Re: [Qemu-devel] [RFC v4 5/9] configure: Enable/disable new qemu_{ld, st} excl insns

[Alex Bennée]

Aurelien Jarno <address@hidden> writes:

> On 2015-08-09 10:51, Alex Bennée wrote:
>> 
>> Aurelien Jarno <address@hidden> writes:
>> 
>> > On 2015-08-09 09:11, Alex Bennée wrote:
>> >> 
>> >> Aurelien Jarno <address@hidden> writes:
>> >> 
>> >> > On 2015-08-07 19:03, Alvise Rigo wrote:
>> >> >> Introduce the new --enable-tcg-ldst-excl configure option to enable the
>> >> >> LL/SC operations only for those backends that support them.
>> >> >> 
>> >> >> Suggested-by: Jani Kokkonen <address@hidden>
>> >> >> Suggested-by: Claudio Fontana <address@hidden>
>> >> >> Signed-off-by: Alvise Rigo <address@hidden>
>> >> >> ---
>> >> >>  configure | 21 +++++++++++++++++++++
>> >> >>  1 file changed, 21 insertions(+)
>> >> >
>> >> > We have seen that for this kind of patch, it's better to add support in
>> >> > all backends, otherwise it takes ages to get all the backends converted.
>> >> > I think you should involve the backend maintainers. I can try to provide
>> >> > the corresponding patches for mips and ia64.
>> >> 
>> >> We discussed this on the last MTTCG call and agree. However we will need
>> >> help from the other TCG maintainers for the backends. The changes should
>> >> be fairly mechanical though.
>> >> 
>> >> However in the spirit of keeping trees building in the meantime should
>> >> we change this from a configure option to just a static option for each
>> >> given backend as it is converted?
>> >
>> > I am not even sure we need a static option. I guess providing we are
>> > doing that early enough in the 2.5 cycle, we can just add the new ops
>> > and start using them. Of course we should put the backends maintainers
>> > in the loop so they can fix that quickly and don't get a surprise weeks
>> > afters.
>> >
>> > That said, please see my other email, I am not sure we actually need to
>> > modify backends, I think we can implement these new "ops" through
>> > tcg-runtime.
>> 
>> We still need to ensure "normal" ld/st operations trip the exclusive bit
>> though.
>
> Isn't that taken care of by the TLB_EXCL flag, causing pages in
> exclusive mode to always go through the slow path? Then we should not
> need to modify the backends as they already check for non zero bits in
> this area.

Ahh I see. Yes of course that sounds like a good idea. 

-- 
Alex Bennée