Re: [PATCH] Add a text_poke syscall

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

On 11/18/2013 04:27 PM, Andi Kleen wrote:
> 
> Proposed man page:
> 
> NAME
> 	text_poke - Safely modify running instructions (x86)
> 
> SYNOPSYS
> 	int text_poke(void *addr, const void *opcode, size_t len,
> 	              void (*handler)(void), int timeout);
> 
> DESCRIPTION
> 	The text_poke system allows to safely modify code that may
> 	be currently executing in parallel on other threads.
> 	Patch the instruction at addr with the new instructions
> 	at opcode of length len. The target instruction will temporarily
> 	be patched with a break point, before it is replaced
> 	with the final replacement instruction. When the break point
> 	hits the code handler will be called in the context
> 	of the thread. The handler does not save any registers
> 	and cannot return. Typically it would consist of the
> 	original instruction and then a jump to after the original
> 	instruction. The handler is only needed during the
> 	patching process and can be overwritten once the syscall
> 	returns. timeout defines an optional timout to indicate
> 	to the kernel how long the patching could be delayed.
> 	Right now it has to be 0.
> 

I think I would prefer an interface which took a list of patch points,
or implemented only the aspects which are impossible to do in user space.

All we really need in the kernel is the IPI broadcasts - the rest can be
done in user space, including intercepting SIGTRAP.  For userspace it is
probably the best to just put a thread to sleep until the patching is
done, which can be done with a futex.

One advantage with doing this in userspace is that the kernel doesn't
have to be responsible avoiding holding a thread due to a slightly
different SIGTRAP -- it will all come out after the signal handler is
restored, anyway.

That being said, the user space code would really need to be librarized.

	-hpa