Merge tag 'drm-xe-fixes-2026-03-19' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-fixes

Driver Changes:
- A number of teardown fixes (Daniele, Matt Brost, Zhanjun, Ashutosh)
- Skip over non-leaf PTE for PRL generation  (Brian)
- Fix an unitialized variable (Umesh)
- Fix a missing runtime PM reference (Sanjay)

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

From: Thomas Hellstrom <[email protected]>
Link: https://patch.msgid.link/abxj4_dBHYBiSvDG@fedora

Author: airlied

drivers/gpu/drm/xe/xe_ggtt.c

@@ -313,6 +313,8 @@ static void dev_fini_ggtt(void *arg)
 {
 	struct xe_ggtt *ggtt = arg;
 
+	scoped_guard(mutex, &ggtt->lock)
+		ggtt->flags &= ~XE_GGTT_FLAGS_ONLINE;
 	drain_workqueue(ggtt->wq);
 }
 
@@ -377,6 +379,7 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt)
 	if (err)
 		return err;
 
+	ggtt->flags |= XE_GGTT_FLAGS_ONLINE;
 	err = devm_add_action_or_reset(xe->drm.dev, dev_fini_ggtt, ggtt);
 	if (err)
 		return err;
@@ -410,13 +413,10 @@ static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt)
 static void ggtt_node_remove(struct xe_ggtt_node *node)
 {
 	struct xe_ggtt *ggtt = node->ggtt;
-	struct xe_device *xe = tile_to_xe(ggtt->tile);
 	bool bound;
-	int idx;
-
-	bound = drm_dev_enter(&xe->drm, &idx);
 
 	mutex_lock(&ggtt->lock);
+	bound = ggtt->flags & XE_GGTT_FLAGS_ONLINE;
 	if (bound)
 		xe_ggtt_clear(ggtt, node->base.start, node->base.size);
 	drm_mm_remove_node(&node->base);
@@ -429,8 +429,6 @@ static void ggtt_node_remove(struct xe_ggtt_node *node)
 	if (node->invalidate_on_remove)
 		xe_ggtt_invalidate(ggtt);
 
-	drm_dev_exit(idx);
-
 free_node:
 	xe_ggtt_node_fini(node);
 }

drivers/gpu/drm/xe/xe_ggtt_types.h

@@ -28,11 +28,14 @@ struct xe_ggtt {
 	/** @size: Total usable size of this GGTT */
 	u64 size;
 
-#define XE_GGTT_FLAGS_64K BIT(0)
+#define XE_GGTT_FLAGS_64K       BIT(0)
+#define XE_GGTT_FLAGS_ONLINE	BIT(1)
 	/**
 	 * @flags: Flags for this GGTT
 	 * Acceptable flags:
 	 * - %XE_GGTT_FLAGS_64K - if PTE size is 64K. Otherwise, regular is 4K.
+	 * - %XE_GGTT_FLAGS_ONLINE - is GGTT online, protected by ggtt->lock
+	 *   after init
 	 */
 	unsigned int flags;
 	/** @scratch: Internal object allocation used as a scratch page */

drivers/gpu/drm/xe/xe_gt_ccs_mode.c

@@ -12,6 +12,7 @@
 #include "xe_gt_printk.h"
 #include "xe_gt_sysfs.h"
 #include "xe_mmio.h"
+#include "xe_pm.h"
 #include "xe_sriov.h"
 
 static void __xe_gt_apply_ccs_mode(struct xe_gt *gt, u32 num_engines)
@@ -150,6 +151,7 @@ ccs_mode_store(struct device *kdev, struct device_attribute *attr,
 		xe_gt_info(gt, "Setting compute mode to %d\n", num_engines);
 		gt->ccs_mode = num_engines;
 		xe_gt_record_user_engines(gt);
+		guard(xe_pm_runtime)(xe);
 		xe_gt_reset(gt);
 	}
 

drivers/gpu/drm/xe/xe_guc.c

@@ -1124,22 +1124,22 @@ static int guc_wait_ucode(struct xe_guc *guc)
 	struct xe_guc_pc *guc_pc = &gt->uc.guc.pc;
 	u32 before_freq, act_freq, cur_freq;
 	u32 status = 0, tries = 0;
+	int load_result, ret;
 	ktime_t before;
 	u64 delta_ms;
-	int ret;
 
 	before_freq = xe_guc_pc_get_act_freq(guc_pc);
 	before = ktime_get();
 
