xiaomi-sm8450/android_kernel_xiaomi_sm8450
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

drm/i915: Pull scheduling under standalone lock

Browse Source 
Currently, the backend scheduling code abuses struct_mutex into order to
have a global lock to manipulate a temporary list (without widespread
allocation) and to protect against list modifications. This is an
extraneous coupling to struct_mutex and further can not extend beyond
the local device.

Pull all the code that needs to be under the one true lock into
i915_scheduler.c, and make it so.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181001144755.7978-2-chris@chris-wilson.co.uk
This commit is contained in:
Chris Wilson
2018-10-01 15:47:54 +01:00
parent b16c765122
commit e2f3496e93
8 changed files with 411 additions and 361 deletions
Download Patch File Download Diff File Expand all files Collapse all files

268
drivers/gpu/drm/i915/intel_lrc.c
View File

@@ -259,102 +259,6 @@ intel_lr_context_descriptor_update(struct i915_gem_context *ctx,
ce->lrc_desc = desc;
}
static void assert_priolists(struct intel_engine_execlists * const execlists,
long queue_priority)
{
struct rb_node *rb;
long last_prio, i;
if (!IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
return;
GEM_BUG_ON(rb_first_cached(&execlists->queue) !=
rb_first(&execlists->queue.rb_root));
last_prio = (queue_priority >> I915_USER_PRIORITY_SHIFT) + 1;
for (rb = rb_first_cached(&execlists->queue); rb; rb = rb_next(rb)) {
struct i915_priolist *p = to_priolist(rb);
GEM_BUG_ON(p->priority >= last_prio);
last_prio = p->priority;
GEM_BUG_ON(!p->used);
for (i = 0; i < ARRAY_SIZE(p->requests); i++) {
if (list_empty(&p->requests[i]))
continue;
GEM_BUG_ON(!(p->used & BIT(i)));
}
}
}
static struct list_head *
lookup_priolist(struct intel_engine_cs *engine, int prio)
{
struct intel_engine_execlists * const execlists = &engine->execlists;
struct i915_priolist *p;
struct rb_node **parent, *rb;
bool first = true;
int idx, i;
assert_priolists(execlists, INT_MAX);
/* buckets sorted from highest [in slot 0] to lowest priority */
idx = I915_PRIORITY_COUNT - (prio & I915_PRIORITY_MASK) - 1;
prio >>= I915_USER_PRIORITY_SHIFT;
if (unlikely(execlists->no_priolist))
prio = I915_PRIORITY_NORMAL;
find_priolist:
/* most positive priority is scheduled first, equal priorities fifo */
rb = NULL;
parent = &execlists->queue.rb_root.rb_node;
while (*parent) {
rb = *parent;
p = to_priolist(rb);
if (prio > p->priority) {
parent = &rb->rb_left;
} else if (prio < p->priority) {
parent = &rb->rb_right;
first = false;
} else {
goto out;
}
}
if (prio == I915_PRIORITY_NORMAL) {
p = &execlists->default_priolist;
} else {
p = kmem_cache_alloc(engine->i915->priorities, GFP_ATOMIC);
/* Convert an allocation failure to a priority bump */
if (unlikely(!p)) {
prio = I915_PRIORITY_NORMAL; /* recurses just once */
/* To maintain ordering with all rendering, after an
* allocation failure we have to disable all scheduling.
* Requests will then be executed in fifo, and schedule
* will ensure that dependencies are emitted in fifo.
* There will be still some reordering with existing
* requests, so if userspace lied about their
* dependencies that reordering may be visible.
*/
execlists->no_priolist = true;
goto find_priolist;
}
}
p->priority = prio;
for (i = 0; i < ARRAY_SIZE(p->requests); i++)
INIT_LIST_HEAD(&p->requests[i]);
rb_link_node(&p->node, rb, parent);
rb_insert_color_cached(&p->node, &execlists->queue, first);
p->used = 0;
out:
p->used |= BIT(idx);
return &p->requests[idx];
}
static void unwind_wa_tail(struct i915_request *rq)
{
rq->tail = intel_ring_wrap(rq->ring, rq->wa_tail - WA_TAIL_BYTES);
@@ -381,7 +285,7 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
GEM_BUG_ON(rq_prio(rq) == I915_PRIORITY_INVALID);
if (rq_prio(rq) != prio) {
prio = rq_prio(rq);
pl = lookup_priolist(engine, prio);
pl = i915_sched_lookup_priolist(engine, prio);
}
GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root));
@@ -398,7 +302,7 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
if (!(prio & I915_PRIORITY_NEWCLIENT)) {
prio |= I915_PRIORITY_NEWCLIENT;
list_move_tail(&active->sched.link,
lookup_priolist(engine, prio));
i915_sched_lookup_priolist(engine, prio));
}
}
@@ -792,7 +696,6 @@ done:
*/
execlists->queue_priority =
port != execlists->port ? rq_prio(last) : INT_MIN;
assert_priolists(execlists, execlists->queue_priority);
if (submit) {
port_assign(port, last);
@@ -1119,12 +1022,7 @@ static void queue_request(struct intel_engine_cs *engine,
struct i915_sched_node *node,
int prio)
{
list_add_tail(&node->link, lookup_priolist(engine, prio));
}
static void __update_queue(struct intel_engine_cs *engine, int prio)
{
engine->execlists.queue_priority = prio;
list_add_tail(&node->link, i915_sched_lookup_priolist(engine, prio));
}
static void __submit_queue_imm(struct intel_engine_cs *engine)
@@ -1143,7 +1041,7 @@ static void __submit_queue_imm(struct intel_engine_cs *engine)
static void submit_queue(struct intel_engine_cs *engine, int prio)
{
if (prio > engine->execlists.queue_priority) {
__update_queue(engine, prio);
engine->execlists.queue_priority = prio;
__submit_queue_imm(engine);
}
