Re: [PATCH v2 1/2] vfio/mdev: add version attribute for mdev device

[Alex Williamson <alex.williamson@redhat.com>]

On Tue, 14 May 2019 09:43:44 +0200
Erik Skultety <eskultet@redhat.com> wrote:

> On Tue, May 14, 2019 at 03:32:19AM -0400, Yan Zhao wrote:
> > On Tue, May 14, 2019 at 03:20:40PM +0800, Erik Skultety wrote:  
> > > On Tue, May 14, 2019 at 02:12:35AM -0400, Yan Zhao wrote:  
> > > > On Mon, May 13, 2019 at 09:28:04PM +0800, Erik Skultety wrote:  
> > > > > On Fri, May 10, 2019 at 11:48:38AM +0200, Cornelia Huck wrote:  
> > > > > > On Fri, 10 May 2019 10:36:09 +0100
> > > > > > "Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:
> > > > > >  
> > > > > > > * Cornelia Huck (cohuck@redhat.com) wrote:  
> > > > > > > > On Thu, 9 May 2019 17:48:26 +0100
> > > > > > > > "Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:
> > > > > > > >  
> > > > > > > > > * Cornelia Huck (cohuck@redhat.com) wrote:  
> > > > > > > > > > On Thu, 9 May 2019 16:48:57 +0100
> > > > > > > > > > "Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:
> > > > > > > > > >  
> > > > > > > > > > > * Cornelia Huck (cohuck@redhat.com) wrote:  
> > > > > > > > > > > > On Tue, 7 May 2019 15:18:26 -0600
> > > > > > > > > > > > Alex Williamson <alex.williamson@redhat.com> wrote:
> > > > > > > > > > > >  
> > > > > > > > > > > > > On Sun,  5 May 2019 21:49:04 -0400
> > > > > > > > > > > > > Yan Zhao <yan.y.zhao@intel.com> wrote:  
> > > > > > > > > > > >  
> > > > > > > > > > > > > > +  Errno:
> > > > > > > > > > > > > > +  If vendor driver wants to claim a mdev device incompatible to all other mdev
> > > > > > > > > > > > > > +  devices, it should not register version attribute for this mdev device. But if
> > > > > > > > > > > > > > +  a vendor driver has already registered version attribute and it wants to claim
> > > > > > > > > > > > > > +  a mdev device incompatible to all other mdev devices, it needs to return
> > > > > > > > > > > > > > +  -ENODEV on access to this mdev device's version attribute.
> > > > > > > > > > > > > > +  If a mdev device is only incompatible to certain mdev devices, write of
> > > > > > > > > > > > > > +  incompatible mdev devices's version strings to its version attribute should
> > > > > > > > > > > > > > +  return -EINVAL;  
> > > > > > > > > > > > >
> > > > > > > > > > > > > I think it's best not to define the specific errno returned for a
> > > > > > > > > > > > > specific situation, let the vendor driver decide, userspace simply
> > > > > > > > > > > > > needs to know that an errno on read indicates the device does not
> > > > > > > > > > > > > support migration version comparison and that an errno on write
> > > > > > > > > > > > > indicates the devices are incompatible or the target doesn't support
> > > > > > > > > > > > > migration versions.  
> > > > > > > > > > > >
> > > > > > > > > > > > I think I have to disagree here: It's probably valuable to have an
> > > > > > > > > > > > agreed error for 'cannot migrate at all' vs 'cannot migrate between
> > > > > > > > > > > > those two particular devices'. Userspace might want to do different
> > > > > > > > > > > > things (e.g. trying with different device pairs).  
> > > > > > > > > > >
> > > > > > > > > > > Trying to stuff these things down an errno seems a bad idea; we can't
> > > > > > > > > > > get much information that way.  
> > > > > > > > > >
> > > > > > > > > > So, what would be a reasonable approach? Userspace should first read
> > > > > > > > > > the version attributes on both devices (to find out whether migration
> > > > > > > > > > is supported at all), and only then figure out via writing whether they
> > > > > > > > > > are compatible?
> > > > > > > > > >
> > > > > > > > > > (Or just go ahead and try, if it does not care about the reason.)  
> > > > > > > > >
> > > > > > > > > Well, I'm OK with something like writing to test whether it's
> > > > > > > > > compatible, it's just we need a better way of saying 'no'.
> > > > > > > > > I'm not sure if that involves reading back from somewhere after
> > > > > > > > > the write or what.  
> > > > > > > >
> > > > > > > > Hm, so I basically see two ways of doing that:
> > > > > > > > - standardize on some error codes... problem: error codes can be hard
> > > > > > > >   to fit to reasons
> > > > > > > > - make the error available in some attribute that can be read
