On 12/20/23 20:40, Luis Chamberlain wrote:
On Tue, Dec 19, 2023 at 01:26:49PM -0800, Luis Chamberlain wrote:
On Wed, Nov 22, 2023 at 11:18:10PM +0100, deller@xxxxxxxxxx wrote:
From: Helge Deller <deller@xxxxxx>
My questions:
- Am I wrong with my analysis?
This would typically of course depend on the arch, but whether it helps
is what I would like to see with real numbers rather then speculation.
Howeer, I don't expect some of these are hot paths except maybe the
table lookups. So could you look at some perf stat differences
without / with alignment ? Other than bootup live patching would be
a good test case. We have selftest for modules, the script in selftests
tools/testing/selftests/kmod/kmod.sh is pretty aggressive, but the live
patching tests might be better suited.
- What does people see on other architectures?
- Does it make sense to add a compile- and runtime-check, like the patch below, to the kernel?
The chatty aspects really depend on the above results.
Aren't there some archs where an unaligned access would actually crash?
Why hasn't that happened?
I've gone down through memory lane and we have discussed this before.
It would seem this misalignment should not affect performance, and this
should not be an issue unless you have a buggy exception hanlder. We
actually ran into one before. Please refer to merge commit
e74acdf55da6649dd30be5b621a93b71cbe7f3f9
Yes, this is the second time I stumbled over this issue.
But let's continue discussing in the other mail thread...
Helge
