Merge tag 'drm-xe-next-fixes-2024-07-18' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-next

- Xe_exec ioctl minor fix on sync entry cleanup upon error (Ashutosh)
- SRIOV: limit VF LMEM provisioning (Michal)
- Wedge mode fixes (Brost)

Signed-off-by: Dave Airlie <[email protected]>

From: Rodrigo Vivi <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

Author: airlied

drivers/gpu/drm/xe/xe_device.c

@@ -854,6 +854,13 @@ u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address)
 	return address & GENMASK_ULL(xe->info.va_bits - 1, 0);
 }
 
+static void xe_device_wedged_fini(struct drm_device *drm, void *arg)
+{
+	struct xe_device *xe = arg;
+
+	xe_pm_runtime_put(xe);
+}
+
 /**
  * xe_device_declare_wedged - Declare device wedged
  * @xe: xe device instance
@@ -870,11 +877,21 @@ u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address)
  */
 void xe_device_declare_wedged(struct xe_device *xe)
 {
+	struct xe_gt *gt;
+	u8 id;
+
 	if (xe->wedged.mode == 0) {
 		drm_dbg(&xe->drm, "Wedged mode is forcibly disabled\n");
 		return;
 	}
 
+	if (drmm_add_action_or_reset(&xe->drm, xe_device_wedged_fini, xe)) {
+		drm_err(&xe->drm, "Failed to register xe_device_wedged_fini clean-up. Although device is wedged.\n");
+		return;
+	}
+
+	xe_pm_runtime_get_noresume(xe);
+
 	if (!atomic_xchg(&xe->wedged.flag, 1)) {
 		xe->needs_flr_on_fini = true;
 		drm_err(&xe->drm,
@@ -883,4 +900,7 @@ void xe_device_declare_wedged(struct xe_device *xe)
 			"Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/xe/kernel/issues/new\n",
 			dev_name(xe->drm.dev));
 	}
+
+	for_each_gt(gt, xe, id)
+		xe_gt_declare_wedged(gt);
 }

drivers/gpu/drm/xe/xe_exec.c

@@ -118,7 +118,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	u64 addresses[XE_HW_ENGINE_MAX_INSTANCE];
 	struct drm_gpuvm_exec vm_exec = {.extra.fn = xe_exec_fn};
 	struct drm_exec *exec = &vm_exec.exec;
-	u32 i, num_syncs = 0, num_ufence = 0;
+	u32 i, num_syncs, num_ufence = 0;
 	struct xe_sched_job *job;
 	struct xe_vm *vm;
 	bool write_locked, skip_retry = false;
@@ -156,15 +156,15 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 
 	vm = q->vm;
 
-	for (i = 0; i < args->num_syncs; i++) {
-		err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs++],
-					  &syncs_user[i], SYNC_PARSE_FLAG_EXEC |
+	for (num_syncs = 0; num_syncs < args->num_syncs; num_syncs++) {
+		err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs],
+					  &syncs_user[num_syncs], SYNC_PARSE_FLAG_EXEC |
 					  (xe_vm_in_lr_mode(vm) ?
 					   SYNC_PARSE_FLAG_LR_MODE : 0));
 		if (err)
 			goto err_syncs;
 
-		if (xe_sync_is_ufence(&syncs[i]))
+		if (xe_sync_is_ufence(&syncs[num_syncs]))
 			num_ufence++;
 	}
 
@@ -325,8 +325,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	if (err == -EAGAIN && !skip_retry)
 		goto retry;
 err_syncs:
-	for (i = 0; i < num_syncs; i++)
-		xe_sync_entry_cleanup(&syncs[i]);
+	while (num_syncs--)
+		xe_sync_entry_cleanup(&syncs[num_syncs]);
 	kfree(syncs);
 err_exec_queue:
 	xe_exec_queue_put(q);

drivers/gpu/drm/xe/xe_gt.c

@@ -904,3 +904,18 @@ struct xe_hw_engine *xe_gt_any_hw_engine(struct xe_gt *gt)
 
 	return NULL;
 }
+
+/**
+ * xe_gt_declare_wedged() - Declare GT wedged
+ * @gt: the GT object
+ *
+ * Wedge the GT which stops all submission, saves desired debug state, and
+ * cleans up anything which could timeout.
+ */
+void xe_gt_declare_wedged(struct xe_gt *gt)
+{
+	xe_gt_assert(gt, gt_to_xe(gt)->wedged.mode);
+
+	xe_uc_declare_wedged(&gt->uc);
+	xe_gt_tlb_invalidation_reset(gt);
+}

drivers/gpu/drm/xe/xe_gt.h

@@ -37,6 +37,7 @@ struct xe_gt *xe_gt_alloc(struct xe_tile *tile);
 int xe_gt_init_hwconfig(struct xe_gt *gt);
 int xe_gt_init_early(struct xe_gt *gt);
 int xe_gt_init(struct xe_gt *gt);
+void xe_gt_declare_wedged(struct xe_gt *gt);
 int xe_gt_record_default_lrcs(struct xe_gt *gt);
 
