]> git.baikalelectronics.ru Git - kernel.git/commitdiff
drm/i915: Drop GEM context as a direct link from i915_request
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 20 Dec 2019 10:12:29 +0000 (10:12 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Fri, 20 Dec 2019 10:52:21 +0000 (10:52 +0000)
Keep the intel_context as being the primary state for i915_request, with
the GEM context a backpointer from the low level state for the rarer
cases we need client information. Our goal is to remove such references
to clients from the backend, and leave the HW submission agnostic to
client interfaces and self-contained.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Andi Shyti <andi.shyti@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191220101230.256839-1-chris@chris-wilson.co.uk
21 files changed:
drivers/gpu/drm/i915/gem/i915_gem_context.c
drivers/gpu/drm/i915/gem/i915_gem_context.h
drivers/gpu/drm/i915/gem/i915_gem_context_types.h
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
drivers/gpu/drm/i915/gt/intel_context.c
drivers/gpu/drm/i915/gt/intel_context.h
drivers/gpu/drm/i915/gt/intel_context_types.h
drivers/gpu/drm/i915/gt/intel_engine_cs.c
drivers/gpu/drm/i915/gt/intel_lrc.c
drivers/gpu/drm/i915/gt/intel_reset.c
drivers/gpu/drm/i915/gt/intel_ring_submission.c
drivers/gpu/drm/i915/gt/selftest_lrc.c
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
drivers/gpu/drm/i915/gvt/scheduler.c
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/i915_gpu_error.c
drivers/gpu/drm/i915/i915_perf.c
drivers/gpu/drm/i915/i915_request.c
drivers/gpu/drm/i915/i915_request.h
drivers/gpu/drm/i915/i915_scheduler.c

index 6618c0c6506c915477af8f243b8f0d301e487b0b..42585a20a9ae139d66ab0938e43164221efa8abd 100644 (file)
@@ -69,6 +69,7 @@
 
 #include <drm/i915_drm.h>
 
+#include "gt/intel_context.h"
 #include "gt/intel_engine_heartbeat.h"
 #include "gt/intel_engine_pm.h"
 #include "gt/intel_engine_user.h"
@@ -423,15 +424,6 @@ static void kill_context(struct i915_gem_context *ctx)
        struct i915_gem_engines_iter it;
        struct intel_context *ce;
 
-       /*
-        * If we are already banned, it was due to a guilty request causing
-        * a reset and the entire context being evicted from the GPU.
-        */
-       if (i915_gem_context_is_banned(ctx))
-               return;
-
-       i915_gem_context_set_banned(ctx);
-
        /*
         * Map the user's engine back to the actual engines; one virtual
         * engine will be mapped to multiple engines, and using ctx->engine[]
@@ -442,6 +434,9 @@ static void kill_context(struct i915_gem_context *ctx)
        for_each_gem_engine(ce, __context_engines_static(ctx), it) {
                struct intel_engine_cs *engine;
 
+               if (intel_context_set_banned(ce))
+                       continue;
+
                /*
                 * Check the current active state of this context; if we
                 * are currently executing on the GPU we need to evict
@@ -1093,7 +1088,7 @@ static void set_ppgtt_barrier(void *data)
 
 static int emit_ppgtt_update(struct i915_request *rq, void *data)
 {
-       struct i915_address_space *vm = rq->hw_context->vm;
+       struct i915_address_space *vm = rq->context->vm;
        struct intel_engine_cs *engine = rq->engine;
        u32 base = engine->mmio_base;
        u32 *cs;
index 18e50a769a6ea220fc8f63a761b60dcfc0d13304..69932899803e368fe4c9f75fc4fa55f68d50e2d8 100644 (file)
@@ -91,26 +91,6 @@ static inline void i915_gem_context_clear_persistence(struct i915_gem_context *c
        clear_bit(UCONTEXT_PERSISTENCE, &ctx->user_flags);
 }
 
-static inline bool i915_gem_context_is_banned(const struct i915_gem_context *ctx)
-{
-       return test_bit(CONTEXT_BANNED, &ctx->flags);
-}
-
-static inline void i915_gem_context_set_banned(struct i915_gem_context *ctx)
-{
-       set_bit(CONTEXT_BANNED, &ctx->flags);
-}
-
-static inline bool i915_gem_context_force_single_submission(const struct i915_gem_context *ctx)
-{
-       return test_bit(CONTEXT_FORCE_SINGLE_SUBMISSION, &ctx->flags);
-}
-
-static inline void i915_gem_context_set_force_single_submission(struct i915_gem_context *ctx)
-{
-       __set_bit(CONTEXT_FORCE_SINGLE_SUBMISSION, &ctx->flags);
-}
-
 static inline bool
 i915_gem_context_user_engines(const struct i915_gem_context *ctx)
 {
@@ -129,24 +109,6 @@ i915_gem_context_clear_user_engines(struct i915_gem_context *ctx)
        clear_bit(CONTEXT_USER_ENGINES, &ctx->flags);
 }
 
-static inline bool
-i915_gem_context_nopreempt(const struct i915_gem_context *ctx)
-{
-       return test_bit(CONTEXT_NOPREEMPT, &ctx->flags);
-}
-
-static inline void
-i915_gem_context_set_nopreempt(struct i915_gem_context *ctx)
-{
-       set_bit(CONTEXT_NOPREEMPT, &ctx->flags);
-}
-
-static inline void
-i915_gem_context_clear_nopreempt(struct i915_gem_context *ctx)
-{
-       clear_bit(CONTEXT_NOPREEMPT, &ctx->flags);
-}
-
 static inline bool i915_gem_context_is_kernel(struct i915_gem_context *ctx)
 {
        return !ctx->file_priv;
index 69df5459c350ab50aeac482e529aa9e989aec859..017ca803ab47e62760b35968e09568976a5563b5 100644 (file)
@@ -134,11 +134,8 @@ struct i915_gem_context {
         * @flags: small set of booleans
         */
        unsigned long flags;
-#define CONTEXT_BANNED                 0
-#define CONTEXT_CLOSED                 1
-#define CONTEXT_FORCE_SINGLE_SUBMISSION        2
-#define CONTEXT_USER_ENGINES           3
-#define CONTEXT_NOPREEMPT              4
+#define CONTEXT_CLOSED                 0
+#define CONTEXT_USER_ENGINES           1
 
        struct mutex mutex;
 
index 7d07131aa3f7df8c95520b559bfb130178a43da8..cbd2bcade3c8d60335e648712a2b1447c16c6415 100644 (file)
@@ -730,9 +730,6 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
        unsigned int i, batch;
        int err;
 
