[PATCH] Fix recent ARM remote-MCJIT regression

[David Tweed]

Thanks for looking at this Tim. Once things are back up here I'll verify
that patch here. One minor question: you've conditionalized the cache
invalidation to those architectures which definitely need it, but doesn't it
turn in to a no-op on arch's that don't, so we could just put it there
unconditionally to simplify the code?

Cheers,
Dave

-----Original Message-----
From: llvm-commits-bounces at cs.uiuc.edu
[mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Tim Northover
Sent: 19 May 2013 13:23
To: llvm-commits; Renato Golin; Kaylor, Andrew
Subject: [PATCH] Fix recent ARM remote-MCJIT regression

Hi,

I've been looking at the recent failures on ARM MCJIT tests. There are
two issues:

1. Long-term known problems with how relocations are handled.
2. Regressions caused by recent faffing about with applyPermissions
(probably).

I'm still working on the first problem, but meanwhile I think I've
spotted the cause of the second. lli's RemoteTarget calls
setExecutable after copying code across, but not any kind of cache
invalidation.

In fact, on Darwin platforms it seems to be setExecutable's
responsibility to invalidate the cache, so this patch does similarly
on other Unix-like systems.

I think this removes the need for a universal "don't do remote MCJIT
on ARM". It also seems to allow some individual tests to pass so I've
removed the XFAIL from them.

Ok to commit?

Cheers.

Tim.