This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: [RFC] [PATCH 2.6.37-rc5-tip 13/20] 13: x86: x86 specific probe handling
- From: Roland McGrath <roland at redhat dot com>
- To: Peter Zijlstra <peterz at infradead dot org>
- Cc: Srikar Dronamraju <srikar at linux dot vnet dot ibm dot com>, Ingo Molnar <mingo at elte dot hu>, Steven Rostedt <rostedt at goodmis dot org>, Arnaldo Carvalho de Melo <acme at infradead dot org>, Linus Torvalds <torvalds at linux-foundation dot org>, Masami Hiramatsu <masami dot hiramatsu dot pt at hitachi dot com>, Christoph Hellwig <hch at infradead dot org>, Andi Kleen <andi at firstfloor dot org>, Oleg Nesterov <oleg at redhat dot com>, Andrew Morton <akpm at linux-foundation dot org>, SystemTap <systemtap at sources dot redhat dot com>, Jim Keniston <jkenisto at linux dot vnet dot ibm dot com>, Frederic Weisbecker <fweisbec at gmail dot com>, Ananth N Mavinakayanahalli <ananth at in dot ibm dot com>, LKML <linux-kernel at vger dot kernel dot org>, "Paul E. McKenney" <paulmck at linux dot vnet dot ibm dot com>
- Date: Fri, 28 Jan 2011 10:23:11 -0800 (PST)
- Subject: Re: [RFC] [PATCH 2.6.37-rc5-tip 13/20] 13: x86: x86 specific probe handling
- References: <20101216095714.23751.52601.sendpatchset@localhost6.localdomain6> <20101216095947.23751.75003.sendpatchset@localhost6.localdomain6> <1295963783.28776.1061.camel@laptop> <20110127094041.GR19725@linux.vnet.ibm.com> <1296123733.15234.53.camel@laptop> <20110127191146.DB22F180999@magilla.sf.frob.com> <20110128045721.GV19725@linux.vnet.ibm.com> <20110128062316.9515018099A@magilla.sf.frob.com> <1296203798.15234.249.camel@laptop>
> And reset the hardware back to block step when done, and provide the
> actual break blockstep would have.
Oh, sure, that too. If you're that ambitious, then the place to start
first is with plain single-step working right. When TF was already set
(either via user_enable_single_step, so TIF_SINGLESTEP is set, or just from
user mode, so it and TIF_FORCED_TF are not set, but TF is in the user
state's eflags) and you hit a uprobe, then after servicing the uprobe and
stepping over the copied original instruction and restoring the PC to where
it should be, you should let the trap turn into a SIGTRAP as normal rather
than swallowing it.
To support block-step correctly, you have to do something more clever.
If block-step was enabled (TIF_BLOCKSTEP set), then you need to figure
out which of two things is the right one to do. If the copied original
instruction uprobes just single-stepped over is one that would trigger
block-step, then you should treat it as if plain single-step were
enabled, i.e. let that SIGTRAP go as above. If not, then you should
swallow the signal, re-enable block-step and set TF (i.e. do the work of
user_enable_block_step) before resuming. You have to decide which case
it is based on instruction analysis. If it's a control-flow instruction
(including the syscall instructions), then it would trigger block-step.
IIRC a conditional branch instruction triggers it only if the branch is
taken (check the book), so you have to notice that too.
Thanks,
Roland