}
@@ -1166,162 +1064,6 @@ static void execlists_submit_request(struct i915_request *request)
spin_unlock_irqrestore(&engine->timeline.lock, flags);
}
static struct i915_request *sched_to_request(struct i915_sched_node *node)
{
return container_of(node, struct i915_request, sched);
}
static struct intel_engine_cs *
sched_lock_engine(struct i915_sched_node *node, struct intel_engine_cs *locked)
{
struct intel_engine_cs *engine = sched_to_request(node)->engine;
GEM_BUG_ON(!locked);
if (engine != locked) {
spin_unlock(&locked->timeline.lock);
spin_lock(&engine->timeline.lock);
}
return engine;
}
static void execlists_schedule(struct i915_request *request,
const struct i915_sched_attr *attr)
{
struct list_head *uninitialized_var(pl);
struct intel_engine_cs *engine, *last;
struct i915_dependency *dep, *p;
struct i915_dependency stack;
const int prio = attr->priority;
LIST_HEAD(dfs);
GEM_BUG_ON(prio == I915_PRIORITY_INVALID);
if (i915_request_completed(request))
return;
if (prio <= READ_ONCE(request->sched.attr.priority))
return;
/* Need BKL in order to use the temporary link inside i915_dependency */
lockdep_assert_held(&request->i915->drm.struct_mutex);
stack.signaler = &request->sched;
list_add(&stack.dfs_link, &dfs);
/*
* Recursively bump all dependent priorities to match the new request.
*
* A naive approach would be to use recursion:
* static void update_priorities(struct i915_sched_node *node, prio) {
* list_for_each_entry(dep, &node->signalers_list, signal_link)
* update_priorities(dep->signal, prio)
* queue_request(node);
* }
* but that may have unlimited recursion depth and so runs a very
* real risk of overunning the kernel stack. Instead, we build
* a flat list of all dependencies starting with the current request.
* As we walk the list of dependencies, we add all of its dependencies
* to the end of the list (this may include an already visited
* request) and continue to walk onwards onto the new dependencies. The
* end result is a topological list of requests in reverse order, the
* last element in the list is the request we must execute first.
*/
list_for_each_entry(dep, &dfs, dfs_link) {
struct i915_sched_node *node = dep->signaler;
/*
* Within an engine, there can be no cycle, but we may
* refer to the same dependency chain multiple times
* (redundant dependencies are not eliminated) and across
* engines.
*/
list_for_each_entry(p, &node->signalers_list, signal_link) {
GEM_BUG_ON(p == dep); /* no cycles! */
if (i915_sched_node_signaled(p->signaler))
continue;
GEM_BUG_ON(p->signaler->attr.priority < node->attr.priority);
if (prio > READ_ONCE(p->signaler->attr.priority))
list_move_tail(&p->dfs_link, &dfs);
}
}
/*
* If we didn't need to bump any existing priorities, and we haven't
* yet submitted this request (i.e. there is no potential race with
* execlists_submit_request()), we can set our own priority and skip
* acquiring the engine locks.
*/
if (request->sched.attr.priority == I915_PRIORITY_INVALID) {
GEM_BUG_ON(!list_empty(&request->sched.link));
request->sched.attr = *attr;
if (stack.dfs_link.next == stack.dfs_link.prev)
return;
__list_del_entry(&stack.dfs_link);
}
last = NULL;
engine = request->engine;
spin_lock_irq(&engine->timeline.lock);
/* Fifo and depth-first replacement ensure our deps execute before us */
list_for_each_entry_safe_reverse(dep, p, &dfs, dfs_link) {
struct i915_sched_node *node = dep->signaler;
INIT_LIST_HEAD(&dep->dfs_link);
engine = sched_lock_engine(node, engine);
/* Recheck after acquiring the engine->timeline.lock */
if (prio <= node->attr.priority)
continue;
if (i915_sched_node_signaled(node))
continue;
node->attr.priority = prio;
if (!list_empty(&node->link)) {
if (last != engine) {
pl = lookup_priolist(engine, prio);
last = engine;
}
list_move_tail(&node->link, pl);
} else {
/*
* If the request is not in the priolist queue because
* it is not yet runnable, then it doesn't contribute
* to our preemption decisions. On the other hand,
* if the request is on the HW, it too is not in the
* queue; but in that case we may still need to reorder
* the inflight requests.
*/
if (!i915_sw_fence_done(&sched_to_request(node)->submit))
continue;
}
if (prio <= engine->execlists.queue_priority)
continue;
/*
* If we are already the currently executing context, don't
* bother evaluating if we should preempt ourselves.
*/
if (sched_to_request(node)->global_seqno &&
i915_seqno_passed(port_request(engine->execlists.port)->global_seqno,
sched_to_request(node)->global_seqno))
continue;
/* Defer (tasklet) submission until after all of our updates. */
__update_queue(engine, prio);
tasklet_hi_schedule(&engine->execlists.tasklet);
}
spin_unlock_irq(&engine->timeline.lock);
}
static void execlists_context_destroy(struct intel_context *ce)
{
GEM_BUG_ON(ce->pin_count);
@@ -2359,7 +2101,7 @@ void intel_execlists_set_default_submission(struct intel_engine_cs *engine)
{
engine->submit_request = execlists_submit_request;
engine->cancel_requests = execlists_cancel_requests;
engine->schedule = execlists_schedule;
engine->schedule = i915_schedule;
engine->execlists.tasklet.func = execlists_submission_tasklet;
engine->reset.prepare = execlists_reset_prepare;