Re: [PATCH] block: add debug when block operations finish

[Kevin Wolf <kwolf@redhat.com>]

Am 09.09.2026 um 21:56 hat Jorge Merlino geschrieben:
> Add new debug tracepoints when block read and write operations finish to
> be able to measure latency.
> 
> Signed-off-by: Jorge Merlino <jorge.merlino@canonical.com>

The subject line should mention trace points. Maybe just "block: Add
tracepoiont at end of read/write"?

> This patch adds debug tracepoints when block read and write operations
> finish. The log includes the parameters used when the operation was
> initiated so that the start and finish messages can be easily
> correlated. Also the return value of the operation is logged.
> 
> This can be used to analyze latency issues with block operations.

Yes, this seems useful.

I wonder if we shouldn't make sure that the start and end traces always
come in pairs. Specifically...

>  block/block-backend.c | 2 ++
>  block/trace-events    | 2 ++
>  2 files changed, 4 insertions(+)
> 
> diff --git a/block/block-backend.c b/block/block-backend.c
> index 37ba7e9fc4..75d3cd1949 100644
> --- a/block/block-backend.c
> +++ b/block/block-backend.c
> @@ -1364,6 +1364,7 @@ blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,

Adding more context here:

    trace_blk_co_preadv(blk, bs, offset, bytes, flags);

    ret = blk_check_byte_request(blk, offset, bytes);
    if (ret < 0) {
        return ret;
    }

Should this error return also call trace_blk_co_preadv_done()?

    bdrv_inc_in_flight(bs);

    /* throttling disk I/O */
    if (qatomic_load_acquire(&tgm->throttle_state)) {
        throttle_group_co_io_limits_intercept(tgm, bytes, THROTTLE_READ);
    }

>  
>      ret = bdrv_co_preadv_part(blk->root, offset, bytes, qiov, qiov_offset,
>                                flags);
> +    trace_blk_co_preadv_done(blk, bs, offset, bytes, ret);
>      bdrv_dec_in_flight(bs);
>      return ret;
>  }
> @@ -1442,6 +1443,7 @@ blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
>  
>      ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
>                                 flags);
> +    trace_blk_co_pwritev_done(blk, bs, offset, bytes, ret);
>      bdrv_dec_in_flight(bs);
>      return ret;
>  }

Same question for writes.

Kevin