Re: [PATCH kvmtool 6/9] disk/aio: Fix AIO thread

[Andre Przywara <andre.przywara@xxxxxxx>]

On Mon, 18 Feb 2019 13:06:59 +0000
Jean-Philippe Brucker <jean-philippe.brucker@xxxxxxx> wrote:

> Currently when the kernel completes a batch of AIO requests and signals it
> via eventfd, we retrieve at most AIO_MAX events (256), and ignore the
> rest. Call io_getevents() again in case more events are pending.

Did you encounter this case in a test or is this just theoretical?

But I am just curious, the patch looks alright:

> Signed-off-by: Jean-Philippe Brucker <jean-philippe.brucker@xxxxxxx>

Reviewed-by: Andre Przywara <andre.przywara@xxxxxxx>

Cheers,
Andre.

> ---
>  disk/aio.c | 21 ++++++++++++++++-----
>  1 file changed, 16 insertions(+), 5 deletions(-)
> 
> diff --git a/disk/aio.c b/disk/aio.c
> index 007415c69..1fcf36857 100644
> --- a/disk/aio.c
> +++ b/disk/aio.c
> @@ -66,20 +66,31 @@ ssize_t raw_image__write_async(struct disk_image *disk, u64 sector,
>  			   offset, disk->evt, param);
>  }
>  
> -static void *disk_aio_thread(void *param)
> +static int disk_aio_get_events(struct disk_image *disk)
>  {
> -	struct disk_image *disk = param;
>  	struct io_event event[AIO_MAX];
>  	struct timespec notime = {0};
>  	int nr, i;
> +
> +	do {
> +		nr = io_getevents(disk->ctx, 1, ARRAY_SIZE(event), event, &notime);
> +		for (i = 0; i < nr; i++)
> +			disk->disk_req_cb(event[i].data, event[i].res);
> +	} while (nr > 0);
> +
> +	return 0;
> +}
> +
> +static void *disk_aio_thread(void *param)
> +{
> +	struct disk_image *disk = param;
>  	u64 dummy;
>  
>  	kvm__set_thread_name("disk-image-io");
>  
>  	while (read(disk->evt, &dummy, sizeof(dummy)) > 0) {
> -		nr = io_getevents(disk->ctx, 1, ARRAY_SIZE(event), event, &notime);
> -		for (i = 0; i < nr; i++)
> -			disk->disk_req_cb(event[i].data, event[i].res);
> +		if (disk_aio_get_events(disk))
> +			break;
>  	}
>  
>  	return NULL;