const struct intel_engine_cs *engine,
u32 head);
+static int lrc_ring_mi_mode(const struct intel_engine_cs *engine)
+{
+ if (INTEL_GEN(engine->i915) >= 12)
+ return 0x60;
+ else if (INTEL_GEN(engine->i915) >= 9)
+ return 0x54;
+ else if (engine->class == RENDER_CLASS)
+ return 0x58;
+ else
+ return -1;
+}
+
+static int lrc_ring_gpr0(const struct intel_engine_cs *engine)
+{
+ if (INTEL_GEN(engine->i915) >= 12)
+ return 0x74;
+ else if (INTEL_GEN(engine->i915) >= 9)
+ return 0x68;
+ else if (engine->class == RENDER_CLASS)
+ return 0xd8;
+ else
+ return -1;
+}
+
+static int lrc_ring_wa_bb_per_ctx(const struct intel_engine_cs *engine)
+{
+ if (INTEL_GEN(engine->i915) >= 12)
+ return 0x12;
+ else if (INTEL_GEN(engine->i915) >= 9 || engine->class == RENDER_CLASS)
+ return 0x18;
+ else
+ return -1;
+}
+
+static int lrc_ring_indirect_ptr(const struct intel_engine_cs *engine)
+{
+ int x;
+
+ x = lrc_ring_wa_bb_per_ctx(engine);
+ if (x < 0)
+ return x;
+
+ return x + 2;
+}
+
+static int lrc_ring_indirect_offset(const struct intel_engine_cs *engine)
+{
+ int x;
+
+ x = lrc_ring_indirect_ptr(engine);
+ if (x < 0)
+ return x;
+
+ return x + 2;
+}
+
+static int lrc_ring_cmd_buf_cctl(const struct intel_engine_cs *engine)
+{
+ if (engine->class != RENDER_CLASS)
+ return -1;
+
+ if (INTEL_GEN(engine->i915) >= 12)
+ return 0xb6;
+ else if (INTEL_GEN(engine->i915) >= 11)
+ return 0xaa;
+ else
+ return -1;
+}
+
+static u32
+lrc_ring_indirect_offset_default(const struct intel_engine_cs *engine)
+{
+ switch (INTEL_GEN(engine->i915)) {
+ default:
+ MISSING_CASE(INTEL_GEN(engine->i915));
+ fallthrough;
+ case 12:
+ return GEN12_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
+ case 11:
+ return GEN11_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
+ case 10:
+ return GEN10_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
+ case 9:
+ return GEN9_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
+ case 8:
+ return GEN8_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
+ }
+}
+
+static void
+lrc_ring_setup_indirect_ctx(u32 *regs,
+ const struct intel_engine_cs *engine,
+ u32 ctx_bb_ggtt_addr,
+ u32 size)
+{
+ GEM_BUG_ON(!size);
+ GEM_BUG_ON(!IS_ALIGNED(size, CACHELINE_BYTES));
+ GEM_BUG_ON(lrc_ring_indirect_ptr(engine) == -1);
+ regs[lrc_ring_indirect_ptr(engine) + 1] =
+ ctx_bb_ggtt_addr | (size / CACHELINE_BYTES);
+
+ GEM_BUG_ON(lrc_ring_indirect_offset(engine) == -1);
+ regs[lrc_ring_indirect_offset(engine) + 1] =
+ lrc_ring_indirect_offset_default(engine) << 6;
+}
+
+static u32 intel_context_get_runtime(const struct intel_context *ce)
+{
+ /*
+ * We can use either ppHWSP[16] which is recorded before the context
+ * switch (and so excludes the cost of context switches) or use the
+ * value from the context image itself, which is saved/restored earlier
+ * and so includes the cost of the save.
