From: Chris Wilson Date: Tue, 28 May 2019 09:29:54 +0000 (+0100) Subject: drm/i915: Move GEM client throttling to its own file X-Git-Tag: v5.3-rc1~81^2~16^2~195 X-Git-Url: https://asedeno.scripts.mit.edu/gitweb/?a=commitdiff_plain;h=446e2d16a131ed2319ba92d9aa61db5164aa25d2;p=linux.git drm/i915: Move GEM client throttling to its own file Continuing the decluttering of i915_gem.c by moving the client self throttling into its own file. Signed-off-by: Chris Wilson Reviewed-by: Mika Kuoppala Link: https://patchwork.freedesktop.org/patch/msgid/20190528092956.14910-13-chris@chris-wilson.co.uk --- diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile index e1ed526b96bf..83588e9840f8 100644 --- a/drivers/gpu/drm/i915/Makefile +++ b/drivers/gpu/drm/i915/Makefile @@ -104,6 +104,7 @@ gem-y += \ gem/i915_gem_shmem.o \ gem/i915_gem_shrinker.o \ gem/i915_gem_stolen.o \ + gem/i915_gem_throttle.o \ gem/i915_gem_tiling.o \ gem/i915_gem_userptr.o \ gem/i915_gem_wait.o \ diff --git a/drivers/gpu/drm/i915/gem/i915_gem_throttle.c b/drivers/gpu/drm/i915/gem/i915_gem_throttle.c new file mode 100644 index 000000000000..adb3074d9ce2 --- /dev/null +++ b/drivers/gpu/drm/i915/gem/i915_gem_throttle.c @@ -0,0 +1,73 @@ +/* + * SPDX-License-Identifier: MIT + * + * Copyright © 2014-2016 Intel Corporation + */ + +#include + +#include + +#include "i915_drv.h" +#include "i915_gem_ioctls.h" +#include "i915_gem_object.h" + +/* + * 20ms is a fairly arbitrary limit (greater than the average frame time) + * chosen to prevent the CPU getting more than a frame ahead of the GPU + * (when using lax throttling for the frontbuffer). We also use it to + * offer free GPU waitboosts for severely congested workloads. + */ +#define DRM_I915_THROTTLE_JIFFIES msecs_to_jiffies(20) + +/* + * Throttle our rendering by waiting until the ring has completed our requests + * emitted over 20 msec ago. + * + * Note that if we were to use the current jiffies each time around the loop, + * we wouldn't escape the function with any frames outstanding if the time to + * render a frame was over 20ms. + * + * This should get us reasonable parallelism between CPU and GPU but also + * relatively low latency when blocking on a particular request to finish. + */ +int +i915_gem_throttle_ioctl(struct drm_device *dev, void *data, + struct drm_file *file) +{ + struct drm_i915_file_private *file_priv = file->driver_priv; + unsigned long recent_enough = jiffies - DRM_I915_THROTTLE_JIFFIES; + struct i915_request *request, *target = NULL; + long ret; + + /* ABI: return -EIO if already wedged */ + ret = i915_terminally_wedged(to_i915(dev)); + if (ret) + return ret; + + spin_lock(&file_priv->mm.lock); + list_for_each_entry(request, &file_priv->mm.request_list, client_link) { + if (time_after_eq(request->emitted_jiffies, recent_enough)) + break; + + if (target) { + list_del(&target->client_link); + target->file_priv = NULL; + } + + target = request; + } + if (target) + i915_request_get(target); + spin_unlock(&file_priv->mm.lock); + + if (!target) + return 0; + + ret = i915_request_wait(target, + I915_WAIT_INTERRUPTIBLE, + MAX_SCHEDULE_TIMEOUT); + i915_request_put(target); + + return ret < 0 ? ret : 0; +} diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 652b5edefaae..4eb8f6a181c1 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -213,12 +213,6 @@ struct drm_i915_file_private { struct { spinlock_t lock; struct list_head request_list; -/* 20ms is a fairly arbitrary limit (greater than the average frame time) - * chosen to prevent the CPU getting more than a frame ahead of the GPU - * (when using lax throttling for the frontbuffer). We also use it to - * offer free GPU waitboosts for severely congested workloads. - */ -#define DRM_I915_THROTTLE_JIFFIES msecs_to_jiffies(20) } mm; struct idr context_idr; diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 6af448736030..e5aafbeb1d19 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -1012,57 +1012,6 @@ int i915_gem_wait_for_idle(struct drm_i915_private *i915, return 0; } -/* Throttle our rendering by waiting until the ring has completed our requests - * emitted over 20 msec ago. - * - * Note that if we were to use the current jiffies each time around the loop, - * we wouldn't escape the function with any frames outstanding if the time to - * render a frame was over 20ms. - * - * This should get us reasonable parallelism between CPU and GPU but also - * relatively low latency when blocking on a particular request to finish. - */ -static int -i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file) -{ - struct drm_i915_private *dev_priv = to_i915(dev); - struct drm_i915_file_private *file_priv = file->driver_priv; - unsigned long recent_enough = jiffies - DRM_I915_THROTTLE_JIFFIES; - struct i915_request *request, *target = NULL; - long ret; - - /* ABI: return -EIO if already wedged */ - ret = i915_terminally_wedged(dev_priv); - if (ret) - return ret; - - spin_lock(&file_priv->mm.lock); - list_for_each_entry(request, &file_priv->mm.request_list, client_link) { - if (time_after_eq(request->emitted_jiffies, recent_enough)) - break; - - if (target) { - list_del(&target->client_link); - target->file_priv = NULL; - } - - target = request; - } - if (target) - i915_request_get(target); - spin_unlock(&file_priv->mm.lock); - - if (target == NULL) - return 0; - - ret = i915_request_wait(target, - I915_WAIT_INTERRUPTIBLE, - MAX_SCHEDULE_TIMEOUT); - i915_request_put(target); - - return ret < 0 ? ret : 0; -} - struct i915_vma * i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, const struct i915_ggtt_view *view, @@ -1142,13 +1091,6 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, return vma; } -int -i915_gem_throttle_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - return i915_gem_ring_throttle(dev, file_priv); -} - int i915_gem_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)