> > > > > > > >
> > > > > > > > I'm not sure how we can serialize the readback with the last write,
> > > > > > > > though (this looks inherently racy).
> > > > > > > >
> > > > > > > > How important is detailed error reporting here?  
> > > > > > >
> > > > > > > I think we need something, otherwise we're just going to get vague
> > > > > > > user reports of 'but my VM doesn't migrate'; I'd like the error to be
> > > > > > > good enough to point most users to something they can understand
> > > > > > > (e.g. wrong card family/too old a driver etc).  
> > > > > >
> > > > > > Ok, that sounds like a reasonable point. Not that I have a better idea
> > > > > > how to achieve that, though... we could also log a more verbose error
> > > > > > message to the kernel log, but that's not necessarily where a user will
> > > > > > look first.  
> > > > >
> > > > > In case of libvirt checking the compatibility, it won't matter how good the
> > > > > error message in the kernel log is and regardless of how many error states you
> > > > > want to handle, libvirt's only limited to errno here, since we're going to do
> > > > > plain read/write, so our internal error message returned to the user is only
> > > > > going to contain what the errno says - okay, of course we can (and we DO)
> > > > > provide libvirt specific string, further specifying the error but like I
> > > > > mentioned, depending on how many error cases we want to distinguish this may be
> > > > > hard for anyone to figure out solely on the error code, as apps will most
> > > > > probably not parse the
> > > > > logs.
> > > > >
> > > > > Regards,
> > > > > Erik  
> > > > hi Erik
> > > > do you mean you are agreeing on defining common errors and only returning errno?  
> > >
> > > In a sense, yes. While it is highly desirable to have logs with descriptive
> > > messages which will help in troubleshooting tremendously, I wanted to point out
> > > that spending time with error logs may not be that worthwhile especially since
> > > most apps (like libvirt) will solely rely on using read(3)/write(3) to sysfs.
> > > That means that we're limited by the errnos available, so apart from
> > > reporting the generic system message we can't any more magic in terms of the
> > > error messages, so the driver needs to assure that a proper message is
> > > propagated to the journal and at best libvirt can direct the user (consumer) to
> > > look through the system logs for more info. I also agree with the point
> > > mentioned above that defining a specific errno is IMO not the way to go, as
> > > these would be just too specific for the read(3)/write(3) use case.
> > >
> > > That said, from libvirt POV as a consumer, I'd expect there to be truly only 2
> > > errors (I believe Alex has mentioned something similar in one of his responses
> > > in one of the threads):
> > >     a) read error indicating that an mdev type doesn't support migration
> > >         - I assume if one type doesn't support migration, none of the other
> > >           types exposed on the parent device do, is that a fair assumption?

I'd prefer not to make this assumption.  Let's leave open the
possibility that (for whatever reason) a vendor may choose to support
migration on some types, but not others.

> > >     b) write error indicating that the mdev types are incompatible for
> > >     migration
> > >
> > > Regards,
> > > Erik  
> > Thanks for this explanation.
> > so, can we arrive at below agreements?
> >
> > 1. "not to define the specific errno returned for a specific situation,
> > let the vendor driver decide, userspace simply needs to know that an errno on
> > read indicates the device does not support migration version comparison and
> > that an errno on write indicates the devices are incompatible or the target
> > doesn't support migration versions. "
> > 2. vendor driver should log detailed error reasons in kernel log.  
> 
> That would be my take on this, yes, but I open to hear any other suggestions and
> ideas I couldn't think of as well.

Kernel logging tends to be rather ineffective, it's surprisingly
difficult to get users to look in dmesg and it's not really a good
choice for scraping diagnostic information either.  I'd probably leave
this to vendor driver's discretion at this point.  Thanks,

Alex