i386: inline restore_fpu
Quoted 10 lines · Linus Torvalds
/*
* The "nop" is needed to make the instructions the same
* length.
*/
#define restore_fpu(tsk) \
alternative_input( \
"nop ; frstor %1", \
"fxrstor %1", \
X86_FEATURE_FXSR, \
"m" ((tsk)->thread.i387.fsave))Probably a very minor point, but shouldn't it be
"m" ((tsk)->thread.i387.fxsave))
because that's the largest possible operand that could end up being read?
Quoted 3 lines
Now, we should do the same for "fnsave ; fwait" vs "fxsave ; fnclex" too, but I was lazy. If somebody wants to try that, it would need an "
alternative_output()" define but should otherwise be trivial too.
Is that function called __save_init_fpu because it saves and then
initializes the fpu? Unlike fsave, fxsave does not init the fpu after it saves the state; to make the behavior match it should be "fxsave ; fninit"
__
Chuck
Quoted 6 lines · Chuck Ebbert
Probably a very minor point, but shouldn't it be
"m" ((tsk)->thread.i387.fxsave))because that's the largest possible operand that could end up being read?
Yes, I ended up fixing that already (along with handling the fxsave case too).
Quoted 3 lines
Is that function called
__save_init_fpubecause it saves and then
initializes the fpu? Unlike fsave, fxsave does not init the fpu after it saves the state; to make the behavior match it should be "fxsave ; fninit"
The "init" part is really just "reset_exceptions" - we don't care about the rest of the state, and the naming is historical (ie it's really called "init" because "fnsave" does the reset by re-initializing everything, ie there's an implied "fninit").
So for the fxsave path, we just use fnclex, since that's faster than a full fninit.
Linus