Re: [Qemu-devel] [RFC PATCH 04/33] blockjob: Introduce block_job_ratelimit_get_delay()

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 3006 bytes --]

On 04/24/2018 10:24 AM, Kevin Wolf wrote:
> This gets us rid of more direct accesses to BlockJob fields from the
> job drivers.
> 
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> ---

> +++ b/include/block/blockjob_int.h
> @@ -166,6 +166,14 @@ void block_job_sleep_ns(BlockJob *job, int64_t ns);
>  void block_job_yield(BlockJob *job);
>  
>  /**
> + * block_job_ratelimit_get_delay:
> + *
> + * Calculate and return delay for the next request in ns. See the docuemntation

s/docuemntation/documentation/

> + * of ratelimit_calculate_delay() for details.
> + */
> +int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n);

Signed return type matches ratelimit_calculate_delay, but what does a
negative return really mean?  Hmm, it looks like it can't actually
return a negative, other than if qemu_clock_get_ns() returns a negative
on some sort of error (which appears to be unlikely - we don't even
bother checking for it, even though it would mess things up if it did
happen) - all other return paths are 0 or a subtraction between time
values that can't overflow a 64-bit integer in our lifetime.  Should we
audit the source to flip the callers to use an unsigned value, perhaps
as a separate cleanup?

> +
> +/**
>   * block_job_pause_all:
>   *
>   * Asynchronously pause all jobs.
> diff --git a/block/backup.c b/block/backup.c
> index 8468fd9f94..3f3ec6e408 100644
> --- a/block/backup.c
> +++ b/block/backup.c
> @@ -325,21 +325,17 @@ static void backup_complete(BlockJob *job, void *opaque)
>  
>  static bool coroutine_fn yield_and_check(BackupBlockJob *job)
>  {
> +    uint64_t delay_ns;
> +
>      if (block_job_is_cancelled(&job->common)) {
>          return true;
>      }
>  
> -    /* we need to yield so that bdrv_drain_all() returns.
> -     * (without, VM does not reboot)
> -     */
> -    if (job->common.speed) {
> -        uint64_t delay_ns = ratelimit_calculate_delay(&job->common.limit,
> -                                                      job->bytes_read);
> -        job->bytes_read = 0;
> -        block_job_sleep_ns(&job->common, delay_ns);
> -    } else {
> -        block_job_sleep_ns(&job->common, 0);
> -    }
> +    /* we need to yield even for delay_ns = 0 so that bdrv_drain_all() returns.
> +     * (without it, the VM does not reboot) */

As long as you are touching this, is it worth further improving the grammar?

/* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
return; without a yield, the VM would not reboot. */

> +    delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read);
> +    job->bytes_read = 0;
> +    block_job_sleep_ns(&job->common, delay_ns);
>  
>      if (block_job_is_cancelled(&job->common)) {
>          return true;

With typo fixes,
Reviewed-by: Eric Blake <eblake@redhat.com>

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 619 bytes --]