]> git.karo-electronics.de Git - karo-tx-linux.git/blobdiff - drivers/gpu/drm/i915/intel_ringbuffer.h
drm/i915: Avoid accessing request->timeline outside of its lifetime
[karo-tx-linux.git] / drivers / gpu / drm / i915 / intel_ringbuffer.h
index 12cb7ed90014bd7a1b974a026b73487d0d3e2d1a..642b54692d0d9697b3c8c8117dc96f84ac18a957 100644 (file)
@@ -3,6 +3,8 @@
 
 #include <linux/hashtable.h>
 #include "i915_gem_batch_pool.h"
+#include "i915_gem_request.h"
+#include "i915_gem_timeline.h"
 
 #define I915_CMD_HASH_ORDER 9
 
  */
 #define I915_RING_FREE_SPACE 64
 
-struct  intel_hw_status_page {
-       u32             *page_addr;
-       unsigned int    gfx_addr;
-       struct          drm_i915_gem_object *obj;
+struct intel_hw_status_page {
+       struct i915_vma *vma;
+       u32 *page_addr;
+       u32 ggtt_offset;
 };
 
-#define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
-#define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
+#define I915_READ_TAIL(engine) I915_READ(RING_TAIL((engine)->mmio_base))
+#define I915_WRITE_TAIL(engine, val) I915_WRITE(RING_TAIL((engine)->mmio_base), val)
 
-#define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
-#define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
+#define I915_READ_START(engine) I915_READ(RING_START((engine)->mmio_base))
+#define I915_WRITE_START(engine, val) I915_WRITE(RING_START((engine)->mmio_base), val)
 
-#define I915_READ_HEAD(ring)  I915_READ(RING_HEAD((ring)->mmio_base))
-#define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
+#define I915_READ_HEAD(engine)  I915_READ(RING_HEAD((engine)->mmio_base))
+#define I915_WRITE_HEAD(engine, val) I915_WRITE(RING_HEAD((engine)->mmio_base), val)
 
-#define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
-#define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
+#define I915_READ_CTL(engine) I915_READ(RING_CTL((engine)->mmio_base))
+#define I915_WRITE_CTL(engine, val) I915_WRITE(RING_CTL((engine)->mmio_base), val)
 
-#define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
-#define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
+#define I915_READ_IMR(engine) I915_READ(RING_IMR((engine)->mmio_base))
+#define I915_WRITE_IMR(engine, val) I915_WRITE(RING_IMR((engine)->mmio_base), val)
 
-#define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base))
-#define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val)
+#define I915_READ_MODE(engine) I915_READ(RING_MI_MODE((engine)->mmio_base))
+#define I915_WRITE_MODE(engine, val) I915_WRITE(RING_MI_MODE((engine)->mmio_base), val)
 
 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
  * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
@@ -56,13 +58,13 @@ struct  intel_hw_status_page {
 #define GEN8_SEMAPHORE_OFFSET(__from, __to)                         \
        (((__from) * I915_NUM_ENGINES  + (__to)) * gen8_semaphore_seqno_size)
 #define GEN8_SIGNAL_OFFSET(__ring, to)                      \
-       (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
+       (dev_priv->semaphore->node.start + \
         GEN8_SEMAPHORE_OFFSET((__ring)->id, (to)))
 #define GEN8_WAIT_OFFSET(__ring, from)                      \
-       (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
+       (dev_priv->semaphore->node.start + \
         GEN8_SEMAPHORE_OFFSET(from, (__ring)->id))
 
-enum intel_ring_hangcheck_action {
+enum intel_engine_hangcheck_action {
        HANGCHECK_IDLE = 0,
        HANGCHECK_WAIT,
        HANGCHECK_ACTIVE,
@@ -72,23 +74,49 @@ enum intel_ring_hangcheck_action {
 
