From 1d054c9b8457b56a651109fac21f56f46ccd46b2 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Tue, 9 Nov 2021 10:11:02 -0800 Subject: drm/msm: Drop priv->lastctx cur_ctx_seqno already does the same thing, but handles the edge cases where a refcnt'd context can live after lastclose. So let's not have two ways to do the same thing. Signed-off-by: Rob Clark Reviewed-by: Akhil P Oommen Link: https://lore.kernel.org/r/20211109181117.591148-3-robdclark@gmail.com Signed-off-by: Rob Clark --- drivers/gpu/drm/msm/msm_gpu.h | 11 +++++++++++ 1 file changed, 11 insertions(+) (limited to 'drivers/gpu/drm/msm/msm_gpu.h') diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index 48ea2de911f1..f33f2c6aab47 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -144,6 +144,17 @@ struct msm_gpu { struct msm_ringbuffer *rb[MSM_GPU_MAX_RINGS]; int nr_rings; + /** + * cur_ctx_seqno: + * + * The ctx->seqno value of the last context to submit rendering, + * and the one with current pgtables installed (for generations + * that support per-context pgtables). Tracked by seqno rather + * than pointer value to avoid dangling pointers, and cases where + * a ctx can be freed and a new one created with the same address. + */ + int cur_ctx_seqno; + /* * List of GEM active objects on this gpu. Protected by * msm_drm_private::mm_lock -- cgit v1.2.3-70-g09d2 From c28e2f2b417ed747bfbc5f900c87f3ec9cc6b25e Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Tue, 9 Nov 2021 10:11:03 -0800 Subject: drm/msm: Remove struct_mutex usage The remaining struct_mutex usage is just to serialize various gpu related things (submit/retire/recover/fault/etc), so replace struct_mutex with gpu->lock. Signed-off-by: Rob Clark Link: https://lore.kernel.org/r/20211109181117.591148-4-robdclark@gmail.com Signed-off-by: Rob Clark --- drivers/gpu/drm/msm/adreno/a5xx_debugfs.c | 4 ++-- drivers/gpu/drm/msm/adreno/adreno_device.c | 4 ++-- drivers/gpu/drm/msm/msm_debugfs.c | 12 ++++++------ drivers/gpu/drm/msm/msm_gpu.c | 14 +++++++------- drivers/gpu/drm/msm/msm_gpu.h | 20 +++++++++++++++----- drivers/gpu/drm/msm/msm_perf.c | 9 ++++++--- drivers/gpu/drm/msm/msm_rd.c | 16 +++++++++------- drivers/gpu/drm/msm/msm_ringbuffer.c | 4 ++-- 8 files changed, 49 insertions(+), 34 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gpu.h') diff --git a/drivers/gpu/drm/msm/adreno/a5xx_debugfs.c b/drivers/gpu/drm/msm/adreno/a5xx_debugfs.c index dd593ec2bc56..6bd397a85834 100644 --- a/drivers/gpu/drm/msm/adreno/a5xx_debugfs.c +++ b/drivers/gpu/drm/msm/adreno/a5xx_debugfs.c @@ -107,7 +107,7 @@ reset_set(void *data, u64 val) * try to reset an active GPU. */ - mutex_lock(&dev->struct_mutex); + mutex_lock(&gpu->lock); release_firmware(adreno_gpu->fw[ADRENO_FW_PM4]); adreno_gpu->fw[ADRENO_FW_PM4] = NULL; @@ -133,7 +133,7 @@ reset_set(void *data, u64 val) gpu->funcs->recover(gpu); pm_runtime_put_sync(&gpu->pdev->dev); - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); return 0; } diff --git a/drivers/gpu/drm/msm/adreno/adreno_device.c b/drivers/gpu/drm/msm/adreno/adreno_device.c index 2a6ce76656aa..9e01ccc800a6 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_device.c +++ b/drivers/gpu/drm/msm/adreno/adreno_device.c @@ -408,9 +408,9 @@ struct msm_gpu *adreno_load_gpu(struct drm_device *dev) return NULL; } - mutex_lock(&dev->struct_mutex); + mutex_lock(&gpu->lock); ret = msm_gpu_hw_init(gpu); - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); pm_runtime_put_autosuspend(&pdev->dev); if (ret) { DRM_DEV_ERROR(dev->dev, "gpu hw init failed: %d\n", ret); diff --git a/drivers/gpu/drm/msm/msm_debugfs.c b/drivers/gpu/drm/msm/msm_debugfs.c index 698e86f5b960..6a99e8b5d25d 100644 --- a/drivers/gpu/drm/msm/msm_debugfs.c +++ b/drivers/gpu/drm/msm/msm_debugfs.c @@ -29,14 +29,14 @@ static int msm_gpu_show(struct seq_file *m, void *arg) struct msm_gpu *gpu = priv->gpu; int ret; - ret = mutex_lock_interruptible(&show_priv->dev->struct_mutex); + ret = mutex_lock_interruptible(&gpu->lock); if (ret) return ret; drm_printf(&p, "%s Status:\n", gpu->name); gpu->funcs->show(gpu, show_priv->state, &p); - mutex_unlock(&show_priv->dev->struct_mutex); + mutex_unlock(&gpu->lock); return 0; } @@ -48,9 +48,9 @@ static int msm_gpu_release(struct inode *inode, struct file *file) struct msm_drm_private *priv = show_priv->dev->dev_private; struct msm_gpu *gpu = priv->gpu; - mutex_lock(&show_priv->dev->struct_mutex); + mutex_lock(&gpu->lock); gpu->funcs->gpu_state_put(show_priv->state); - mutex_unlock(&show_priv->dev->struct_mutex); + mutex_unlock(&gpu->lock); kfree(show_priv); @@ -72,7 +72,7 @@ static int msm_gpu_open(struct inode *inode, struct file *file) if (!show_priv) return -ENOMEM; - ret = mutex_lock_interruptible(&dev->struct_mutex); + ret = mutex_lock_interruptible(&gpu->lock); if (ret) goto free_priv; @@ -81,7 +81,7 @@ static int msm_gpu_open(struct inode *inode, struct file *file) show_priv->state = gpu->funcs->gpu_state_get(gpu); pm_runtime_put_sync(&gpu->pdev->dev); - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); if (IS_ERR(show_priv->state)) { ret = PTR_ERR(show_priv->state); diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c index 3dfc58e6498f..13de1241d595 100644 --- a/drivers/gpu/drm/msm/msm_gpu.c +++ b/drivers/gpu/drm/msm/msm_gpu.c @@ -150,7 +150,7 @@ int msm_gpu_hw_init(struct msm_gpu *gpu) { int ret; - WARN_ON(!mutex_is_locked(&gpu->dev->struct_mutex)); + WARN_ON(!mutex_is_locked(&gpu->lock)); if (!gpu->needs_hw_init) return 0; @@ -361,7 +361,7 @@ static void recover_worker(struct kthread_work *work) char *comm = NULL, *cmd = NULL; int i; - mutex_lock(&dev->struct_mutex); + mutex_lock(&gpu->lock); DRM_DEV_ERROR(dev->dev, "%s: hangcheck recover!\n", gpu->name); @@ -442,7 +442,7 @@ static void recover_worker(struct kthread_work *work) } } - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); msm_gpu_retire(gpu); } @@ -450,12 +450,11 @@ static void recover_worker(struct kthread_work *work) static void fault_worker(struct kthread_work *work) { struct msm_gpu *gpu = container_of(work, struct msm_gpu, fault_work); - struct drm_device *dev = gpu->dev; struct msm_gem_submit *submit; struct msm_ringbuffer *cur_ring = gpu->funcs->active_ring(gpu); char *comm = NULL, *cmd = NULL; - mutex_lock(&dev->struct_mutex); + mutex_lock(&gpu->lock); submit = find_submit(cur_ring, cur_ring->memptrs->fence + 1); if (submit && submit->fault_dumped) @@ -490,7 +489,7 @@ resume_smmu: memset(&gpu->fault_info, 