Commit ce3d6554 authored by Matthew Brost's avatar Matthew Brost
Browse files

drm/xe: Add GT stats ktime helpers 



Normalize GT stats that record execution periods in code paths by
adding helpers to perform the ktime calculation. Use these helpers in
the SVM code.

Suggested-by: default avatarFrancois Dugast <francois.dugast@intel.com>
Signed-off-by: default avatarMatthew Brost <matthew.brost@intel.com>
Reviewed-by: default avatarThomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/20251212182847.1683222-7-matthew.brost@intel.com
parent 4ac9048d

drivers/gpu/drm/xe/xe_gt_stats.h

+32 −0
Original line number Diff line number Diff line
@@ -6,6 +6,8 @@ 
#ifndef _XE_GT_STATS_H_

#define _XE_GT_STATS_H_



#include <linux/ktime.h>



#include "xe_gt_stats_types.h"



struct xe_gt;
@@ -23,4 +25,34 @@ xe_gt_stats_incr(struct xe_gt *gt, const enum xe_gt_stats_id id, 
}



#endif



/**

 * xe_gt_stats_ktime_us_delta() - Get delta in microseconds between now and a

 * start time

 * @start: Start time

 *

 * Helper for GT stats to get delta in microseconds between now and a start

 * time, compiles out if GT stats are disabled.

 *

 * Return: Delta in microseconds between now and a start time

 */

static inline s64 xe_gt_stats_ktime_us_delta(ktime_t start)

{

	return IS_ENABLED(CONFIG_DEBUG_FS) ?

		ktime_us_delta(ktime_get(), start) : 0;

}



/**

 * xe_gt_stats_ktime_get() - Get current ktime

 *

 * Helper for GT stats to get current ktime, compiles out if GT stats are

 * disabled.

 *

 * Return: Get current ktime

 */

static inline ktime_t xe_gt_stats_ktime_get(void)

{

	return IS_ENABLED(CONFIG_DEBUG_FS) ? ktime_get() : 0;

}



#endif

drivers/gpu/drm/xe/xe_svm.c

+9 −20
Original line number Diff line number Diff line
@@ -176,24 +176,13 @@ xe_svm_range_notifier_event_end(struct xe_vm *vm, struct drm_gpusvm_range *r, 
						   mmu_range);

}



static s64 xe_svm_stats_ktime_us_delta(ktime_t start)

{

	return IS_ENABLED(CONFIG_DEBUG_FS) ?

		ktime_us_delta(ktime_get(), start) : 0;

}



static void xe_svm_tlb_inval_us_stats_incr(struct xe_gt *gt, ktime_t start)

{

	s64 us_delta = xe_svm_stats_ktime_us_delta(start);

	s64 us_delta = xe_gt_stats_ktime_us_delta(start);



	xe_gt_stats_incr(gt, XE_GT_STATS_ID_SVM_TLB_INVAL_US, us_delta);

}



static ktime_t xe_svm_stats_ktime_get(void)

{

	return IS_ENABLED(CONFIG_DEBUG_FS) ? ktime_get() : 0;

}



static void xe_svm_invalidate(struct drm_gpusvm *gpusvm,

			      struct drm_gpusvm_notifier *notifier,

			      const struct mmu_notifier_range *mmu_range)
@@ -202,7 +191,7 @@ static void xe_svm_invalidate(struct drm_gpusvm *gpusvm, 
	struct xe_device *xe = vm->xe;

	struct drm_gpusvm_range *r, *first;

	struct xe_tile *tile;

	ktime_t start = xe_svm_stats_ktime_get();

	ktime_t start = xe_gt_stats_ktime_get();

	u64 adj_start = mmu_range->start, adj_end = mmu_range->end;

	u8 tile_mask = 0, id;

	long err;
@@ -442,7 +431,7 @@ static void xe_svm_copy_us_stats_incr(struct xe_gt *gt, 
				      unsigned long npages,

				      ktime_t start)

{

	s64 us_delta = xe_svm_stats_ktime_us_delta(start);

	s64 us_delta = xe_gt_stats_ktime_us_delta(start);



	if (dir == XE_SVM_COPY_TO_VRAM) {

		switch (npages) {
@@ -494,7 +483,7 @@ static int xe_svm_copy(struct page **pages, 
	u64 vram_addr = XE_VRAM_ADDR_INVALID;

	int err = 0, pos = 0;

	bool sram = dir == XE_SVM_COPY_TO_SRAM;

	ktime_t start = xe_svm_stats_ktime_get();

	ktime_t start = xe_gt_stats_ktime_get();



	/*

	 * This flow is complex: it locates physically contiguous device pages,
@@ -986,7 +975,7 @@ static void xe_svm_range_##elem##_us_stats_incr(struct xe_gt *gt, \ 
						struct xe_svm_range *range, \

						ktime_t start) \

{ \

	s64 us_delta = xe_svm_stats_ktime_us_delta(start); \

	s64 us_delta = xe_gt_stats_ktime_us_delta(start); \

\

	switch (xe_svm_range_size(range)) { \

	case SZ_4K: \
@@ -1031,7 +1020,7 @@ static int __xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma, 
	struct drm_pagemap *dpagemap;

	struct xe_tile *tile = gt_to_tile(gt);

	int migrate_try_count = ctx.devmem_only ? 3 : 1;

	ktime_t start = xe_svm_stats_ktime_get(), bind_start, get_pages_start;

	ktime_t start = xe_gt_stats_ktime_get(), bind_start, get_pages_start;

	int err;



	lockdep_assert_held_write(&vm->lock);
@@ -1070,7 +1059,7 @@ static int __xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma, 


	if (--migrate_try_count >= 0 &&

	    xe_svm_range_needs_migrate_to_vram(range, vma, !!dpagemap || ctx.devmem_only)) {

		ktime_t migrate_start = xe_svm_stats_ktime_get();

		ktime_t migrate_start = xe_gt_stats_ktime_get();



		/* TODO : For multi-device dpagemap will be used to find the

		 * remote tile and remote device. Will need to modify
@@ -1107,7 +1096,7 @@ static int __xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma, 
	}



get_pages:

	get_pages_start = xe_svm_stats_ktime_get();

	get_pages_start = xe_gt_stats_ktime_get();



	range_debug(range, "GET PAGES");

	err = xe_svm_range_get_pages(vm, range, &ctx);
@@ -1134,7 +1123,7 @@ static int __xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma, 
	xe_svm_range_get_pages_us_stats_incr(gt, range, get_pages_start);

	range_debug(range, "PAGE FAULT - BIND");



	bind_start = xe_svm_stats_ktime_get();

	bind_start = xe_gt_stats_ktime_get();

	xe_validation_guard(&vctx, &vm->xe->val, &exec, (struct xe_val_flags) {}, err) {

		err = xe_vm_drm_exec_lock(vm, &exec);

		drm_exec_retry_on_contention(&exec);