On Fri, 29 Aug 2025 09:50:12 -0700
Linus Torvalds [off-list ref] wrote:

On Fri, 29 Aug 2025 at 09:42, Linus Torvalds
[off-list ref] wrote:

quoted

Just use the hash. Don't do anything to it. Don't mess with it.  

In fact, at actual stack trace time, don't even do the hashing. Just
save the raw pointer value (but as a *value*, not as a pointer: we
absolutely do *not* want people to think that the random value can be
used as a 'struct file' *: it needs to be a plain unsigned long, not
some kernel pointer).

Then the hashing can happen when you expose those entries to user
space (in the "print" stage). At that point you can do that

       hash = siphash_1u64(value, secret);

thing.

That will likely help I$ and D$ too, since you won't be accessing the
secret hashing data randomly, but do it only at trace output time
(presumably in a fairly tight loop at that point).

Note, the ring buffer can be mapped to user space. So anything written into
the buffer is already exposed. The "at trace output time" is done by user
space, not the kernel (except when using "trace" and "trace_pipe" files).

-- Steve