On 27/01/2026 12:02, Jinjie Ruan wrote:
>
> On 2026/1/27 17:43, Kevin Brodsky wrote:
>> On 27/01/2026 04:01, Jinjie Ruan wrote:
>>>> I understand that you're gradually making the arch code more similar to
>>>> the core code so that we can switch over to it, but I'm struggling to
>>>> understand why syscall_trace_enter() takes the 'syscall' argument.
>>>>
>>>> Even the core code just seems to use it as a local variable, which it
>>>> overrides before it ever uses it. What am I missing?
>>> Hi,
>>>
>>> You're absolutely right. The 'syscall' parameter is indeed treated as a
>>> local variable and gets overridden before any real use. Should we
>>> refactor to remove the parameter entirely in generic entry?
>> I noticed this as well, removing it from the generic function would make
>> sense. AFAICT that removal could be propagated quite far in fact:
>> syscall_enter_from_user_mode_work(), syscall_enter_from_user_mode(),
>> even arch implementation (do_syscall_64() on x86).
> Not really, it is the default return value of
> syscall_enter_from_user_mode_work() as below, so we only need to remove
> the parameter in syscall_trace_enter().
>
> static __always_inline long syscall_enter_from_user_mode_work(struct
> pt_regs *regs, long syscall)
> {
> unsigned long work = READ_ONCE(current_thread_info()->syscall_work);
>
> if (work & SYSCALL_WORK_ENTER)
> syscall = syscall_trace_enter(regs, work);
>
> return syscall;
> }
You're right, hadn't realised the call to syscall_trace_enter() was
conditional.
- Kevin