 #define HANGCHECK_SCORE_RING_HUNG 31
 
-struct intel_ring_hangcheck {
+#define I915_MAX_SLICES        3
+#define I915_MAX_SUBSLICES 3
+
+#define instdone_slice_mask(dev_priv__) \
+       (INTEL_GEN(dev_priv__) == 7 ? \
+        1 : INTEL_INFO(dev_priv__)->sseu.slice_mask)
+
+#define instdone_subslice_mask(dev_priv__) \
+       (INTEL_GEN(dev_priv__) == 7 ? \
+        1 : INTEL_INFO(dev_priv__)->sseu.subslice_mask)
+
+#define for_each_instdone_slice_subslice(dev_priv__, slice__, subslice__) \
+       for ((slice__) = 0, (subslice__) = 0; \
+            (slice__) < I915_MAX_SLICES; \
+            (subslice__) = ((subslice__) + 1) < I915_MAX_SUBSLICES ? (subslice__) + 1 : 0, \
+              (slice__) += ((subslice__) == 0)) \
+               for_each_if((BIT(slice__) & instdone_slice_mask(dev_priv__)) && \
+                           (BIT(subslice__) & instdone_subslice_mask(dev_priv__)))
+
+struct intel_instdone {
+       u32 instdone;
+       /* The following exist only in the RCS engine */
+       u32 slice_common;
+       u32 sampler[I915_MAX_SLICES][I915_MAX_SUBSLICES];
+       u32 row[I915_MAX_SLICES][I915_MAX_SUBSLICES];
+};
+
+struct intel_engine_hangcheck {
        u64 acthd;
-       unsigned long user_interrupts;
        u32 seqno;
        int score;
-       enum intel_ring_hangcheck_action action;
+       enum intel_engine_hangcheck_action action;
        int deadlock;
-       u32 instdone[I915_NUM_INSTDONE_REG];
+       struct intel_instdone instdone;
 };
 
-struct intel_ringbuffer {
-       struct drm_i915_gem_object *obj;
-       void __iomem *virtual_start;
+struct intel_ring {
        struct i915_vma *vma;
+       void *vaddr;
 
        struct intel_engine_cs *engine;
-       struct list_head link;
+
+       struct list_head request_list;
 
        u32 head;
        u32 tail;
@@ -121,15 +149,16 @@ struct drm_i915_reg_table;
  *    an option for future use.
  *  size: size of the batch in DWORDS
  */
-struct  i915_ctx_workarounds {
+struct i915_ctx_workarounds {
        struct i915_wa_ctx_bb {
                u32 offset;
                u32 size;
        } indirect_ctx, per_ctx;
-       struct drm_i915_gem_object *obj;
+       struct i915_vma *vma;
 };
 
 struct drm_i915_gem_request;
+struct intel_render_state;
 
 struct intel_engine_cs {
        struct drm_i915_private *i915;
@@ -141,14 +170,22 @@ struct intel_engine_cs {
                VCS2,   /* Keep instances of the same type engine together. */
                VECS
        } id;
-#define I915_NUM_ENGINES 5
 #define _VCS(n) (VCS + (n))
        unsigned int exec_id;
-       unsigned int hw_id;
-       unsigned int guc_id; /* XXX same as hw_id? */
+       enum intel_engine_hw_id {
+               RCS_HW = 0,
+               VCS_HW,
+               BCS_HW,
+               VECS_HW,
+               VCS2_HW
+       } hw_id;
+       enum intel_engine_hw_id guc_id; /* XXX same as hw_id? */
        u32             mmio_base;
-       struct intel_ringbuffer *buffer;
-       struct list_head buffers;
+       unsigned int irq_shift;
+       struct intel_ring *buffer;
+       struct intel_timeline *timeline;
+
+       struct intel_render_state *render_state;
 
        /* Rather than have every client wait upon all user interrupts,
         * with the herd waking after every interrupt and each doing the
@@ -167,17 +204,19 @@ struct intel_engine_cs {
         * the overhead of waking that client is much preferred.
         */
        struct intel_breadcrumbs {
-               struct task_struct *irq_seqno_bh; /* bh for user interrupts */
-               unsigned long irq_wakeups;
+               struct task_struct __rcu *irq_seqno_bh; /* bh for interrupts */
                bool irq_posted;
 
