[Sound-open-firmware] [PATCH] sof: Turn IRQs off globally when send command to pipeline component

[Pierre-Louis Bossart]

On 3/27/18 4:17 AM, Wu Zhigang wrote:
> When pipeline processing the command, we have to turn off the IRQs.
> If not, the DMA IRQ will interrupt the pipeline state set process,
> when pipeline is scheduled, the states inconsistence will cause xrun.

we disable the irqs but then will walk the graph from host to DAI with 
component_op_downstream to pass a COMPS_OPS_CMD with in turn will call 
all .cmd callbacks from all components

This could take a significant amount of time and lead to interrupts 
unrelated to that specific pipeline not being serviced, leading to xruns.

what am I missing?

> 
> Signed-off-by: Wu Zhigang <zhigang.wu at linux.intel.com>
> 
> ---
> Tested with
> apl-gpmrb board
> 
> kernel: https://github.com/plbossart/sound.git  branch: topic/sof-v4.14
> 	95d77adbaadc44b28c8975a3714f70824d1a8529
> 
> firmware: git://git.alsa-project.org/sound-open-firmware.git  branch: 1.1-stable
>          210989dffeea811de2370fccb7cf5d53106b1e6e
> 
> tools:	branch: 1.1-stable
> 	cc91c73aa3e91eea35abdeb76d578b97f718feff
> ---
>   src/audio/pipeline.c | 5 +++--
>   1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/src/audio/pipeline.c b/src/audio/pipeline.c
> index 1b7f248..23be503 100644
> --- a/src/audio/pipeline.c
> +++ b/src/audio/pipeline.c
> @@ -627,6 +627,7 @@ int pipeline_cmd(struct pipeline *p, struct comp_dev *host, int cmd,
>   {
>   	struct op_data op_data;
>   	int ret;
> +	uint32_t flags;
>   
>   	trace_pipe("cmd");
>   
> @@ -635,7 +636,7 @@ int pipeline_cmd(struct pipeline *p, struct comp_dev *host, int cmd,
>   	op_data.cmd = cmd;
>   	op_data.cmd_data = data;
>   
> -	spin_lock(&p->lock);
> +	spin_lock_irq(&p->lock, flags);
>   
>   	if (host->params.direction == SOF_IPC_STREAM_PLAYBACK) {
>   		/* send cmd downstream from host to DAI */
> @@ -651,7 +652,7 @@ int pipeline_cmd(struct pipeline *p, struct comp_dev *host, int cmd,
>   		trace_error_value(cmd);
>   	}
>   
> -	spin_unlock(&p->lock);
> +	spin_unlock_irq(&p->lock, flags);
>   	return ret;
>   }
>   
>