+ */
+ return READ_ONCE(ce->lrc_reg_state[CTX_TIMESTAMP]);
+}
+
static void mark_eio(struct i915_request *rq)
{
if (i915_request_completed(rq))
#define REG16(x) \
(((x) >> 9) | BIT(7) | BUILD_BUG_ON_ZERO(x >= 0x10000)), \
(((x) >> 2) & 0x7f)
-#define END(x) 0, (x)
+#define END(total_state_size) 0, (total_state_size)
{
const u32 base = engine->mmio_base;
if (flags & POSTED)
*regs |= MI_LRI_FORCE_POSTED;
if (INTEL_GEN(engine->i915) >= 11)
- *regs |= MI_LRI_CS_MMIO;
+ *regs |= MI_LRI_LRM_CS_MMIO;
regs++;
GEM_BUG_ON(!count);
NOP(6),
LRI(1, 0),
REG(0x0c8),
+ NOP(3 + 9 + 1),
+
+ LRI(51, POSTED),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG(0x028),
+ REG(0x09c),
+ REG(0x0c0),
+ REG(0x178),
+ REG(0x17c),
+ REG16(0x358),
+ REG(0x170),
+ REG(0x150),
+ REG(0x154),
+ REG(0x158),
+ REG16(0x41c),
+ REG16(0x600),
+ REG16(0x604),
+ REG16(0x608),
+ REG16(0x60c),
+ REG16(0x610),
+ REG16(0x614),
+ REG16(0x618),
+ REG16(0x61c),
+ REG16(0x620),
+ REG16(0x624),
+ REG16(0x628),
+ REG16(0x62c),
+ REG16(0x630),
+ REG16(0x634),
+ REG16(0x638),
+ REG16(0x63c),
+ REG16(0x640),
+ REG16(0x644),
+ REG16(0x648),
+ REG16(0x64c),
+ REG16(0x650),
+ REG16(0x654),
+ REG16(0x658),
+ REG16(0x65c),
+ REG16(0x660),
+ REG16(0x664),
+ REG16(0x668),
+ REG16(0x66c),
+ REG16(0x670),
+ REG16(0x674),
+ REG16(0x678),
+ REG16(0x67c),
+ REG(0x068),
+ REG(0x084),
+ NOP(1),
- END(80)
+ END(192)
};
#undef END
write_sequnlock_irqrestore(&engine->stats.lock, flags);
}
-static int lrc_ring_mi_mode(const struct intel_engine_cs *engine)
-{
- if (INTEL_GEN(engine->i915) >= 12)
- return 0x60;
- else if (INTEL_GEN(engine->i915) >= 9)
- return 0x54;
- else if (engine->class == RENDER_CLASS)
- return 0x58;
- else
- return -1;
-}
-
static void
execlists_check_context(const struct intel_context *ce,
const struct intel_engine_cs *engine)
if (engine->pinned_default_state)
memcpy(regs, /* skip restoring the vanilla PPHWSP */
- engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE,
+ engine->pinned_default_state + LRC_STATE_OFFSET,
engine->context_size - PAGE_SIZE);
execlists_init_reg_state(regs, ce, engine, ce->ring, false);
+ ce->runtime.last = intel_context_get_runtime(ce);
}
static void reset_active(struct i915_request *rq,
ce->lrc_desc |= CTX_DESC_FORCE_RESTORE;
}
-static u32 intel_context_get_runtime(const struct intel_context *ce)
-{
- /*
- * We can use either ppHWSP[16] which is recorded before the context
- * switch (and so excludes the cost of context switches) or use the
- * value from the context image itself, which is saved/restored earlier
- * and so includes the cost of the save.
- */
- return READ_ONCE(ce->lrc_reg_state[CTX_TIMESTAMP]);
-}
-
static void st_update_runtime_underflow(struct intel_context *ce, s32 dt)
{
#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
}
}
+static __maybe_unused char *
+dump_port(char *buf, int buflen, const char *prefix, struct i915_request *rq)
+{
+ if (!rq)
+ return "";
+
+ snprintf(buf, buflen, "%s%llx:%lld%s prio %d",
+ prefix,
+ rq->fence.context, rq->fence.seqno,
+ i915_request_completed(rq) ? "!" :
+ i915_request_started(rq) ? "*" :
+ "",
+ rq_prio(rq));
+
+ return buf;
+}
+
static __maybe_unused void
trace_ports(const struct intel_engine_execlists *execlists,
const char *msg,
{
const struct intel_engine_cs *engine =
container_of(execlists, typeof(*engine), execlists);
+ char __maybe_unused p0[40], p1[40];
if (!ports[0])
return;
- ENGINE_TRACE(engine, "%s { %llx:%lld%s, %llx:%lld }\n", msg,
- ports[0]->fence.context,
- ports[0]->fence.seqno,
- i915_request_completed(ports[0]) ? "!" :
- i915_request_started(ports[0]) ? "*" :
- "",
- ports[1] ? ports[1]->fence.context : 0,
- ports[1] ? ports[1]->fence.seqno : 0);
+ ENGINE_TRACE(engine, "%s { %s%s }\n", msg,
+ dump_port(p0, sizeof(p0), "", ports[0]),
+ dump_port(p1, sizeof(p1), ", ", ports[1]));
}
static inline bool
}
static bool
-need_timeslice(struct intel_engine_cs *engine, const struct i915_request *rq)
+need_timeslice(const struct intel_engine_cs *engine,
+ const struct i915_request *rq)
{
int hint;
return hint >= effective_prio(rq);
}
+static bool
+timeslice_yield(const struct intel_engine_execlists *el,
+ const struct i915_request *rq)
+{
+ /*
+ * Once bitten, forever smitten!
