[PATCH 0/1][focal/linux-azure] Azure 5.4: vDSO broken with 5.4.0-1075-azure

Tim Gardner tim.gardner at canonical.com
Mon Jun 6 15:13:25 UTC 2022

BugLink: https://bugs.launchpad.net/bugs/1977753

SRU Justification

vDSO works with the 5.4.0-1074-azure kernel, but is broken in 5.4.0-1075-azure and newer due to this commit

Please revert the commit 5c899e1bcda02c755a5327353ed66f0768771644.

How to verify vDSO is working: "strace date 2>&1 | grep time" should not show the time related syscalls, e.g. clock_gettime()
With 5.4.0-1075-azure and newer, I see the time related syscalls.

More info:
The buggy commit 5c899e1bcda02c755a5327353ed66f0768771644 was backported from the mainline commit:
e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO differences inline")
but due to the differences between the mainline kernel and the 5.4.0-107x-azure kernel, we actually should not simply remove hv_set_clocksource_vdso().

FWIW, the mainline commit is actually buggy and is fixed by this later mainline path:

commit 3486d2c9be652a31033363bdd50391b0c8a8fe21
Author: Vitaly Kuznetsov vkuznets at redhat.com
Date: Thu May 13 09:32:46 2021 +0200

clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86

Mohammed reports (https://bugzilla.kernel.org/show_bug.cgi?id=213029)
the commit e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO
differences inline") broke vDSO on x86. The problem appears to be that
VDSO_CLOCKMODE_HVCLOCK is an enum value in 'enum vdso_clock_mode' and
'#ifdef VDSO_CLOCKMODE_HVCLOCK' branch evaluates to false (it is not
a define).

Use a dedicated HAVE_VDSO_CLOCKMODE_HVCLOCK define instead.

Fixes: e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO differences inline")
Reported-by: Mohammed Gamal mgamal at redhat.com
Suggested-by: Thomas Gleixner tglx at linutronix.de
Signed-off-by: Vitaly Kuznetsov vkuznets at redhat.com
Signed-off-by: Thomas Gleixner tglx at linutronix.de
Reviewed-by: Michael Kelley mikelley at microsoft.com
Link: https://lore.kernel.org/r/20210513073246.1715070-1-vkuznets@redhat.com

PS, Ubunt 20.04’s 5.13.0-1026-azure is not affected.

After some discussion it was decided that the revert is the best solution.

[Test Case]

"strace date 2>&1 | grep time" should no longer show the clock_gettime() syscall.
also Microsoft tested

[Where things could go wrong]

vDSO entry point could stop working altogether

[Other Info]

SF: #00338309

More information about the kernel-team mailing list