Re: [Qemu-devel] [PATCH 1/9] s390x/css: fix cc handling for XSCH

[Cornelia Huck <cohuck@redhat.com>]

On Thu, 31 Aug 2017 09:32:49 +0200
Thomas Huth <thuth@redhat.com> wrote:

> On 31.08.2017 08:38, Cornelia Huck wrote:
> > On Thu, 31 Aug 2017 07:51:17 +0200
> > Thomas Huth <thuth@redhat.com> wrote:
> >   
> >> On 30.08.2017 18:36, Halil Pasic wrote:  
> >>> The function ioinst_handle_xsch is presenting cc 2 when it's supposed to
> >>> present cc 1 and the other way around, because css_do_xsch has the error
> >>> codes mixed up. Fixing the error codes also fixes the priority.
> >>>
> >>> Let us fix this.    
> >>
> >> (Nit: In case you respin, I'd suggest to remove the last sentence. You
> >> already used "fix" two times in the previous one)  
> > 
> > I can also remove it on applying, if Halil agrees (I have not yet
> > reviewed it, though).
> >   
> >>  
> >>> Signed-off-by: Halil Pasic <pasic@linux.vnet.ibm.com>
> >>> Reported-by: Pierre Morel<pmorel@linux.vnet.ibm.com>    
> >>
> >> Space missing -------------^  
> > 
> > And I can also add that space on applying :)
> >   
> >>  
> >>> ---
> >>>  hw/s390x/css.c | 4 ++--
> >>>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/hw/s390x/css.c b/hw/s390x/css.c
> >>> index 1880b1a0ff..a50fb0727e 100644
> >>> --- a/hw/s390x/css.c
> >>> +++ b/hw/s390x/css.c
> >>> @@ -1281,12 +1281,12 @@ int css_do_xsch(SubchDev *sch)
> >>>          (!(s->ctrl &
> >>>             (SCSW_ACTL_RESUME_PEND | SCSW_ACTL_START_PEND | SCSW_ACTL_SUSP))) ||
> >>>          (s->ctrl & SCSW_ACTL_SUBCH_ACTIVE)) {
> >>> -        ret = -EINPROGRESS;
> >>> +        ret = -EBUSY;
> >>>          goto out;
> >>>      }
> >>>  
> >>>      if (s->ctrl & SCSW_CTRL_MASK_STCTL) {
> >>> -        ret = -EBUSY;
> >>> +        ret = -EINPROGRESS;
> >>>          goto out;
> >>>      }    
> >>
> >> Using both, EBUSY and EINPROGRESS as error codes sounds very confusing
> >> to me here ... what's the difference between busy and in-progress? So
> >> while you're at it, maybe you could replace the code for CC 2 ("CANCEL
> >> SUBCHANNEL not applicable") with a different error code?  
> > 
> > IIRC, I used these two as they matched my idea of what happens best
> > (there's a difference between "subchannel is busy" and "the I/O is
> > already in progress, too late to cancel"). "xsch not applicable" is
> > very hard to translate to an Unix error code :/  
> 
> OK, the codes make more sense with your description ==> Maybe simply add
> a proper comment for each of the return codes?

Taking a step back and looking at the other I/O instructions and their
implementation in qemu:

- For those instructions that can set it, cc 1 is set when the
  subchannel is status pending. That's usually the "default" branch in
  ioinst.c.
- cc 2 is set when the subchannel is "busy" (or, in the case of xsch,
  "not applicable for xsch"... sigh) This is supposed to be handled via
  -EBUSY.

So, there are actually two problems with the current implementation of
xsch:

- The return codes are switched around, which this patch fixes.
- But "status pending" should also take precedence over "not
  applicable", if I read the PoP correctly, so the second if needs to
  be moved up.

And yes, it makes sense do add some comments...