+ *
+ * If the active context ever busy-waited on a semaphore,
+ * it will be treated as a hog until the end of its timeslice (i.e.
+ * until it is scheduled out and replaced by a new submission,
+ * possibly even its own lite-restore). The HW only sends an interrupt
+ * on the first miss, and we do know if that semaphore has been
+ * signaled, or even if it is now stuck on another semaphore. Play
+ * safe, yield if it might be stuck -- it will be given a fresh
+ * timeslice in the near future.
+ */
+ return upper_32_bits(rq->context->lrc_desc) == READ_ONCE(el->yield);
+}
+
+static bool
+timeslice_expired(const struct intel_engine_execlists *el,
+ const struct i915_request *rq)
+{
+ return timer_expired(&el->timer) || timeslice_yield(el, rq);
+}
+
static int
switch_prio(struct intel_engine_cs *engine, const struct i915_request *rq)
{
return READ_ONCE(engine->props.timeslice_duration_ms);
}
-static unsigned long
-active_timeslice(const struct intel_engine_cs *engine)
+static unsigned long active_timeslice(const struct intel_engine_cs *engine)
{
const struct intel_engine_execlists *execlists = &engine->execlists;
const struct i915_request *rq = *execlists->active;
static void set_timeslice(struct intel_engine_cs *engine)
{
+ unsigned long duration;
+
if (!intel_engine_has_timeslices(engine))
return;
- set_timer_ms(&engine->execlists.timer, active_timeslice(engine));
+ duration = active_timeslice(engine);
+ ENGINE_TRACE(engine, "bump timeslicing, interval:%lu", duration);
+
+ set_timer_ms(&engine->execlists.timer, duration);
}
static void start_timeslice(struct intel_engine_cs *engine)
{
struct intel_engine_execlists *execlists = &engine->execlists;
- int prio = queue_prio(execlists);
+ const int prio = queue_prio(execlists);
+ unsigned long duration;
+
+ if (!intel_engine_has_timeslices(engine))
+ return;
WRITE_ONCE(execlists->switch_priority_hint, prio);
if (prio == INT_MIN)
if (timer_pending(&execlists->timer))
return;
- set_timer_ms(&execlists->timer, timeslice(engine));
+ duration = timeslice(engine);
+ ENGINE_TRACE(engine,
+ "start timeslicing, prio:%d, interval:%lu",
+ prio, duration);
+
+ set_timer_ms(&execlists->timer, duration);
}
static void record_preemption(struct intel_engine_execlists *execlists)
* of trouble.
*/
active = READ_ONCE(execlists->active);
- while ((last = *active) && i915_request_completed(last))
- active++;
- if (last) {
+ /*
+ * In theory we can skip over completed contexts that have not
+ * yet been processed by events (as those events are in flight):
+ *
+ * while ((last = *active) && i915_request_completed(last))
+ * active++;
+ *
+ * However, the GPU cannot handle this as it will ultimately
+ * find itself trying to jump back into a context it has just
+ * completed and barf.
+ */
+
+ if ((last = *active)) {
if (need_preempt(engine, last, rb)) {
+ if (i915_request_completed(last)) {
+ tasklet_hi_schedule(&execlists->tasklet);
+ return;
+ }
+
ENGINE_TRACE(engine,
"preempting last=%llx:%lld, prio=%d, hint=%d\n",
last->fence.context,
last = NULL;
} else if (need_timeslice(engine, last) &&
- timer_expired(&engine->execlists.timer)) {
+ timeslice_expired(execlists, last)) {
+ if (i915_request_completed(last)) {
+ tasklet_hi_schedule(&execlists->tasklet);
+ return;
+ }
+
ENGINE_TRACE(engine,
- "expired last=%llx:%lld, prio=%d, hint=%d\n",
+ "expired last=%llx:%lld, prio=%d, hint=%d, yield?=%s\n",
last->fence.context,
last->fence.seqno,
last->sched.attr.priority,
- execlists->queue_priority_hint);
+ execlists->queue_priority_hint,
+ yesno(timeslice_yield(execlists, last)));
ring_set_paused(engine, 1);
defer_active(engine);
}
clear_ports(port + 1, last_port - port);
+ WRITE_ONCE(execlists->yield, -1);
execlists_submit_ports(engine);
set_preempt_timeout(engine, *active);
} else {
if (promote) {
struct i915_request * const *old = execlists->active;
- GEM_BUG_ON(!assert_pending_valid(execlists, "promote"));
-
ring_set_paused(engine, 0);
/* Point active to the new ELSP; prevent overwriting */
execlists_schedule_out(*old++);
/* switch pending to inflight */
+ GEM_BUG_ON(!assert_pending_valid(execlists, "promote"));
memcpy(execlists->inflight,
execlists->pending,
execlists_num_ports(execlists) *
* We rely on the hardware being strongly
* ordered, that the breadcrumb write is
* coherent (visible from the CPU) before the
- * user interrupt and CSB is processed.
