drm/xe: Introduce the wedged_mode debugfs

So, the wedged mode can be selected per device at runtime,
before the tests or before reproducing the issue.

v2: - s/busted/wedged
    - some locking consistency

v3: - remove mutex
    - toggle guc reset policy on any mode change

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240423221817.1285081-4-rodrigo.vivi@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drivers/gpu/drm/xe/xe_debugfs.c

@@ -12,6 +12,8 @@
 #include "xe_bo.h"
 #include "xe_device.h"
 #include "xe_gt_debugfs.h"
+#include "xe_gt_printk.h"
+#include "xe_guc_ads.h"
 #include "xe_pm.h"
 #include "xe_sriov.h"
 #include "xe_step.h"
@@ -117,6 +119,56 @@ static const struct file_operations forcewake_all_fops = {
 	.release = forcewake_release,
 };
 
+static ssize_t wedged_mode_show(struct file *f, char __user *ubuf,
+				size_t size, loff_t *pos)
+{
+	struct xe_device *xe = file_inode(f)->i_private;
+	char buf[32];
+	int len = 0;
+
+	len = scnprintf(buf, sizeof(buf), "%d\n", xe->wedged.mode);
+
+	return simple_read_from_buffer(ubuf, size, pos, buf, len);
+}
+
+static ssize_t wedged_mode_set(struct file *f, const char __user *ubuf,
+			       size_t size, loff_t *pos)
+{
+	struct xe_device *xe = file_inode(f)->i_private;
+	struct xe_gt *gt;
+	u32 wedged_mode;
+	ssize_t ret;
+	u8 id;
+
+	ret = kstrtouint_from_user(ubuf, size, 0, &wedged_mode);
+	if (ret)
+		return ret;
+
+	if (wedged_mode > 2)
+		return -EINVAL;
+
+	if (xe->wedged.mode == wedged_mode)
+		return 0;
+
+	xe->wedged.mode = wedged_mode;
+
+	for_each_gt(gt, xe, id) {
+		ret = xe_guc_ads_scheduler_policy_toggle_reset(&gt->uc.guc.ads);
+		if (ret) {
+			xe_gt_err(gt, "Failed to update GuC ADS scheduler policy. GuC may still cause engine reset even with wedged_mode=2\n");
+			return -EIO;
+		}
+	}
+
+	return size;
+}
+
+static const struct file_operations wedged_mode_fops = {
+	.owner = THIS_MODULE,
+	.read = wedged_mode_show,
+	.write = wedged_mode_set,
+};
+
 void xe_debugfs_register(struct xe_device *xe)
 {
 	struct ttm_device *bdev = &xe->ttm;
@@ -134,6 +186,9 @@ void xe_debugfs_register(struct xe_device *xe)
 	debugfs_create_file("forcewake_all", 0400, root, xe,
 			    &forcewake_all_fops);
 
+	debugfs_create_file("wedged_mode", 0400, root, xe,
+			    &wedged_mode_fops);
+
 	for (mem_type = XE_PL_VRAM0; mem_type <= XE_PL_VRAM1; ++mem_type) {
 		man = ttm_manager_type(bdev, mem_type);
 

drivers/gpu/drm/xe/xe_device.c

@@ -506,6 +506,8 @@ int xe_device_probe_early(struct xe_device *xe)
 	if (err)
 		return err;
 
+	xe->wedged.mode = xe_modparam.wedged_mode;
+
 	return 0;
 }
 
@@ -769,7 +771,7 @@ u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address)
  * xe_device_declare_wedged - Declare device wedged
  * @xe: xe device instance
  *
- * This is a final state that can only be cleared with a module
+ * This is a final state that can only be cleared with a mudule
  * re-probe (unbind + bind).
  * In this state every IOCTL will be blocked so the GT cannot be used.
  * In general it will be called upon any critical error such as gt reset
@@ -781,10 +783,12 @@ u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address)
  */
 void xe_device_declare_wedged(struct xe_device *xe)
 {
-	if (xe_modparam.wedged_mode == 0)
+	if (xe->wedged.mode == 0) {
+		drm_dbg(&xe->drm, "Wedged mode is forcebly disabled\n");
 		return;
+	}
 
