c) 2022-2025 Intel Corporation
>>>>> + */
>>>>> +
>>>>> +#ifndef _XE_GT_PAGEFAULT_TYPES_H_
>>>>> +#define _XE_GT_PAGEFAULT_TYPES_H_
>>>>> +
>>>>
>>>> don't forget to
>>>>
>>>> #include <linux/types.h>
>>>
>>> That explains why the kernel failed to compile on CI.  It was compiling just
>>> fine locally, so that's why I missed this.
>>>
>>>>
>>>>> +/**
>>>>> + * struct xe_gt_pagefault - Structure of pagefaults returned by the
>>>>> + * pagefault handler
>>>>> + */
>>>>> +struct xe_gt_pagefault {
>>>>> +	/** @page_addr: faulted address of this pagefault */
>>>>> +	u64 page_addr;
>>>>> +	/** @asid: ASID of this pagefault */
>>>>> +	u32 asid;
>>>>> +	/** @pdata: PDATA of this pagefault */
>>>>> +	u16 pdata;
>>>>> +	/** @vfid: VFID of this pagefault */
>>>>> +	u8 vfid;
>>>>
>>>> btw, IIRC the VFID from the descriptor will be zero'ed
>>>> does it make sense to keep it here?
>>>
>>> Is the argument that every time pf->vfid is accessed, it's guaranteed to be
>>
>> it's FIELD_GET(PFD_VFID, desc->dw2) guaranteed to be zero
>>
>>> zero?  I can't counter that claim, but wouldn't it be safer to keep reporting
>>> the VFID in case we ever hit a case where it's no longer zero?
>>
>> it can't be non-zero, look at GuC ABI spec that says:
>>
>> ".. with the VF number being zero’d out in the descriptor."
>>
>>>
>>> Also, did we know it would always be zero when we were making the
>>> pagefault struct originally?  If so, why did we include the vfid originally?
>>
>> dunno, ask authors, it was pushed part of the initial commit
>> dd08ebf6c352 that was *not* going through normal review cycle
>>
>>>
>>>>
>>>>> +	/**
>>>>> +	 * @access_type: access type of this pagefault, as a value
>>>>> +	 * from xe_gt_pagefault_access_type
>>>>> +	 */
>>>>> +	u8 access_type;
>>>>> +	/**
>>>>> +	 * @fault_type: fault type of this pagefault, as a value
>>>>> +	 * from xe_gt_pagefault_fault_type
>>>>> +	 */
>>>>> +	u8 fault_type;
>>>>> +	/** @fault_level: fault level of this pagefault */
>>>>> +	u8 fault_level;
>>>>> +	/** @engine_class: engine class this pagefault was reported on */
>>>>> +	u8 engine_class;
>>>>> +	/** @engine_instance: engine instance this pagefault was reported on */
>>>>> +	u8 engine_instance;
>>>>> +	/** @fault_unsuccessful: flag for if the pagefault recovered or not */
>>>>> +	u8 fault_unsuccessful;
>>>>> +	/** @prefetch: unused */
>>>>> +	bool prefetch;
>>>>> +	/** @trva_fault: is set if this is a TRTT fault */
>>>>> +	bool trva_fault;
>>>>> +};
>>>>> +
>>>>> +/**
>>>>> + * enum xe_gt_pagefault_access_type - Access type reported to the xe_gt_pagefault
>>>>> + * struct.  Saved to xe_gt_pagefault at access_type
>>>>
>>>> this seems to be copied from G2H descriptor as-is.
>>>> so shouldn't this be part of the GuC ABI?
>>>> or based on HW ABI if GuC is just a proxy
>>>
>>> What information should I be including in the kernel docs for these enums?
>>
>> are you asking about xe enums or GuC ABI defs/enums?
>>
>> for the xe enums, we can say this is what HW/GuC provides
>> for the HW defs, we can likely drop kernel-doc
>> we can just point to the Bspec in the commit msg
> 
> Could you please provide the bspec number for the pagefault
> access and fault types?  I tried searching for them on my end but
> wasn't able to find them.

try this one

Bspec: 77412

> -Jonathan Cavitt
> 
>>
>>> -Jonathan Cavitt
>>>
>>>>
>>>>> + */
>>>>> +enum xe_gt_pagefault_access_type {
>>>>> +	XE_GT_PAGEFAULT_ACCESS_TYPE_READ = 0,
>>>>> +	XE_GT_PAGEFAULT_ACCESS_TYPE_WRITE = 1,
>>>>> +	XE_GT_PAGEFAULT_ACCESS_TYPE_ATOMIC = 2,
>>>>> +	XE_GT_PAGEFAULT_ACCESS_TYPE_RESERVED = 3,
>>>>> +};
>>>>> +
>>>>> +/**
>>>>> + * enum xe_gt_pagefault_fault_type - Fault type reported to the xe_gt_pagefault
>>>>> + * struct.  Saved to xe_gt_pagefault at fault_type
>>>>
>>>> ditto
>>>>
>>>>> + */
>>>>> +enum xe_gt_pagefault_fault_type {
>>>>> +	XE_GT_PAGEFAULT_TYPE_NOT_PRESENT = 0,
>>>>> +	XE_GT_PAGEFAULT_TYPE_WRITE_ACCESS_VIOLATION = 1,
>>>>> +	XE_GT_PAGEFAULT_TYPE_ATOMIC_ACCESS_VIOLATION = 2,
>>>>> +};
>>>>> +
>>>>> +#endif
>>>>
>>>>
>>
>>