+ * user interrupt is processed. One might assume
+ * that the breadcrumb write being before the
+ * user interrupt and the CS event for the context
+ * switch would therefore be before the CS event
+ * itself...
*/
if (GEM_SHOW_DEBUG() &&
- !i915_request_completed(*execlists->active) &&
- !reset_in_progress(execlists)) {
- struct i915_request *rq __maybe_unused =
- *execlists->active;
+ !i915_request_completed(*execlists->active)) {
+ struct i915_request *rq = *execlists->active;
const u32 *regs __maybe_unused =
rq->context->lrc_reg_state;
+ ENGINE_TRACE(engine,
+ "context completed before request!\n");
ENGINE_TRACE(engine,
"ring:{start:0x%08x, head:%04x, tail:%04x, ctl:%08x, mode:%08x}\n",
ENGINE_READ(engine, RING_START),
regs[CTX_RING_START],
regs[CTX_RING_HEAD],
regs[CTX_RING_TAIL]);
-
- GEM_BUG_ON("context completed before request");
}
execlists_schedule_out(*execlists->active++);
return NULL;
}
+static struct i915_request *
+active_context(struct intel_engine_cs *engine, u32 ccid)
+{
+ const struct intel_engine_execlists * const el = &engine->execlists;
+ struct i915_request * const *port, *rq;
+
+ /*
+ * Use the most recent result from process_csb(), but just in case
+ * we trigger an error (via interrupt) before the first CS event has
+ * been written, peek at the next submission.
+ */
+
+ for (port = el->active; (rq = *port); port++) {
+ if (upper_32_bits(rq->context->lrc_desc) == ccid) {
+ ENGINE_TRACE(engine,
+ "ccid found at active:%zd\n",
+ port - el->active);
+ return rq;
+ }
+ }
+
+ for (port = el->pending; (rq = *port); port++) {
+ if (upper_32_bits(rq->context->lrc_desc) == ccid) {
+ ENGINE_TRACE(engine,
+ "ccid found at pending:%zd\n",
+ port - el->pending);
+ return rq;
+ }
+ }
+
+ ENGINE_TRACE(engine, "ccid:%x not found\n", ccid);
+ return NULL;
+}
+
+static u32 active_ccid(struct intel_engine_cs *engine)
+{
+ return ENGINE_READ_FW(engine, RING_EXECLIST_STATUS_HI);
+}
+
static bool execlists_capture(struct intel_engine_cs *engine)
{
struct execlists_capture *cap;
return true;
spin_lock_irq(&engine->active.lock);
- cap->rq = execlists_active(&engine->execlists);
+ cap->rq = active_context(engine, active_ccid(engine));
if (cap->rq) {
cap->rq = active_request(cap->rq->context->timeline, cap->rq);
cap->rq = i915_request_get_rcu(cap->rq);
if (reset_in_progress(execlists))
return; /* defer until we restart the engine following reset */
- if (execlists->tasklet.func == execlists_submission_tasklet)
- __execlists_submission_tasklet(engine);
- else
- tasklet_hi_schedule(&execlists->tasklet);
+ /* Hopefully we clear execlists->pending[] to let us through */
+ if (READ_ONCE(execlists->pending[0]) &&
+ tasklet_trylock(&execlists->tasklet)) {
+ process_csb(engine);
+ tasklet_unlock(&execlists->tasklet);
+ }
+
+ __execlists_submission_tasklet(engine);
}
static void submit_queue(struct intel_engine_cs *engine,
vaddr += engine->context_size;
if (memchr_inv(vaddr, CONTEXT_REDZONE, I915_GTT_PAGE_SIZE))
- dev_err_once(engine->i915->drm.dev,
+ drm_err_once(&engine->i915->drm,
"%s context redzone overwritten!\n",
engine->name);
}
static void execlists_context_unpin(struct intel_context *ce)
