Using the coda_mutex lock to serialize hardware access would cause "INFO: possible circular locking dependency detected" lockdep warnings. Since the possible locking paths are hard to follow, serialize hardware access with a single workqueue thread. Ultimately the workqueue could be converted to only do register setup and readout for per-command work items. Using the initialized context property, SEQ_END is only queued in coda_release when needed.
Signed-off-by: Philipp Zabel <p.za...@pengutronix.de> --- drivers/media/platform/coda.c | 162 +++++++++++++++++++----------------------- 1 file changed, 74 insertions(+), 88 deletions(-) diff --git a/drivers/media/platform/coda.c b/drivers/media/platform/coda.c index 965b0d9..bc5b6c0 100644 --- a/drivers/media/platform/coda.c +++ b/drivers/media/platform/coda.c @@ -147,11 +147,11 @@ struct coda_dev { spinlock_t irqlock; struct mutex dev_mutex; struct mutex coda_mutex; + struct workqueue_struct *workqueue; struct v4l2_m2m_dev *m2m_dev; struct vb2_alloc_ctx *alloc_ctx; struct list_head instances; unsigned long instance_mask; - struct delayed_work timeout; }; struct coda_params { @@ -198,7 +198,9 @@ struct coda_ctx { struct coda_dev *dev; struct mutex buffer_mutex; struct list_head list; - struct work_struct skip_run; + struct work_struct pic_run_work; + struct work_struct seq_end_work; + struct completion completion; int aborting; int initialized; int streamon_out; @@ -1009,13 +1011,6 @@ static const struct v4l2_ioctl_ops coda_ioctl_ops = { static int coda_start_decoding(struct coda_ctx *ctx); -static void coda_skip_run(struct work_struct *work) -{ - struct coda_ctx *ctx = container_of(work, struct coda_ctx, skip_run); - - v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->m2m_ctx); -} - static inline int coda_get_bitstream_payload(struct coda_ctx *ctx) { return kfifo_len(&ctx->bitstream_fifo); @@ -1170,7 +1165,7 @@ static int coda_prepare_decode(struct coda_ctx *ctx) v4l2_dbg(1, coda_debug, &dev->v4l2_dev, "bitstream payload: %d, skipping\n", coda_get_bitstream_payload(ctx)); - schedule_work(&ctx->skip_run); + v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->m2m_ctx); return -EAGAIN; } @@ -1179,7 +1174,7 @@ static int coda_prepare_decode(struct coda_ctx *ctx) int ret = coda_start_decoding(ctx); if (ret < 0) { v4l2_err(&dev->v4l2_dev, "failed to start decoding\n"); - schedule_work(&ctx->skip_run); + v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->m2m_ctx); return -EAGAIN; } else { ctx->initialized = 1; @@ -1387,24 +1382,48 @@ static void coda_device_run(void *m2m_priv) { struct coda_ctx *ctx = m2m_priv; struct coda_dev *dev = ctx->dev; - int ret; + + queue_work(dev->workqueue, &ctx->pic_run_work); +} + +static void coda_free_framebuffers(struct coda_ctx *ctx); +static void coda_free_context_buffers(struct coda_ctx *ctx); + +static void coda_seq_end_work(struct work_struct *work) +{ + struct coda_ctx *ctx = container_of(work, struct coda_ctx, seq_end_work); + struct coda_dev *dev = ctx->dev; mutex_lock(&ctx->buffer_mutex); + mutex_lock(&dev->coda_mutex); - /* - * If streamoff dequeued all buffers before we could get the lock, - * just bail out immediately. - */ - if ((!v4l2_m2m_num_src_bufs_ready(ctx->m2m_ctx) && - ctx->inst_type != CODA_INST_DECODER) || - !v4l2_m2m_num_dst_bufs_ready(ctx->m2m_ctx)) { - v4l2_dbg(1, coda_debug, &dev->v4l2_dev, - "%d: device_run without buffers\n", ctx->idx); - mutex_unlock(&ctx->buffer_mutex); - schedule_work(&ctx->skip_run); - return; + v4l2_dbg(1, coda_debug, &dev->v4l2_dev, + "%d: %s: sent command 'SEQ_END' to coda\n", ctx->idx, __func__); + if (coda_command_sync(ctx, CODA_COMMAND_SEQ_END)) { + v4l2_err(&dev->v4l2_dev, + "CODA_COMMAND_SEQ_END failed\n"); } + kfifo_init(&ctx->bitstream_fifo, + ctx->bitstream.vaddr, ctx->bitstream.size); + + coda_free_framebuffers(ctx); + coda_free_context_buffers(ctx); + + mutex_unlock(&dev->coda_mutex); + mutex_unlock(&ctx->buffer_mutex); +} + +static void coda_finish_decode(struct coda_ctx *ctx); +static void coda_finish_encode(struct coda_ctx *ctx); + +static void coda_pic_run_work(struct work_struct *work) +{ + struct coda_ctx *ctx = container_of(work, struct coda_ctx, pic_run_work); + struct coda_dev *dev = ctx->dev; + int ret; + + mutex_lock(&ctx->buffer_mutex); mutex_lock(&dev->coda_mutex); if (ctx->inst_type == CODA_INST_DECODER) { @@ -1423,12 +1442,26 @@ static void coda_device_run(void *m2m_priv) coda_write(dev, ctx->iram_info.axi_sram_use, CODA7_REG_BIT_AXI_SRAM_USE); - /* 1 second timeout in case CODA locks up */ - schedule_delayed_work(&dev->timeout, HZ); - if (ctx->inst_type == CODA_INST_DECODER) coda_kfifo_sync_to_device_full(ctx); coda_command_async(ctx, CODA_COMMAND_PIC_RUN); + + if (!wait_for_completion_timeout(&ctx->completion, msecs_to_jiffies(1000))) { + dev_err(&dev->plat_dev->dev, "CODA PIC_RUN timeout\n"); + } else if (!ctx->aborting) { + if (ctx->inst_type == CODA_INST_DECODER) + coda_finish_decode(ctx); + else + coda_finish_encode(ctx); + } + + if (ctx->aborting || (!ctx->streamon_cap && !ctx->streamon_out)) + queue_work(dev->workqueue, &ctx->seq_end_work); + + mutex_unlock(&dev->coda_mutex); + mutex_unlock(&ctx->buffer_mutex); + + v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->m2m_ctx); } static int coda_job_ready(void *m2m_priv) @@ -2785,7 +2818,9 @@ static int coda_open(struct file *file) } set_bit(idx, &dev->instance_mask); - INIT_WORK(&ctx->skip_run, coda_skip_run); + init_completion(&ctx->completion); + INIT_WORK(&ctx->pic_run_work, coda_pic_run_work); + INIT_WORK(&ctx->seq_end_work, coda_seq_end_work); v4l2_fh_init(&ctx->fh, video_devdata(file)); file->private_data = &ctx->fh; v4l2_fh_add(&ctx->fh); @@ -2888,16 +2923,10 @@ static int coda_release(struct file *file) v4l2_m2m_ctx_release(ctx->m2m_ctx); /* In case the instance was not running, we still need to call SEQ_END */ - mutex_lock(&dev->coda_mutex); - v4l2_dbg(1, coda_debug, &dev->v4l2_dev, - "%s: sent command 'SEQ_END' to coda\n", __func__); - if (coda_command_sync(ctx, CODA_COMMAND_SEQ_END)) { - v4l2_err(&dev->v4l2_dev, - "CODA_COMMAND_SEQ_END failed\n"); - mutex_unlock(&dev->coda_mutex); - return -ETIMEDOUT; + if (ctx->initialized) { + queue_work(dev->workqueue, &ctx->seq_end_work); + flush_work(&ctx->seq_end_work); } - mutex_unlock(&dev->coda_mutex); coda_free_framebuffers(ctx); @@ -3195,8 +3224,6 @@ static irqreturn_t coda_irq_handler(int irq, void *data) struct coda_dev *dev = data; struct coda_ctx *ctx; - cancel_delayed_work(&dev->timeout); - /* read status register to attend the IRQ */ coda_read(dev, CODA_REG_BIT_INT_STATUS); coda_write(dev, CODA_REG_BIT_INT_CLEAR_SET, @@ -3212,7 +3239,6 @@ static irqreturn_t coda_irq_handler(int irq, void *data) if (ctx->aborting) { v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev, "task has been aborted\n"); - goto out; } if (coda_isbusy(ctx->dev)) { @@ -3221,57 +3247,11 @@ static irqreturn_t coda_irq_handler(int irq, void *data) return IRQ_NONE; } - if (ctx->inst_type == CODA_INST_DECODER) - coda_finish_decode(ctx); - else - coda_finish_encode(ctx); - -out: - if (ctx->aborting || (!ctx->streamon_cap && !ctx->streamon_out)) { - v4l2_dbg(1, coda_debug, &dev->v4l2_dev, - "%s: sent command 'SEQ_END' to coda\n", __func__); - if (coda_command_sync(ctx, CODA_COMMAND_SEQ_END)) { - v4l2_err(&dev->v4l2_dev, - "CODA_COMMAND_SEQ_END failed\n"); - } - - kfifo_init(&ctx->bitstream_fifo, - ctx->bitstream.vaddr, ctx->bitstream.size); - - coda_free_framebuffers(ctx); - coda_free_context_buffers(ctx); - } - - mutex_unlock(&dev->coda_mutex); - mutex_unlock(&ctx->buffer_mutex); - - v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->m2m_ctx); + complete(&ctx->completion); return IRQ_HANDLED; } -static void coda_timeout(struct work_struct *work) -{ - struct coda_ctx *ctx; - struct coda_dev *dev = container_of(to_delayed_work(work), - struct coda_dev, timeout); - - dev_err(&dev->plat_dev->dev, "CODA PIC_RUN timeout, stopping all streams\n"); - - mutex_lock(&dev->dev_mutex); - list_for_each_entry(ctx, &dev->instances, list) { - if (mutex_is_locked(&ctx->buffer_mutex)) - mutex_unlock(&ctx->buffer_mutex); - v4l2_m2m_streamoff(NULL, ctx->m2m_ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT); - v4l2_m2m_streamoff(NULL, ctx->m2m_ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE); - } - mutex_unlock(&dev->dev_mutex); - - mutex_unlock(&dev->coda_mutex); - ctx = v4l2_m2m_get_curr_priv(dev->m2m_dev); - v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->m2m_ctx); -} - static u32 coda_supported_firmwares[] = { CODA_FIRMWARE_VERNUM(CODA_DX6, 2, 2, 5), CODA_FIRMWARE_VERNUM(CODA_7541, 1, 4, 50), @@ -3575,7 +3555,6 @@ static int coda_probe(struct platform_device *pdev) spin_lock_init(&dev->irqlock); INIT_LIST_HEAD(&dev->instances); - INIT_DELAYED_WORK(&dev->timeout, coda_timeout); dev->plat_dev = pdev; dev->clk_per = devm_clk_get(&pdev->dev, "per"); @@ -3682,6 +3661,12 @@ static int coda_probe(struct platform_device *pdev) return -ENOMEM; } + dev->workqueue = alloc_workqueue("coda", WQ_UNBOUND | WQ_MEM_RECLAIM, 1); + if (!dev->workqueue) { + dev_err(&pdev->dev, "unable to alloc workqueue\n"); + return -ENOMEM; + } + platform_set_drvdata(pdev, dev); return coda_firmware_request(dev); @@ -3697,6 +3682,7 @@ static int coda_remove(struct platform_device *pdev) if (dev->alloc_ctx) vb2_dma_contig_cleanup_ctx(dev->alloc_ctx); v4l2_device_unregister(&dev->v4l2_dev); + destroy_workqueue(dev->workqueue); if (dev->iram.vaddr) gen_pool_free(dev->iram_pool, (unsigned long)dev->iram.vaddr, dev->iram.size); -- 2.0.0 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html