[PATCH] tracing/syscalls: ignore numbers outside NR_syscalls' range

[Nathan Lynch]

On 10/30/2014 06:35 AM, Russell King - ARM Linux wrote:
> On Thu, Oct 30, 2014 at 07:30:28AM -0400, Steven Rostedt wrote:
>> On Thu, 30 Oct 2014 11:14:41 +0000
>> Russell King - ARM Linux <linux at arm.linux.org.uk> wrote:
>>
>>
>>> We have always had syscall number range of 0x900000 or so.  The tracing
>>> design does not expect that.  Therefore, the tracing design did not take
>>> account of ARM when it was created.  Therefore, it's up to the tracing
>>> people to decide how to properly fit their ill-designed subsystem into
>>> one of the popular and well-established kernel architectures - or at
>>> least suggest a way to work around this issue.
>>>
>>
>>
>> Fine, lets define a MAX_SYSCALL_NR that is by default NR_syscalls, but
>> an architecture can override it.
>>
>> In trace_syscalls.c, where the checks are done, have this:
>>
>> #ifndef MAX_SYSCALL_NR
>> # define MAX_SYSCALL_NR NR_syscalls
>> #endif
>>
>> change all the checks to test against MAX_SYSCALL_NR instead of
>> NR_syscalls.
>>
>> Then in arch/arm/include/asm/syscall.h have:
>>
>> #define MAX_SYSCALL_NR 0xa00000
>>
>> or whatever would be the highest syscall number for ARM.
> 
> Or do we just ignore the high "special" ARM syscalls and treat them (from
> the tracing point of view) as non-syscalls, avoiding the allocation of
> something around 1.2MB for the syscall bitmap.  I really don't know, I
> don't use any of this tracing stuff, so it isn't something I care about.
> 
> Maybe those who do use the facility should have an input here?

I checked strace and it knows about ARM's high syscalls.  I wouldn't
want to go from casually using strace to digging deeper with ftrace only
to get the impression that syscalls are disappearing.