Re: [Xen-devel] Ping: Re: [PATCH 2/3] VT-d: correct dma_msi_set_affinity()

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 15.12.16 at 13:52, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 15/12/16 09:54, Jan Beulich wrote:
>>>>> On 09.12.16 at 09:47, <JBeulich@xxxxxxxx> wrote:
>>>>>> On 08.12.16 at 18:33, <andrew.cooper3@xxxxxxxxxx> wrote:
>>>> On 08/12/16 16:01, Jan Beulich wrote:
>>>>> That commit ("VT-d: use msi_compose_msg()) together with 15aa6c6748
>>>> Which commit?
>>> Oops - initially I had intended the title to include the hash: 83cd2038fe.
>>> I've adjusted the text.
>>>
>>>>> ("amd iommu: use base platform MSI implementation") introducing the use
>>>>> of a per-CPU scratch CPU mask went too far: dma_msi_set_affinity() may,
>>>>> at least in theory, be called in interrupt context, and hence the use
>>>>> of that scratch variable is not correct.
>>>>>
>>>>> Since the function overwrites the destination information anyway,
>>>>> allow msi_compose_msg() to be called with a NULL CPU mask, avoiding the
>>>>> use of that scratch variable.
>>>> Which function overwrites what?  I can't see dma_msi_set_affinity()
>>>> doing anything to clobber msg.dest32, so I don't understand why this
>>>> change is correct.
>>> msg.dest32 simply isn't being used. msg is local to that function, so
>>> all that matters is which fields the function consumes. Is uses only
>>> address and data, and updates address to properly specify the
>>> intended destination. To guard against stale data (in
>>> iommu->msi.msg), it may be reasonable to nevertheless set dest32
>>> before storing msg into that field.
>> Any further thoughts here? Do I need to resend with that one line
>> added?
> 
> msg is tiny.  I'd suggest just initialising to 0 at the start of the
> function.

Why? msi_compose_msg() does exactly that as its very first action.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel