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: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Fri, 25 Jun 2021 17:54:39 +0700
From:   Strontium <strntydog@...il.com>
To:     Ilya Lipnitskiy <ilya.lipnitskiy@...il.com>,
        "Maciej W. Rozycki" <macro@...am.me.uk>
Cc:     Thomas Bogendoerfer <tsbogend@...ha.franken.de>,
        Wei Li <liwei391@...wei.com>,
        Tiezhu Yang <yangtiezhu@...ngson.cn>,
        linux-mips@...r.kernel.org,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Felix Fietkau <nbd@....name>
Subject: Re: [PATCH] MIPS: add support for buggy MT7621S core detection

On 8/4/21 1:49 am, Ilya Lipnitskiy wrote:
> On Wed, Apr 7, 2021 at 6:49 AM Maciej W. Rozycki <macro@...am.me.uk> wrote:
>> On Mon, 5 Apr 2021, Ilya Lipnitskiy wrote:
>>
>>> Thanks for the comments. Including asm/bugs.h in asm/mips-cps.h led to
>>> some circular dependencies when I tried it, but I will try again based
>>> on your feedback - indeed it would be much cleaner to have this logic
>>> in mips_cps_numcores. The only wrinkle is that mips_cps_numcores may
>>> return a different value on MT7621 after the cores have started due to
>>> CPULAUNCH flags changing, but nobody calls mips_cps_numcores later
>>> anyway, so it's a moot point today. I will clean up the change and
>>> resend.
>>  Hmm, I don't know this system, but by the look of the code it queries
>> launch[2], which I gather refers to the VPE #0 of an inexistent core #1,
>> so why would the structure change given that there is no corresponding
>> silicon?
> The structure would change only on the dual-core flavor of MT7621, the
> single-core would always report 1 core, but on the dual-core, if
> somebody were to call mips_cps_numcores after the second core had
> already started, mips_cps_numcores would return 1 instead of 2. I
> think it may be feasible to fix it by checking other flags, but there
> is no use case for that today, so I'd rather keep this hacky logic to
> a minimum.
>
> Ilya
>
>
Actually,  I am currently struggling with a side effect of this approach
in the original OpenWrt version of this method, although i think this
version will suffer from the same effect. 

When you kexec the kernel from a previously running kernel, it only
detects a single core.  I am about to disable it entirely, as i really
need to be able to run kexec on a MT7621 platform.

I have instrumented the code with some debug to prove it is the case:

Boot from u-boot:

[    0.000000] nclusters = 1
[    0.000000] VPE topology
[    0.000000] cl = 0
[    0.000000] {
[    0.000000] ncores = 2
[    0.000000] cpulaunch.pc = 000000ff
[    0.000000] cpulaunch.gp = 0000ff00
[    0.000000] cpulaunch.sp = 0000ffff
[    0.000000] cpulaunch.a0 = 08000800
[    0.000000] cpulaunch.flags = 00000020
[    0.000000] plat_cpu_core_present(0) = true
[    0.000000] core_vpes = 2
[    0.000000] 2
[    0.000000] cpulaunch.pc = 000000ff
[    0.000000] cpulaunch.gp = 0000ff00
[    0.000000] cpulaunch.sp = 0000ffff
[    0.000000] cpulaunch.a0 = 08000800
[    0.000000] cpulaunch.flags = 00000020
[    0.000000] plat_cpu_core_present(1) = true
[    0.000000] core_vpes = 2
[    0.000000] ,2} total 4

Boot from kexec:

[    0.000000] nclusters = 1
[    0.000000] VPE topology
[    0.000000] cl = 0
[    0.000000] {
[    0.000000] ncores = 2
[    0.000000] cpulaunch.pc = 00000000
[    0.000000] cpulaunch.gp = 00000000
[    0.000000] cpulaunch.sp = 00000000
[    0.000000] cpulaunch.a0 = 00000000
[    0.000000] cpulaunch.flags = 00000000
[    0.000000] plat_cpu_core_present(0) = true
[    0.000000] core_vpes = 2
[    0.000000] 2
[    0.000000] cpulaunch.pc = 00000000
[    0.000000] cpulaunch.gp = 00000000
[    0.000000] cpulaunch.sp = 00000000
[    0.000000] cpulaunch.a0 = 00000000
[    0.000000] cpulaunch.flags = 00000000} total 2

Steven

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