-       if (unlikely(i915_gem_context_is_banned(eb->gem_context)))
-               return -EIO;
-
        INIT_LIST_HEAD(&eb->relocs);
        INIT_LIST_HEAD(&eb->unbound);
 
@@ -2175,7 +2172,7 @@ static int eb_submit(struct i915_execbuffer *eb)
                        return err;
        }
 
-       if (i915_gem_context_nopreempt(eb->gem_context))
+       if (intel_context_nopreempt(eb->context))
                eb->request->flags |= I915_REQUEST_NOPREEMPT;
 
        return 0;
@@ -2261,6 +2258,9 @@ static int __eb_pin_engine(struct i915_execbuffer *eb, struct intel_context *ce)
        if (err)
                return err;
 
+       if (unlikely(intel_context_is_banned(ce)))
+               return -EIO;
+
        /*
         * Pinning the contexts may generate requests in order to acquire
         * GGTT space, so do this first before we reserve a seqno for
index 2979f0fd9270f9f3820489ef64985f78d1080cf2..0ba524a414c68d66558afe24674613a2229451ee 100644 (file)
@@ -281,7 +281,7 @@ bool i915_request_enable_breadcrumb(struct i915_request *rq)
 
        if (test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags)) {
                struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
-               struct intel_context *ce = rq->hw_context;
+               struct intel_context *ce = rq->context;
                struct list_head *pos;
 
                spin_lock(&b->irq_lock);
@@ -338,7 +338,7 @@ void i915_request_cancel_breadcrumb(struct i915_request *rq)
         */
        spin_lock(&b->irq_lock);
        if (test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags)) {
-               struct intel_context *ce = rq->hw_context;
+               struct intel_context *ce = rq->context;
 
                list_del(&rq->signal_link);
                if (list_empty(&ce->signals))
index b1e346d2d35f509ff0716386467ed321b8a2a596..ae0dc40031df9d2aa5d592f43989b358cc459248 100644 (file)
@@ -302,7 +302,7 @@ int intel_context_prepare_remote_request(struct intel_context *ce,
        int err;
 
        /* Only suitable for use in remotely modifying this context */
-       GEM_BUG_ON(rq->hw_context == ce);
+       GEM_BUG_ON(rq->context == ce);
 
        if (rcu_access_pointer(rq->timeline) != tl) { /* timeline sharing! */
                /* Queue this switch after current activity by this context. */
index b39eb1fcfbca3b84f58a742acc8390679f44fe42..65389cb8a565c51bf5714c5fd1bbeb096f0f767a 100644 (file)
@@ -7,7 +7,9 @@
 #ifndef __INTEL_CONTEXT_H__
 #define __INTEL_CONTEXT_H__
 
+#include <linux/bitops.h>
 #include <linux/lockdep.h>
+#include <linux/types.h>
 
 #include "i915_active.h"
 #include "intel_context_types.h"
@@ -160,4 +162,44 @@ static inline struct intel_ring *__intel_context_ring_size(u64 sz)
        return u64_to_ptr(struct intel_ring, sz);
 }
 
+static inline bool intel_context_is_banned(const struct intel_context *ce)
+{
+       return test_bit(CONTEXT_BANNED, &ce->flags);
+}
+
+static inline bool intel_context_set_banned(struct intel_context *ce)
+{
+       return test_and_set_bit(CONTEXT_BANNED, &ce->flags);
+}
+
+static inline bool
+intel_context_force_single_submission(const struct intel_context *ce)
+{
+       return test_bit(CONTEXT_FORCE_SINGLE_SUBMISSION, &ce->flags);
+}
+
+static inline void
+intel_context_set_single_submission(struct intel_context *ce)
+{
+       __set_bit(CONTEXT_FORCE_SINGLE_SUBMISSION, &ce->flags);
+}
+
+static inline bool
+intel_context_nopreempt(const struct intel_context *ce)
+{
+       return test_bit(CONTEXT_NOPREEMPT, &ce->flags);
+}
+
+static inline void
+intel_context_set_nopreempt(struct intel_context *ce)
+{
+       set_bit(CONTEXT_NOPREEMPT, &ce->flags);
+}
+
+static inline void
+intel_context_clear_nopreempt(struct intel_context *ce)
+{
+       clear_bit(CONTEXT_NOPREEMPT, &ce->flags);
+}
+
 #endif /* __INTEL_CONTEXT_H__ */
index d1204cc899a337a44efc8c4faa5561a1a1dde206..597448f6e98bf4c3744780f086b54bbd942615d6 100644 (file)
@@ -54,8 +54,11 @@ struct intel_context {
        struct intel_timeline *timeline;
 
        unsigned long flags;
-#define CONTEXT_ALLOC_BIT 0
-#define CONTEXT_VALID_BIT 1
+#define CONTEXT_ALLOC_BIT              0
+#define CONTEXT_VALID_BIT              1
+#define CONTEXT_BANNED                 2
+#define CONTEXT_FORCE_SINGLE_SUBMISSION        3
+#define CONTEXT_NOPREEMPT              4
 
        u32 *lrc_reg_state;
        u64 lrc_desc;
index 6dd18f93d45c867c5f7bb370e8a8db55bf3f2646..e091b3366eaea74e39ce0c2de556987e73815396 100644 (file)
@@ -1523,9 +1523,9 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 
                print_request_ring(m, rq);
 