0, sizeof(gpu->fault_info)); gpu->aspace->mmu->funcs->resume_translation(gpu->aspace->mmu); - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); } static void hangcheck_timer_reset(struct msm_gpu *gpu) @@ -733,7 +732,7 @@ void msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) struct msm_ringbuffer *ring = submit->ring; unsigned long flags; - WARN_ON(!mutex_is_locked(&dev->struct_mutex)); + WARN_ON(!mutex_is_locked(&gpu->lock)); pm_runtime_get_sync(&gpu->pdev->dev); @@ -848,6 +847,7 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, INIT_LIST_HEAD(&gpu->active_list); mutex_init(&gpu->active_lock); + mutex_init(&gpu->lock); kthread_init_work(&gpu->retire_work, retire_worker); kthread_init_work(&gpu->recover_work, recover_worker); kthread_init_work(&gpu->fault_work, fault_worker); diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index f33f2c6aab47..0078a925fb64 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -161,13 +161,23 @@ struct msm_gpu { */ struct list_head active_list; + /** + * lock: + * + * General lock for serializing all the gpu things. + * + * TODO move to per-ring locking where feasible (ie. submit/retire + * path, etc) + */ + struct mutex lock; + /** * active_submits: * * The number of submitted but not yet retired submits, used to * determine transitions between active and idle. * - * Protected by lock + * Protected by active_lock */ int active_submits; @@ -548,28 +558,28 @@ static inline struct msm_gpu_state *msm_gpu_crashstate_get(struct msm_gpu *gpu) { struct msm_gpu_state *state = NULL; - mutex_lock(&gpu->dev->struct_mutex); + mutex_lock(&gpu->lock); if (gpu->crashstate) { kref_get(&gpu->crashstate->ref); state = gpu->crashstate; } - mutex_unlock(&gpu->dev->struct_mutex); + mutex_unlock(&gpu->lock); return state; } static inline void msm_gpu_crashstate_put(struct msm_gpu *gpu) { - mutex_lock(&gpu->dev->struct_mutex); + mutex_lock(&gpu->lock); if (gpu->crashstate) { if (gpu->funcs->gpu_state_put(gpu->crashstate)) gpu->crashstate = NULL; } - mutex_unlock(&gpu->dev->struct_mutex); + mutex_unlock(&gpu->lock); } /* diff --git a/drivers/gpu/drm/msm/msm_perf.c b/drivers/gpu/drm/msm/msm_perf.c index 3a27153eef08..3d3da79fec2a 100644 --- a/drivers/gpu/drm/msm/msm_perf.c +++ b/drivers/gpu/drm/msm/msm_perf.c @@ -155,9 +155,12 @@ static int perf_open(struct inode *inode, struct file *file) struct msm_gpu *gpu = priv->gpu; int ret = 0; - mutex_lock(&dev->struct_mutex); + if (!gpu) + return -ENODEV; - if (perf->open || !gpu) { + mutex_lock(&gpu->lock); + + if (perf->open) { ret = -EBUSY; goto out; } @@ -171,7 +174,7 @@ static int perf_open(struct inode *inode, struct file *file) perf->next_jiffies = jiffies + SAMPLE_TIME; out: - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); return ret; } diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c index b55398a34fa4..81432ec07012 100644 --- a/drivers/gpu/drm/msm/msm_rd.c +++ b/drivers/gpu/drm/msm/msm_rd.c @@ -86,7 +86,7 @@ struct msm_rd_state { struct msm_gem_submit *submit; /* fifo access is synchronized on the producer side by - * struct_mutex held by submit code (otherwise we could + * gpu->lock held by submit code (otherwise we could * end up w/ cmds logged in different order than