Re: [DRAFT] mm/kprobes: Add generic kprobe_fault_handler() fallback definition

From: Anshuman Khandual
Date: Wed Jul 03 2019 - 10:10:45 EST




On 07/03/2019 06:29 PM, Guenter Roeck wrote:
> On 7/2/19 10:35 PM, Anshuman Khandual wrote:
>>
>>
>> On 07/01/2019 06:58 PM, Guenter Roeck wrote:
>>> On 7/1/19 2:35 AM, Anshuman Khandual wrote:
>>>> Architectures like parisc enable CONFIG_KROBES without having a definition
>>>> for kprobe_fault_handler() which results in a build failure. Arch needs to
>>>> provide kprobe_fault_handler() as it is platform specific and cannot have
>>>> a generic working alternative. But in the event when platform lacks such a
>>>> definition there needs to be a fallback.
>>>>
>>>> This adds a stub kprobe_fault_handler() definition which not only prevents
>>>> a build failure but also makes sure that kprobe_page_fault() if called will
>>>> always return negative in absence of a sane platform specific alternative.
>>>>
>>>> While here wrap kprobe_page_fault() in CONFIG_KPROBES. This enables stud
>>>> definitions for generic kporbe_fault_handler() and kprobes_built_in() can
>>>> just be dropped. Only on x86 it needs to be added back locally as it gets
>>>> used in a !CONFIG_KPROBES function do_general_protection().
>>>>
>>>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
>>>> ---
>>>> I am planning to go with approach unless we just want to implement a stub
>>>> definition for parisc to get around the build problem for now.
>>>>
>>>> Hello Guenter,
>>>>
>>>> Could you please test this in your parisc setup. Thank you.
>>>>
>>>
>>> With this patch applied on top of next-20190628, parisc:allmodconfig builds
>>> correctly. I scheduled a full build for tonight for all architectures.
>>
>> How did that come along ? Did this pass all build tests ?
>>
>
> Let's say it didn't find any failures related to this patch. I built on top of
> next-20190701 which was quite badly broken for other reasons. Unfortunately,
> next-20190702 is much worse, so retesting would not add any value at this time.
> I'd say go for it.
>
> Guenter
>

Sure thanks, will post it out soon.