drm/i915: Start chopping up the GPU error capture

In the near future, we will want to start a GPU error capture from a new
context, from inside the softirq region of a forced preemption. To do
so requires us to break up the monolithic error capture to provide new
entry points with finer control; in particular focusing on one
engine/gt, and being able to compose an error state from little pieces
of HW capture.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Andi Shyti <andi.shyti@intel.com>
Acked-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200110123059.1348712-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gt/intel_engine.h

@@ -202,7 +202,7 @@ void intel_engine_set_hwsp_writemask(struct intel_engine_cs *engine, u32 mask);
 u64 intel_engine_get_active_head(const struct intel_engine_cs *engine);
 u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine);
 
-void intel_engine_get_instdone(struct intel_engine_cs *engine,
+void intel_engine_get_instdone(const struct intel_engine_cs *engine,
 			       struct intel_instdone *instdone);
 
 void intel_engine_init_execlists(struct intel_engine_cs *engine);

drivers/gpu/drm/i915/gt/intel_engine_cs.c

@@ -914,8 +914,8 @@ const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
 }
 
 static u32
-read_subslice_reg(struct intel_engine_cs *engine, int slice, int subslice,
+read_subslice_reg(const struct intel_engine_cs *engine,
-		  i915_reg_t reg)
+		  int slice, int subslice, i915_reg_t reg)
 {
 	struct drm_i915_private *i915 = engine->i915;
 	struct intel_uncore *uncore = engine->uncore;
@@ -959,7 +959,7 @@ read_subslice_reg(struct intel_engine_cs *engine, int slice, int subslice,
 }
 
 /* NB: please notice the memset */
-void intel_engine_get_instdone(struct intel_engine_cs *engine,
+void intel_engine_get_instdone(const struct intel_engine_cs *engine,
 			       struct intel_instdone *instdone)
 {
 	struct drm_i915_private *i915 = engine->i915;

drivers/gpu/drm/i915/gt/intel_ggtt.c

@@ -495,6 +495,7 @@ static void cleanup_init_ggtt(struct i915_ggtt *ggtt)
 	ggtt_release_guc_top(ggtt);
 	if (drm_mm_node_allocated(&ggtt->error_capture))
 		drm_mm_remove_node(&ggtt->error_capture);
+	mutex_destroy(&ggtt->error_mutex);
 }
 
 static int init_ggtt(struct i915_ggtt *ggtt)
@@ -526,6 +527,7 @@ static int init_ggtt(struct i915_ggtt *ggtt)
 	if (ret)
 		return ret;
 
+	mutex_init(&ggtt->error_mutex);
 	if (ggtt->mappable_end) {
 		/* Reserve a mappable slot for our lockless error capture */
 		ret = drm_mm_insert_node_in_range(&ggtt->vm.mm,
@@ -716,6 +718,7 @@ static void ggtt_cleanup_hw(struct i915_ggtt *ggtt)
 
 	if (drm_mm_node_allocated(&ggtt->error_capture))
 		drm_mm_remove_node(&ggtt->error_capture);
+	mutex_destroy(&ggtt->error_mutex);
 
 	ggtt_release_guc_top(ggtt);
 	intel_vgt_deballoon(ggtt);

drivers/gpu/drm/i915/gt/intel_gtt.h

@@ -345,6 +345,7 @@ struct i915_ggtt {
 	/* Manual runtime pm autosuspend delay for user GGTT mmaps */
 	struct intel_wakeref_auto userfault_wakeref;
 
