mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Steven Rostedt <rostedt@goodmis.org>
To: Nicolai Stange <nstange@suse.de>
Cc: Ingo Molnar <mingo@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	"H. Peter Anvin" <hpa@zytor.com>,
	Josh Poimboeuf <jpoimboe@redhat.com>,
	Jiri Kosina <jikos@kernel.org>, Miroslav Benes <mbenes@suse.cz>,
	Petr Mladek <pmladek@suse.com>,
	live-patching@vger.kernel.org, x86@kernel.org,
	linux-kernel@vger.kernel.org,
	Andy Lutomirski <luto@amacapital.net>
Subject: Re: [RFC PATCH 1/1] x86/ftrace: make ftrace_int3_handler() not to skip fops invocation
Date: Tue, 23 Apr 2019 19:50:23 -0400	[thread overview]
Message-ID: <20190423195023.425902dc@gandalf.local.home> (raw)
In-Reply-To: <87ef5sfuzu.fsf@suse.de>

On Tue, 23 Apr 2019 20:15:49 +0200
Nicolai Stange <nstange@suse.de> wrote:

> > This can be made much simpler by making a hardcoded ftrace_int3_tramp
> > that does the following:
> >
> > ftrace_int3_tramp
> > 	push	%r11
> > 	jmp	ftrace_caller  
> 
> 
> But wouldn't this mean that ftrace_caller could nest if the breakpoint
> in question happened to be placed at ftrace_call? Infinite recursion set
> aside, the ip value determined from inner calls based on the on-stack
> return address would be wrong, AFAICS. Or am I missing something here?

I had a reply here, but I think you explained what I just explained
(and then deleted) below ;-)

> 
> 
> > The ftrace_caller will either call a single ftrace callback, if there's
> > only a single ops registered, or it will call the loop function that
> > iterates over all the ftrace_ops registered and will call each function
> > that matches the ftrace_ops hashes.
> >
> > In either case, it's what we want.  
> 
> Ok, under the assumption that my above point is valid, this patch could
> still get simplified a lot by having two trampolines:
> 
> 1.) Your ftrace_int3_tramp from above, to be used if the patched insn is
>     some mcount call site. The live patching fops will need
>     ftrace_regs_caller though. So,
> 
> 	ftrace_int3_tramp_regs_caller:
> 		push %r11
>                 jmp ftrace_regs_caller
> 
> 2.) Another one redirecting control flow to ftrace_ops_list_func(). It's
>     going to be used when the int3 is found at ftrace_call or
>     ftrace_regs_call resp. their counterparts in some ftrace_ops'
>     trampoline.
> 
> 	ftrace_int3_tramp_list_func:
> 		push %r11
>                 jmp ftrace_ops_list_func

Yes, I wrote a reply basically stating something similar, but then
deleted it after reading what you wrote here!

> 
> ftrace_int3_handler() would then distinguish the following cases:
> 1.) ip == ftrace_graph_call => ignore, i.e. skip the insn

OK, because this transition would be from "call function graph" to
"nop" or the other way. Either case, one would always be a nop.

> 2.) is_ftrace_caller(ip) => ftrace_int3_tramp_list_func
> 3.) ftrace_location(ip) => ftrace_int3_tramp_regs_caller
> 
> ftrace_location() is getting invoked from ftrace_int3_handler() already,
> so there wouldn't be any additional cost.
> 
> If that makes sense to you, I'd prepare a patch...

Yes it does.

> 
> 
> > The ftrace_int3_tramp will simply simulate the call ftrace_caller that
> > would be there as the default caller if more than one function is set
> > to it.
> >
> > For 32 bit, we could add 4 variables on the thread_info and make 4
> > trampolines, one for each context (normal, softirq, irq and NMI), and
> > have them use the variable stored in the thread_info for that location:
> >
> > ftrace_int3_tramp_normal
> > 	push %eax # just for space
> > 	push %eax
> > 	mov whatever_to_get_thread_info, %eax
> > 	mov normal(%eax), %eax
> > 	mov %eax, 4(%esp)
> > 	pop %eax
> > 	jmp ftrace_caller
> >
> > ftrace_int3_tramp_softiqr
> > 	push %eax # just for space
> > 	push %eax
> > 	mov whatever_to_get_thread_info, %eax
> > 	mov softirq(%eax), %eax
> > 	mov %eax, 4(%esp)
> > 	pop %eax
> > 	jmp ftrace_caller
> >
> > etc..
> >
> >
> > A bit crazier for 32 bit but it can still be done. ;-)  
> 
> Ok, but currently CONFIG_HAVE_LIVEPATCH=n for x86 && !x86_64,
> so I'd rather not invest too much energy into screwing 32 bit up ;)
>

Probably not something you care about, but something that I do. Which
means it will have to go on my TODO list. I care about missed functions
being called. This means, if you have something tracing a function, and
then enable something else to trace that same function, you might miss
the first one.

-- Steve

  reply	other threads:[~2019-04-23 23:50 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-07-26 10:40 [RFC PATCH 0/1] x86/ftrace: fix live patching vs. tracing race Nicolai Stange
2018-07-26 10:40 ` [RFC PATCH 1/1] x86/ftrace: make ftrace_int3_handler() not to skip fops invocation Nicolai Stange
2019-04-19 20:05   ` Steven Rostedt
2019-04-23 18:15     ` Nicolai Stange
2019-04-23 23:50       ` Steven Rostedt [this message]
2019-04-24  6:20         ` Nicolai Stange
2019-04-24 12:35           ` Steven Rostedt
2018-07-26 14:23 ` [RFC PATCH 0/1] x86/ftrace: fix live patching vs. tracing race Steven Rostedt

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20190423195023.425902dc@gandalf.local.home \
    --to=rostedt@goodmis.org \
    --cc=hpa@zytor.com \
    --cc=jikos@kernel.org \
    --cc=jpoimboe@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=live-patching@vger.kernel.org \
    --cc=luto@amacapital.net \
    --cc=mbenes@suse.cz \
    --cc=mingo@redhat.com \
    --cc=nstange@suse.de \
    --cc=pmladek@suse.com \
    --cc=tglx@linutronix.de \
    --cc=x86@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®