-               if (rq->hw_context->lrc_reg_state) {
+               if (rq->context->lrc_reg_state) {
                        drm_printf(m, "Logical Ring Context:\n");
-                       hexdump(m, rq->hw_context->lrc_reg_state, PAGE_SIZE);
+                       hexdump(m, rq->context->lrc_reg_state, PAGE_SIZE);
                }
        }
        spin_unlock_irqrestore(&engine->active.lock, flags);
@@ -1586,7 +1586,7 @@ int intel_enable_engine_stats(struct intel_engine_cs *engine)
 
                for (port = execlists->pending; (rq = *port); port++) {
                        /* Exclude any contexts already counted in active */
-                       if (!intel_context_inflight_count(rq->hw_context))
+                       if (!intel_context_inflight_count(rq->context))
                                engine->stats.active++;
                }
 
index b7da6d7e7e2a289163e995f42ba5614bec6778ad..ec93c47c4bdd2fc406dd0028ae5e19a59d684c10 100644 (file)
@@ -880,7 +880,7 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine)
                        list_move(&rq->sched.link, pl);
                        active = rq;
                } else {
-                       struct intel_engine_cs *owner = rq->hw_context->engine;
+                       struct intel_engine_cs *owner = rq->context->engine;
 
                        /*
                         * Decouple the virtual breadcrumb before moving it
@@ -1051,7 +1051,7 @@ static void restore_default_state(struct intel_context *ce,
 static void reset_active(struct i915_request *rq,
                         struct intel_engine_cs *engine)
 {
-       struct intel_context * const ce = rq->hw_context;
+       struct intel_context * const ce = rq->context;
        u32 head;
 
        /*
@@ -1092,11 +1092,11 @@ static inline struct intel_engine_cs *
 __execlists_schedule_in(struct i915_request *rq)
 {
        struct intel_engine_cs * const engine = rq->engine;
-       struct intel_context * const ce = rq->hw_context;
+       struct intel_context * const ce = rq->context;
 
        intel_context_get(ce);
 
-       if (unlikely(i915_gem_context_is_banned(ce->gem_context)))
+       if (unlikely(intel_context_is_banned(ce)))
                reset_active(rq, engine);
 
        if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
@@ -1124,7 +1124,7 @@ __execlists_schedule_in(struct i915_request *rq)
 static inline struct i915_request *
 execlists_schedule_in(struct i915_request *rq, int idx)
 {
-       struct intel_context * const ce = rq->hw_context;
+       struct intel_context * const ce = rq->context;
        struct intel_engine_cs *old;
 
        GEM_BUG_ON(!intel_engine_pm_is_awake(rq->engine));
@@ -1155,7 +1155,7 @@ static inline void
 __execlists_schedule_out(struct i915_request *rq,
                         struct intel_engine_cs * const engine)
 {
-       struct intel_context * const ce = rq->hw_context;
+       struct intel_context * const ce = rq->context;
 
        /*
         * NB process_csb() is not under the engine->active.lock and hence
@@ -1193,7 +1193,7 @@ __execlists_schedule_out(struct i915_request *rq,
 static inline void
 execlists_schedule_out(struct i915_request *rq)
 {
-       struct intel_context * const ce = rq->hw_context;
+       struct intel_context * const ce = rq->context;
        struct intel_engine_cs *cur, *old;
 
        trace_i915_request_out(rq);
@@ -1210,7 +1210,7 @@ execlists_schedule_out(struct i915_request *rq)
 
 static u64 execlists_update_context(struct i915_request *rq)
 {
-       struct intel_context *ce = rq->hw_context;
+       struct intel_context *ce = rq->context;
        u64 desc = ce->lrc_desc;
        u32 tail;
 
@@ -1311,13 +1311,13 @@ assert_pending_valid(const struct intel_engine_execlists *execlists,
                GEM_BUG_ON(!kref_read(&rq->fence.refcount));
                GEM_BUG_ON(!i915_request_is_active(rq));
 
-               if (ce == rq->hw_context) {
+               if (ce == rq->context) {
                        GEM_TRACE_ERR("Dup context:%llx in pending[%zd]\n",
                                      ce->timeline->fence_context,
                                      port - execlists->pending);
                        return false;
                }
-               ce = rq->hw_context;
+               ce = rq->context;
 
                /* Hold tightly onto the lock to prevent concurrent retires! */
                if (!spin_trylock_irqsave(&rq->lock, flags))
@@ -1326,8 +1326,7 @@ assert_pending_valid(const struct intel_engine_execlists *execlists,
                if (i915_request_completed(rq))
                        goto unlock;
 
-               if (i915_active_is_idle(&ce->active) &&
-                   !i915_gem_context_is_kernel(ce->gem_context)) {
+               if (i915_active_is_idle(&ce->active) && ce->gem_context) {
                        GEM_TRACE_ERR("Inactive context:%llx in pending[%zd]\n",
                                      ce->timeline->fence_context,
                                      port - execlists->pending);
@@ -1399,7 +1398,7 @@ static void execlists_submit_ports(struct intel_engine_cs *engine)
 static bool ctx_single_port_submission(const struct intel_context *ce)
 {
        return (IS_ENABLED(CONFIG_DRM_I915_GVT) &&
-               i915_gem_context_force_single_submission(ce->gem_context));
+               intel_context_force_single_submission(ce));
 }
 
 static bool can_merge_ctx(const struct intel_context *prev,
@@ -1435,7 +1434,7 @@ static bool can_merge_rq(const struct i915_request *prev,
                     (I915_REQUEST_NOPREEMPT | I915_REQUEST_SENTINEL)))
                return false;
 