-               spinlock_t lock; /* protects the lists of requests */
+               spinlock_t lock; /* protects the lists of requests; irqsafe */
                struct rb_root waiters; /* sorted by retirement, priority */
                struct rb_root signals; /* sorted by retirement */
                struct intel_wait *first_wait; /* oldest waiter by retirement */
                struct task_struct *signaler; /* used for fence signalling */
                struct drm_i915_gem_request *first_signal;
                struct timer_list fake_irq; /* used after a missed interrupt */
+               struct timer_list hangcheck; /* detect missed interrupts */
+
+               unsigned long timeout;
 
                bool irq_enabled : 1;
                bool rpm_wakelock : 1;
@@ -192,36 +231,50 @@ struct intel_engine_cs {
 
        struct intel_hw_status_page status_page;
        struct i915_ctx_workarounds wa_ctx;
+       struct i915_vma *scratch;
 
        u32             irq_keep_mask; /* always keep these interrupts */
        u32             irq_enable_mask; /* bitmask to enable ring interrupt */
-       void            (*irq_enable)(struct intel_engine_cs *ring);
-       void            (*irq_disable)(struct intel_engine_cs *ring);
+       void            (*irq_enable)(struct intel_engine_cs *engine);
+       void            (*irq_disable)(struct intel_engine_cs *engine);
 
-       int             (*init_hw)(struct intel_engine_cs *ring);
+       int             (*init_hw)(struct intel_engine_cs *engine);
+       void            (*reset_hw)(struct intel_engine_cs *engine,
+                                   struct drm_i915_gem_request *req);
 
        int             (*init_context)(struct drm_i915_gem_request *req);
 
-       void            (*write_tail)(struct intel_engine_cs *ring,
-                                     u32 value);
-       int __must_check (*flush)(struct drm_i915_gem_request *req,
-                                 u32   invalidate_domains,
-                                 u32   flush_domains);
-       int             (*add_request)(struct drm_i915_gem_request *req);
+       int             (*emit_flush)(struct drm_i915_gem_request *request,
+                                     u32 mode);
+#define EMIT_INVALIDATE        BIT(0)
+#define EMIT_FLUSH     BIT(1)
+#define EMIT_BARRIER   (EMIT_INVALIDATE | EMIT_FLUSH)
+       int             (*emit_bb_start)(struct drm_i915_gem_request *req,
+                                        u64 offset, u32 length,
+                                        unsigned int dispatch_flags);
+#define I915_DISPATCH_SECURE BIT(0)
+#define I915_DISPATCH_PINNED BIT(1)
+#define I915_DISPATCH_RS     BIT(2)
+       void            (*emit_breadcrumb)(struct drm_i915_gem_request *req,
+                                          u32 *out);
+       int             emit_breadcrumb_sz;
+
+       /* Pass the request to the hardware queue (e.g. directly into
+        * the legacy ringbuffer or to the end of an execlist).
+        *
+        * This is called from an atomic context with irqs disabled; must
+        * be irq safe.
+        */
+       void            (*submit_request)(struct drm_i915_gem_request *req);
+
        /* Some chipsets are not quite as coherent as advertised and need
         * an expensive kick to force a true read of the up-to-date seqno.
         * However, the up-to-date seqno is not always required and the last
         * seen value is good enough. Note that the seqno will always be
         * monotonic, even if not coherent.
         */
-       void            (*irq_seqno_barrier)(struct intel_engine_cs *ring);
-       int             (*dispatch_execbuffer)(struct drm_i915_gem_request *req,
-                                              u64 offset, u32 length,
-                                              unsigned dispatch_flags);
-#define I915_DISPATCH_SECURE 0x1
-#define I915_DISPATCH_PINNED 0x2
-#define I915_DISPATCH_RS     0x4
-       void            (*cleanup)(struct intel_engine_cs *ring);
+       void            (*irq_seqno_barrier)(struct intel_engine_cs *engine);
+       void            (*cleanup)(struct intel_engine_cs *engine);
 
        /* GEN8 signal/wait table - never trust comments!
         *        signal to     signal to    signal to   signal to      signal to
@@ -261,84 +314,47 @@ struct intel_engine_cs {
         *  ie. transpose of f(x, y)
         */
        struct {
-               u32     sync_seqno[I915_NUM_ENGINES-1];
-
                union {
+#define GEN6_SEMAPHORE_LAST    VECS_HW
+#define GEN6_NUM_SEMAPHORES    (GEN6_SEMAPHORE_LAST + 1)
+#define GEN6_SEMAPHORES_MASK   GENMASK(GEN6_SEMAPHORE_LAST, 0)
                        struct {
                                /* our mbox written by others */
-                               u32             wait[I915_NUM_ENGINES];
+                               u32             wait[GEN6_NUM_SEMAPHORES];
                                /* mboxes this ring signals to */
-                               i915_reg_t      signal[I915_NUM_ENGINES];
+                               i915_reg_t      signal[GEN6_NUM_SEMAPHORES];
                        } mbox;
                        u64             signal_ggtt[I915_NUM_ENGINES];
                };
 
