Re: [RFC][PATCH] tracing/syscalls: Have ia32 compat syscalls show raw format

["H. Peter Anvin" <hpa@linux.intel.com>]

On 02/12/2013 11:39 AM, H. Peter Anvin wrote:
> On 02/12/2013 10:42 AM, Steven Rostedt wrote:
>>
>> And currently the output is just plain broken. This isn't a hack. You
>> should have seen my first attempt. Now THAT was a hack! My first attempt
>> was extremely intrusive, and required a lot of arch changes. But then I
>> realized it was too much, and found that I could do the same thing
>> pretty much completely contained within just the tracing code itself.
>>
>> I know you feel that the syscall tracing is broken/hack/whatever. But it
>> exists as of today, and yes, there's lots of users out there depending
>> on it.
>>
> 
> I am getting extremely frustrated with this cycle:
> 
> 1. "We should have done <X> but we did <Y> because <X> was too
>     hard/required arch changes/..."
> 2. "Well, <Y> is broken, but people rely on it.  We should have done
>     <X> but now it is too hard/breaks legacy/... so let's do <Z>..."
> 3. Lather, rinse, repeat.
> 
> The whole system with trace metadata seems to be broken at the core,
> *exactly* because it intercepts at a different place than the one which
> has a well-defined ABI and every hack, kluge and patch which doesn't fix
> that fundamental design error will just make it worse and just kicks the
> can further down the road.
> 

As to why I care: I care about the number of ways we present ABIs to
userspace.  The current tracing code takes the kernel internal
implementation and makes it an ABI -- that ties the hands of kernel
developers forever, because we can't know what we break if we redesign it.

	-hpa