Re: [PATCH] ARM: ftrace: Ensure code modifications are synchronisedacross all cpus

[Steven Rostedt]

On Fri, 2012-12-07 at 19:02 +0000, Will Deacon wrote:

> For ARMv7, there are small subsets of instructions for ARM and Thumb which
> are guaranteed to be atomic wrt concurrent modification and execution of
> the instruction stream between different processors:
> 
> Thumb:	The 16-bit encodings of the B, NOP, BKPT, and SVC instructions.
> ARM:	The B, BL, NOP, BKPT, SVC, HVC, and SMC instructions.
> 
> but before your eyes light up at the presence of the BKPT instruction in
> that list; we don't actually use that in Linux and instead leave it for
> external (i.e. JTAG) debuggers so that they can operate without getting
> tangled up with spurious traps from the OS. Linux actually picks its own
> undefined instructions, which are obviously not included in the lists above.

My eyes actually lit up with the B instruction :-)  As Jon showed, we
could use a 16bit jump instead. Add the B to jump over the other half of
the call (to all places that you want to modify). Send a sync to all
CPUs to flush their caches. Modify the other half of the call, send
another sync, and then modify the first half.

> 
> Also note that the 16-bit limitation for Thumb instructions above can
> actually be used to modify *half* of a BL instruction but, to keep things
> exciting, the PC-relative immediate is split across the two halves. However,
> you could in theory mess around with bottom 10 bits or so, depending on the
> exact encoding...
> 
> Obviously this doesn't preclude the need for cache maintenance on both D and
> I side, but the atomicity guarantees are as I've described above.

Right. Thanks for the update.

-- Steve


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/