-       if (!can_merge_ctx(prev->hw_context, next->hw_context))
+       if (!can_merge_ctx(prev->context, next->context))
                return false;
 
        return true;
@@ -1622,7 +1621,7 @@ static unsigned long active_preempt_timeout(struct intel_engine_cs *engine)
                return 0;
 
        /* Force a fast reset for terminated contexts (ignoring sysfs!) */
-       if (unlikely(i915_gem_context_is_banned(rq->gem_context)))
+       if (unlikely(intel_context_is_banned(rq->context)))
                return 1;
 
        return READ_ONCE(engine->props.preempt_timeout_ms);
@@ -1730,7 +1729,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
                         * tendency to ignore us rewinding the TAIL to the
                         * end of an earlier request.
                         */
-                       last->hw_context->lrc_desc |= CTX_DESC_FORCE_RESTORE;
+                       last->context->lrc_desc |= CTX_DESC_FORCE_RESTORE;
                        last = NULL;
                } else if (need_timeslice(engine, last) &&
                           timer_expired(&engine->execlists.timer)) {
@@ -1802,7 +1801,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 
                GEM_BUG_ON(rq != ve->request);
                GEM_BUG_ON(rq->engine != &ve->base);
-               GEM_BUG_ON(rq->hw_context != &ve->context);
+               GEM_BUG_ON(rq->context != &ve->context);
 
                if (rq_prio(rq) >= queue_prio(execlists)) {
                        if (!virtual_matches(ve, rq, engine)) {
@@ -1921,7 +1920,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
                                 * same LRCA, i.e. we must submit 2 different
                                 * contexts if we submit 2 ELSP.
                                 */
-                               if (last->hw_context == rq->hw_context)
+                               if (last->context == rq->context)
                                        goto done;
 
                                if (i915_request_has_sentinel(last))
@@ -1934,8 +1933,8 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
                                 * the same context (even though a different
                                 * request) to the second port.
                                 */
-                               if (ctx_single_port_submission(last->hw_context) ||
-                                   ctx_single_port_submission(rq->hw_context))
+                               if (ctx_single_port_submission(last->context) ||
+                                   ctx_single_port_submission(rq->context))
                                        goto done;
 
                                merge = false;
@@ -1949,8 +1948,8 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
                                }
 
                                GEM_BUG_ON(last &&
-                                          !can_merge_ctx(last->hw_context,
-                                                         rq->hw_context));
+                                          !can_merge_ctx(last->context,
+                                                         rq->context));
 
                                submit = true;
                                last = rq;
@@ -2564,7 +2563,7 @@ static int execlists_request_alloc(struct i915_request *request)
 {
        int ret;
 
-       GEM_BUG_ON(!intel_context_is_pinned(request->hw_context));
+       GEM_BUG_ON(!intel_context_is_pinned(request->context));
 
        /*
         * Flush enough space to reduce the likelihood of waiting after
@@ -3071,7 +3070,7 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled)
        /* We still have requests in-flight; the engine should be active */
        GEM_BUG_ON(!intel_engine_pm_is_awake(engine));
 
-       ce = rq->hw_context;
+       ce = rq->context;
        GEM_BUG_ON(!i915_vma_is_pinned(ce->state));
 
        if (i915_request_completed(rq)) {
index f3d1e921fba64972202eb345cca2a98158d07c96..3d293be64fc2f7d31b634fb8800f86cf1961a2b9 100644 (file)
@@ -41,27 +41,30 @@ static void rmw_clear_fw(struct intel_uncore *uncore, i915_reg_t reg, u32 clr)
 static void engine_skip_context(struct i915_request *rq)
 {
        struct intel_engine_cs *engine = rq->engine;
-       struct i915_gem_context *hung_ctx = rq->gem_context;
+       struct intel_context *hung_ctx = rq->context;
 
        if (!i915_request_is_active(rq))
                return;
 
        lockdep_assert_held(&engine->active.lock);
        list_for_each_entry_continue(rq, &engine->active.requests, sched.link)
-               if (rq->gem_context == hung_ctx)
+               if (rq->context == hung_ctx)
                        i915_request_skip(rq, -EIO);
 }
 
-static void client_mark_guilty(struct drm_i915_file_private *file_priv,
-                              const struct i915_gem_context *ctx)
+static void client_mark_guilty(struct i915_request *rq, bool banned)
 {
-       unsigned int score;
+       struct i915_gem_context *ctx = rq->context->gem_context;
+       struct drm_i915_file_private *file_priv = ctx->file_priv;
        unsigned long prev_hang;
+       unsigned int score;
 
-       if (i915_gem_context_is_banned(ctx))
+       if (IS_ERR_OR_NULL(file_priv))
+               return;
+
+       score = 0;
+       if (banned)
                score = I915_CLIENT_SCORE_CONTEXT_BAN;
-       else
-               score = 0;
 
        prev_hang = xchg(&file_priv->hang_timestamp, jiffies);
        if (time_before(jiffies, prev_hang + I915_CLIENT_FAST_HANG_JIFFIES))
@@ -76,14 +79,15 @@ static void client_mark_guilty(struct drm_i915_file_private *file_priv,
        }
 }
 
-static bool context_mark_guilty(struct i915_gem_context *ctx)
+static bool mark_guilty(struct i915_request *rq)
 {
+       struct i915_gem_context *ctx = rq->context->gem_context;
        unsigned long prev_hang;
        bool banned;
        int i;
 
        if (i915_gem_context_is_closed(ctx)) {
-               i915_gem_context_set_banned(ctx);
+               intel_context_set_banned(rq->context);
                return true;
        }
 