                /* AKA wait() */
-               int     (*sync_to)(struct drm_i915_gem_request *to_req,
-                                  struct intel_engine_cs *from,
-                                  u32 seqno);
-               int     (*signal)(struct drm_i915_gem_request *signaller_req,
-                                 /* num_dwords needed by caller */
-                                 unsigned int num_dwords);
+               int     (*sync_to)(struct drm_i915_gem_request *req,
+                                  struct drm_i915_gem_request *signal);
+               u32     *(*signal)(struct drm_i915_gem_request *req, u32 *out);
        } semaphore;
 
        /* Execlists */
        struct tasklet_struct irq_tasklet;
        spinlock_t execlist_lock; /* used inside tasklet, use spin_lock_bh */
+       struct execlist_port {
+               struct drm_i915_gem_request *request;
+               unsigned int count;
+       } execlist_port[2];
        struct list_head execlist_queue;
        unsigned int fw_domains;
-       unsigned int next_context_status_buffer;
-       unsigned int idle_lite_restore_wa;
        bool disable_lite_restore_wa;
+       bool preempt_wa;
        u32 ctx_desc_template;
-       int             (*emit_request)(struct drm_i915_gem_request *request);
-       int             (*emit_flush)(struct drm_i915_gem_request *request,
-                                     u32 invalidate_domains,
-                                     u32 flush_domains);
-       int             (*emit_bb_start)(struct drm_i915_gem_request *req,
-                                        u64 offset, unsigned dispatch_flags);
-
-       /**
-        * List of objects currently involved in rendering from the
-        * ringbuffer.
-        *
-        * Includes buffers having the contents of their GPU caches
-        * flushed, not necessarily primitives.  last_read_req
-        * represents when the rendering involved will be completed.
-        *
-        * A reference is held on the buffer while on this list.
-        */
-       struct list_head active_list;
-
-       /**
-        * List of breadcrumbs associated with GPU requests currently
-        * outstanding.
-        */
-       struct list_head request_list;
-
-       /**
-        * Seqno of request most recently submitted to request_list.
-        * Used exclusively by hang checker to avoid grabbing lock while
-        * inspecting request list.
-        */
-       u32 last_submitted_seqno;
-
-       bool gpu_caches_dirty;
 
        struct i915_gem_context *last_context;
 
-       struct intel_ring_hangcheck hangcheck;
-
-       struct {
-               struct drm_i915_gem_object *obj;
-               u32 gtt_offset;
-       } scratch;
+       struct intel_engine_hangcheck hangcheck;
 
        bool needs_cmd_parser;
 
        /*
         * Table of commands the command parser needs to know about
-        * for this ring.
+        * for this engine.
         */
        DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
 
@@ -352,48 +368,21 @@ struct intel_engine_cs {
         * Returns the bitmask for the length field of the specified command.
         * Return 0 for an unrecognized/invalid command.
         *
-        * If the command parser finds an entry for a command in the ring's
+        * If the command parser finds an entry for a command in the engine's
         * cmd_tables, it gets the command's length based on the table entry.
-        * If not, it calls this function to determine the per-ring length field
-        * encoding for the command (i.e. certain opcode ranges use certain bits
-        * to encode the command length in the header).
+        * If not, it calls this function to determine the per-engine length
+        * field encoding for the command (i.e. different opcode ranges use
+        * certain bits to encode the command length in the header).
         */
        u32 (*get_cmd_length_mask)(u32 cmd_header);
 };
 
-static inline bool
-intel_engine_initialized(const struct intel_engine_cs *engine)
-{
-       return engine->i915 != NULL;
-}
-
 static inline unsigned
 intel_engine_flag(const struct intel_engine_cs *engine)
 {
        return 1 << engine->id;
 }
 
