linux/drivers/gpu/drm/i915/gt/intel_engine_pm.c
Chris Wilson ce476c80b8 drm/i915: Keep contexts pinned until after the next kernel context switch
We need to keep the context image pinned in memory until after the GPU
has finished writing into it. Since it continues to write as we signal
the final breadcrumb, we need to keep it pinned until the request after
it is complete. Currently we know the order in which requests execute on
each engine, and so to remove that presumption we need to identify a
request/context-switch we know must occur after our completion. Any
request queued after the signal must imply a context switch, for
simplicity we use a fresh request from the kernel context.

The sequence of operations for keeping the context pinned until saved is:

 - On context activation, we preallocate a node for each physical engine
   the context may operate on. This is to avoid allocations during
   unpinning, which may be from inside FS_RECLAIM context (aka the
   shrinker)

 - On context deactivation on retirement of the last active request (which
   is before we know the context has been saved), we add the
   preallocated node onto a barrier list on each engine

 - On engine idling, we emit a switch to kernel context. When this
   switch completes, we know that all previous contexts must have been
   saved, and so on retiring this request we can finally unpin all the
   contexts that were marked as deactivated prior to the switch.

We can enhance this in future by flushing all the idle contexts on a
regular heartbeat pulse of a switch to kernel context, which will also
be used to check for hung engines.

v2: intel_context_active_acquire/_release

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190614164606.15633-1-chris@chris-wilson.co.uk
2019-06-14 19:03:32 +01:00

166 lines
4.3 KiB
C

/*
* SPDX-License-Identifier: MIT
*
* Copyright © 2019 Intel Corporation
*/
#include "i915_drv.h"
#include "intel_engine.h"
#include "intel_engine_pm.h"
#include "intel_gt_pm.h"
static int __engine_unpark(struct intel_wakeref *wf)
{
struct intel_engine_cs *engine =
container_of(wf, typeof(*engine), wakeref);
void *map;
GEM_TRACE("%s\n", engine->name);
intel_gt_pm_get(engine->i915);
/* Pin the default state for fast resets from atomic context. */
map = NULL;
if (engine->default_state)
map = i915_gem_object_pin_map(engine->default_state,
I915_MAP_WB);
if (!IS_ERR_OR_NULL(map))
engine->pinned_default_state = map;
if (engine->unpark)
engine->unpark(engine);
intel_engine_init_hangcheck(engine);
return 0;
}
void intel_engine_pm_get(struct intel_engine_cs *engine)
{
intel_wakeref_get(&engine->i915->runtime_pm, &engine->wakeref, __engine_unpark);
}
void intel_engine_park(struct intel_engine_cs *engine)
{
/*
* We are committed now to parking this engine, make sure there
* will be no more interrupts arriving later and the engine
* is truly idle.
*/
if (wait_for(intel_engine_is_idle(engine), 10)) {
struct drm_printer p = drm_debug_printer(__func__);
dev_err(engine->i915->drm.dev,
"%s is not idle before parking\n",
engine->name);
intel_engine_dump(engine, &p, NULL);
}
}
static bool switch_to_kernel_context(struct intel_engine_cs *engine)
{
struct i915_request *rq;
/* Already inside the kernel context, safe to power down. */
if (engine->wakeref_serial == engine->serial)
return true;
/* GPU is pointing to the void, as good as in the kernel context. */
if (i915_reset_failed(engine->i915))
return true;
/*
* Note, we do this without taking the timeline->mutex. We cannot
* as we may be called while retiring the kernel context and so
* already underneath the timeline->mutex. Instead we rely on the
* exclusive property of the __engine_park that prevents anyone
* else from creating a request on this engine. This also requires
* that the ring is empty and we avoid any waits while constructing
* the context, as they assume protection by the timeline->mutex.
* This should hold true as we can only park the engine after
* retiring the last request, thus all rings should be empty and
* all timelines idle.
*/
rq = __i915_request_create(engine->kernel_context, GFP_NOWAIT);
if (IS_ERR(rq))
/* Context switch failed, hope for the best! Maybe reset? */
return true;
/* Check again on the next retirement. */
engine->wakeref_serial = engine->serial + 1;
i915_request_add_barriers(rq);
__i915_request_commit(rq);
return false;
}
static int __engine_park(struct intel_wakeref *wf)
{
struct intel_engine_cs *engine =
container_of(wf, typeof(*engine), wakeref);
/*
* If one and only one request is completed between pm events,
* we know that we are inside the kernel context and it is
* safe to power down. (We are paranoid in case that runtime
* suspend causes corruption to the active context image, and
* want to avoid that impacting userspace.)
*/
if (!switch_to_kernel_context(engine))
return -EBUSY;
GEM_TRACE("%s\n", engine->name);
intel_engine_disarm_breadcrumbs(engine);
/* Must be reset upon idling, or we may miss the busy wakeup. */
GEM_BUG_ON(engine->execlists.queue_priority_hint != INT_MIN);
if (engine->park)
engine->park(engine);
if (engine->pinned_default_state) {
i915_gem_object_unpin_map(engine->default_state);
engine->pinned_default_state = NULL;
}
engine->execlists.no_priolist = false;
intel_gt_pm_put(engine->i915);
return 0;
}
void intel_engine_pm_put(struct intel_engine_cs *engine)
{
intel_wakeref_put(&engine->i915->runtime_pm, &engine->wakeref, __engine_park);
}
void intel_engine_init__pm(struct intel_engine_cs *engine)
{
intel_wakeref_init(&engine->wakeref);
}
int intel_engines_resume(struct drm_i915_private *i915)
{
struct intel_engine_cs *engine;
enum intel_engine_id id;
int err = 0;
intel_gt_pm_get(i915);
for_each_engine(engine, i915, id) {
intel_engine_pm_get(engine);
engine->serial++; /* kernel context lost */
err = engine->resume(engine);
intel_engine_pm_put(engine);
if (err) {
dev_err(i915->drm.dev,
"Failed to restart %s (%d)\n",
engine->name, err);
break;
}
}
intel_gt_pm_put(i915);
return err;
}