{
- check_redzone((void *)ce->lrc_reg_state - LRC_STATE_PN * PAGE_SIZE,
+ check_redzone((void *)ce->lrc_reg_state - LRC_STATE_OFFSET,
ce->engine);
i915_gem_object_unpin_map(ce->state->obj);
}
+static u32 *
+gen12_emit_timestamp_wa(const struct intel_context *ce, u32 *cs)
+{
+ *cs++ = MI_LOAD_REGISTER_MEM_GEN8 |
+ MI_SRM_LRM_GLOBAL_GTT |
+ MI_LRI_LRM_CS_MMIO;
+ *cs++ = i915_mmio_reg_offset(GEN8_RING_CS_GPR(0, 0));
+ *cs++ = i915_ggtt_offset(ce->state) + LRC_STATE_OFFSET +
+ CTX_TIMESTAMP * sizeof(u32);
+ *cs++ = 0;
+
+ *cs++ = MI_LOAD_REGISTER_REG |
+ MI_LRR_SOURCE_CS_MMIO |
+ MI_LRI_LRM_CS_MMIO;
+ *cs++ = i915_mmio_reg_offset(GEN8_RING_CS_GPR(0, 0));
+ *cs++ = i915_mmio_reg_offset(RING_CTX_TIMESTAMP(0));
+
+ *cs++ = MI_LOAD_REGISTER_REG |
+ MI_LRR_SOURCE_CS_MMIO |
+ MI_LRI_LRM_CS_MMIO;
+ *cs++ = i915_mmio_reg_offset(GEN8_RING_CS_GPR(0, 0));
+ *cs++ = i915_mmio_reg_offset(RING_CTX_TIMESTAMP(0));
+
+ return cs;
+}
+
+static u32 *
+gen12_emit_restore_scratch(const struct intel_context *ce, u32 *cs)
+{
+ GEM_BUG_ON(lrc_ring_gpr0(ce->engine) == -1);
+
+ *cs++ = MI_LOAD_REGISTER_MEM_GEN8 |
+ MI_SRM_LRM_GLOBAL_GTT |
+ MI_LRI_LRM_CS_MMIO;
+ *cs++ = i915_mmio_reg_offset(GEN8_RING_CS_GPR(0, 0));
+ *cs++ = i915_ggtt_offset(ce->state) + LRC_STATE_OFFSET +
+ (lrc_ring_gpr0(ce->engine) + 1) * sizeof(u32);
+ *cs++ = 0;
+
+ return cs;
+}
+
+static u32 *
+gen12_emit_cmd_buf_wa(const struct intel_context *ce, u32 *cs)
+{
+ GEM_BUG_ON(lrc_ring_cmd_buf_cctl(ce->engine) == -1);
+
+ *cs++ = MI_LOAD_REGISTER_MEM_GEN8 |
+ MI_SRM_LRM_GLOBAL_GTT |
+ MI_LRI_LRM_CS_MMIO;
+ *cs++ = i915_mmio_reg_offset(GEN8_RING_CS_GPR(0, 0));
+ *cs++ = i915_ggtt_offset(ce->state) + LRC_STATE_OFFSET +
+ (lrc_ring_cmd_buf_cctl(ce->engine) + 1) * sizeof(u32);
+ *cs++ = 0;
+
+ *cs++ = MI_LOAD_REGISTER_REG |
+ MI_LRR_SOURCE_CS_MMIO |
+ MI_LRI_LRM_CS_MMIO;
+ *cs++ = i915_mmio_reg_offset(GEN8_RING_CS_GPR(0, 0));
+ *cs++ = i915_mmio_reg_offset(RING_CMD_BUF_CCTL(0));
+
+ return cs;
+}
+
+static u32 *
+gen12_emit_indirect_ctx_rcs(const struct intel_context *ce, u32 *cs)
+{
+ cs = gen12_emit_timestamp_wa(ce, cs);
+ cs = gen12_emit_cmd_buf_wa(ce, cs);
+ cs = gen12_emit_restore_scratch(ce, cs);
+
+ return cs;
+}
+
+static u32 *
+gen12_emit_indirect_ctx_xcs(const struct intel_context *ce, u32 *cs)
+{
+ cs = gen12_emit_timestamp_wa(ce, cs);
+ cs = gen12_emit_restore_scratch(ce, cs);
+
+ return cs;
+}
+
+static inline u32 context_wa_bb_offset(const struct intel_context *ce)
+{
+ return PAGE_SIZE * ce->wa_bb_page;
+}
+
+static u32 *context_indirect_bb(const struct intel_context *ce)
+{
+ void *ptr;
+
+ GEM_BUG_ON(!ce->wa_bb_page);
+
+ ptr = ce->lrc_reg_state;
+ ptr -= LRC_STATE_OFFSET; /* back to start of context image */
+ ptr += context_wa_bb_offset(ce);
+
+ return ptr;
+}
+
+static void
+setup_indirect_ctx_bb(const struct intel_context *ce,
+ const struct intel_engine_cs *engine,
+ u32 *(*emit)(const struct intel_context *, u32 *))
+{
+ u32 * const start = context_indirect_bb(ce);
+ u32 *cs;
+
+ cs = emit(ce, start);
+ GEM_BUG_ON(cs - start > I915_GTT_PAGE_SIZE / sizeof(*cs));
+ while ((unsigned long)cs % CACHELINE_BYTES)
