[Crash-utility] [PATCH v2] Handle blk_mq_ctx member changes for kernels v5.16-rc1~75^2~44

[lijiang]

On Mon, Dec 27, 2021 at 10:24 AM HAGIO KAZUHITO(萩尾　一仁)
<k-hagio-ab at nec.com> wrote:
>
> Hi Lianbo,
>
> thanks for the update,
>
> -----Original Message-----
> > Subject: [PATCH v2] Handle blk_mq_ctx member changes for kernels v5.16-rc1~75^2~44
>
> only "5.16-rc1" is fine, as we have the commit id below.
>
> > Kernel commit <9a14d6ce4135> ("block: remove debugfs blk_mq_ctx
> > dispatched/merged/completed attributes") removed the member
> > rq_dispatched and rq_completed from struct blk_mq_ctx. Without
> > this patch, crash will fail with the following error:
> >
> > crash> dev -d
> > MAJOR GENDISK            NAME       REQUEST_QUEUE      TOTAL ASYNC  SYNC
> >
> > dev: invalid structure member offset: blk_mq_ctx_rq_dispatched
> >      FILE: dev.c  LINE: 4229  FUNCTION: get_one_mctx_diskio()
> >
> > Signed-off-by: Lianbo Jiang <lijiang at redhat.com>
> > ---
> > change since v1:
> > [1] display prompt information "(not supported)" instead of the
> > statistical result if kernel does not support it
> >
> >  dev.c | 56 ++++++++++++++++++++++++++++++++++++++------------------
> >  1 file changed, 38 insertions(+), 18 deletions(-)
> >
> > diff --git a/dev.c b/dev.c
> > index effe789f38d8..c5e38e1c5376 100644
> > --- a/dev.c
> > +++ b/dev.c
> > @@ -4246,6 +4246,10 @@ get_mq_diskio(unsigned long q, unsigned long *mq_count)
> >       unsigned long mctx_addr;
> >       struct diskio tmp;
> >
> > +     if (!MEMBER_EXISTS("blk_mq_ctx", "rq_dispatched") &&
> > +             !MEMBER_EXISTS("blk_mq_ctx", "rq_completed"))
> > +             return;
> > +
>
> It's better to use INVALID_MEMBER() here, because they are in the offset_table and
> it doesn't call gdb function.  And, both of them should be gone on 5.16 and later,
> but let's use 'or' for safe:
>
>         if (INVALID_MEMBER(blk_mq_ctx_rq_dispatched) ||
>             INVALID_MEMBER(blk_mq_ctx_rq_completed))
>                 return;
>
> >       memset(&tmp, 0x00, sizeof(struct diskio));
> >
> >       readmem(q + OFFSET(request_queue_queue_ctx), KVADDR, &queue_ctx,
> > @@ -4475,24 +4479,40 @@ display_one_diskio(struct iter *i, unsigned long gendisk, ulong flags)
> >               && (io.read + io.write == 0))
> >               return;
> >
> > -     fprintf(fp, "%s%s%s  %s%s%s%s  %s%5d%s%s%s%s%s",
> > -             mkstring(buf0, 5, RJUST|INT_DEC, (char *)(unsigned long)major),
> > -             space(MINSPACE),
> > -             mkstring(buf1, VADDR_PRLEN, LJUST|LONG_HEX, (char *)gendisk),
> > -             space(MINSPACE),
> > -             mkstring(buf2, 10, LJUST, disk_name),
> > -             space(MINSPACE),
> > -             mkstring(buf3, VADDR_PRLEN <= 11 ? 11 : VADDR_PRLEN,
> > -                      LJUST|LONG_HEX, (char *)queue_addr),
> > -             space(MINSPACE),
> > -             io.read + io.write,
> > -             space(MINSPACE),
> > -             mkstring(buf4, 5, RJUST|INT_DEC,
> > -                     (char *)(unsigned long)io.read),
> > -             space(MINSPACE),
> > -             mkstring(buf5, 5, RJUST|INT_DEC,
> > -                     (char *)(unsigned long)io.write),
> > -             space(MINSPACE));
> > +     if (!MEMBER_EXISTS("blk_mq_ctx", "rq_dispatched") &&
> > +             !MEMBER_EXISTS("blk_mq_ctx", "rq_completed"))
>
> Same as above.
>
> And this stops displaying the statistics of all devices on 5.16 kernel.
> I think that we can stop only mq devices, so how about this?
>
>         if (use_mq_interface(queue_addr) &&
>             (INVALID_MEMBER(blk_mq_ctx_rq_dispatched) ||
>              INVALID_MEMBER(blk_mq_ctx_rq_completed)))
>
> With this, on a 5.16-rc2 kernel:
>
> crash-dev> dev -d
> MAJOR GENDISK            NAME       REQUEST_QUEUE      TOTAL ASYNC  SYNC
>     8 ffff92bbd102e400   sdb        ffff92bbf04e3678     (not supported)
>     8 ffff92bbd1014400   sdc        ffff92bbf04e26e8     (not supported)
>     8 ffff92bbc9bff200   sdd        ffff92bbf04e07c8     (not supported)
>     8 ffff92bbd9676200   sda        ffff92bbf04e0000     (not supported)
>    11 ffff92c347da5e00   sr0        ffff92bbd9c3e528     (not supported)
>   253 ffff92bbca726e00   dm-0       ffff92bbcad7dd60       0     0     0
>   253 ffff92bbe03f1000   dm-1       ffff92bbe1d587c8       0     0     0
>   253 ffff92bbe02fa800   dm-2       ffff92bc01b11f20       0     0     0
>
> where dm-X devices don't use mq interface.
>

Looks better.

> I've modified the patch and attached, you can merge it if you are OK.
>
Will apply this one after some tests.

Thanks.
Lianbo