-static inline u32
-intel_ring_sync_index(struct intel_engine_cs *engine,
-                     struct intel_engine_cs *other)
-{
-       int idx;
-
-       /*
-        * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
-        * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
-        * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
-        * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
-        * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
-        */
-
-       idx = (other - engine) - 1;
-       if (idx < 0)
-               idx += I915_NUM_ENGINES;
-
-       return idx;
-}
-
 static inline void
 intel_flush_status_page(struct intel_engine_cs *engine, int reg)
 {
@@ -437,67 +426,93 @@ intel_write_status_page(struct intel_engine_cs *engine,
 #define I915_GEM_HWS_SCRATCH_INDEX     0x40
 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
 
-struct intel_ringbuffer *
-intel_engine_create_ringbuffer(struct intel_engine_cs *engine, int size);
-int intel_pin_and_map_ringbuffer_obj(struct drm_i915_private *dev_priv,
-                                    struct intel_ringbuffer *ringbuf);
-void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
-void intel_ringbuffer_free(struct intel_ringbuffer *ring);
+struct intel_ring *
+intel_engine_create_ring(struct intel_engine_cs *engine, int size);
+int intel_ring_pin(struct intel_ring *ring);
+void intel_ring_unpin(struct intel_ring *ring);
+void intel_ring_free(struct intel_ring *ring);
+
+void intel_engine_stop(struct intel_engine_cs *engine);
+void intel_engine_cleanup(struct intel_engine_cs *engine);
 
-void intel_stop_engine(struct intel_engine_cs *engine);
-void intel_cleanup_engine(struct intel_engine_cs *engine);
+void intel_legacy_submission_resume(struct drm_i915_private *dev_priv);
 
 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
 
 int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n);
 int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req);
-static inline void intel_ring_emit(struct intel_engine_cs *engine,
-                                  u32 data)
+
+static inline void intel_ring_emit(struct intel_ring *ring, u32 data)
+{
+       *(uint32_t *)(ring->vaddr + ring->tail) = data;
+       ring->tail += 4;
+}
+
+static inline void intel_ring_emit_reg(struct intel_ring *ring, i915_reg_t reg)
 {
-       struct intel_ringbuffer *ringbuf = engine->buffer;
-       iowrite32(data, ringbuf->virtual_start + ringbuf->tail);
-       ringbuf->tail += 4;
+       intel_ring_emit(ring, i915_mmio_reg_offset(reg));
 }
-static inline void intel_ring_emit_reg(struct intel_engine_cs *engine,
-                                      i915_reg_t reg)
+
+static inline void intel_ring_advance(struct intel_ring *ring)
 {
-       intel_ring_emit(engine, i915_mmio_reg_offset(reg));
+       /* Dummy function.
+        *
+        * This serves as a placeholder in the code so that the reader
+        * can compare against the preceding intel_ring_begin() and
+        * check that the number of dwords emitted matches the space
+        * reserved for the command packet (i.e. the value passed to
+        * intel_ring_begin()).
+        */
 }
-static inline void intel_ring_advance(struct intel_engine_cs *engine)
+
+static inline u32 intel_ring_offset(struct intel_ring *ring, void *addr)
 {
-       struct intel_ringbuffer *ringbuf = engine->buffer;
-       ringbuf->tail &= ringbuf->size - 1;
+       /* Don't write ring->size (equivalent to 0) as that hangs some GPUs. */
+       u32 offset = addr - ring->vaddr;
+       return offset & (ring->size - 1);
 }
+
 int __intel_ring_space(int head, int tail, int size);
-void intel_ring_update_space(struct intel_ringbuffer *ringbuf);
+void intel_ring_update_space(struct intel_ring *ring);
+
+void intel_engine_init_global_seqno(struct intel_engine_cs *engine, u32 seqno);
 
-int __must_check intel_engine_idle(struct intel_engine_cs *engine);
-void intel_ring_init_seqno(struct intel_engine_cs *engine, u32 seqno);
-int intel_ring_flush_all_caches(struct drm_i915_gem_request *req);
-int intel_ring_invalidate_all_caches(struct drm_i915_gem_request *req);
+void intel_engine_setup_common(struct intel_engine_cs *engine);
+int intel_engine_init_common(struct intel_engine_cs *engine);
+int intel_engine_create_scratch(struct intel_engine_cs *engine, int size);
+void intel_engine_cleanup_common(struct intel_engine_cs *engine);
 
