Thread (55 messages) 55 messages, 8 authors, 2025-11-06

Re: [PATCH v4 23/35] vdso/datastore: Map pages through struct page

From: Thomas Weißschuh <hidden>
Date: 2025-11-04 15:43:41
Also in: linux-arm-kernel, linux-kselftest, linux-mips, linux-s390, lkml, loongarch, sparclinux

On Tue, Nov 04, 2025 at 01:14:03PM +0000, Mark Brown wrote:
On Tue, Nov 04, 2025 at 09:44:38AM +0100, Marek Szyprowski wrote:
quoted
On 03.11.2025 16:24, Mark Brown wrote:
quoted
quoted
We do have some other serious breakage affecting arm64 in -next which
are making it hard to get a clear picture of which platforms are
affected, at least the FVP and O6 are unaffected by those other issues
(due to using MTE on platforms that don't have it, those platforms do
have MTE).
quoted
I got almost the same result while bisecting on ARM 32bit Exynos-based 
boards, so the issue with this patchset is not fully ARM64 specific. For 
some reasons it also doesn't affect all systems though. It is even 
worse, because it affected only a subset of boards, but different for 
each tested commit. The observed failure looks exactly the same:
I've now got the results for this specific commit, it looks like every
single arm64 system is failing.  I didn't test any other architectures.
Which one do you mean exactly with "this specific commit"?

6a011a228293 ("vdso/datastore: Map pages through struct page")
10d91dac2ea5 ("vdso/datastore: Allocate data pages dynamically")
quoted
Then I've tested it on ARM64bit (RaspberrryPi3b+ board) and got the 
following panic on 6a011a228293 ("vdso/datastore: Map pages through 
struct page") commit:
I'm seeing the same thing on at least some of the systems - this is with
arm64 defconfig (I suspect that's what Marek is doing too).  For
example:
I can now reproduce this issue, too. Even on QEMU. But it goes away,
as soon as the second commit is applied. The two commits should be merged.
That could also explain the weird bisection results, landing on either one
of the commits. However I can't reproduce the "silent freeze" issue on commit
10d91dac2ea5 ("vdso/datastore: Allocate data pages dynamically") on my
Raspberry Pi 3b+.

Any chance I could get remote access to one of your test machines?
I don't have access to the exact machines and that should reduce the chance
of chasing down dead ends.


Thomas
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help