they * were executed). And read_lock synchronizes the reads */ @@ -181,9 +181,12 @@ static int rd_open(struct inode *inode, struct file *file) uint32_t gpu_id; int ret = 0; - mutex_lock(&dev->struct_mutex); + if (!gpu) + return -ENODEV; - if (rd->open || !gpu) { + mutex_lock(&gpu->lock); + + if (rd->open) { ret = -EBUSY; goto out; } @@ -200,7 +203,7 @@ static int rd_open(struct inode *inode, struct file *file) rd_write_section(rd, RD_GPU_ID, &gpu_id, sizeof(gpu_id)); out: - mutex_unlock(&dev->struct_mutex); + mutex_unlock(&gpu->lock); return ret; } @@ -340,11 +343,10 @@ out_unlock: msm_gem_unlock(&obj->base); } -/* called under struct_mutex */ +/* called under gpu->lock */ void msm_rd_dump_submit(struct msm_rd_state *rd, struct msm_gem_submit *submit, const char *fmt, ...) { - struct drm_device *dev = submit->dev; struct task_struct *task; char msg[256]; int i, n; @@ -355,7 +357,7 @@ void msm_rd_dump_submit(struct msm_rd_state *rd, struct msm_gem_submit *submit, /* writing into fifo is serialized by caller, and * rd->read_lock is used to serialize the reads */ - WARN_ON(!mutex_is_locked(&dev->struct_mutex)); + WARN_ON(!mutex_is_locked(&submit->gpu->lock)); if (fmt) { va_list args; diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c index 652b1dedd7c1..3bbf574c3bdc 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.c +++ b/drivers/gpu/drm/msm/msm_ringbuffer.c @@ -21,11 +21,11 @@ static struct dma_fence *msm_job_run(struct drm_sched_job *job) pm_runtime_get_sync(&gpu->pdev->dev); /* TODO move submit path over to using a per-ring lock.. */ - mutex_lock(&gpu->dev->struct_mutex); + mutex_lock(&gpu->lock); msm_gpu_submit(gpu, submit); - mutex_unlock(&gpu->dev->struct_mutex); + mutex_unlock(&gpu->lock); pm_runtime_put(&gpu->pdev->dev); -- cgit v1.2.3-70-g09d2 From 5f3aee4ceb5b8607e9a4b752c2547667200e19e2 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Tue, 9 Nov 2021 10:11:04 -0800 Subject: drm/msm: Handle fence rollover Add some helpers for fence comparision, which handle rollover properly, and stop open coding fence seqno comparisions. Signed-off-by: Rob Clark Reviewed-by: Akhil P Oommen Link: https://lore.kernel.org/r/20211109181117.591148-5-robdclark@gmail.com Signed-off-by: Rob Clark --- drivers/gpu/drm/msm/msm_drv.c | 2 +- drivers/gpu/drm/msm/msm_fence.h | 12 ++++++++++++ drivers/gpu/drm/msm/msm_gpu.c | 6 +++--- drivers/gpu/drm/msm/msm_gpu.h | 2 +- 4 files changed, 17 insertions(+), 5 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gpu.h') diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index b097bc3ffbe1..91cc4512bafb 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -967,7 +967,7 @@ static int wait_fence(struct msm_gpu_submitqueue *queue, uint32_t fence_id, struct dma_fence *fence; int ret; - if (fence_id > queue->last_fence) { + if (fence_after(fence_id, queue->last_fence)) { DRM_ERROR_RATELIMITED("waiting on invalid fence: %u (of %u)\n", fence_id, queue->last_fence); return -EINVAL; diff --git a/drivers/gpu/drm/msm/msm_fence.h b/drivers/gpu/drm/msm/msm_fence.h index 4783db528bcc..17ee3822b423 100644 --- a/drivers/gpu/drm/msm/msm_fence.h +++ b/drivers/gpu/drm/msm/msm_fence.h @@ -60,4 +60,16 @@ void msm_update_fence(struct msm_fence_context *fctx, uint32_t fence); struct dma_fence * msm_fence_alloc(struct msm_fence_context *fctx); +static inline bool +fence_before(uint32_t a, uint32_t b) +{ + return (int32_t)(a - b) < 0; +} + +static inline bool +fence_after(uint32_t a, uint32_t b) +{ + return (int32_t)(a - b) > 0; +} + #endif diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c index 13de1241d595..0f78c2615272 100644 --- a/drivers/gpu/drm/msm/msm_gpu.c +++ b/drivers/gpu/drm/msm/msm_gpu.c @@ -172,7 +172,7 @@ static void update_fences(struct msm_gpu *gpu, struct msm_ringbuffer *ring, spin_lock_irqsave(&ring->submit_lock, flags); list_for_each_entry(submit, &ring->submits, node) { - if (submit->seqno > fence) + if (fence_after(submit->seqno, fence)) break; msm_update_fence(submit->ring->fctx, @@ -509,7 +509,7 @@ static void hangcheck_handler(struct timer_list *t) if (fence != ring->hangcheck_fence) { /* some progress has been made.. ya! */ ring->hangcheck_fence = fence; - } else if (fence < ring->seqno) { + } else if (fence_before(fence, ring->seqno)) { /* no progress and not done.. hung! */ ring->hangcheck_fence = fence; DRM_DEV_ERROR(dev->dev, "%s: hangcheck detected gpu lockup rb %d!\n", @@ -523,7 +523,7 @@ static void hangcheck_handler(struct timer_list *t) } /* if still more pending work, reset the hangcheck timer: */ - if (ring->seqno > ring->hangcheck_fence) + if (fence_after(ring->seqno, ring->hangcheck_fence)) hangcheck_timer_reset(gpu); /* workaround for missing irq: */ diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index 0078a925fb64..37b3ecb33cbd 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -262,7 +262,7 @@ static inline bool msm_gpu_active(struct msm_gpu *gpu) for (i = 0; i < gpu->nr_rings; i++) { struct msm_ringbuffer *ring = gpu->rb[i]; - if (ring->seqno > ring->memptrs->fence) + if (fence_after(ring->seqno, ring->memptrs->fence)) return true; } -- cgit v1.2.3-70-g09d2 From 7c0ffcd40b161f56a7cea585297d64aeaf4d44a9 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Sat, 20 Nov 2021 12:01:03 -0800 Subject: drm/msm/gpu: Respect PM QoS constraints Re-work the boost and idle clamping to use PM QoS requests instead, so they get aggreggated with other requests (such as cooling device). This does have the minor side-effect that devfreq sysfs min_freq/ max_freq files now reflect the boost and idle clamping, as they show (despite what they are documented to show) the aggregated min/max freq. Fixing that in devfreq does not look straightforward after considering that OPPs can be dynamically added/removed. However writes to the sysfs files still behave as expected. v2: Use 64b math to avoid potential 32b overflow Signed-off-by: Rob Clark Link: https://lore.kernel.org/r/20211120200103.1051459-3-robdclark@gmail.com Signed-off-by: Rob Clark --- drivers/gpu/drm/msm/msm_gpu.h | 33 +++++++++---- drivers/gpu/drm/msm/msm_gpu_devfreq.c | 89 +++++++++++++++++++---------------- 2 files changed, 71 insertions(+), 51 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gpu.h') diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index 37b3ecb33cbd..445c6bfd4b6b 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -87,6 +87,21 @@ struct msm_gpu_devfreq { /** devfreq: devfreq instance */ struct devfreq *devfreq; + /** + * idle_constraint: + * + * A PM QoS constraint to limit max freq while the GPU is idle. + */ + struct dev_pm_qos_request idle_freq; + + /** + * boost_constraint: + * + * A PM QoS constraint to boost min freq for a period of time + * until the boost expires. + */ + struct dev_pm_qos_request boost_freq; + /** * busy_cycles: * @@ -103,22 +118,19 @@ struct msm_gpu_devfreq { ktime_t idle_time; /** - * idle_freq: + * idle_work: * - * Shadow frequency used while the GPU is idle. From the PoV of - * the devfreq governor, we are continuing to sample busyness and - * adjust frequency while the GPU is idle, but we use this shadow - * value as the GPU is actually clamped to minimum frequency while - * it is inactive. + * Used to delay clamping to idle freq on active->idle transition. */ - unsigned long idle_freq; + struct msm_hrtimer_work idle_work; /** - * idle_work: + * boost_work: * - * Used to delay clamping to idle freq on active->idle transition. + * Used to reset the boost_constraint after the boost period has + * elapsed */ - struct msm_hrtimer_work idle_work; + struct msm_hrtimer_work boost_work; }; struct msm_gpu { @@ -522,6 +534,7 @@ void msm_devfreq_init(struct msm_gpu *gpu); void msm_devfreq_cleanup(struct msm_gpu *gpu); void msm_devfreq_resume(struct msm_gpu *gpu); void msm_devfreq_suspend(struct msm_gpu *gpu); +void msm_devfreq_boost(struct msm_gpu *gpu, unsigned factor); void msm_devfreq_active(struct msm_gpu *gpu); void msm_devfreq_idle(struct msm_gpu *gpu); diff --git a/drivers/gpu/drm/msm/msm_gpu_devfreq.c b/drivers/gpu/drm/msm/msm_gpu_devfreq.c index d232f74d2346..62405e980925 100644 --- a/drivers/gpu/drm/msm/msm_gpu_devfreq.c +++ b/drivers/gpu/drm/msm/msm_gpu_devfreq.c @@ -9,6 +9,7 @@ #include #include +#include /* * Power Management: @@ -25,17 +26,6 @@ static int msm_devfreq_target(struct device *dev, unsigned long *freq, * to something that actually is in the opp table: */ opp = devfreq_recommended_opp(dev, freq, flags); - - /* - * If the GPU is idle, devfreq is not aware, so just ignore - * it's requests - */ - if (gpu->devfreq.idle_freq) { - gpu->devfreq.idle_freq = *freq; - dev_pm_opp_put(opp); - return 0; - } - if (IS_ERR(opp)) return PTR_ERR(opp); @@ -53,9 +43,6 @@ static int msm_devfreq_target(struct device *dev, unsigned long *freq, static unsigned long get_freq(struct msm_gpu *gpu) { - if (gpu->devfreq.idle_freq) - return gpu->devfreq.idle_freq; - if (gpu->funcs->gpu_get_freq) return gpu->funcs->gpu_get_freq(gpu); @@ -93,6 +80,7 @@ static struct devfreq_dev_profile msm_devfreq_profile = { .get_cur_freq = msm_devfreq_get_cur_freq, }; +static void msm_devfreq_boost_work(struct kthread_work *work); static void msm_devfreq_idle_work(struct kthread_work *work); void msm_devfreq_init(struct msm_gpu *gpu) @@ -103,6 +91,12 @@ void msm_devfreq_init(struct msm_gpu *gpu) if (!gpu->funcs->gpu_busy) return; + dev_pm_qos_add_request(&gpu->pdev->dev, &df->idle_freq, + DEV_PM_QOS_MAX_FREQUENCY, + PM_QOS_MAX_FREQUENCY_DEFAULT_VALUE); + dev_pm_qos_add_request(&gpu->pdev->dev, &df->boost_freq, + DEV_PM_QOS_MIN_FREQUENCY, 0); + msm_devfreq_profile.initial_freq = gpu->fast_rate; /* @@ -133,13 +127,19 @@ void msm_devfreq_init(struct