+	struct mutex error_mutex;
 	struct drm_mm_node error_capture;
 	struct drm_mm_node uc_fw;
 };

drivers/gpu/drm/i915/gt/intel_reset.c

@@ -1230,7 +1230,7 @@ void intel_gt_handle_error(struct intel_gt *gt,
 	engine_mask &= INTEL_INFO(gt->i915)->engine_mask;
 
 	if (flags & I915_ERROR_CAPTURE) {
-		i915_capture_error_state(gt->i915, engine_mask, msg);
+		i915_capture_error_state(gt->i915);
 		intel_gt_clear_error_registers(gt, engine_mask);
 	}
 

drivers/gpu/drm/i915/gt/selftest_hangcheck.c

@@ -1498,7 +1498,7 @@ static int igt_handle_error(void *arg)
 	struct intel_engine_cs *engine = gt->engine[RCS0];
 	struct hang h;
 	struct i915_request *rq;
-	struct i915_gpu_state *error;
+	struct i915_gpu_coredump *error;
 	int err;
 
 	/* Check that we can issue a global GPU and engine reset */

drivers/gpu/drm/i915/i915_debugfs.c

@@ -685,7 +685,7 @@ static int i915_gem_fence_regs_info(struct seq_file *m, void *data)
 static ssize_t gpu_state_read(struct file *file, char __user *ubuf,
 			      size_t count, loff_t *pos)
 {
-	struct i915_gpu_state *error;
+	struct i915_gpu_coredump *error;
 	ssize_t ret;
 	void *buf;
 
@@ -698,7 +698,7 @@ static ssize_t gpu_state_read(struct file *file, char __user *ubuf,
 	if (!buf)
 		return -ENOMEM;
 
-	ret = i915_gpu_state_copy_to_buffer(error, buf, *pos, count);
+	ret = i915_gpu_coredump_copy_to_buffer(error, buf, *pos, count);
 	if (ret <= 0)
 		goto out;
 
@@ -714,19 +714,19 @@ out:
 
 static int gpu_state_release(struct inode *inode, struct file *file)
 {
-	i915_gpu_state_put(file->private_data);
+	i915_gpu_coredump_put(file->private_data);
 	return 0;
 }
 
 static int i915_gpu_info_open(struct inode *inode, struct file *file)
 {
 	struct drm_i915_private *i915 = inode->i_private;
-	struct i915_gpu_state *gpu;
+	struct i915_gpu_coredump *gpu;
 	intel_wakeref_t wakeref;
 
 	gpu = NULL;
 	with_intel_runtime_pm(&i915->runtime_pm, wakeref)
-		gpu = i915_capture_gpu_state(i915);
+		gpu = i915_gpu_coredump(i915);
 	if (IS_ERR(gpu))
 		return PTR_ERR(gpu);
 
@@ -748,7 +748,7 @@ i915_error_state_write(struct file *filp,
 		       size_t cnt,
 		       loff_t *ppos)
 {
-	struct i915_gpu_state *error = filp->private_data;
+	struct i915_gpu_coredump *error = filp->private_data;
 
 	if (!error)
 		return 0;
@@ -761,7 +761,7 @@ i915_error_state_write(struct file *filp,
 
 static int i915_error_state_open(struct inode *inode, struct file *file)
 {
-	struct i915_gpu_state *error;
+	struct i915_gpu_coredump *error;
 
 	error = i915_first_error_state(inode->i_private);
 	if (IS_ERR(error))

drivers/gpu/drm/i915/i915_drv.h

@@ -1874,7 +1874,7 @@ static inline u32 i915_reset_count(struct i915_gpu_error *error)
 }
 
 static inline u32 i915_reset_engine_count(struct i915_gpu_error *error,
-					  struct intel_engine_cs *engine)
+					  const struct intel_engine_cs *engine)
 {
 	return atomic_read(&error->reset_engine_count[engine->uabi_class]);
 }

drivers/gpu/drm/i915/i915_gpu_error.h

@@ -25,69 +25,41 @@
 #include "i915_scheduler.h"
 
 struct drm_i915_private;
+struct i915_vma_compress;
+struct intel_engine_capture_vma;
 struct intel_overlay_error_state;
 struct intel_display_error_state;
 
-struct i915_gpu_state {
+struct i915_vma_coredump {
-	struct kref ref;
+	struct i915_vma_coredump *next;
-	ktime_t time;
-	ktime_t boottime;
-	ktime_t uptime;
-	unsigned long capture;
 
