Re: [PATCH v6 2/6] iommufd: Add IOMMU_HWPT_INVALIDATE

[Nicolin Chen <nicolinc@nvidia.com>]

On Tue, Nov 28, 2023 at 08:03:35AM +0000, Tian, Kevin wrote:
> > From: Nicolin Chen <nicolinc@nvidia.com>
> > Sent: Tuesday, November 28, 2023 3:53 AM
> >
> > On Fri, Nov 24, 2023 at 02:36:29AM +0000, Tian, Kevin wrote:
> >
> > > > > > > > >> + * @out_driver_error_code: Report a driver speicifc error code
> > > > upon
> > > > > > > > failure.
> > > > > > > > >> + *                         It's optional, driver has a choice to fill it or
> > > > > > > > >> + *                         not.
> > > > > > > > >
> > > > > > > > > Being optional how does the user tell whether the code is filled
> > or
> > > > not?
> > > > > >
> > > > > > Well, naming it "error_code" indicates zero means no error while
> > > > > > non-zero means something? An error return from this ioctl could
> > > > > > also tell the user space to look up for this driver error code,
> > > > > > if it ever cares.
> > > > >
> > > > > probably over-thinking but I'm not sure whether zero is guaranteed to
> > > > > mean no error in all implementations...
> > > >
> > > > Well, you are right. Usually HW conveniently raises a flag in a
> > > > register to indicate something wrong, yet it is probably unsafe
> > > > to say it definitely.
> > > >
> > >
> > > this reminds me one open. What about an implementation having
> > > a hierarchical error code layout e.g. one main error register with
> > > each bit representing an error category then multiple error code
> > > registers each for one error category? In this case probably
> > > a single out_driver_error_code cannot carry that raw information.
> >
> > Hmm, good point.
> >
> > > Instead the iommu driver may need to define a customized error
> > > code convention in uapi header which is converted from the
> > > raw error information.
> > >
> > > From this angle should we simply say that the error code definition
> > > must be included in the uapi header? If raw error information can
> > > be carried by this field then this hw can simply say that the error
> > > code format is same as the hw spec defines.
> > >
> > > With that explicit information then the viommu can easily tell
> > > whether error code is filled or not based on its own convention.
> >
> > That'd be to put this error_code field into the driver uAPI
> > structure right?
> >
> > I also thought about making this out_driver_error_code per HW.
> > Yet, an error can be either per array or per entry/quest. The
> > array-related error should be reported in the array structure
> > that is a core uAPI, v.s. the per-HW entry structure. Though
> > we could still report an array error in the entry structure
> > at the first entry (or indexed by "array->entry_num")?
> >
> 
> why would there be an array error? array is just a software
> entity containing actual HW invalidation cmds. If there is
> any error with the array itself it should be reported via
> ioctl errno.

User array reading is a software operation, but kernel array
reading is a hardware operation that can raise an error when
the memory location to the array is incorrect or so.

With that being said, I think errno (-EIO) could do the job,
as you suggested too.

Thanks
Nic

> Jason, how about your opinion? I didn't spot big issues
> except this one. Hope it can make into 6.8.