Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 20 May 2013 09:48:46 -0400
From:      John Baldwin <jhb@freebsd.org>
To:        freebsd-amd64@freebsd.org
Cc:        Konstantin Belousov <kostikbel@gmail.com>, amd64@freebsd.org, current@freebsd.org
Subject:   Re: Fix amd64 ddb hardware watchpoints for SMP
Message-ID:  <201305200948.46211.jhb@freebsd.org>
In-Reply-To: <20130516064127.GC3047@kib.kiev.ua>
References:  <20130516064127.GC3047@kib.kiev.ua>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thursday, May 16, 2013 2:41:27 am Konstantin Belousov wrote:
> The ddb use of hardware watchpoints on the x86 architectures is known to
> be lacking. There are at least two known problems. One is the improper
> interaction with the user-mode debuggers which use debug registers.
> Another is that ddb only loads the debug registers for the watchpoint
> into the CPU which is executing ddb code, not setting up non-current
> processors.
> 
> Not touching the first problem for now, I want to fix the second issue,
> since as is, hardware watchpoints are useless on SMP. Patch below makes
> the stopped processors to load the debug registers after resuming from
> the stop handler, if directed by ddb.
> 
> Also the patch contains two other commands for ddb which made my
> exprerience with debugger on amd64 better. The 'show pginfo[/p] addr'
> command dumps the vm_page_t information, either by vm_page address, or,
> if the /p modifier is specified, by the physical page address. The 'show
> phys2dmap addr' command translates physical address into the directly
> mapped address, which can be accessed from ddb then.

This looks fine to me.  It would be nice to fix i386 as well to be consistent.
I would commit the new DDB commands as a separate patch from the watchpoint
fixes.

-- 
John Baldwin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201305200948.46211.jhb>