linux-kernel - Re: x86/entry vs kgdb

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: <20200526162806.GD325280@hirez.programming.kicks-ass.net>
Date:   Tue, 26 May 2020 18:28:06 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Daniel Thompson <daniel.thompson@...aro.org>
Cc:     x86@...nel.org, sumit.garg@...aro.org, jason.wessel@...driver.com,
        dianders@...omium.org, kgdb-bugreport@...ts.sourceforge.net,
        linux-kernel@...r.kernel.org, pmladek@...e.com,
        sergey.senozhatsky@...il.com, will@...nel.org,
        laijs@...ux.alibaba.com
Subject: Re: x86/entry vs kgdb

On Tue, May 26, 2020 at 05:16:21PM +0100, Daniel Thompson wrote:
> On Mon, May 25, 2020 at 11:18:32AM +0200, Peter Zijlstra wrote:
> > On Mon, May 25, 2020 at 10:36:05AM +0200, Peter Zijlstra wrote:
> > > Hi!
> > > 
> > > Since you seem to care about kgdb, I figured you might want to fix this
> > > before I mark it broken on x86 (we've been considering doing that for a
> > > while).
> > > 
> > > AFAICT the whole debugreg usage of kgdb-x86_64 is completely hosed; it
> > > doesn't respsect the normal exclusion zones as per arch_build_bp_info().
> > > 
> > > That is, breakpoints must never be in:
> > > 
> > >   - in the cpu_entry_area
> > >   - in .entry.text
> > >   - in .noinstr.text
> > >   - in anything else marked NOKPROBE
> > > 
> > > by not respecting these constraints it is trivial to completely and
> > > utterly hose the machine. The entry rework that is current underway will
> > > explicitly not deal with #DB triggering in any of those places.
> > 
> > This also very much includes single stepping those bits.  Which KGDB
> > obviously also does not respects.
> 
> For breakpoints there's already a pre-poke validation hook that
> architectures can override if they want to. I can modify the default
> implementation to include checking the nokprobe list.

Excellent, and I suppose the arch callback should be changed to share
code with arch_build_bp_info(), which Lai was extending here:

  https://lkml.kernel.org/r/20200526014221.2119-1-laijs@linux.alibaba.com

> Stepping is a bit more complex. There are hooks for some of the
> underlying work but not pre-step validation hook. I'll see if we can add
> one.

That'd be great; because where we're going getting this wrong is
insta-fail.

Another point to look at is the whole dbg_is_early; I suspect that's
similarly wrecked, the entry code isn't more robust early on.