Re: [Intel-xe] [PATCH] drm/xe: Introduce xe_ASSERT macros

[Jani Nikula <jani.nikula@intel.com>]

On Tue, 08 Aug 2023, Michal Wajdeczko <michal.wajdeczko@intel.com> wrote:
> On 08.08.2023 09:45, Jani Nikula wrote:
>> On Mon, 07 Aug 2023, Michal Wajdeczko <michal.wajdeczko@intel.com> wrote:
>>> As we are moving away from the controversial XE_BUG_ON macro,
>>> relying just on WARN_ON or drm_err does not cover the cases
>>> where we want to annotate functions with additional detailed
>>> debug checks to assert that all prerequisites are satisfied,
>>> without paying footprint or performance penalty on non-debug
>>> builds, where all misuses introduced during code integration
>>> were already fixed.
>>>
>>> Introduce family of xe_ASSERT macros that try to follow classic
>>> assert() utility and can be compiled out on non-debug builds.
>>>
>>> Macros are based on drm_WARN, but unlikely to origin, disallow
>>> use in expressions since we will compile that code out.
>> 
>> Final bikeshedding... thought of the inconsistency between:
>> 
>>  - drm_WARN_ON <-> xe_ASSERT
>>  - drm_WARN <-> xe_ASSERT_MSG
>
> but use of _MSG suffix is consistent with:
>
>  - BUILD_BUG_ON(cond)         <-> xe_ASSERT(cond)
>  - BUILD_BUG_ON_MSG(cond,msg) <-> xe_ASSERT_MSG(cond,msg)
>
>> 
>> and wondering if xe_ASSERT_ON and xe_ASSERT would be more in line.
>
> note that unlike WARN, ASSERT warns if the condition is false, thus
> using _ON suffix could be misleading IMO

Agreed on both, please disregard my comment.

Thanks,
Jani.