-int intel_init_pipe_control(struct intel_engine_cs *engine, int size);
-void intel_fini_pipe_control(struct intel_engine_cs *engine);
+int intel_init_render_ring_buffer(struct intel_engine_cs *engine);
+int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine);
+int intel_init_bsd2_ring_buffer(struct intel_engine_cs *engine);
+int intel_init_blt_ring_buffer(struct intel_engine_cs *engine);
+int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine);
 
-int intel_init_render_ring_buffer(struct drm_device *dev);
-int intel_init_bsd_ring_buffer(struct drm_device *dev);
-int intel_init_bsd2_ring_buffer(struct drm_device *dev);
-int intel_init_blt_ring_buffer(struct drm_device *dev);
-int intel_init_vebox_ring_buffer(struct drm_device *dev);
+u64 intel_engine_get_active_head(struct intel_engine_cs *engine);
+u64 intel_engine_get_last_batch_head(struct intel_engine_cs *engine);
 
-u64 intel_ring_get_active_head(struct intel_engine_cs *engine);
 static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine)
 {
        return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
 }
 
-int init_workarounds_ring(struct intel_engine_cs *engine);
-
-static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
+static inline u32 intel_engine_last_submit(struct intel_engine_cs *engine)
 {
-       return ringbuf->tail;
+       /* We are only peeking at the tail of the submit queue (and not the
+        * queue itself) in order to gain a hint as to the current active
+        * state of the engine. Callers are not expected to be taking
+        * engine->timeline->lock, nor are they expected to be concerned
+        * wtih serialising this hint with anything, so document it as
+        * a hint and nothing more.
+        */
+       return READ_ONCE(engine->timeline->last_submitted_seqno);
 }
 
+int init_workarounds_ring(struct intel_engine_cs *engine);
+
+void intel_engine_get_instdone(struct intel_engine_cs *engine,
+                              struct intel_instdone *instdone);
+
 /*
  * Arbitrary size for largest possible 'add request' sequence. The code paths
  * are complex and variable. Empirical measurement shows that the worst case
@@ -509,21 +524,10 @@ static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
 
 static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
 {
-       return engine->status_page.gfx_addr + I915_GEM_HWS_INDEX_ADDR;
+       return engine->status_page.ggtt_offset + I915_GEM_HWS_INDEX_ADDR;
 }
 
 /* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
-struct intel_wait {
-       struct rb_node node;
-       struct task_struct *tsk;
-       u32 seqno;
-};
-
-struct intel_signal_node {
-       struct rb_node node;
-       struct intel_wait wait;
-};
-
 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
 
 static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
@@ -543,29 +547,36 @@ void intel_engine_remove_wait(struct intel_engine_cs *engine,
                              struct intel_wait *wait);
 void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
 
-static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
+static inline bool intel_engine_has_waiter(const struct intel_engine_cs *engine)
 {
-       return READ_ONCE(engine->breadcrumbs.irq_seqno_bh);
+       return rcu_access_pointer(engine->breadcrumbs.irq_seqno_bh);
 }
 
-static inline bool intel_engine_wakeup(struct intel_engine_cs *engine)
+static inline bool intel_engine_wakeup(const struct intel_engine_cs *engine)
 {
        bool wakeup = false;
-       struct task_struct *tsk = READ_ONCE(engine->breadcrumbs.irq_seqno_bh);
+
        /* Note that for this not to dangerously chase a dangling pointer,
-        * the caller is responsible for ensure that the task remain valid for
-        * wake_up_process() i.e. that the RCU grace period cannot expire.
+        * we must hold the rcu_read_lock here.
         *
         * Also note that tsk is likely to be in !TASK_RUNNING state so an
         * early test for tsk->state != TASK_RUNNING before wake_up_process()
         * is unlikely to be beneficial.
         */
-       if (tsk)
-               wakeup = wake_up_process(tsk);
+       if (intel_engine_has_waiter(engine)) {
+               struct task_struct *tsk;
+
+               rcu_read_lock();
+               tsk = rcu_dereference(engine->breadcrumbs.irq_seqno_bh);
+               if (tsk)
+                       wakeup = wake_up_process(tsk);
+               rcu_read_unlock();
+       }
+
        return wakeup;
 }
 
-void intel_engine_enable_fake_irq(struct intel_engine_cs *engine);
+void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine);
 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
 unsigned int intel_kick_waiters(struct drm_i915_private *i915);
 unsigned int intel_kick_signalers(struct drm_i915_private *i915);