Re: [RFC] mm: Generalize notify_page_fault()

[Anshuman Khandual]

On 05/31/2019 11:18 PM, Matthew Wilcox wrote:
> On Fri, May 31, 2019 at 02:17:43PM +0530, Anshuman Khandual wrote:
>> On 05/30/2019 07:09 PM, Matthew Wilcox wrote:
>>> On Thu, May 30, 2019 at 05:31:15PM +0530, Anshuman Khandual wrote:
>>>> On 05/30/2019 04:36 PM, Matthew Wilcox wrote:
>>>>> The two handle preemption differently.  Why is x86 wrong and this one
>>>>> correct?
>>>>
>>>> Here it expects context to be already non-preemptible where as the proposed
>>>> generic function makes it non-preemptible with a preempt_[disable|enable]()
>>>> pair for the required code section, irrespective of it's present state. Is
>>>> not this better ?
>>>
>>> git log -p arch/x86/mm/fault.c
>>>
>>> search for 'kprobes'.
>>>
>>> tell me what you think.
>>
>> Are you referring to these following commits
>>
>> a980c0ef9f6d ("x86/kprobes: Refactor kprobes_fault() like kprobe_exceptions_notify()")
>> b506a9d08bae ("x86: code clarification patch to Kprobes arch code")
>>
>> In particular the later one (b506a9d08bae). It explains how the invoking context
>> in itself should be non-preemptible for the kprobes processing context irrespective
>> of whether kprobe_running() or perhaps smp_processor_id() is safe or not. Hence it
>> does not make much sense to continue when original invoking context is preemptible.
>> Instead just bail out earlier. This seems to be making more sense than preempt
>> disable-enable pair. If there are no concerns about this change from other platforms,
>> I will change the preemption behavior in proposed generic function next time around.
> 
> Exactly.
> 
> So, any of the arch maintainers know of a reason they behave differently
> from x86 in this regard?  Or can Anshuman use the x86 implementation
> for all the architectures supporting kprobes?

So the generic notify_page_fault() will be like this.

int __kprobes notify_page_fault(struct pt_regs *regs, unsigned int trap)
{
        int ret = 0;

        /*
         * To be potentially processing a kprobe fault and to be allowed
         * to call kprobe_running(), we have to be non-preemptible.
         */
        if (kprobes_built_in() && !preemptible() && !user_mode(regs)) {
                if (kprobe_running() && kprobe_fault_handler(regs, trap))
                        ret = 1;
        }
        return ret;
}