-	struct drm_i915_private *i915;
+	char name[20];
 
-	char error_msg[128];
+	u64 gtt_offset;
-	bool simulated;
+	u64 gtt_size;
-	bool awake;
+	u32 gtt_page_sizes;
-	bool wakelock;
-	bool suspended;
-	int iommu;
-	u32 reset_count;
-	u32 suspend_count;
-	struct intel_device_info device_info;
-	struct intel_runtime_info runtime_info;
-	struct intel_driver_caps driver_caps;
-	struct i915_params params;
 
-	struct i915_error_uc {
+	int num_pages;
-		struct intel_uc_fw guc_fw;
+	int page_count;
-		struct intel_uc_fw huc_fw;
+	int unused;
-		struct drm_i915_error_object *guc_log;
+	u32 *pages[0];
-	} uc;
+};
 
-	/* Generic register state */
+struct i915_request_coredump {
-	u32 eir;
+	unsigned long flags;
-	u32 pgtbl_er;
+	pid_t pid;
-	u32 ier;
+	u32 context;
-	u32 gtier[6], ngtier;
+	u32 seqno;
-	u32 ccid;
+	u32 start;
-	u32 derrmr;
+	u32 head;
-	u32 forcewake;
+	u32 tail;
-	u32 error; /* gen6+ */
+	struct i915_sched_attr sched_attr;
-	u32 err_int; /* gen7 */
+};
-	u32 fault_data0; /* gen8, gen9 */
-	u32 fault_data1; /* gen8, gen9 */
-	u32 done_reg;
-	u32 gac_eco;
-	u32 gam_ecochk;
-	u32 gab_ctl;
-	u32 gfx_mode;
-	u32 gtt_cache;
-	u32 aux_err; /* gen12 */
-	u32 sfc_done[GEN12_SFC_DONE_MAX]; /* gen12 */
-	u32 gam_done; /* gen12 */
 
-	u32 nfence;
+struct intel_engine_coredump {
-	u64 fence[I915_MAX_NUM_FENCES];
-	struct intel_overlay_error_state *overlay;
-	struct intel_display_error_state *display;
-
-	struct drm_i915_error_engine {
 	const struct intel_engine_cs *engine;
 
-		/* Software tracked state */
+	bool simulated;
-		bool idle;
 	int num_requests;
 	u32 reset_count;
 
@@ -99,6 +71,7 @@ struct i915_gpu_state {
 	u32 cpu_ring_tail;
 
 	/* Register state */
+	u32 ccid;
 	u32 start;
 	u32 tail;
 	u32 head;
@@ -117,7 +90,7 @@ struct i915_gpu_state {
 	u32 rc_psmi; /* sleep state */
 	struct intel_instdone instdone;
 
-		struct drm_i915_error_context {
+	struct i915_gem_context_coredump {
 		char comm[TASK_COMM_LEN];
 		pid_t pid;
 		int active;
@@ -125,33 +98,9 @@ struct i915_gpu_state {
 		struct i915_sched_attr sched_attr;
 	} context;
 
-		struct drm_i915_error_object {
+	struct i915_vma_coredump *vma;
-			u64 gtt_offset;
-			u64 gtt_size;
-			u32 gtt_page_sizes;
-			int num_pages;
-			int page_count;
-			int unused;
-			u32 *pages[0];
-		} *ringbuffer, *batchbuffer, *wa_batchbuffer, *ctx, *hws_page;
 
-		struct drm_i915_error_object **user_bo;
+	struct i915_request_coredump *requests, execlist[EXECLIST_MAX_PORTS];
-		long user_bo_count;
-
-		struct drm_i915_error_object *wa_ctx;
-		struct drm_i915_error_object *default_state;
-
-		struct drm_i915_error_request {
-			unsigned long flags;
-			long jiffies;
-			pid_t pid;
-			u32 context;
-			u32 seqno;
-			u32 start;
-			u32 head;
-			u32 tail;
-			struct i915_sched_attr sched_attr;
-		} *requests, execlist[EXECLIST_MAX_PORTS];
 	unsigned int num_ports;
 
 	struct {
@@ -162,8 +111,75 @@ struct i915_gpu_state {
 		};
 	} vm_info;
 