+ *cs++ = MI_NOOP;
+
+ lrc_ring_setup_indirect_ctx(ce->lrc_reg_state, engine,
+ i915_ggtt_offset(ce->state) +
+ context_wa_bb_offset(ce),
+ (cs - start) * sizeof(*cs));
+}
+
static void
__execlists_update_reg_state(const struct intel_context *ce,
const struct intel_engine_cs *engine,
i915_oa_init_reg_state(ce, engine);
}
+
+ if (ce->wa_bb_page) {
+ u32 *(*fn)(const struct intel_context *ce, u32 *cs);
+
+ fn = gen12_emit_indirect_ctx_xcs;
+ if (ce->engine->class == RENDER_CLASS)
+ fn = gen12_emit_indirect_ctx_rcs;
+
+ /* Mutually exclusive wrt to global indirect bb */
+ GEM_BUG_ON(engine->wa_ctx.indirect_ctx.size);
+ setup_indirect_ctx_bb(ce, engine, fn);
+ }
}
static int
return PTR_ERR(vaddr);
ce->lrc_desc = lrc_descriptor(ce, engine) | CTX_DESC_FORCE_RESTORE;
- ce->lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE;
+ ce->lrc_reg_state = vaddr + LRC_STATE_OFFSET;
__execlists_update_reg_state(ce, engine, ce->ring->tail);
return 0;
ret = lrc_setup_wa_ctx(engine);
if (ret) {
- DRM_DEBUG_DRIVER("Failed to setup context WA page: %d\n", ret);
+ drm_dbg(&engine->i915->drm,
+ "Failed to setup context WA page: %d\n", ret);
return ret;
}
return ret;
}
+static void reset_csb_pointers(struct intel_engine_cs *engine)
+{
+ struct intel_engine_execlists * const execlists = &engine->execlists;
+ const unsigned int reset_value = execlists->csb_size - 1;
+
+ ring_set_paused(engine, 0);
+
+ /*
+ * After a reset, the HW starts writing into CSB entry [0]. We
+ * therefore have to set our HEAD pointer back one entry so that
+ * the *first* entry we check is entry 0. To complicate this further,
+ * as we don't wait for the first interrupt after reset, we have to
+ * fake the HW write to point back to the last entry so that our
+ * inline comparison of our cached head position against the last HW
+ * write works even before the first interrupt.
+ */
+ execlists->csb_head = reset_value;
+ WRITE_ONCE(*execlists->csb_write, reset_value);
+ wmb(); /* Make sure this is visible to HW (paranoia?) */
+
+ /*
+ * Sometimes Icelake forgets to reset its pointers on a GPU reset.
+ * Bludgeon them with a mmio update to be sure.
+ */
+ ENGINE_WRITE(engine, RING_CONTEXT_STATUS_PTR,
+ reset_value << 8 | reset_value);
+ ENGINE_POSTING_READ(engine, RING_CONTEXT_STATUS_PTR);
+
+ invalidate_csb_entries(&execlists->csb_status[0],
+ &execlists->csb_status[reset_value]);
+}
+
+static void execlists_sanitize(struct intel_engine_cs *engine)
+{
+ /*
+ * Poison residual state on resume, in case the suspend didn't!
+ *
+ * We have to assume that across suspend/resume (or other loss
+ * of control) that the contents of our pinned buffers has been
+ * lost, replaced by garbage. Since this doesn't always happen,
+ * let's poison such state so that we more quickly spot when
+ * we falsely assume it has been preserved.
+ */
+ if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
+ memset(engine->status_page.addr, POISON_INUSE, PAGE_SIZE);
+
+ reset_csb_pointers(engine);
+
+ /*
+ * The kernel_context HWSP is stored in the status_page. As above,
+ * that may be lost on resume/initialisation, and so we need to
+ * reset the value in the HWSP.
