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]
Message-ID: <61ce1a14-7bac-fea8-b065-83a1c0704258@intel.com>
Date:   Wed, 3 Aug 2022 07:49:34 -0700
From:   Dave Hansen <dave.hansen@...el.com>
To:     Rik van Riel <riel@...riel.com>, linux-kernel@...r.kernel.org
Cc:     x86@...r.kernel.org, kernel-team@...com,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        Thomas Gleixner <tglx@...utronix.de>, Dave Jones <dsj@...com>,
        Andy Lutomirski <luto@...nel.org>
Subject: Re: [PATCH] x86,mm: print likely CPU at segfault time

On 8/2/22 13:09, Rik van Riel wrote:
> Add a printk to show_signal_msg() to print the CPU, core, and socket

Nit:     ^ printk(), please

> --- a/arch/x86/mm/fault.c
> +++ b/arch/x86/mm/fault.c
> @@ -782,6 +782,12 @@ show_signal_msg(struct pt_regs *regs, unsigned long error_code,
>  
>  	print_vma_addr(KERN_CONT " in ", regs->ip);
>  
> +	printk(KERN_CONT " on CPU %d (core %d, socket %d)",
> +	       raw_smp_processor_id(),
> +	       topology_core_id(raw_smp_processor_id()),
> +	       topology_physical_package_id(raw_smp_processor_id()));

This seems totally sane to me.  I have found myself looking through
customer-provided *oopses* more than once trying to figure out if the
same CPU cores were at fault.  This extends that to userspace crashes
too.  I've also found myself trying to map back from logical CPU numbers
to core and package.

One nit: Preempt is enabled here, right?  I understand that this thing
is fundamentally racy, but if we did:

	int cpu = READ_ONCE(raw_smp_processor_id());

it would make it internally *consistent*.  Without that, we could
theoretically get three different raw_smp_processor_id()'s.  It might
even make the code look a wee bit nicer.

The changelog here is great, but  couple of comments would also be nice:

	/* This is a racy snapshot, but is better than nothing: */
	int cpu = READ_ONCE(raw_smp_processor_id());
...
	/*
	 * Dump the likely CPU where the fatal segfault happened.  This
	 * can help help identify buggy pieces of hardware.
	 */
	printk(KERN_CONT " on CPU %d (core %d, socket %d)", cpu,
	       topology_core_id(cpu),
	       topology_physical_package_id(cpu));

If you want to wait a bit and see if you get any other comments, this
seems like something we can suck in after the merge window.

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