-		struct drm_i915_error_engine *next;
+	struct intel_engine_coredump *next;
-	} *engine;
+};
+
+struct intel_gt_coredump {
+	const struct intel_gt *_gt;
+	bool awake;
+	bool simulated;
+
+	/* Generic register state */
+	u32 eir;
+	u32 pgtbl_er;
+	u32 ier;
+	u32 gtier[6], ngtier;
+	u32 derrmr;
+	u32 forcewake;
+	u32 error; /* gen6+ */
+	u32 err_int; /* gen7 */
+	u32 fault_data0; /* gen8, gen9 */
+	u32 fault_data1; /* gen8, gen9 */
+	u32 done_reg;
+	u32 gac_eco;
+	u32 gam_ecochk;
+	u32 gab_ctl;
+	u32 gfx_mode;
+	u32 gtt_cache;
+	u32 aux_err; /* gen12 */
+	u32 sfc_done[GEN12_SFC_DONE_MAX]; /* gen12 */
+	u32 gam_done; /* gen12 */
+
+	u32 nfence;
+	u64 fence[I915_MAX_NUM_FENCES];
+
+	struct intel_engine_coredump *engine;
+
+	struct intel_uc_coredump {
+		struct intel_uc_fw guc_fw;
+		struct intel_uc_fw huc_fw;
+		struct i915_vma_coredump *guc_log;
+	} *uc;
+
+	struct intel_gt_coredump *next;
+};
+
+struct i915_gpu_coredump {
+	struct kref ref;
+	ktime_t time;
+	ktime_t boottime;
+	ktime_t uptime;
+	unsigned long capture;
+
+	struct drm_i915_private *i915;
+
+	struct intel_gt_coredump *gt;
+
+	char error_msg[128];
+	bool simulated;
+	bool wakelock;
+	bool suspended;
+	int iommu;
+	u32 reset_count;
+	u32 suspend_count;
+
+	struct intel_device_info device_info;
+	struct intel_runtime_info runtime_info;
+	struct intel_driver_caps driver_caps;
+	struct i915_params params;
+
+	struct intel_overlay_error_state *overlay;
+	struct intel_display_error_state *display;
 
 	struct scatterlist *sgl, *fit;
 };
@@ -172,7 +188,7 @@ struct i915_gpu_error {
 	/* For reset and error_state handling. */
 	spinlock_t lock;
 	/* Protected by the above dev->gpu_error.lock. */
-	struct i915_gpu_state *first_error;
+	struct i915_gpu_coredump *first_error;
 
 	atomic_t pending_fb_pin;
 
@@ -200,29 +216,54 @@ struct drm_i915_error_state_buf {
 __printf(2, 3)
 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...);
 
-struct i915_gpu_state *i915_capture_gpu_state(struct drm_i915_private *i915);
+struct i915_gpu_coredump *i915_gpu_coredump(struct drm_i915_private *i915);
-void i915_capture_error_state(struct drm_i915_private *dev_priv,
+void i915_capture_error_state(struct drm_i915_private *dev_priv);
-			      intel_engine_mask_t engine_mask,
-			      const char *error_msg);
 
