Re: [Xen-devel] [PATCH v3] x86/HVM: p2m_ram_ro is incompatible with device pass-through

[Jan Beulich <jbeulich@xxxxxxxx>]

On 01.10.2019 20:00, Andrew Cooper wrote:
> On 01/10/2019 10:07, Jan Beulich wrote:
>> The write-discard property of the type can't be represented in IOMMU
>> page table entries. Make sure the respective checks / tracking can't
>> race, by utilizing the domain lock. The other sides of the sharing/
>> paging/log-dirty exclusion checks should subsequently perhaps also be
>> put under that lock then.
>>
>> This also fixes an unguarded d->arch.hvm access.
>>
>> Take the opportunity and also convert neighboring bool_t to bool in
>> struct hvm_domain.
>>
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
>> Reviewed-by: Paul Durrant <paul.durrant@xxxxxxxxxx>
> 
> Leaving aside the logdirty change which is an automatic no from me in
> this form,

There's no log-dirty change here, that line is just patch context.
All the patch does is also include that case in the comment next to
the check.

>  I can't see how this change avoids regressing the status quo.
> 
> Qemu really does set this flag for option ROMs, which will typically be
> the iPXE image for net booting.  Currently, netbooting an HVM guest with
> device passed through to it does work, because it is part of XenServers
> basic testing.
> 
> It is entirely possible that this only "works" as long as DMA doesn't
> touch the option ROM, and while this might not be ideal, it seems to be
> less bad behaviour than breaking existing VMs using this configuration.

Hmm, yes, I have to admit I didn't consider this case, mis-remembering
that HVMMEM_ram_ro would be more "special" than it really is. I guess
I'll withdraw the patch then.

Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel