[media] coda/imx-vdoa: always wait for job completion
authorLucas Stach <l.stach@pengutronix.de>
Wed, 5 Apr 2017 13:09:55 +0000 (10:09 -0300)
committerMauro Carvalho Chehab <mchehab@s-opensource.com>
Tue, 20 Jun 2017 12:04:21 +0000 (09:04 -0300)
As long as only one CODA context is running we get alternating device_run()
and wait_for_completion() calls, but when more then one CODA context is
active, other VDOA slots can be inserted between those calls for one context.

Make sure to wait on job completion before running a different context and
before destroying the currently active context.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Hans Verkuil <hansverk@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
drivers/media/platform/coda/imx-vdoa.c

index 669a4c82f1ffa4c79b5e47173536faedecf52c34..df9b71621420cccd8712015d91d9e7d93424ba0d 100644 (file)
@@ -101,6 +101,8 @@ struct vdoa_ctx {
        struct vdoa_data        *vdoa;
        struct completion       completion;
        struct vdoa_q_data      q_data[2];
+       unsigned int            submitted_job;
+       unsigned int            completed_job;
 };
 
 static irqreturn_t vdoa_irq_handler(int irq, void *data)
@@ -114,7 +116,7 @@ static irqreturn_t vdoa_irq_handler(int irq, void *data)
 
        curr_ctx = vdoa->curr_ctx;
        if (!curr_ctx) {
-               dev_dbg(vdoa->dev,
+               dev_warn(vdoa->dev,
                        "Instance released before the end of transaction\n");
                return IRQ_HANDLED;
        }
@@ -127,19 +129,44 @@ static irqreturn_t vdoa_irq_handler(int irq, void *data)
        } else if (!(val & VDOAIST_EOT)) {
                dev_warn(vdoa->dev, "Spurious interrupt\n");
        }
+       curr_ctx->completed_job++;
        complete(&curr_ctx->completion);
 
        return IRQ_HANDLED;
 }
 
+int vdoa_wait_for_completion(struct vdoa_ctx *ctx)
+{
+       struct vdoa_data *vdoa = ctx->vdoa;
+
+       if (ctx->submitted_job == ctx->completed_job)
+               return 0;
+
+       if (!wait_for_completion_timeout(&ctx->completion,
+                                        msecs_to_jiffies(300))) {
+               dev_err(vdoa->dev,
+                       "Timeout waiting for transfer result\n");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(vdoa_wait_for_completion);
+
 void vdoa_device_run(struct vdoa_ctx *ctx, dma_addr_t dst, dma_addr_t src)
 {
        struct vdoa_q_data *src_q_data, *dst_q_data;
        struct vdoa_data *vdoa = ctx->vdoa;
        u32 val;
 
+       if (vdoa->curr_ctx)
+               vdoa_wait_for_completion(vdoa->curr_ctx);
+
        vdoa->curr_ctx = ctx;
 
+       reinit_completion(&ctx->completion);
+       ctx->submitted_job++;
+
        src_q_data = &ctx->q_data[V4L2_M2M_SRC];
        dst_q_data = &ctx->q_data[V4L2_M2M_DST];
 
@@ -177,21 +204,6 @@ void vdoa_device_run(struct vdoa_ctx *ctx, dma_addr_t dst, dma_addr_t src)
 }
 EXPORT_SYMBOL(vdoa_device_run);
 
-int vdoa_wait_for_completion(struct vdoa_ctx *ctx)
-{
-       struct vdoa_data *vdoa = ctx->vdoa;
-
-       if (!wait_for_completion_timeout(&ctx->completion,
-                                        msecs_to_jiffies(300))) {
-               dev_err(vdoa->dev,
-                       "Timeout waiting for transfer result\n");
-               return -ETIMEDOUT;
-       }
-
-       return 0;
-}
-EXPORT_SYMBOL(vdoa_wait_for_completion);
-
 struct vdoa_ctx *vdoa_context_create(struct vdoa_data *vdoa)
 {
        struct vdoa_ctx *ctx;
@@ -218,6 +230,11 @@ void vdoa_context_destroy(struct vdoa_ctx *ctx)
 {
        struct vdoa_data *vdoa = ctx->vdoa;
 
+       if (vdoa->curr_ctx == ctx) {
+               vdoa_wait_for_completion(vdoa->curr_ctx);
+               vdoa->curr_ctx = NULL;
+       }
+
        clk_disable_unprepare(vdoa->vdoa_clk);
        kfree(ctx);
 }