Re: question about save_xstate_sig() - WHY DOES THIS WORK?

[Rik van Riel <riel@redhat.com>]

On 01/29/2015 03:45 PM, Oleg Nesterov wrote:
> On 01/27, Rik van Riel wrote:
>>
>> On 01/27/2015 02:40 PM, Oleg Nesterov wrote:
>>>>>
>>>>> - Is unlazy_fpu()->__save_init_fpu() safe wrt
>>>>> __kernel_fpu_begin() from irq?
>>
>> It looks like it should be safe, as long as __save_init_fpu()
>> knows that the task no longer has the FPU after __kernel_fpu_end(),
>> so it does not try to save the kernel FPU state to the user's
>> task->thread.fpu.state->xstate
> 
> Not sure this is enough, but...
> 
>> The caveat here is that __kernel_fpu_begin()/__kernel_fpu_end()
>> needs to be kept from running during unlazy_fpu().
> 
> Yes,
> 
>> This means interrupted_kernel_fpu_idle and/or irq_fpu_usable
>> need to check whether preemption is disabled, and lock out
>> __kernel_fpu_begin() when preemption is disabled.
> 
> But we already have kernel_fpu_disable/enable. unlazy_cpu() can use
> it to avoid the race ?

I suspect this will be fine, if __kernel_fpu_end()
from IRQ context always restores the FPU context,
or calls stts, so things like save_init_fpu() will
either continue where they left off, or trap and
then continue where they left off.

__kernel_fpu_end() from process context can be
lazier, something I can work on in my next version
of the "defer FPU loading to kernel -> user space
boundary" patch series.

>> I can certainly merge unlazy_fpu() and save_init_fpu() into the
>> same function, but I am not sure whether or not it should call
>> __thread_fpu_end() - it looks like that would be desirable in some
>> cases, but not in others...
> 
> I _think_ that we never actually want __thread_fpu_end(), although it
> doesn't really hurt if !eager. Probably ulazy/save should do
> 
> 	if (!__save_init_fpu())
> 		__thread_fpu_end();

There is at least one case where we want __thread_fpu_end(),
and that is xstateregs_set.  I got this by moving the
__thread_fpu_end() call from save_init_fpu() into init_fpu().

I am not sure about __math_error. I suspect we may need
__thread_fpu_end() in there so the math state can be
re-initialized if the task catches SIGFPE and continues.

On the other hand, I do not see code in there that
actually does that at the moment...

Let me send my RFC patch to clean up & merge
unlazy_fpu and save_init_fpu() in the next email.