@@ -110,18 +114,17 @@ static bool context_mark_guilty(struct i915_gem_context *ctx)
        if (banned) {
                DRM_DEBUG_DRIVER("context %s: guilty %d, banned\n",
                                 ctx->name, atomic_read(&ctx->guilty_count));
-               i915_gem_context_set_banned(ctx);
+               intel_context_set_banned(rq->context);
        }
 
-       if (!IS_ERR_OR_NULL(ctx->file_priv))
-               client_mark_guilty(ctx->file_priv, ctx);
+       client_mark_guilty(rq, banned);
 
        return banned;
 }
 
-static void context_mark_innocent(struct i915_gem_context *ctx)
+static void mark_innocent(struct i915_request *rq)
 {
-       atomic_inc(&ctx->active_count);
+       atomic_inc(&rq->context->gem_context->active_count);
 }
 
 void __i915_request_reset(struct i915_request *rq, bool guilty)
@@ -137,11 +140,11 @@ void __i915_request_reset(struct i915_request *rq, bool guilty)
        rcu_read_lock(); /* protect the GEM context */
        if (guilty) {
                i915_request_skip(rq, -EIO);
-               if (context_mark_guilty(rq->gem_context))
+               if (mark_guilty(rq))
                        engine_skip_context(rq);
        } else {
                dma_fence_set_error(&rq->fence, -EAGAIN);
-               context_mark_innocent(rq->gem_context);
+               mark_innocent(rq);
        }
        rcu_read_unlock();
 }
index b61658601c86ec904a6f372e49186bc3a525e586..a5d30d1468f7150986320b2a051dbecbfb4f841b 100644 (file)
@@ -1480,7 +1480,7 @@ static inline int mi_set_context(struct i915_request *rq, u32 flags)
 
        *cs++ = MI_NOOP;
        *cs++ = MI_SET_CONTEXT;
-       *cs++ = i915_ggtt_offset(rq->hw_context->state) | flags;
+       *cs++ = i915_ggtt_offset(rq->context->state) | flags;
        /*
         * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP
         * WaMiSetContext_Hang:snb,ivb,vlv
@@ -1550,7 +1550,7 @@ static int remap_l3_slice(struct i915_request *rq, int slice)
 
 static int remap_l3(struct i915_request *rq)
 {
-       struct i915_gem_context *ctx = rq->gem_context;
+       struct i915_gem_context *ctx = rq->context->gem_context;
        int i, err;
 
        if (!ctx->remap_slice)
@@ -1597,7 +1597,7 @@ static int switch_mm(struct i915_request *rq, struct i915_address_space *vm)
 
 static int switch_context(struct i915_request *rq)
 {
-       struct intel_context *ce = rq->hw_context;
+       struct intel_context *ce = rq->context;
        int ret;
 
        GEM_BUG_ON(HAS_EXECLISTS(rq->i915));
@@ -1631,7 +1631,7 @@ static int ring_request_alloc(struct i915_request *request)
 {
        int ret;
 
-       GEM_BUG_ON(!intel_context_is_pinned(request->hw_context));
+       GEM_BUG_ON(!intel_context_is_pinned(request->context));
        GEM_BUG_ON(i915_request_timeline(request)->has_initial_breadcrumb);
 
        /*
index ac8b9116d307249e105a40cf1d7c5ba04d27ef8d..619e34813ade2f6d58a4611a0dcc7c1ee43fb75c 100644 (file)
@@ -1195,13 +1195,13 @@ static int __cancel_active0(struct live_preempt_cancel *arg)
                                __func__, arg->engine->name))
                return -EIO;
 
-       clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags);
        rq = spinner_create_request(&arg->a.spin,
                                    arg->a.ctx, arg->engine,
                                    MI_ARB_CHECK);
        if (IS_ERR(rq))
                return PTR_ERR(rq);
 
+       clear_bit(CONTEXT_BANNED, &rq->context->flags);
        i915_request_get(rq);
        i915_request_add(rq);
        if (!igt_wait_for_spinner(&arg->a.spin, rq)) {
@@ -1209,7 +1209,7 @@ static int __cancel_active0(struct live_preempt_cancel *arg)
                goto out;
        }
 
-       i915_gem_context_set_banned(arg->a.ctx);
+       intel_context_set_banned(rq->context);
        err = intel_engine_pulse(arg->engine);
        if (err)
                goto out;
@@ -1244,13 +1244,13 @@ static int __cancel_active1(struct live_preempt_cancel *arg)
                                __func__, arg->engine->name))
                return -EIO;
 
-       clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags);
        rq[0] = spinner_create_request(&arg->a.spin,
                                       arg->a.ctx, arg->engine,
                                       MI_NOOP); /* no preemption */
        if (IS_ERR(rq[0]))
                return PTR_ERR(rq[0]);
 
+       clear_bit(CONTEXT_BANNED, &rq[0]->context->flags);
        i915_request_get(rq[0]);
        i915_request_add(rq[0]);
        if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) {
@@ -1258,7 +1258,6 @@ static int __cancel_active1(struct live_preempt_cancel *arg)
                goto out;
        }
 
-       clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags);
        rq[1] = spinner_create_request(&arg->b.spin,
                                       arg->b.ctx, arg->engine,
                                       MI_ARB_CHECK);
@@ -1267,13 +1266,14 @@ static int __cancel_active1(struct live_preempt_cancel *arg)
                goto out;
        }
 
+       clear_bit(CONTEXT_BANNED, &rq[1]->context->flags);
        i915_request_get(rq[1]);
        err = i915_request_await_dma_fence(rq[1], &rq[0]->fence);
        i915_request_add(rq[1]);
        if (err)
                goto out;
 
-       i915_gem_context_set_banned(arg->b.ctx);
+       intel_context_set_banned(rq[1]->context);
        err = intel_engine_pulse(arg->engine);
        if (err)
                goto out;
@@ -1316,13 +1316,13 @@ static int __cancel_queued(struct live_preempt_cancel *arg)
                                __func__, arg->engine->name))
                return -EIO;
 
-       clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags);
        rq[0] = spinner_create_request(&arg->a.spin,
                                       arg->a.ctx, arg->engine,
                                       MI_ARB_CHECK);
        if (IS_ERR(rq[0]))
                return PTR_ERR(rq[0]);
 
+       clear_bit(CONTEXT_BANNED, &rq[0]->context->flags);
        i915_request_get(rq[0]);
        i915_request_add(rq[0]);
        if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) {
@@ -1330,13 +1330,13 @@ static int __cancel_queued(struct live_preempt_cancel *arg)
                goto out;
        }
 
-       clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags);
        rq[1] = igt_request_alloc(arg->b.ctx, arg->engine);
        if (IS_ERR(rq[1])) {
                err = PTR_ERR(rq[1]);
                goto out;
        }
 
+       clear_bit(CONTEXT_BANNED, &rq[1]->context->flags);
        i915_request_get(rq[1]);
        err = i915_request_await_dma_fence(rq[1], &rq[0]->fence);
        i915_request_add(rq[1]);
@@ -1357,7 +1357,7 @@ static int __cancel_queued(struct live_preempt_cancel *arg)
        if (err)
                goto out;
 
-       i915_gem_context_set_banned(arg->a.ctx);
+       intel_context_set_banned(rq[2]->context);
        err = intel_engine_pulse(arg->engine);
        if (err)
                goto out;
@@ -1404,13 +1404,13 @@ static int __cancel_hostile(struct live_preempt_cancel *arg)
                return 0;
 
        GEM_TRACE("%s(%s)\n", __func__, arg->engine->name);
-       clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags);
        rq = spinner_create_request(&arg->a.spin,
                                    arg->a.ctx, arg->engine,
                                    MI_NOOP); /* preemption disabled */
        if (IS_ERR(rq))
                return PTR_ERR(rq);
 
+       clear_bit(CONTEXT_BANNED, &rq->context->flags);
        i915_request_get(rq);
        i915_request_add(rq);
        if (!igt_wait_for_spinner(&arg->a.spin, rq)) {
@@ -1418,7 +1418,7 @@ static int __cancel_hostile(struct live_preempt_cancel *arg)
                goto out;
        }
 
-       i915_gem_context_set_banned(arg->a.ctx);
+       intel_context_set_banned(rq->context);
        err = intel_engine_pulse(arg->engine); /* force reset */
        if (err)
                goto out;
index 44a7d2e736a7d9f7595f526663495719668bbb63..007636221a7169e39b603f3ed63a1b947523d9e6 100644 (file)
@@ -217,7 +217,7 @@ static void guc_wq_item_append(struct intel_guc *guc,
 static void guc_add_request(struct intel_guc *guc, struct i915_request *rq)
 {
        struct intel_engine_cs *engine = rq->engine;
-       u32 ctx_desc = lower_32_bits(rq->hw_context->lrc_desc);
+       u32 ctx_desc = lower_32_bits(rq->context->lrc_desc);
        u32 ring_tail = intel_ring_set_tail(rq->ring, rq->tail) / sizeof(u64);
 
        guc_wq_item_append(guc, engine->guc_id, ctx_desc,
@@ -315,7 +315,7 @@ static void __guc_dequeue(struct intel_engine_cs *engine)
                int i;
 
                priolist_for_each_request_consume(rq, rn, p, i) {
-                       if (last && rq->hw_context != last->hw_context) {
+                       if (last && rq->context != last->context) {
                                if (port == last_port)
                                        goto done;
 
@@ -420,7 +420,7 @@ static void guc_reset(struct intel_engine_cs *engine, bool stalled)
                stalled = false;
 
        __i915_request_reset(rq, stalled);
-       intel_lr_context_reset(engine, rq->hw_context, rq->head, stalled);
+       intel_lr_context_reset(engine, rq->context, rq->head, stalled);
 
 out_unlock:
        spin_unlock_irqrestore(&engine->active.lock, flags);
index 5b2a7d072ec9b656546225b8a5db915d2c805383..228c66534e21caf9fc0ff9a53c707108d4386322 100644 (file)
@@ -59,7 +59,7 @@ static void set_context_pdp_root_pointer(
 static void update_shadow_pdps(struct intel_vgpu_workload *workload)
 {
        struct drm_i915_gem_object *ctx_obj =
-               workload->req->hw_context->state->obj;
+               workload->req->context->state->obj;
        struct execlist_ring_context *shadow_ring_context;
        struct page *page;
 
@@ -130,7 +130,7 @@ static int populate_shadow_context(struct intel_vgpu_workload *workload)
        struct intel_gvt *gvt = vgpu->gvt;
        int ring_id = workload->ring_id;
        struct drm_i915_gem_object *ctx_obj =
-               workload->req->hw_context->state->obj;
+               workload->req->context->state->obj;
        struct execlist_ring_context *shadow_ring_context;
        struct page *page;
        void *dst;
@@ -205,9 +205,9 @@ static int populate_shadow_context(struct intel_vgpu_workload *workload)
        return 0;
 }
 
-static inline bool is_gvt_request(struct i915_request *req)
+static inline bool is_gvt_request(struct i915_request *rq)
 {
-       return i915_gem_context_force_single_submission(req->gem_context);
+       return intel_context_force_single_submission(rq->context);
 }
 
 static void save_ring_hw_state(struct intel_vgpu *vgpu, int ring_id)
@@ -307,7 +307,7 @@ static int copy_workload_to_ring_buffer(struct intel_vgpu_workload *workload)
        u32 *cs;
        int err;
 