-	ret = poll_timeout_us(ret = guc_load_done(gt, &status, &tries), ret,
+	ret = poll_timeout_us(load_result = guc_load_done(gt, &status, &tries), load_result,
 			      10 * USEC_PER_MSEC,
 			      GUC_LOAD_TIMEOUT_SEC * USEC_PER_SEC, false);
 
 	delta_ms = ktime_to_ms(ktime_sub(ktime_get(), before));
 	act_freq = xe_guc_pc_get_act_freq(guc_pc);
 	cur_freq = xe_guc_pc_get_cur_freq_fw(guc_pc);
 
-	if (ret) {
+	if (ret || load_result <= 0) {
 		xe_gt_err(gt, "load failed: status = 0x%08X, time = %lldms, freq = %dMHz (req %dMHz)\n",
 			  status, delta_ms, xe_guc_pc_get_act_freq(guc_pc),
 			  xe_guc_pc_get_cur_freq_fw(guc_pc));
@@ -1347,15 +1347,37 @@ int xe_guc_enable_communication(struct xe_guc *guc)
 	return 0;
 }
 
-int xe_guc_suspend(struct xe_guc *guc)
+/**
+ * xe_guc_softreset() - Soft reset GuC
+ * @guc: The GuC object
+ *
+ * Send soft reset command to GuC through mmio send.
+ *
+ * Return: 0 if success, otherwise error code
+ */
+int xe_guc_softreset(struct xe_guc *guc)
 {
-	struct xe_gt *gt = guc_to_gt(guc);
 	u32 action[] = {
 		XE_GUC_ACTION_CLIENT_SOFT_RESET,
 	};
 	int ret;
 
+	if (!xe_uc_fw_is_running(&guc->fw))
+		return 0;
+
 	ret = xe_guc_mmio_send(guc, action, ARRAY_SIZE(action));
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+int xe_guc_suspend(struct xe_guc *guc)
+{
+	struct xe_gt *gt = guc_to_gt(guc);
+	int ret;
+
+	ret = xe_guc_softreset(guc);
 	if (ret) {
 		xe_gt_err(gt, "GuC suspend failed: %pe\n", ERR_PTR(ret));
 		return ret;

drivers/gpu/drm/xe/xe_guc.h

@@ -44,6 +44,7 @@ int xe_guc_opt_in_features_enable(struct xe_guc *guc);
 void xe_guc_runtime_suspend(struct xe_guc *guc);
 void xe_guc_runtime_resume(struct xe_guc *guc);
 int xe_guc_suspend(struct xe_guc *guc);
+int xe_guc_softreset(struct xe_guc *guc);
 void xe_guc_notify(struct xe_guc *guc);
 int xe_guc_auth_huc(struct xe_guc *guc, u32 rsa_addr);
 int xe_guc_mmio_send(struct xe_guc *guc, const u32 *request, u32 len);

drivers/gpu/drm/xe/xe_guc_ct.c

@@ -345,6 +345,7 @@ static void guc_action_disable_ct(void *arg)
 {
 	struct xe_guc_ct *ct = arg;
 
+	xe_guc_ct_stop(ct);
 	guc_ct_change_state(ct, XE_GUC_CT_STATE_DISABLED);
 }
 

drivers/gpu/drm/xe/xe_guc_submit.c

@@ -48,6 +48,8 @@
 
 #define XE_GUC_EXEC_QUEUE_CGP_CONTEXT_ERROR_LEN		6
 
+static int guc_submit_reset_prepare(struct xe_guc *guc);
+
 static struct xe_guc *
 exec_queue_to_guc(struct xe_exec_queue *q)
 {
@@ -239,7 +241,7 @@ static bool exec_queue_killed_or_banned_or_wedged(struct xe_exec_queue *q)
 		 EXEC_QUEUE_STATE_BANNED));
 }
 