 /**

drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c

@@ -1543,6 +1543,7 @@ static u64 pf_estimate_fair_lmem(struct xe_gt *gt, unsigned int num_vfs)
 	u64 fair;
 
 	fair = div_u64(available, num_vfs);
+	fair = rounddown_pow_of_two(fair);	/* XXX: ttm_vram_mgr & drm_buddy limitation */
 	fair = ALIGN_DOWN(fair, alignment);
 #ifdef MAX_FAIR_LMEM
 	fair = min_t(u64, MAX_FAIR_LMEM, fair);

drivers/gpu/drm/xe/xe_guc.c

@@ -1178,3 +1178,19 @@ void xe_guc_print_info(struct xe_guc *guc, struct drm_printer *p)
 	xe_guc_ct_print(&guc->ct, p, false);
 	xe_guc_submit_print(guc, p);
 }
+
+/**
+ * xe_guc_declare_wedged() - Declare GuC wedged
+ * @guc: the GuC object
+ *
+ * Wedge the GuC which stops all submission, saves desired debug state, and
+ * cleans up anything which could timeout.
+ */
+void xe_guc_declare_wedged(struct xe_guc *guc)
+{
+	xe_gt_assert(guc_to_gt(guc), guc_to_xe(guc)->wedged.mode);
+
+	xe_guc_reset_prepare(guc);
+	xe_guc_ct_stop(&guc->ct);
+	xe_guc_submit_wedge(guc);
+}

drivers/gpu/drm/xe/xe_guc.h

@@ -37,6 +37,7 @@ void xe_guc_reset_wait(struct xe_guc *guc);
 void xe_guc_stop_prepare(struct xe_guc *guc);
 void xe_guc_stop(struct xe_guc *guc);
 int xe_guc_start(struct xe_guc *guc);
+void xe_guc_declare_wedged(struct xe_guc *guc);
 
 static inline u16 xe_engine_class_to_guc_class(enum xe_engine_class class)
 {

drivers/gpu/drm/xe/xe_guc_submit.c

@@ -861,36 +861,47 @@ static void xe_guc_exec_queue_trigger_cleanup(struct xe_exec_queue *q)
 		xe_sched_tdr_queue_imm(&q->guc->sched);
 }
 
-static bool guc_submit_hint_wedged(struct xe_guc *guc)
+/**
+ * xe_guc_submit_wedge() - Wedge GuC submission
+ * @guc: the GuC object
+ *
+ * Save exec queue's registered with GuC state by taking a ref to each queue.
+ * Register a DRMM handler to drop refs upon driver unload.
+ */
+void xe_guc_submit_wedge(struct xe_guc *guc)
 {
 	struct xe_device *xe = guc_to_xe(guc);
 	struct xe_exec_queue *q;
 	unsigned long index;
 	int err;
 
-	if (xe->wedged.mode != 2)
-		return false;
-
-	if (xe_device_wedged(xe))
-		return true;
-
-	xe_device_declare_wedged(xe);
-
-	xe_guc_submit_reset_prepare(guc);
-	xe_guc_ct_stop(&guc->ct);
+	xe_gt_assert(guc_to_gt(guc), guc_to_xe(guc)->wedged.mode);
 
 	err = drmm_add_action_or_reset(&guc_to_xe(guc)->drm,
 				       guc_submit_wedged_fini, guc);
 	if (err) {
 		drm_err(&xe->drm, "Failed to register xe_guc_submit clean-up on wedged.mode=2. Although device is wedged.\n");
-		return true; /* Device is wedged anyway */
+		return;
 	}
 
 	mutex_lock(&guc->submission_state.lock);
 	xa_for_each(&guc->submission_state.exec_queue_lookup, index, q)
 		if (xe_exec_queue_get_unless_zero(q))
 			set_exec_queue_wedged(q);
 	mutex_unlock(&guc->submission_state.lock);
+}
+
+static bool guc_submit_hint_wedged(struct xe_guc *guc)
+{
+	struct xe_device *xe = guc_to_xe(guc);
+
+	if (xe->wedged.mode != 2)
+		return false;
+
+	if (xe_device_wedged(xe))
+		return true;
+
+	xe_device_declare_wedged(xe);
 
 	return true;
 }
@@ -1677,7 +1688,8 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc)
 
 void xe_guc_submit_reset_wait(struct xe_guc *guc)
 {
-	wait_event(guc->ct.wq, !guc_read_stopped(guc));
+	wait_event(guc->ct.wq, xe_device_wedged(guc_to_xe(guc)) ||
+		   !guc_read_stopped(guc));
 }
 
 void xe_guc_submit_stop(struct xe_guc *guc)

drivers/gpu/drm/xe/xe_guc_submit.h

@@ -18,6 +18,7 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc);
 void xe_guc_submit_reset_wait(struct xe_guc *guc);
 void xe_guc_submit_stop(struct xe_guc *guc);
 int xe_guc_submit_start(struct xe_guc *guc);
+void xe_guc_submit_wedge(struct xe_guc *guc);
 
 int xe_guc_sched_done_handler(struct xe_guc *guc, u32 *msg, u32 len);
 int xe_guc_deregister_done_handler(struct xe_guc *guc, u32 *msg, u32 len);

drivers/gpu/drm/xe/xe_uc.c

@@ -300,3 +300,17 @@ void xe_uc_remove(struct xe_uc *uc)
 {
 	xe_gsc_remove(&uc->gsc);
 }
+
+/**
+ * xe_uc_declare_wedged() - Declare UC wedged
+ * @uc: the UC object
+ *
+ * Wedge the UC which stops all submission, saves desired debug state, and
+ * cleans up anything which could timeout.
+ */
+void xe_uc_declare_wedged(struct xe_uc *uc)
+{
+	xe_gt_assert(uc_to_gt(uc), uc_to_xe(uc)->wedged.mode);
+
+	xe_guc_declare_wedged(&uc->guc);
+}