-       if (IS_GEN(req->i915, 9) && is_inhibit_context(req->hw_context))
+       if (IS_GEN(req->i915, 9) && is_inhibit_context(req->context))
                intel_vgpu_restore_inhibit_context(vgpu, req);
 
        /*
@@ -363,11 +363,10 @@ static void release_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
 }
 
 static void set_context_ppgtt_from_shadow(struct intel_vgpu_workload *workload,
-                                         struct i915_gem_context *ctx)
+                                         struct intel_context *ce)
 {
        struct intel_vgpu_mm *mm = workload->shadow_mm;
-       struct i915_ppgtt *ppgtt =
-               i915_vm_to_ppgtt(i915_gem_context_get_vm_rcu(ctx));
+       struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(ce->vm);
        int i = 0;
 
        if (mm->ppgtt_mm.root_entry_type == GTT_TYPE_PPGTT_ROOT_L4_ENTRY) {
@@ -380,8 +379,6 @@ static void set_context_ppgtt_from_shadow(struct intel_vgpu_workload *workload,
                        px_dma(pd) = mm->ppgtt_mm.shadow_pdps[i];
                }
        }
-
-       i915_vm_put(&ppgtt->vm);
 }
 
 static int
@@ -529,7 +526,7 @@ static void update_wa_ctx_2_shadow_ctx(struct intel_shadow_wa_ctx *wa_ctx)
                container_of(wa_ctx, struct intel_vgpu_workload, wa_ctx);
        struct i915_request *rq = workload->req;
        struct execlist_ring_context *shadow_ring_context =
-               (struct execlist_ring_context *)rq->hw_context->lrc_reg_state;
+               (struct execlist_ring_context *)rq->context->lrc_reg_state;
 
        shadow_ring_context->bb_per_ctx_ptr.val =
                (shadow_ring_context->bb_per_ctx_ptr.val &
@@ -628,7 +625,7 @@ static int prepare_workload(struct intel_vgpu_workload *workload)
 
        update_shadow_pdps(workload);
 
-       set_context_ppgtt_from_shadow(workload, s->shadow[ring]->gem_context);
+       set_context_ppgtt_from_shadow(workload, s->shadow[ring]);
 
        ret = intel_vgpu_sync_oos_pages(workload->vgpu);
        if (ret) {
@@ -787,7 +784,7 @@ static void update_guest_context(struct intel_vgpu_workload *workload)
        struct i915_request *rq = workload->req;
        struct intel_vgpu *vgpu = workload->vgpu;
        struct intel_gvt *gvt = vgpu->gvt;
-       struct drm_i915_gem_object *ctx_obj = rq->hw_context->state->obj;
+       struct drm_i915_gem_object *ctx_obj = rq->context->state->obj;
        struct execlist_ring_context *shadow_ring_context;
        struct page *page;
        void *src;
@@ -1232,8 +1229,6 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
        if (IS_ERR(ctx))
                return PTR_ERR(ctx);
 
-       i915_gem_context_set_force_single_submission(ctx);
-
        ppgtt = i915_vm_to_ppgtt(i915_gem_context_get_vm_rcu(ctx));
        i915_context_ppgtt_root_save(s, ppgtt);
 
@@ -1249,6 +1244,8 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
                        goto out_shadow_ctx;
                }
 
+               intel_context_set_single_submission(ce);
+
                if (!USES_GUC_SUBMISSION(i915)) { /* Max ring buffer size */
                        const unsigned int ring_size = 512 * SZ_4K;
 
index f19c678ebefcbc1e3b891c6f9920e18ba973127a..6e7010089e23823e55053a8dc215884516722527 100644 (file)
@@ -1168,19 +1168,18 @@ err_rq:
                if (!rq)
                        continue;
 
-               GEM_BUG_ON(!test_bit(CONTEXT_ALLOC_BIT,
-                                    &rq->hw_context->flags));
-               state = rq->hw_context->state;
+               GEM_BUG_ON(!test_bit(CONTEXT_ALLOC_BIT, &rq->context->flags));
+               state = rq->context->state;
                if (!state)
                        continue;
 
                /* Serialise with retirement on another CPU */
-               err = __intel_context_flush_retire(rq->hw_context);
+               err = __intel_context_flush_retire(rq->context);
                if (err)
                        goto out;
 
                /* We want to be able to unbind the state from the GGTT */
-               GEM_BUG_ON(intel_context_is_pinned(rq->hw_context));
+               GEM_BUG_ON(intel_context_is_pinned(rq->context));
 
                /*
                 * As we will hold a reference to the logical state, it will
@@ -1230,7 +1229,7 @@ out:
                if (!rq)
                        continue;
 
-               ce = rq->hw_context;
+               ce = rq->context;
                i915_request_put(rq);
                intel_context_put(ce);
        }
index 8374d50c0770124c9f5937559d1dfb946788b083..7e2cb063110cfb7f5f8de317fbef3709a0085b45 100644 (file)
@@ -1221,7 +1221,7 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
 static void record_request(const struct i915_request *request,
                           struct drm_i915_error_request *erq)
 {
-       const struct i915_gem_context *ctx = request->gem_context;
+       const struct i915_gem_context *ctx = request->context->gem_context;
 
        erq->flags = request->fence.flags;
        erq->context = request->fence.context;
@@ -1231,7 +1231,7 @@ static void record_request(const struct i915_request *request,
        erq->start = i915_ggtt_offset(request->ring->vma);
        erq->head = request->head;
        erq->tail = request->tail;
-       erq->pid = ctx->pid ? pid_nr(ctx->pid) : 0;
+       erq->pid = ctx && ctx->pid ? pid_nr(ctx->pid) : 0;
 }
 
