Re: [PATCH v7 1/1] vfio/nvgpu: Add vfio pci variant module for grace hopper

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

On Wed, 23 Aug 2023 14:18:52 +0000
Ankit Agrawal <ankita@nvidia.com> wrote:

> >> +
> >> +     /*
> >> +      * Handle read on the BAR2 region. Map to the target device memory
> >> +      * physical address and copy to the request read buffer.
> >> +      */
> >> +     if (copy_to_user(buf, (u8 *)addr + offset, read_count))
> >> +             return -EFAULT;  
> >
> > Just to verify, does this memory allow access of arbitrary alignment
> > and size?  
> 
> Please correct me if I'm wrong, but based on following gdb dump data on
> the corresponding MemoryRegion->ops, unaligned access isn't supported, and
> a read of size upto 8 may be done. 
> 
> (gdb) p/x *(mr->ops)
> $7 = {read = 0xaaab5e0b1c50, write = 0xaaab5e0b1a50, read_with_attrs = 0x0, write_with_attrs = 0x0,
>   endianness = 0x2, valid = {min_access_size = 0x1, max_access_size = 0x8, unaligned = 0x0, accepts = 0x0},
>   impl = {min_access_size = 0x1, max_access_size = 0x8, unaligned = 0x0}}

This is QEMU policy relative to this region, the kernel interface is
not exclusive to QEMU usage.  Thanks,

Alex