lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [day] [month] [year] [list]
Message-ID: <1914d310f3d.12a7d406f898033.7903291218646592039@collabora.com>
Date: Tue, 13 Aug 2024 20:22:26 +0100
From: Robert Beckett <bob.beckett@...labora.com>
To: "Thomas Gleixner" <tglx@...utronix.de>,
	"Peter Zijlstra" <peterz@...radead.org>
Cc: "x86" <x86@...nel.org>, "linux-kernel" <linux-kernel@...r.kernel.org>,
	"Engineering - Kernel" <kernel@...labora.com>
Subject: CONFIG_MITIGATION_CALL_DEPTH_TRACKING

Hi,

After upgrading from 6.1 to 6.5 based kernel we noticed a regression in boot times
on AMD Zen 2 based device.

After some debug, we figured out that CONFIG_MITIGATION_CALL_DEPTH_TRACKING
(or CONFIG_CALL_DEPTH_TRACKING as it was then) seemed to cause the slow down.

Having two 6.5 kernels, one with it enabled and one without, with no other differences
we performed 101 reboots with each version.

It showed a median boot to graphical.target of 8.537s with it enabled and 5.938s with it
disabled as measured via systemd-analyze. This difference (within noise) matches the
regressing in boot times observed after the kernel upgrade.

Post boot runtime comparisons show as [1] that there is nothing between them.

I'd appreciate any advice you can give, especially around these questions:

- The original patch series [1] mentioned runtime performance testing. Does anyone
know of any boot time testing write-ups that may have occurred at the time?

- The help for the config mentions a 5% text size overhead. While I could (perhaps
naively) expect up to 5% more page cache misses etc during boot up, is a ~30% slow
down explainable via this config? (no kernel boot params are specified for mitigations
so all defaults apply)

- I am tempted to conclude that if we want to mitigate retbleed in a consistent way
that we should just disable this option and rely on CONFIG_CPU_IBRS_ENTRY. Is 
there any other option to support accelerated Intel mitigation without penalising 
AMD boot up times due to the overhead?


any thoughts would be greatly appreciated.

Bob


[1] https://lore.kernel.org/all/20220915111039.092790446@infradead.org/

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