 static void engine_record_requests(struct intel_engine_cs *engine,
@@ -1298,7 +1298,10 @@ static void error_record_engine_execlists(const struct intel_engine_cs *engine,
 static bool record_context(struct drm_i915_error_context *e,
                           const struct i915_request *rq)
 {
-       const struct i915_gem_context *ctx = rq->gem_context;
+       const struct i915_gem_context *ctx = rq->context->gem_context;
+
+       if (!ctx)
+               return false;
 
        if (ctx->pid) {
                struct task_struct *task;
@@ -1452,7 +1455,7 @@ gem_record_rings(struct i915_gpu_state *error, struct compress *compress)
                capture = request_record_user_bo(request, ee, capture);
 
                capture = capture_vma(capture,
-                                     request->hw_context->state,
+                                     request->context->state,
                                      &ee->ctx);
 
                capture = capture_vma(capture,
index 3c163a9d69a92ab621ebea6d71eb91f3113e3e77..9d4733927a2c9b30bba810307aed1f42279b3097 100644 (file)
@@ -3112,7 +3112,7 @@ static void i915_perf_enable_locked(struct i915_perf_stream *stream)
                stream->ops->enable(stream);
 
        if (stream->hold_preemption)
-               i915_gem_context_set_nopreempt(stream->ctx);
+               intel_context_set_nopreempt(stream->pinned_ctx);
 }
 
 /**
@@ -3138,7 +3138,7 @@ static void i915_perf_disable_locked(struct i915_perf_stream *stream)
        stream->enabled = false;
 
        if (stream->hold_preemption)
-               i915_gem_context_clear_nopreempt(stream->ctx);
+               intel_context_clear_nopreempt(stream->pinned_ctx);
 
        if (stream->ops->disable)
                stream->ops->disable(stream);
index 2118284b796ee166e37f5f8ffa718ced19a4b0ec..218d20d4e414e3876441a1ba3b07e080f3842b1b 100644 (file)
@@ -62,6 +62,8 @@ static const char *i915_fence_get_driver_name(struct dma_fence *fence)
 
 static const char *i915_fence_get_timeline_name(struct dma_fence *fence)
 {
+       const struct i915_gem_context *ctx;
+
        /*
         * The timeline struct (as part of the ppgtt underneath a context)
         * may be freed when the request is no longer in use by the GPU.
@@ -74,7 +76,11 @@ static const char *i915_fence_get_timeline_name(struct dma_fence *fence)
        if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
                return "signaled";
 
-       return to_request(fence)->gem_context->name ?: "[" DRIVER_NAME "]";
+       ctx = to_request(fence)->context->gem_context;
+       if (!ctx)
+               return "[" DRIVER_NAME "]";
+
+       return ctx->name;
 }
 
 static bool i915_fence_signaled(struct dma_fence *fence)
@@ -269,8 +275,8 @@ bool i915_request_retire(struct i915_request *rq)
        remove_from_client(rq);
        list_del(&rq->link);
 
-       intel_context_exit(rq->hw_context);
-       intel_context_unpin(rq->hw_context);
+       intel_context_exit(rq->context);
+       intel_context_unpin(rq->context);
 
        free_capture_list(rq);
        i915_sched_node_fini(&rq->sched);
@@ -369,7 +375,7 @@ bool __i915_request_submit(struct i915_request *request)
        if (i915_request_completed(request))
                goto xfer;
 
-       if (i915_gem_context_is_banned(request->gem_context))
+       if (intel_context_is_banned(request->context))
                i915_request_skip(request, -EIO);
 
        /*
@@ -648,8 +654,7 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp)
                goto err_free;
 
        rq->i915 = ce->engine->i915;
-       rq->hw_context = ce;
-       rq->gem_context = ce->gem_context;
+       rq->context = ce;
        rq->engine = ce->engine;
        rq->ring = ce->ring;
        rq->execution_mask = ce->engine->mask;
@@ -917,7 +922,7 @@ i915_request_await_request(struct i915_request *to, struct i915_request *from)
                                                       &from->submit,
                                                       I915_FENCE_GFP);
        } else if (intel_engine_has_semaphores(to->engine) &&
-                  to->gem_context->sched.priority >= I915_PRIORITY_NORMAL) {
+                  to->context->gem_context->sched.priority >= I915_PRIORITY_NORMAL) {
                ret = emit_semaphore_wait(to, from, I915_FENCE_GFP);
        } else {
                ret = i915_sw_fence_await_dma_fence(&to->submit,
@@ -1298,7 +1303,7 @@ void __i915_request_queue(struct i915_request *rq,
 
 void i915_request_add(struct i915_request *rq)
 {
-       struct i915_sched_attr attr = rq->gem_context->sched;
+       struct i915_sched_attr attr = rq->context->gem_context->sched;
        struct intel_timeline * const tl = i915_request_timeline(rq);
        struct i915_request *prev;
 
index c18c0bcd0193c25d8989a572d6274ac9e5669341..0e4fe3205ce74c58af52234fff1dcd7d657152c0 100644 (file)
@@ -115,9 +115,8 @@ struct i915_request {
         * i915_request_free() will then decrement the refcount on the
         * context.
         */
-       struct i915_gem_context *gem_context;
        struct intel_engine_cs *engine;
-       struct intel_context *hw_context;
+       struct intel_context *context;
        struct intel_ring *ring;
        struct intel_timeline __rcu *timeline;
        struct list_head signal_link;
index 2bc2aa46a1b92d32123af6c989ae7d916fb80a01..bf87c70bfdd9630b50af237503e5389c478e2715 100644 (file)
@@ -213,7 +213,7 @@ static void kick_submission(struct intel_engine_cs *engine,
         * If we are already the currently executing context, don't
         * bother evaluating if we should preempt ourselves.
         */
-       if (inflight->hw_context == rq->hw_context)
+       if (inflight->context == rq->context)
                goto unlock;
 
        engine->execlists.queue_priority_hint = prio;