Re: [Xen-devel] [PATCH] PVH: vcpu info placement, loadselectors, and remove debug printk.

[Jan Beulich]

>>> On 05.06.13 at 21:17, Mukesh Rathor <mukesh.rathor@xxxxxxxxxx> wrote:
> On Wed, 05 Jun 2013 08:03:12 +0100
> "Jan Beulich" <JBeulich@xxxxxxxx> wrote:
> 
>> >>> On 04.06.13 at 23:53, Mukesh Rathor <mukesh.rathor@xxxxxxxxxx>
>> >>> wrote:
>> > Following OK? :
>> > 
>> >         if (xen_feature(XENFEAT_auto_translated_physmap)) {
>> >                 switch_to_new_gdt(0);
>> > 
>> >                 asm volatile (
>> >                         "pushq %%rax\n"
>> >                         "leaq 1f(%%rip),%%rax\n"
>> >                         "pushq %%rax\n"
>> >                         "lretq\n"
>> >                         "1:\n"
>> >                         : : "a" (__KERNEL_CS) : "memory");
>> > 
>> >                 return;
>> >         }
>> 
>> While generally the choice of using %%rax instead of %0 here is
>> a matter of taste to some degree, I still don't see why you can't
>> use "r" as the constraint here in the first place.
> 
> The compiler mostly picks eax anyways, but good suggestion.
> 
>> Furthermore, assuming this sits in a function guaranteed to not be
>> inlined, this has a latent bug (and if the assumption isn't right, the
>> bug is real) in that the asm() modifies %rax without telling the
>> compiler.
> 
> According to one of the unofficial asm tutorials i've here, the compiler
> knows since it's input and doesn't need to be told. In fact
> it'll barf if added to clobber list.

No, if a tutorial says something like this, it's plain wrong. The
compiler does, as we know, occasionally re-use an unaltered
input in subsequent code (see the hypervisor change
http://xenbits.xen.org/gitweb/?p=xen.git;a=commitdiff;h=84628ee52a427b0f0fe50502eb8ffd0eedad0f03
for a case where this actually caused very bad problems in
practice).

Adding a register to the clobber list that's among the inputs and/or
outputs of course won't work, because it being on the clobber list
means the compiler must not use it _at all_ for any of its own
purposes. That's why you need - in the case here - a dummy output.

Jan

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/