+ */
+ intel_timeline_reset_seqno(engine->kernel_context->timeline);
+}
+
static void enable_error_interrupt(struct intel_engine_cs *engine)
{
u32 status;
status = ENGINE_READ(engine, RING_ESR);
if (unlikely(status)) {
- dev_err(engine->i915->drm.dev,
+ drm_err(&engine->i915->drm,
"engine '%s' resumed still in error: %08x\n",
engine->name, status);
__intel_gt_reset(engine->gt, engine->mask);
bool unexpected = false;
if (ENGINE_READ_FW(engine, RING_MI_MODE) & STOP_RING) {
- DRM_DEBUG_DRIVER("STOP_RING still set in RING_MI_MODE\n");
+ drm_dbg(&engine->i915->drm,
+ "STOP_RING still set in RING_MI_MODE\n");
unexpected = true;
}
*
* FIXME: Wa for more modern gens needs to be validated
*/
+ ring_set_paused(engine, 1);
intel_engine_stop_cs(engine);
}
-static void reset_csb_pointers(struct intel_engine_cs *engine)
-{
- struct intel_engine_execlists * const execlists = &engine->execlists;
- const unsigned int reset_value = execlists->csb_size - 1;
-
- ring_set_paused(engine, 0);
-
- /*
- * After a reset, the HW starts writing into CSB entry [0]. We
- * therefore have to set our HEAD pointer back one entry so that
- * the *first* entry we check is entry 0. To complicate this further,
- * as we don't wait for the first interrupt after reset, we have to
- * fake the HW write to point back to the last entry so that our
- * inline comparison of our cached head position against the last HW
- * write works even before the first interrupt.
- */
- execlists->csb_head = reset_value;
- WRITE_ONCE(*execlists->csb_write, reset_value);
- wmb(); /* Make sure this is visible to HW (paranoia?) */
-
- /*
- * Sometimes Icelake forgets to reset its pointers on a GPU reset.
- * Bludgeon them with a mmio update to be sure.
- */
- ENGINE_WRITE(engine, RING_CONTEXT_STATUS_PTR,
- reset_value << 8 | reset_value);
- ENGINE_POSTING_READ(engine, RING_CONTEXT_STATUS_PTR);
-
- invalidate_csb_entries(&execlists->csb_status[0],
- &execlists->csb_status[reset_value]);
-}
-
static void __reset_stop_ring(u32 *regs, const struct intel_engine_cs *engine)
{
int x;
static void execlists_release(struct intel_engine_cs *engine)
{
+ engine->sanitize = NULL; /* no longer in control, nothing to sanitize */
+
execlists_shutdown(engine);
intel_engine_cleanup_common(engine);
engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT << shift;
engine->irq_keep_mask = GT_CONTEXT_SWITCH_INTERRUPT << shift;
engine->irq_keep_mask |= GT_CS_MASTER_ERROR_INTERRUPT << shift;
+ engine->irq_keep_mask |= GT_WAIT_SEMAPHORE_INTERRUPT << shift;
}
static void rcs_submission_override(struct intel_engine_cs *engine)
* because we only expect rare glitches but nothing
* critical to prevent us from using GPU
*/
- DRM_ERROR("WA batch buffer initialization failed\n");
+ drm_err(&i915->drm, "WA batch buffer initialization failed\n");
if (HAS_LOGICAL_RING_ELSQ(i915)) {
execlists->submit_reg = uncore->regs +
else
execlists->csb_size = GEN11_CSB_ENTRIES;
- reset_csb_pointers(engine);
-
/* Finally, take ownership and responsibility for cleanup! */
+ engine->sanitize = execlists_sanitize;
engine->release = execlists_release;
return 0;
}
-static u32 intel_lr_indirect_ctx_offset(const struct intel_engine_cs *engine)
-{
- u32 indirect_ctx_offset;
-
- switch (INTEL_GEN(engine->i915)) {
- default:
- MISSING_CASE(INTEL_GEN(engine->i915));
- /* fall through */
- case 12:
- indirect_ctx_offset =
- GEN12_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
- break;
- case 11:
- indirect_ctx_offset =
- GEN11_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
- break;
- case 10:
- indirect_ctx_offset =
- GEN10_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
- break;
- case 9:
- indirect_ctx_offset =
- GEN9_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
