Re: [RFC PATCH stable 5.4 0/8] bpf: Fix bpf_probe_read/bpf_probe_read_str helpers

[Greg KH]

On Sun, May 28, 2023 at 10:02:49PM +0200, Jiri Olsa wrote:
> On Fri, May 26, 2023 at 07:54:17PM +0100, Greg KH wrote:
> > On Mon, May 22, 2023 at 10:33:44PM +0200, Jiri Olsa wrote:
> > > hi,
> > > we see broken access to user space with bpf_probe_read/bpf_probe_read_str
> > > helpers on arm64 with 5.4 kernel. The problem is that both helpers try to
> > > read user memory by calling probe_kernel_read, which seems to work on x86
> > > but fails on arm64.
> > 
> > Has this ever worked on arm64 for the 5.4 kernel tree?  If not, it's not
> > really a regression, and so, why not use a newer kernel that has this
> > new feature added to it there?
> > 
> > In other words, what requires you to use the 5.4.y tree and requires
> > feature parity across architectures?
> 
> we have a customer running ok on x86 v5.4, but arm64 is broken with
> the same bpf/user space code

Again why can they not use a newer kernel version?  What forces this
customer to be stuck with a specific kernel version that spans different
processor types?

> upgrade is an option of course, but it's not a big change and we can
> have 5.4 working on arm64 as well

For loads of other reasons, I'd recommend 5.15 or newer for arm64, so
why not use that?

> I can send out the change that will be closer to upstream changes,
> if that's a concern.. with adding the new probe helpers, which I
> guess is not a problem, because it does not change current API

You are trying to add features to a stable kernel that are not
regression fixes, which is something that we generally do not accept
into stable kernels.

thnaks,

greg k-h