Re: [PATCH v7 05/11] arm64/ptrace: Handle ptrace_report_syscall_entry() error

[Kevin Brodsky <kevin.brodsky@arm.com>]

On 21/11/2025 05:15, Jinjie Ruan wrote:
>
> On 2025/11/19 1:12, Kevin Brodsky wrote:
>> On 17/11/2025 14:30, Jinjie Ruan wrote:
>>> The generic entry handle error of ptrace_report_syscall_entry(), but
>>> arm64 not.
>> This suggests that arm64 ignores the error completely, which isn't the
>> case: no syscall will be performed, but tracing will still occur as normal.
>>
>> What this patch seems to be doing is to abort the _enter sequence if
>> ptrace_report_syscall_entry() errors out. The commit title and message
>> should be reworded accordingly.
> You are right,the description is unclear .
>
>>> As the comment said, the calling arch code should abort the system
>> Which comment?
> ptrace_report_syscall_entry()

I found that later, this should be clarified in the commit message.

>>> call and must prevent normal entry so no system call is
>>> made if ptrace_report_syscall_entry() return nonzero.
>> This is already the case since we're calling forget_syscall().
> Yes. it is similar with the generic entry returns NO_SYSCALL.

My point is that this patch is not changing this - arm64 was already
skipping the syscall if ptrace_report_syscall_entry() returns an error.

>>> In preparation for moving arm64 over to the generic entry code,
>>> return early if ptrace_report_syscall_entry() encounters an error.
>>>
>>> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
>>> ---
>>>  arch/arm64/kernel/ptrace.c | 16 +++++++++++-----
>>>  1 file changed, 11 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/arch/arm64/kernel/ptrace.c b/arch/arm64/kernel/ptrace.c
>>> index 95984bbf53db..707951ad5d24 100644
>>> --- a/arch/arm64/kernel/ptrace.c
>>> +++ b/arch/arm64/kernel/ptrace.c
>>> @@ -2317,10 +2317,10 @@ enum ptrace_syscall_dir {
>>>  	PTRACE_SYSCALL_EXIT,
>>>  };
>>>  
>>> -static void report_syscall_enter(struct pt_regs *regs)
>>> +static int report_syscall_enter(struct pt_regs *regs)
>>>  {
>>> -	int regno;
>>>  	unsigned long saved_reg;
>>> +	int regno, ret;
>>>  
>>>  	/*
>>>  	 * We have some ABI weirdness here in the way that we handle syscall
>>> @@ -2342,9 +2342,13 @@ static void report_syscall_enter(struct pt_regs *regs)
>>>  	saved_reg = regs->regs[regno];
>>>  	regs->regs[regno] = PTRACE_SYSCALL_ENTER;
>>>  
>>> -	if (ptrace_report_syscall_entry(regs))
>>> +	ret = ptrace_report_syscall_entry(regs);
>>> +	if (ret)
>>>  		forget_syscall(regs);
>> The generic syscall_trace_enter() doesn't do this (i.e. setting
>> regs->syscallno to NO_SYSCALL). Is that an oversight or do we just not
>> need it? In principle this does have a visible effect (e.g. via
>> REGSET_SYSTEM_CALL).
> We just not need it because the original syscall_trace_enter() need use
> regs->syscallno as the return value, but now we return early by using
> NO_SYSCALL.

Calling forget_syscall() means setting regs->syscallno to NO_SYSCALL. It
is indeed no longer required for the entry sequence, but it does have
visible side effects. For instance, regs->syscallno can be inspected via
ptrace(PTRACE_GETREGSET, REGSET_SYSTEM_CALL). So the question is whether
we need to deviate from the generic path on arm64 (this is mostly a
question for arm64 maintainers).

- Kevin