-	if (!atomic_xchg(&xe->wedged, 1)) {
+	if (!atomic_xchg(&xe->wedged.flag, 1)) {
 		xe->needs_flr_on_fini = true;
 		drm_err(&xe->drm,
 			"CRITICAL: Xe has declared device %s as wedged.\n"

drivers/gpu/drm/xe/xe_device.h

@@ -169,7 +169,7 @@ u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address);
 
 static inline bool xe_device_wedged(struct xe_device *xe)
 {
-	return atomic_read(&xe->wedged);
+	return atomic_read(&xe->wedged.flag);
 }
 
 void xe_device_declare_wedged(struct xe_device *xe);

drivers/gpu/drm/xe/xe_device_types.h

@@ -459,8 +459,13 @@ struct xe_device {
 	/** @needs_flr_on_fini: requests function-reset on fini */
 	bool needs_flr_on_fini;
 
-	/** @wedged: Xe device faced a critical error and is now blocked. */
-	atomic_t wedged;
+	/** @wedged: Struct to control Wedged States and mode */
+	struct {
+		/** @wedged.flag: Xe device faced a critical error and is now blocked. */
+		atomic_t flag;
+		/** @wedged.mode: Mode controlled by kernel parameter and debugfs */
+		int mode;
+	} wedged;
 
 	/* private: */
 

drivers/gpu/drm/xe/xe_guc_ads.c

@@ -9,6 +9,7 @@
 
 #include <generated/xe_wa_oob.h>
 
+#include "abi/guc_actions_abi.h"
 #include "regs/xe_engine_regs.h"
 #include "regs/xe_gt_regs.h"
 #include "regs/xe_guc_regs.h"
@@ -16,11 +17,11 @@
 #include "xe_gt.h"
 #include "xe_gt_ccs_mode.h"
 #include "xe_guc.h"
+#include "xe_guc_ct.h"
 #include "xe_hw_engine.h"
 #include "xe_lrc.h"
 #include "xe_map.h"
 #include "xe_mmio.h"
-#include "xe_module.h"
 #include "xe_platform_types.h"
 #include "xe_wa.h"
 
@@ -441,6 +442,7 @@ int xe_guc_ads_init_post_hwconfig(struct xe_guc_ads *ads)
 
 static void guc_policies_init(struct xe_guc_ads *ads)
 {
+	struct xe_device *xe = ads_to_xe(ads);
 	u32 global_flags = 0;
 
 	ads_blob_write(ads, policies.dpc_promote_time,
@@ -448,7 +450,7 @@ static void guc_policies_init(struct xe_guc_ads *ads)
 	ads_blob_write(ads, policies.max_num_work_items,
 		       GLOBAL_POLICY_MAX_NUM_WI);
 
-	if (xe_modparam.wedged_mode == 2)
+	if (xe->wedged.mode == 2)
 		global_flags |= GLOBAL_POLICY_DISABLE_ENGINE_RESET;
 
 	ads_blob_write(ads, policies.global_flags, global_flags);
@@ -806,3 +808,57 @@ void xe_guc_ads_populate_post_load(struct xe_guc_ads *ads)
 {
 	guc_populate_golden_lrc(ads);
 }
+
+static int guc_ads_action_update_policies(struct xe_guc_ads *ads, u32 policy_offset)
+{
+	struct  xe_guc_ct *ct = &ads_to_guc(ads)->ct;
+	u32 action[] = {
+		XE_GUC_ACTION_GLOBAL_SCHED_POLICY_CHANGE,
+		policy_offset
+	};
+
+	return xe_guc_ct_send(ct, action, ARRAY_SIZE(action), 0, 0);
+}
+
+/**
+ * xe_guc_ads_scheduler_policy_toggle_reset - Toggle reset policy
+ * @ads: Additional data structures object
+ *
+ * This function update the GuC's engine reset policy based on wedged.mode.
+ *
+ * Return: 0 on success, and negative error code otherwise.
+ */
+int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads)
+{
+	struct xe_device *xe = ads_to_xe(ads);
+	struct xe_gt *gt = ads_to_gt(ads);
+	struct xe_tile *tile = gt_to_tile(gt);
+	struct guc_policies *policies;
+	struct xe_bo *bo;
+	int ret = 0;
+
+	policies = kmalloc(sizeof(*policies), GFP_KERNEL);
+	if (!policies)
+		return -ENOMEM;
+
+	policies->dpc_promote_time = ads_blob_read(ads, policies.dpc_promote_time);
+	policies->max_num_work_items = ads_blob_read(ads, policies.max_num_work_items);
+	policies->is_valid = 1;
+	if (xe->wedged.mode == 2)
+		policies->global_flags |= GLOBAL_POLICY_DISABLE_ENGINE_RESET;
+	else
+		policies->global_flags &= ~GLOBAL_POLICY_DISABLE_ENGINE_RESET;
+
+	bo = xe_managed_bo_create_from_data(xe, tile, policies, sizeof(struct guc_policies),
+					    XE_BO_FLAG_VRAM_IF_DGFX(tile) |
+					    XE_BO_FLAG_GGTT);
+	if (IS_ERR(bo)) {
+		ret = PTR_ERR(bo);
+		goto out;
+	}
+
+	ret = guc_ads_action_update_policies(ads, xe_bo_ggtt_addr(bo));
+out:
+	kfree(policies);
+	return ret;
+}