- break;
- case 8:
- indirect_ctx_offset =
- GEN8_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT;
- break;
- }
-
- return indirect_ctx_offset;
-}
-
-
static void init_common_reg_state(u32 * const regs,
const struct intel_engine_cs *engine,
const struct intel_ring *ring,
regs[CTX_CONTEXT_CONTROL] = ctl;
regs[CTX_RING_CTL] = RING_CTL_SIZE(ring->size) | RING_VALID;
+ regs[CTX_TIMESTAMP] = 0;
}
static void init_wa_bb_reg_state(u32 * const regs,
- const struct intel_engine_cs *engine,
- u32 pos_bb_per_ctx)
+ const struct intel_engine_cs *engine)
{
const struct i915_ctx_workarounds * const wa_ctx = &engine->wa_ctx;
if (wa_ctx->per_ctx.size) {
const u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
- regs[pos_bb_per_ctx] =
+ GEM_BUG_ON(lrc_ring_wa_bb_per_ctx(engine) == -1);
+ regs[lrc_ring_wa_bb_per_ctx(engine) + 1] =
(ggtt_offset + wa_ctx->per_ctx.offset) | 0x01;
}
if (wa_ctx->indirect_ctx.size) {
- const u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
-
- regs[pos_bb_per_ctx + 2] =
- (ggtt_offset + wa_ctx->indirect_ctx.offset) |
- (wa_ctx->indirect_ctx.size / CACHELINE_BYTES);
-
- regs[pos_bb_per_ctx + 4] =
- intel_lr_indirect_ctx_offset(engine) << 6;
+ lrc_ring_setup_indirect_ctx(regs, engine,
+ i915_ggtt_offset(wa_ctx->vma) +
+ wa_ctx->indirect_ctx.offset,
+ wa_ctx->indirect_ctx.size);
}
}
init_common_reg_state(regs, engine, ring, inhibit);
init_ppgtt_reg_state(regs, vm_alias(ce->vm));
- init_wa_bb_reg_state(regs, engine,
- INTEL_GEN(engine->i915) >= 12 ?
- GEN12_CTX_BB_PER_CTX_PTR :
- CTX_BB_PER_CTX_PTR);
+ init_wa_bb_reg_state(regs, engine);
__reset_stop_ring(regs, engine);
}
vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB);
if (IS_ERR(vaddr)) {
ret = PTR_ERR(vaddr);
- DRM_DEBUG_DRIVER("Could not map object pages! (%d)\n", ret);
+ drm_dbg(&engine->i915->drm,
+ "Could not map object pages! (%d)\n", ret);
return ret;
}
* The second page of the context object contains some registers which
* must be set up prior to the first execution.
*/
- execlists_init_reg_state(vaddr + LRC_STATE_PN * PAGE_SIZE,
+ execlists_init_reg_state(vaddr + LRC_STATE_OFFSET,
ce, engine, ring, inhibit);
ret = 0;
if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
context_size += I915_GTT_PAGE_SIZE; /* for redzone */
+ if (INTEL_GEN(engine->i915) == 12) {
+ ce->wa_bb_page = context_size / PAGE_SIZE;
+ context_size += PAGE_SIZE;
+ }
+
ctx_obj = i915_gem_object_create_shmem(engine->i915, context_size);
if (IS_ERR(ctx_obj))
return PTR_ERR(ctx_obj);
ret = populate_lr_context(ce, ctx_obj, engine, ring);
if (ret) {
- DRM_DEBUG_DRIVER("Failed to populate LRC: %d\n", ret);
+ drm_dbg(&engine->i915->drm,
+ "Failed to populate LRC: %d\n", ret);
goto error_ring_free;
}
__execlists_context_fini(&ve->context);
intel_context_fini(&ve->context);
+ intel_engine_free_request_pool(&ve->base);
+
kfree(ve->bonds);
kfree(ve);
}
return;
local_irq_disable();
- for (n = 0; READ_ONCE(ve->request) && n < ve->num_siblings; n++) {
- struct intel_engine_cs *sibling = ve->siblings[n];
+ for (n = 0; n < ve->num_siblings; n++) {
+ struct intel_engine_cs *sibling = READ_ONCE(ve->siblings[n]);
struct ve_node * const node = &ve->nodes[sibling->id];
struct rb_node **parent, *rb;
bool first;
+ if (!READ_ONCE(ve->request))
+ break; /* already handled by a sibling's tasklet */
+
if (unlikely(!(mask & sibling->mask))) {
if (!RB_EMPTY_NODE(&node->rb)) {
spin_lock(&sibling->active.lock);
submit_engine:
GEM_BUG_ON(RB_EMPTY_NODE(&node->rb));
node->prio = prio;
- if (first && prio > sibling->execlists.queue_priority_hint) {
- sibling->execlists.queue_priority_hint = prio;
+ if (first && prio > sibling->execlists.queue_priority_hint)
tasklet_hi_schedule(&sibling->execlists.tasklet);
- }
spin_unlock(&sibling->active.lock);
}