-static inline struct i915_gpu_state *
+struct i915_gpu_coredump *
-i915_gpu_state_get(struct i915_gpu_state *gpu)
+i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp);
+
+struct intel_gt_coredump *
+intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp);
+
+struct intel_engine_coredump *
+intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp);
+
+struct intel_engine_capture_vma *
+intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
+				  struct i915_request *rq,
+				  gfp_t gfp);
+
+void intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
+				   struct intel_engine_capture_vma *capture,
+				   struct i915_vma_compress *compress);
+
+struct i915_vma_compress *
+i915_vma_capture_prepare(struct intel_gt_coredump *gt);
+
+void i915_vma_capture_finish(struct intel_gt_coredump *gt,
+			     struct i915_vma_compress *compress);
+
+void i915_error_state_store(struct i915_gpu_coredump *error);
+
+static inline struct i915_gpu_coredump *
+i915_gpu_coredump_get(struct i915_gpu_coredump *gpu)
 {
 	kref_get(&gpu->ref);
 	return gpu;
 }
 
-ssize_t i915_gpu_state_copy_to_buffer(struct i915_gpu_state *error,
+ssize_t
+i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
 				 char *buf, loff_t offset, size_t count);
 
-void __i915_gpu_state_free(struct kref *kref);
+void __i915_gpu_coredump_free(struct kref *kref);
-static inline void i915_gpu_state_put(struct i915_gpu_state *gpu)
+static inline void i915_gpu_coredump_put(struct i915_gpu_coredump *gpu)
 {
 	if (gpu)
-		kref_put(&gpu->ref, __i915_gpu_state_free);
+		kref_put(&gpu->ref, __i915_gpu_coredump_free);
 }
 
-struct i915_gpu_state *i915_first_error_state(struct drm_i915_private *i915);
+struct i915_gpu_coredump *i915_first_error_state(struct drm_i915_private *i915);
 void i915_reset_error_state(struct drm_i915_private *i915);
 void i915_disable_error_state(struct drm_i915_private *i915, int err);
 
@@ -234,7 +275,56 @@ static inline void i915_capture_error_state(struct drm_i915_private *dev_priv,
 {
 }
 
-static inline struct i915_gpu_state *
+static inline struct i915_gpu_coredump *
+i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
+{
+	return NULL;
+}
+
+static inline struct intel_gt_coredump *
+intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
+{
+	return NULL;
+}
+
+static inline struct intel_engine_coredump *
+intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
+{
+	return NULL;
+}
+
+static inline struct intel_engine_capture_vma *
+intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
+				  struct i915_request *rq,
+				  gfp_t gfp)
+{
+	return NULL;
+}
+
+static inline void
+intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
+			      struct intel_engine_capture_vma *capture,
+			      struct i915_vma_compress *compress)
+{
+}
+
+static inline struct i915_vma_compress *
+i915_vma_compress_prepare(struct intel_gt_coredump *gt)
+{
+	return NULL;
+}
+
+void i915_vma_compress_prepare(struct i915_vma_compress *compress)
+{
+}
+
+static inline void
+i915_error_state_store(struct drm_i915_private *i915,
+		       struct i915_gpu_coredump *error)
+{
+}
+
+static inline struct i915_gpu_coredump *
 i915_first_error_state(struct drm_i915_private *i915)
 {
 	return ERR_PTR(-ENODEV);

drivers/gpu/drm/i915/i915_sysfs.c

@@ -498,15 +498,15 @@ static ssize_t error_state_read(struct file *filp, struct kobject *kobj,
 
 	struct device *kdev = kobj_to_dev(kobj);
 	struct drm_i915_private *i915 = kdev_minor_to_i915(kdev);
-	struct i915_gpu_state *gpu;
+	struct i915_gpu_coredump *gpu;
 	ssize_t ret;
 
 	gpu = i915_first_error_state(i915);
 	if (IS_ERR(gpu)) {
 		ret = PTR_ERR(gpu);
 	} else if (gpu) {
-		ret = i915_gpu_state_copy_to_buffer(gpu, buf, off, count);
+		ret = i915_gpu_coredump_copy_to_buffer(gpu, buf, off, count);
-		i915_gpu_state_put(gpu);
+		i915_gpu_coredump_put(gpu);
 	} else {
 		const char *str = "No error state collected\n";
 		size_t len = strlen(str);