drivers/gpu/drm/xe/xe_guc_ads.h

@@ -13,5 +13,6 @@ int xe_guc_ads_init_post_hwconfig(struct xe_guc_ads *ads);
 void xe_guc_ads_populate(struct xe_guc_ads *ads);
 void xe_guc_ads_populate_minimal(struct xe_guc_ads *ads);
 void xe_guc_ads_populate_post_load(struct xe_guc_ads *ads);
+int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads);
 
 #endif

drivers/gpu/drm/xe/xe_guc_submit.c

@@ -35,7 +35,6 @@
 #include "xe_macros.h"
 #include "xe_map.h"
 #include "xe_mocs.h"
-#include "xe_module.h"
 #include "xe_ring_ops_types.h"
 #include "xe_sched_job.h"
 #include "xe_trace.h"
@@ -868,26 +867,38 @@ static void xe_guc_exec_queue_trigger_cleanup(struct xe_exec_queue *q)
 		xe_sched_tdr_queue_imm(&q->guc->sched);
 }
 
-static void guc_submit_wedged(struct xe_guc *guc)
+static bool guc_submit_hint_wedged(struct xe_guc *guc)
 {
+	struct xe_device *xe = guc_to_xe(guc);
 	struct xe_exec_queue *q;
 	unsigned long index;
 	int err;
 
-	xe_device_declare_wedged(guc_to_xe(guc));
+	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);
 
 	err = drmm_add_action_or_reset(&guc_to_xe(guc)->drm,
 				       guc_submit_wedged_fini, guc);
-	if (err)
-		return;
+	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 */
+	}
 
 	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);
+
+	return true;
 }
 
 static void xe_guc_exec_queue_lr_cleanup(struct work_struct *w)
@@ -898,15 +909,12 @@ static void xe_guc_exec_queue_lr_cleanup(struct work_struct *w)
 	struct xe_guc *guc = exec_queue_to_guc(q);
 	struct xe_device *xe = guc_to_xe(guc);
 	struct xe_gpu_scheduler *sched = &ge->sched;
-	bool wedged = xe_device_wedged(xe);
+	bool wedged;
 
 	xe_assert(xe, xe_exec_queue_is_lr(q));
 	trace_xe_exec_queue_lr_cleanup(q);
 
-	if (!wedged && xe_modparam.wedged_mode == 2) {
-		guc_submit_wedged(exec_queue_to_guc(q));
-		wedged = true;
-	}
+	wedged = guc_submit_hint_wedged(exec_queue_to_guc(q));
 
 	/* Kill the run_job / process_msg entry points */
 	xe_sched_submission_stop(sched);
@@ -957,7 +965,7 @@ guc_exec_queue_timedout_job(struct drm_sched_job *drm_job)
 	struct xe_device *xe = guc_to_xe(exec_queue_to_guc(q));
 	int err = -ETIME;
 	int i = 0;
-	bool wedged = xe_device_wedged(xe);
+	bool wedged;
 
 	/*
 	 * TDR has fired before free job worker. Common if exec queue
@@ -981,10 +989,7 @@ guc_exec_queue_timedout_job(struct drm_sched_job *drm_job)
 
 	trace_xe_sched_job_timedout(job);
 
-	if (!wedged && xe_modparam.wedged_mode == 2) {
-		guc_submit_wedged(exec_queue_to_guc(q));
-		wedged = true;
-	}
+	wedged = guc_submit_hint_wedged(exec_queue_to_guc(q));
 
 	/* Kill the run_job entry point */
 	xe_sched_submission_stop(sched);