-static void guc_submit_fini(struct drm_device *drm, void *arg)
+static void guc_submit_sw_fini(struct drm_device *drm, void *arg)
 {
 	struct xe_guc *guc = arg;
 	struct xe_device *xe = guc_to_xe(guc);
@@ -257,6 +259,19 @@ static void guc_submit_fini(struct drm_device *drm, void *arg)
 	xa_destroy(&guc->submission_state.exec_queue_lookup);
 }
 
+static void guc_submit_fini(void *arg)
+{
+	struct xe_guc *guc = arg;
+
+	/* Forcefully kill any remaining exec queues */
+	xe_guc_ct_stop(&guc->ct);
+	guc_submit_reset_prepare(guc);
+	xe_guc_softreset(guc);
+	xe_guc_submit_stop(guc);
+	xe_uc_fw_sanitize(&guc->fw);
+	xe_guc_submit_pause_abort(guc);
+}
+
 static void guc_submit_wedged_fini(void *arg)
 {
 	struct xe_guc *guc = arg;
@@ -326,7 +341,11 @@ int xe_guc_submit_init(struct xe_guc *guc, unsigned int num_ids)
 
 	guc->submission_state.initialized = true;
 
-	return drmm_add_action_or_reset(&xe->drm, guc_submit_fini, guc);
+	err = drmm_add_action_or_reset(&xe->drm, guc_submit_sw_fini, guc);
+	if (err)
+		return err;
+
+	return devm_add_action_or_reset(xe->drm.dev, guc_submit_fini, guc);
 }
 
 /*
@@ -1252,6 +1271,7 @@ static void disable_scheduling_deregister(struct xe_guc *guc,
  */
 void xe_guc_submit_wedge(struct xe_guc *guc)
 {
+	struct xe_device *xe = guc_to_xe(guc);
 	struct xe_gt *gt = guc_to_gt(guc);
 	struct xe_exec_queue *q;
 	unsigned long index;
@@ -1266,20 +1286,28 @@ void xe_guc_submit_wedge(struct xe_guc *guc)
 	if (!guc->submission_state.initialized)
 		return;
 
-	err = devm_add_action_or_reset(guc_to_xe(guc)->drm.dev,
-				       guc_submit_wedged_fini, guc);
-	if (err) {
-		xe_gt_err(gt, "Failed to register clean-up in wedged.mode=%s; "
-			  "Although device is wedged.\n",
-			  xe_wedged_mode_to_string(XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET));
-		return;
-	}
+	if (xe->wedged.mode == 2) {
+		err = devm_add_action_or_reset(guc_to_xe(guc)->drm.dev,
+					       guc_submit_wedged_fini, guc);
+		if (err) {
+			xe_gt_err(gt, "Failed to register clean-up on wedged.mode=2; "
+				  "Although device is wedged.\n");
+			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);
+		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);
+	} else {
+		/* Forcefully kill any remaining exec queues, signal fences */
+		guc_submit_reset_prepare(guc);
+		xe_guc_submit_stop(guc);
+		xe_guc_softreset(guc);
+		xe_uc_fw_sanitize(&guc->fw);
+		xe_guc_submit_pause_abort(guc);
+	}
 }
 
 static bool guc_submit_hint_wedged(struct xe_guc *guc)
@@ -2230,14 +2258,15 @@ static const struct xe_exec_queue_ops guc_exec_queue_ops = {
 static void guc_exec_queue_stop(struct xe_guc *guc, struct xe_exec_queue *q)
 {
 	struct xe_gpu_scheduler *sched = &q->guc->sched;
+	bool do_destroy = false;
 
 	/* Stop scheduling + flush any DRM scheduler operations */
 	xe_sched_submission_stop(sched);
 
