[CRIU] [PATCH 2/3] dump: use breakpoints instead of tracing syscalls (v2)

[Pavel Emelyanov]

On 09/19/2014 06:19 PM, Christopher Covington wrote:
> On 09/19/2014 10:01 AM, Pavel Emelyanov wrote:
>> On 09/17/2014 07:12 PM, Andrey Vagin wrote:
>>> Currently CRIU traces syscalls to catch a moment, when sigreturn() is
>>> called. Now we trace recv(cmd), close(logfd), close(cmdfd), sigreturn().
>>>
>>> We can reduce a number of steps by using hw breakpoints. A breakpoint is
>>> set before sigreturn, so we will need to trace only it.
>>>
>>> v2: In the first version a breakpoint is set after sigreturn. In this
>>> case we have a problem with signals. If a process has pending signals,
>>> it will start to precess them after exiting from sigreturn(), but before
>>> returning to userspace. So the breakpoint will not be triggered.
>>>
>>> And at the end Here are a few numbers how we catch sigreturn.
>>> Before this patch criu executes 36 syscalls and gets 12 signals.
>>> With this patch criu executes 18 syscalls and gets 5 signals.
>>>
>>> Signed-off-by: Andrey Vagin <avagin at openvz.org>
>>
>> OK, applied, thanks a lot :)
>>
>> Christopher, if it's possible, could you help us placing breakpoints
>> into ARM parasite and restorer? :) This is quite good optimization,
>> I think it makes sense to have one on ARM too.
> 
> I will look into this. 

Great! Thanks :)

> Some Linux interfaces that I've used in the past that
> may be available across architectures are:
> 
> Software breakpoint placement with uprobes. This requires debugfs.
> 
> https://lwn.net/Articles/499190/
> 
> Hardware breakpoint placement with perf_event_open.
> 
> https://github.com/deater/perf_event_tests/blob/master/tests/breakpoints/breakpoint_support.c
> 
> http://man7.org/linux/man-pages/man2/perf_event_open.2.html

Don't you know which of them are used by gdb? AFAIK the ones for x86
were taken from there.

Thanks,
Pavel