Re: irq-disabled vs vmap vs text_poke

[Masami Hiramatsu <mhiramat@redhat.com>]

Mathieu Desnoyers wrote:
>>> text_poke should actually use local_irq_disable/enable rather than
>>> local_irq_save/restore because it _must_ be called with interrupts on.
>> Could you tell me why it must be called with irq on?
>>
> 
> Because it uses vmap, but see below..
> 
>>>> Anybody got an idea on how to fix this?
>>> There is probably something wrong with the caller, kprobes, which calls
>>> text_poke with interrupts off.
>> Hmm, kprobe's smoke test caused this problem. Since (un)register_kprobe()
>> may sleep for waiting a mutex, it should not be called with interrupts off.
>> So, it's not text_poke()'s issue nor vmap().
>>
>> BTW, what about using map_vm_area() in text_poke() instead of vmap()?
>> Since text_poke() just maps text pages to alias pages temporarily,
>> I think we don't need to use delayed vunmap().
>>
> 
> As with this patch from you ? Sorry, when it has been initially
> submitted, the discussion turned in a different direction. Please see
> comments inline.

No, sorry for confusing, vm_unmap_ram() is basically same as vunmap().
(The root cause of that bug which I had been reported was not in text_poke())

Here I said is (maybe) improving text_poke() by using map_vm_area() which
simply maps pages to pre-allocated vm_area. And when unmapping, we just
cleanup pte by unmap_kernel_range().
For pre-allocating vm_area, we can use get_vm_area().

If we can use kmap for this purpose, it will be better solution.
However, kmap can not be used for making alias pages.

Thanks,

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America) Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com