>
>> 
>> *shrug*
>> 
>> Either way, this looks fine.
>> 
>> BR,
>> Jani.
>> 
>> 
>> 
>> 
>>>
>>> As we are operating on the xe pointers, we can print additional
>>> information about the device, like tile or GT identifier, that
>>> is not available from generic WARN report:
>>>
>>> [ ] xe 0000:00:02.0: [drm] Assertion `true == false` failed!
>>>     platform: 1 subplatform: 1
>>>     graphics: Xe_LP 12.0 step B0
>>>     media: Xe_M 12.0 step B0
>>>     display: enabled step D0
>>>     tile: 0 VRAM 0 B
>>>     GT: 0 type 1
>>>
>>> [ ] xe 0000:b3:00.0: [drm] Assertion `true == false` failed!
>>>     platform: 7 subplatform: 3
>>>     graphics: Xe_HPG 12.55 step A1
>>>     media: Xe_HPM 12.55 step A1
>>>     display: disabled step **
>>>     tile: 0 VRAM 14.0 GiB
>>>     GT: 0 type 1
>>>
>>> [ ] WARNING: CPU: 0 PID: 2687 at drivers/gpu/drm/xe/xe_device.c:281 xe_device_probe+0x374/0x520 [xe]
>>> [ ] RIP: 0010:xe_device_probe+0x374/0x520 [xe]
>>> [ ] Call Trace:
>>> [ ]  ? __warn+0x7b/0x160
>>> [ ]  ? xe_device_probe+0x374/0x520 [xe]
>>> [ ]  ? report_bug+0x1c3/0x1d0
>>> [ ]  ? handle_bug+0x42/0x70
>>> [ ]  ? exc_invalid_op+0x14/0x70
>>> [ ]  ? asm_exc_invalid_op+0x16/0x20
>>> [ ]  ? xe_device_probe+0x374/0x520 [xe]
>>> [ ]  ? xe_device_probe+0x374/0x520 [xe]
>>> [ ]  xe_pci_probe+0x6e3/0x950 [xe]
>>> [ ]  ? lockdep_hardirqs_on+0xc7/0x140
>>> [ ]  pci_device_probe+0x9e/0x160
>>> [ ]  really_probe+0x19d/0x400
>>>
>>> Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
>>> Cc: Oded Gabbay <ogabbay@kernel.org>
>>> Cc: Jani Nikula <jani.nikula@intel.com>
>>> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
>>> Cc: Matthew Brost <matthew.brost@intel.com>
>>> ---
>>>  drivers/gpu/drm/xe/xe_assert.h | 160 +++++++++++++++++++++++++++++++++
>>>  1 file changed, 160 insertions(+)
>>>  create mode 100644 drivers/gpu/drm/xe/xe_assert.h
>>>
>>> diff --git a/drivers/gpu/drm/xe/xe_assert.h b/drivers/gpu/drm/xe/xe_assert.h
>>> new file mode 100644
>>> index 000000000000..7ea295b7091c
>>> --- /dev/null
>>> +++ b/drivers/gpu/drm/xe/xe_assert.h
>>> @@ -0,0 +1,160 @@
>>> +/* SPDX-License-Identifier: MIT */
>>> +/*
>>> + * Copyright © 2023 Intel Corporation
>>> + */
>>> +
>>> +#ifndef __XE_ASSERT_H__
>>> +#define __XE_ASSERT_H__
>>> +
>>> +#include <linux/string_helpers.h>
>>> +#include <drm/drm_print.h>
>>> +#include "xe_device_types.h"
>>> +#include "xe_step.h"
>>> +
>>> +/**
>>> + * DOC: Xe ASSERTs
>>> + *
>>> + * While Xe driver aims to be simpler than legacy i915 driver it is still
>>> + * complex enough that some changes introduced while adding new functionality
>>> + * could break the existing code.
>>> + *
>>> + * Adding &drm_WARN or &drm_err to catch unwanted programming usage could lead
>>> + * to undesired increased driver footprint and may impact production driver
>>> + * performance as this additional code will be always present.
>>> + *
>>> + * To allow annotate functions with additional detailed debug checks to assert
>>> + * that all prerequisites are satisfied, without worrying about footprint or
>>> + * performance penalty on production builds where all potential misuses
>>> + * introduced during code integration were already fixed, we introduce family
>>> + * of ASSERT macros that try to follow classic assert() utility and can be
>>> + * compiled out on non-debug builds:
>>> + *
>>> + *  * &xe_ASSERT
>>> + *  * &xe_tile_ASSERT
>>> + *  * &xe_gt_ASSERT
>>> + *
>>> + * These macros are based on &drm_WARN, but unlikely to the origin, we disallow
>>> + * use of them in an expressions since we will compile that code out.
>>> + *
>>> + * Note that these macros shall not be used to cover known gaps in the
>>> + * implementation, for such cases use &drm_WARN or &drm_err and provide valid
>>> + * safe fallback.
>>> + *
>>> + * Below code shows how asserts could help in debug to catch unplanned use::
>>> + *
>>> + *	static void one_igfx(struct xe_device *xe)
>>> + *	{
>>> + *		xe_ASSERT(xe, xe->info.is_dgfx == false);
>>> + *		xe_ASSERT(xe, xe->info.tile_count == 1);
>>> + *	}
>>> + *
>>> + *	static void two_dgfx(struct xe_device *xe)
>>> + *	{
>>> + *		xe_ASSERT(xe, xe->info.is_dgfx);
>>> + *		xe_ASSERT(xe, xe->info.tile_count == 2);
>>> + *	}
>>> + *
>>> + *	void foo(struct xe_device *xe)
>>> + *	{
>>> + *		if (xe->info.dgfx)
>>> + *			return two_dgfx(xe);
>>> + *		return one_igfx(xe);
>>> + *	}
>>> + */
>>> +
>>> +#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
>>> +#define __XE_ASSERT_MSG(xe, condition, msg, arg...) ({						\
>>> +	(void)drm_WARN(&(xe)->drm, !(condition), "[" DRM_NAME "] Assertion `%s` failed!\n" msg,	\
>>> +		       __stringify(condition), ## arg);						\
>>> +})
>>> +#else
>>> +#define __XE_ASSERT_MSG(xe, condition, msg, arg...) ({						\
>>> +	typecheck(struct xe_device *, xe);							\
>>> +	BUILD_BUG_ON_INVALID(condition);							\
>>> +})
>>> +#endif
>>> +
>>> +/**
>>> + * xe_ASSERT - warn if condition is false when debugging.
>>> + * @xe: the &struct xe_device pointer to which &condition applies
>>> + * @condition: condition to check
>>> + *
>>> + * xe_ASSERT() uses &drm_WARN to emit a warning and print additional information
>>> + * that could be read from the &xe pointer if provided &condition is false.
>>> + *
>>> + * Contrary to &drm_WARN, xe_ASSERT() is effective only on debug builds
>>> + * (&CONFIG_DRM_XE_DEBUG must be enabled) and cannot be used in expressions
>>> + * or as a condition.
>>> + *
>>> + * See `Xe ASSERTs`_ for general usage guidelines.
>>> + */
>>> +#define xe_ASSERT(xe, condition) xe_ASSERT_MSG((xe), condition, "")
>>> +#define xe_ASSERT_MSG(xe, condition, msg, arg...) ({						\
>>> +	struct xe_device *__xe = (xe);								\
>>> +	__XE_ASSERT_MSG(__xe, condition,							\
>>> +			"platform: %d subplatform: %d\n"					\
>>> +			"graphics: %s %u.%u step %s\n"						\
>>> +			"media: %s %u.%u step %s\n"						\
>>> +			"display: %s step %s\n"							\
>>> +			msg,									\
>>> +			__xe->info.platform, __xe->info.subplatform,				\
>>> +			__xe->info.graphics_name,						\
>>> +			__xe->info.graphics_verx100 / 100,					\
>>> +			__xe->info.graphics_verx100 % 100,					\
>>> +			xe_step_name(__xe->info.step.graphics),					\
>>> +			__xe->info.media_name,							\
>>> +			__xe->info.media_verx100 / 100,						\
>>> +			__xe->info.media_verx100 % 100,						\
>>> +			xe_step_name(__xe->info.step.media),					\
>>> +			str_enabled_disabled(__xe->info.enable_display),			\
>>> +			xe_step_name(__xe->info.step.display),					\
>>> +			## arg);								\
>>> +})
>>> +
>>> +/**
>>> + * xe_tile_ASSERT - warn if condition is false when debugging.
>>> + * @tile: the &struct xe_tile pointer to which &condition applies
>>> + * @condition: condition to check
>>> + *
>>> + * xe_tile_ASSERT() uses &drm_WARN to emit a warning and print additional
>>> + * information that could be read from the &tile pointer if provided &condition
>>> + * is false.
>>> + *
>>> + * Contrary to &drm_WARN, xe_tile_ASSERT() is effective only on debug builds
>>> + * (&CONFIG_DRM_XE_DEBUG must be enabled) and cannot be used in expressions
>>> + * or as a condition.
>>> + *
>>> + * See `Xe ASSERTs`_ for general usage guidelines.
>>> + */
>>> +#define xe_tile_ASSERT(tile, condition) xe_tile_ASSERT_MSG((tile), condition, "")
>>> +#define xe_tile_ASSERT_MSG(tile, condition, msg, arg...) ({					\
>>> +	struct xe_tile *__tile = (tile);							\
>>> +	char __buf[10];										\
>>> +	xe_ASSERT_MSG(tile_to_xe(__tile), condition, "tile: %u VRAM %s\n" msg,			\
>>> +		      __tile->id, ({ string_get_size(__tile->mem.vram.actual_physical_size, 1,	\
>>> +				     STRING_UNITS_2, __buf, sizeof(__buf)); __buf; }), ## arg);	\
>>> +})
>>> +
>>> +/**
>>> + * xe_gt_ASSERT - warn if condition is false when debugging.
>>> + * @gt: the &struct xe_gt pointer to which &condition applies
>>> + * @condition: condition to check
>>> + *
>>> + * xe_gt_ASSERT() uses &drm_WARN to emit a warning and print additional
>>> + * information that could be safetely read from the &gt pointer if provided
>>> + * &condition is false.
>>> + *
>>> + * Contrary to &drm_WARN, xe_gt_ASSERT() is effective only on debug builds
>>> + * (&CONFIG_DRM_XE_DEBUG must be enabled) and cannot be used in expressions
>>> + * or as a condition.
>>> + *
>>> + * See `Xe ASSERTs`_ for general usage guidelines.
>>> + */
>>> +#define xe_gt_ASSERT(gt, condition) xe_gt_ASSERT_MSG((gt), condition, "")
>>> +#define xe_gt_ASSERT_MSG(gt, condition, msg, arg...) ({						\
>>> +	struct xe_gt *__gt = (gt);								\
>>> +	xe_tile_ASSERT_MSG(gt_to_tile(__gt), condition, "GT: %u type %d\n" msg,			\
>>> +			   __gt->info.id, __gt->info.type, ## arg);				\
>>> +})
>>> +
>>> +#endif /* __XE_ASSERT_H__ */
>> 

-- 
Jani Nikula, Intel Open Source Graphics Center