On Wed, Sep 11, 2019 at 2:08 PM Sean Christopherson <sean.j.christopherson@xxxxxxxxx> wrote: > > On Mon, Sep 09, 2019 at 02:19:20PM -0700, Bill Wendling wrote: > > The "mov" instruction to get the error code shouldn't move into a memory > > location. Don't allow the compiler to make this decision. Instead > > specify that only a register is appropriate here. > > I'd prefer the changelog to say something like: > > Remove a bogus memory contraint as x86 does not have a generic > memory-to-memory "mov" instruction. > > Saying "shouldn't move into a memory location" makes it sound like there's > an unwanted side effect when the compiler selects memory, though I suppose > you could argue that a build error is an unwanted side effect :-). > No prob. :-) I'm not familiar with sending these patches via email. Do I need to regenerate the patch and send via "send-email"? (Similar question for the "Reviewed-By" comments.) > Out of curiosity, do any compilers actually generate errors because of > this, or is it simply dead code? > Yeah, clang uses a memory location, probably due to inlining causing register pressure (that's a guess, I didn't explore it too deeply). -bw > > Signed-off-by: Bill Wendling <morbo@xxxxxxxxxx> > > --- > > lib/x86/desc.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/lib/x86/desc.c b/lib/x86/desc.c > > index 5f37cef..451f504 100644 > > --- a/lib/x86/desc.c > > +++ b/lib/x86/desc.c > > @@ -263,7 +263,7 @@ unsigned exception_error_code(void) > > { > > unsigned short error_code; > > > > - asm("mov %%gs:6, %0" : "=rm"(error_code)); > > + asm("mov %%gs:6, %0" : "=r"(error_code)); > > return error_code; > > }