Re: [PATCH v11 04/11] tracing: wprobe: Add watchpoint probe event based on hardware breakpoint

[Jinchao Wang <wangjinchao600@gmail.com>]

On 8/6/2026 8:06 AM, Masami Hiramatsu (Google) wrote:
> On Tue, 4 Aug 2026 01:08:40 +0200
> Peter Zijlstra <peterz@infradead.org> wrote:
> 
>> On Tue, Aug 04, 2026 at 08:00:35AM +0900, Masami Hiramatsu wrote:
>>> On Mon, 3 Aug 2026 09:03:56 +0200
>>> Peter Zijlstra <peterz@infradead.org> wrote:
>>>
>>>> On Sun, Aug 02, 2026 at 05:19:10PM +0900, Masami Hiramatsu (Google) wrote:
>>>>> From: Masami Hiramatsu (Google) <mhiramat@kernel.org>
>>>>>
>>>>> Add a new probe event for the hardware breakpoint called wprobe-event.
>>>>> This wprobe allows user to trace (watch) the memory access at the
>>>>> specified memory address.
>>>>
>>>> s/the memory/the kernel memory/
>>>
>>> Indeed.
>>>
>>>>
>>>>> The new syntax is;
>>>>>
>>>>>  w[:[GROUP/]EVENT] [r|w|rw]@[ADDR|SYM][:SIZE] [FETCH_ARGs]
>>>>>
>>>>> User also can use $addr to fetch the accessed address and $value to fetch
>>>>> the accessed memory value (shorthand for '+0($addr)'). No other variables
>>>>> are supported.
>>>>
>>>> This seems limited to kernel addresses only? Should not also userspace
>>>> addresses be supported?
>>>
>>> Hmm, for x86, yes, since it causes a trap after executing (all page fault
>>> are handled). But other architecture like arm64 may not be able to support
>>> it because it needs to enable watchpoint after major page fault.
>>> At this step, I would like to limit it just for kernel memory.
>>
>> Sure, but suppose you were to add userspace, what would the ABI be?
> 
> Ahh, sorry. It was misreading, I mean it should be only for the kernel address,
> not for user space. So I don't want to add userspace. (I answered the first one)
> 
> Thank you,
> 

Keeping wprobe limited to kernel addresses makes sense to me.
Userspace watchpoints are already supported by the perf hardware-breakpoint API,
and supporting them in wprobe would add unnecessary complexity.

An explicit address check could also make this boundary clear.

>>
>> That is, don't paint yourself in a corner etc.
> 
>