Re: [Qemu-devel] [PATCH 0/3] vfio-ccw: support hsch/csch (kernel part)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, 5 Dec 2018 13:34:11 -0500
Farhan Ali <alifm@xxxxxxxxxxxxx> wrote:

> On 12/05/2018 07:54 AM, Cornelia Huck wrote:
> >> Yeah, that is perfectly clear, but it ain't the complete story. E.g.
> >> are subsequent commands blocking until the preceding command finishes
> >> is part of the interface. And what is good implementation depends on the
> >> answer. What I mean, I first need to understand how things are supposed
> >> to work (together) so I can double check that against the
> >> implementation. Otherwise all I can do is nitpicking.
> >>
> >> To get more tangible: we are in the middle of processing an SSCH request
> >> (e.g. doing the translation) when a HSCH comes in. What should happen?
> >> Should we start processing HSCH after he instruction part of SSCH is
> >> done -- which currently includes translation? Or should we -EBUSY? Or do
> >> we abort START related activities and do the HALT stuff?  
> > I think most of the sorting-out-the-operations stuff should be done by
> > the hardware itself, and we should not really try to enforce anything
> > special in our vfio code.
> > 
> > For your example, it might be best if a hsch is always accepted and
> > send on towards the hardware. Probably best to reflect back -EAGAIN if
> > we're currently processing another instruction from another vcpu, so
> > that the user space caller can retry. Same for ssch, if another ssch is
> > already being processed. We*could*  reflect cc 2 if the fctl bit is
> > already set, but that won't do for csch, so it is probably best to have
> > the hardware figure that out in any case.
> > 
> > If I read the code correctly, we currently reflect -EBUSY and not
> > -EAGAIN if we get a ssch request while already processing another one.
> > QEMU hands that back to the guest as a cc 2, which is not 100% correct.
> > In practice, we don't see this with Linux guests due to locking.
> >   
> 
> If we have a ssch and a csch immediately afterwards from userspace, will 
> we end up issuing csch first and then ssch to the hardware?
> 
> If I understand correctly, the ccw translation as part of the ssch can 
> be a slow operation so it might be possible we issue the csch first?
> In that case we won't actually clear the original start function as 
> intended.

When we start processing the ssch request (translation and so on), we
set the state to BUSY. This means that any csch request will get a
-EBUSY, no overtaking possible. (I think maybe I'll need to check what
this series looks like if I rebase it on top of Pierre's rework, as he
did some changes in the state machine.)

My idea above was to return -EAGAIN instead of -EBUSY, so that user
space can retry the operation.



[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [Kernel Development]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite Info]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Samba]     [Linux Media]     [Device Mapper]

  Powered by Linux