Hi Geert,
Am 09.07.2021 um 19:29 schrieb Geert Uytterhoeven:
Hi Christoph,
On Fri, Jul 9, 2021 at 6:22 AM Christoph Hellwig <hch@xxxxxx> wrote:
On Fri, Jul 09, 2021 at 12:31:45PM +1200, Michael Schmitz wrote:
That patch works fine on a casual test. What you did to
__constant_copy_to_user() does not appear to matter - but I haven't put the
system under any kind of stress yet. I'm a little reluctant to do that
(recovering from a trashed boot disk is a little dicey), I'll probably only
try that with your changes to __constant_copy_to_user() from commit
d36105c942e0 backed out.
As Linus pointed out, small copy_to_user basically doesn't happen as
we have switched all the suspect call sites to just use put_user.
Geert: do you care about __constant_copy_to_user at all, or can we just
kill it (as well as the copy_from_user side)?
If it blocks you, feel free to remove it.
BTW, do you have an idea of how many calls use small sizes?
Just booting to a login prompt: Quite a few with 8 bytes, some with 4
bytes (those from the keyboard driver), the rest is 16, 24, 36 and 92
bytes (a lot of those). Used ratelimiting so probably missed a few.
I'll repeat that again logging only the small sizes, but I think we can
live with using the generic version for those few 4 and 8 byte cases.
Cheers,
Michael
Gr{oetje,eeting}s,
Geert