msm_gpu *gpu) gpu->cooling = NULL; } + msm_hrtimer_work_init(&df->boost_work, gpu->worker, msm_devfreq_boost_work, + CLOCK_MONOTONIC, HRTIMER_MODE_REL); msm_hrtimer_work_init(&df->idle_work, gpu->worker, msm_devfreq_idle_work, CLOCK_MONOTONIC, HRTIMER_MODE_REL); } void msm_devfreq_cleanup(struct msm_gpu *gpu) { + struct msm_gpu_devfreq *df = &gpu->devfreq; + devfreq_cooling_unregister(gpu->cooling); + dev_pm_qos_remove_request(&df->boost_freq); + dev_pm_qos_remove_request(&df->idle_freq); } void msm_devfreq_resume(struct msm_gpu *gpu) @@ -155,12 +155,40 @@ void msm_devfreq_suspend(struct msm_gpu *gpu) devfreq_suspend_device(gpu->devfreq.devfreq); } +static void msm_devfreq_boost_work(struct kthread_work *work) +{ + struct msm_gpu_devfreq *df = container_of(work, + struct msm_gpu_devfreq, boost_work.work); + + dev_pm_qos_update_request(&df->boost_freq, 0); +} + +void msm_devfreq_boost(struct msm_gpu *gpu, unsigned factor) +{ + struct msm_gpu_devfreq *df = &gpu->devfreq; + uint64_t freq; + + freq = get_freq(gpu); + freq *= factor; + + /* + * A nice little trap is that PM QoS operates in terms of KHz, + * while devfreq operates in terms of Hz: + */ + do_div(freq, HZ_PER_KHZ); + + dev_pm_qos_update_request(&df->boost_freq, freq); + + msm_hrtimer_queue_work(&df->boost_work, + ms_to_ktime(msm_devfreq_profile.polling_ms), + HRTIMER_MODE_REL); +} + void msm_devfreq_active(struct msm_gpu *gpu) { struct msm_gpu_devfreq *df = &gpu->devfreq; struct devfreq_dev_status status; unsigned int idle_time; - unsigned long target_freq = df->idle_freq; if (!df->devfreq) return; @@ -170,12 +198,6 @@ void msm_devfreq_active(struct msm_gpu *gpu) */ hrtimer_cancel(&df->idle_work.timer); - /* - * Hold devfreq lock to synchronize with get_dev_status()/ - * target() callbacks - */ - mutex_lock(&df->devfreq->lock); - idle_time = ktime_to_ms(ktime_sub(ktime_get(), df->idle_time)); /* @@ -184,20 +206,17 @@ void msm_devfreq_active(struct msm_gpu *gpu) * the governor to ramp up the freq.. so give some boost */ if (idle_time > msm_devfreq_profile.polling_ms) { - target_freq *= 2; + msm_devfreq_boost(gpu, 2); } - df->idle_freq = 0; - - msm_devfreq_target(&gpu->pdev->dev, &target_freq, 0); + dev_pm_qos_update_request(&df->idle_freq, + PM_QOS_MAX_FREQUENCY_DEFAULT_VALUE); /* * Reset the polling interval so we aren't inconsistent * about freq vs busy/total cycles */ msm_devfreq_get_dev_status(&gpu->pdev->dev, &status); - - mutex_unlock(&df->devfreq->lock); } @@ -206,23 +225,11 @@ static void msm_devfreq_idle_work(struct kthread_work *work) struct msm_gpu_devfreq *df = container_of(work, struct msm_gpu_devfreq, idle_work.work); struct msm_gpu *gpu = container_of(df, struct msm_gpu, devfreq); - unsigned long idle_freq, target_freq = 0; - - /* - * Hold devfreq lock to synchronize with get_dev_status()/ - * target() callbacks - */ - mutex_lock(&df->devfreq->lock); - - idle_freq = get_freq(gpu); - - if (gpu->clamp_to_idle) - msm_devfreq_target(&gpu->pdev->dev, &target_freq, 0); df->idle_time = ktime_get(); - df->idle_freq = idle_freq; - mutex_unlock(&df->devfreq->lock); + if (gpu->clamp_to_idle) + dev_pm_qos_update_request(&df->idle_freq, 0); } void msm_devfreq_idle(struct msm_gpu *gpu) -- cgit v1.2.3-70-g09d2