ARM: Clean up FlushICache.

ARM: Clean up FlushICache.

In theory, the Thumb version of FlushICache doesn't have to be any
different from the ARM version: There is no need to switch to ARM mode
for the svc, and r7 can be used just like any other register.

In practice, GCC uses r7 as a frame pointer in Thumb code, and it
refuses to allow it be used in inline assembly. To avoid complicated
checks to determine whether or not it is possible to use it directly,
the FlushICache function explicitly preserves r7.

R=ulan@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=21949

[ulan, 2014-06-18 14:08:31]

https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc
File src/arm/cpu-arm.cc (right):

https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc#newcode46
src/arm/cpu-arm.cc:46: void __attribute__((naked)) CPU::FlushICache(void* start,
size_t size) {
Do you know if this attribute is supported by clang? I can't find any reference.

[jbramley, 2014-06-18 14:19:37]

https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc
File src/arm/cpu-arm.cc (right):

https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc#newcode46
src/arm/cpu-arm.cc:46: void __attribute__((naked)) CPU::FlushICache(void* start,
size_t size) {
On 2014/06/18 14:08:31, ulan wrote:
> Do you know if this attribute is supported by clang? I can't find any
reference.

I couldn't find a definite source, but the following pages strongly hint that
Clang supports it:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52646
http://comments.gmane.org/gmane.comp.compilers.clang.devel/29171

I also couldn't find reference to it (or any other __attribute__) in Clang's
manual, but it's _not_ included in the list of GCC features that aren't yet
supported:
http://clang.llvm.org/docs/UsersManual.html#gcc-extensions-not-implemented-yet

I was hoping that the try servers would have a Clang builder, but they don't. (I
don't have a Clang tool chain set up here.)

[jbramley, 2014-06-18 14:37:42]

On 2014/06/18 14:19:37, jbramley wrote:
> https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc
> File src/arm/cpu-arm.cc (right):
> 
> https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc#newcode46
> src/arm/cpu-arm.cc:46: void __attribute__((naked)) CPU::FlushICache(void*
start,
> size_t size) {
> On 2014/06/18 14:08:31, ulan wrote:
> > Do you know if this attribute is supported by clang? I can't find any
> reference.
> 
> I couldn't find a definite source, but the following pages strongly hint that
> Clang supports it:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52646
> http://comments.gmane.org/gmane.comp.compilers.clang.devel/29171
> 
> I also couldn't find reference to it (or any other __attribute__) in Clang's
> manual, but it's _not_ included in the list of GCC features that aren't yet
> supported:
> http://clang.llvm.org/docs/UsersManual.html#gcc-extensions-not-implemented-yet
> 
> I was hoping that the try servers would have a Clang builder, but they don't.
(I
> don't have a Clang tool chain set up here.)

I can't get Ubuntu's 'clang' package to cross-compile V8 for ARM, but a simple
test with __attribute__((naked)) works:

[test.c]
void __attribute__((naked)) test(void) {
  __asm__ __volatile__("nop\n" : : : "memory");
}

[Benedikt Meurer, 2014-06-18 16:52:30]

On 2014/06/18 14:37:42, jbramley wrote:
> On 2014/06/18 14:19:37, jbramley wrote:
> > https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc
> > File src/arm/cpu-arm.cc (right):
> > 
> >
https://codereview.chromium.org/335133002/diff/1/src/arm/cpu-arm.cc#newcode46
> > src/arm/cpu-arm.cc:46: void __attribute__((naked)) CPU::FlushICache(void*
> start,
> > size_t size) {
> > On 2014/06/18 14:08:31, ulan wrote:
> > > Do you know if this attribute is supported by clang? I can't find any
> > reference.
> > 
> > I couldn't find a definite source, but the following pages strongly hint
that
> > Clang supports it:
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52646
> > http://comments.gmane.org/gmane.comp.compilers.clang.devel/29171
> > 
> > I also couldn't find reference to it (or any other __attribute__) in Clang's
> > manual, but it's _not_ included in the list of GCC features that aren't yet
> > supported:
> >
http://clang.llvm.org/docs/UsersManual.html#gcc-extensions-not-implemented-yet
> > 
> > I was hoping that the try servers would have a Clang builder, but they
don't.
> (I
> > don't have a Clang tool chain set up here.)
> 
> I can't get Ubuntu's 'clang' package to cross-compile V8 for ARM, but a simple
> test with __attribute__((naked)) works:
> 
> [test.c]
> void __attribute__((naked)) test(void) {
>   __asm__ __volatile__("nop\n" : : : "memory");
> }

I'm really against relying on the availability of the naked attribute.

[jbramley, 2014-06-23 09:31:00]

https://codereview.chromium.org/335133002/diff/20001/src/arm/cpu-arm.cc
File src/arm/cpu-arm.cc (right):

https://codereview.chromium.org/335133002/diff/20001/src/arm/cpu-arm.cc#newco...
src/arm/cpu-arm.cc:61: : "memory", "cc");
Sorry, "cc" shouldn't be in this list. (It's left over from an early
implementation where I checked for zero length in the assembly.)

[jbramley, 2014-06-23 09:58:03]

Ok, is this better? It takes one or two more instructions than before, but it
avoids the 'naked' attribute.

[ulan, 2014-06-23 11:42:44]

lgtm

[jbramley, 2014-06-23 17:30:07]

Committed patchset #3 manually as r21949 (presubmit successful).