 	/* Clean up lost G2H + reset engine state */
 	if (exec_queue_registered(q)) {
 		if (exec_queue_destroyed(q))
-			__guc_exec_queue_destroy(guc, q);
+			do_destroy = true;
 	}
 	if (q->guc->suspend_pending) {
 		set_exec_queue_suspended(q);
@@ -2273,18 +2302,15 @@ static void guc_exec_queue_stop(struct xe_guc *guc, struct xe_exec_queue *q)
 			xe_guc_exec_queue_trigger_cleanup(q);
 		}
 	}
+
+	if (do_destroy)
+		__guc_exec_queue_destroy(guc, q);
 }
 
-int xe_guc_submit_reset_prepare(struct xe_guc *guc)
+static int guc_submit_reset_prepare(struct xe_guc *guc)
 {
 	int ret;
 
-	if (xe_gt_WARN_ON(guc_to_gt(guc), vf_recovery(guc)))
-		return 0;
-
-	if (!guc->submission_state.initialized)
-		return 0;
-
 	/*
 	 * Using an atomic here rather than submission_state.lock as this
 	 * function can be called while holding the CT lock (engine reset
@@ -2299,6 +2325,17 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc)
 	return ret;
 }
 
+int xe_guc_submit_reset_prepare(struct xe_guc *guc)
+{
+	if (xe_gt_WARN_ON(guc_to_gt(guc), vf_recovery(guc)))
+		return 0;
+
+	if (!guc->submission_state.initialized)
+		return 0;
+
+	return guc_submit_reset_prepare(guc);
+}
+
 void xe_guc_submit_reset_wait(struct xe_guc *guc)
 {
 	wait_event(guc->ct.wq, xe_device_wedged(guc_to_xe(guc)) ||
@@ -2695,8 +2732,7 @@ void xe_guc_submit_pause_abort(struct xe_guc *guc)
 			continue;
 
 		xe_sched_submission_start(sched);
-		if (exec_queue_killed_or_banned_or_wedged(q))
-			xe_guc_exec_queue_trigger_cleanup(q);
+		guc_exec_queue_kill(q);
 	}
 	mutex_unlock(&guc->submission_state.lock);
 }

drivers/gpu/drm/xe/xe_lrc.c

@@ -2413,14 +2413,14 @@ static int get_ctx_timestamp(struct xe_lrc *lrc, u32 engine_id, u64 *reg_ctx_ts)
  * @lrc: Pointer to the lrc.
  *
  * Return latest ctx timestamp. With support for active contexts, the
- * calculation may bb slightly racy, so follow a read-again logic to ensure that
+ * calculation may be slightly racy, so follow a read-again logic to ensure that
  * the context is still active before returning the right timestamp.
  *
  * Returns: New ctx timestamp value
  */
 u64 xe_lrc_timestamp(struct xe_lrc *lrc)
 {
-	u64 lrc_ts, reg_ts, new_ts;
+	u64 lrc_ts, reg_ts, new_ts = lrc->ctx_timestamp;
 	u32 engine_id;
 
 	lrc_ts = xe_lrc_ctx_timestamp(lrc);

drivers/gpu/drm/xe/xe_oa.c

@@ -543,8 +543,7 @@ static ssize_t xe_oa_read(struct file *file, char __user *buf,
 	size_t offset = 0;
 	int ret;
 
-	/* Can't read from disabled streams */
-	if (!stream->enabled || !stream->sample)
+	if (!stream->sample)
 		return -EINVAL;
 
 	if (!(file->f_flags & O_NONBLOCK)) {
@@ -1460,6 +1459,10 @@ static void xe_oa_stream_disable(struct xe_oa_stream *stream)
 
 	if (stream->sample)
 		hrtimer_cancel(&stream->poll_check_timer);
+
+	/* Update stream->oa_buffer.tail to allow any final reports to be read */
+	if (xe_oa_buffer_check_unlocked(stream))
+		wake_up(&stream->poll_wq);
 }
 
 static int xe_oa_enable_preempt_timeslice(struct xe_oa_stream *stream)