linux-kernel - Re: [RFC PATCH 2/5] entry: Add calls for save/restore auxiliary pt

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] [thread-next>] [day] [month] [year] [list]

Message-ID: <YvKpi/CVHko50PNQ@iweiny-desk3>
Date:   Tue, 9 Aug 2022 11:38:03 -0700
From:   Ira Weiny <ira.weiny@...el.com>
To:     Borislav Petkov <bp@...en8.de>
CC:     Rik van Riel <riel@...riel.com>,
        Dave Hansen <dave.hansen@...el.com>,
        Dave Hansen <dave.hansen@...ux.intel.com>, <x86@...nel.org>,
        <linux-kernel@...r.kernel.org>, <kernel-team@...com>
Subject: Re: [RFC PATCH 2/5] entry: Add calls for save/restore auxiliary
 pt_regs

On Tue, Aug 09, 2022 at 02:05:15PM +0200, Borislav Petkov wrote:
> On Fri, Aug 05, 2022 at 10:30:06AM -0700, ira.weiny@...el.com wrote:
> > From: Ira Weiny <ira.weiny@...el.com>
> > 
> > Some architectures have auxiliary pt_regs space available to store
> > information on the stack during exceptions.  This information is easier
> > to obtain and store within C code rather than in arch specific assembly.
> 
> There are others?

Other archs?  not now.

> 
> Because I would've done this whole thing in arch/x86/ only...

Thomas did a lot of work to make the entry code generic.  So I was keeping that
work consistent.  This also helps to ensure I did not miss any places.

> 
> > diff --git a/kernel/entry/common.c b/kernel/entry/common.c
> > index 8c0f334c4b75..a70a0f314aee 100644
> > --- a/kernel/entry/common.c
> > +++ b/kernel/entry/common.c
> > @@ -317,7 +317,7 @@ noinstr irqentry_state_t irqentry_enter(struct pt_regs *regs)
> >  
> >  	if (user_mode(regs)) {
> >  		irqentry_enter_from_user_mode(regs);
> > -		return ret;
> > +		goto aux_save;
> 
> Why do you have to goto and do the instrumentation sandwitch around it
> at the goto label?
> 
> Why not simply do
> 
> 	if (user_mode(regs)) {
> 		irqentry_enter_from_user_mode(regs);
> 		arch_save_aux_pt_regs(regs);
> 		return ret;

I don't believe this is correct because instrumentation is not enabled here.
See below.[1]

There are 3 exit paths from irqentry_enter().  All must call
arch_save_aux_pt_regs().  I felt the maintenance of the code would be easier if
I did not scatter those calls through the function and simply exited 1
place.[1]

FWICT calling instrumentation_{begin,end}() is a noop in production code.  So
there is no real cost to calling instrumentation_begin() -> end -> begin ->
end.  And the goto seemed low enough overhead.  Given the current discussion I
admit I may have made the wrong choice.  But I think I would add some comments
to the below to help future developers.

Ira

[1]

--- a/kernel/entry/common.c
+++ b/kernel/entry/common.c
@@ -317,6 +317,9 @@ noinstr irqentry_state_t irqentry_enter(struct pt_regs *regs)

        if (user_mode(regs)) {
                irqentry_enter_from_user_mode(regs);
+               instrumentation_begin();
+               arch_save_aux_pt_regs(regs);
+               instrumentation_end();
                return ret;
        }

@@ -353,6 +356,7 @@ noinstr irqentry_state_t irqentry_enter(struct pt_regs *regs)
                ct_irq_enter();
                instrumentation_begin();
                trace_hardirqs_off_finish();
+               arch_save_aux_pt_regs(regs);
                instrumentation_end();

                ret.exit_rcu = true;
@@ -369,6 +373,7 @@ noinstr irqentry_state_t irqentry_enter(struct pt_regs *regs)
        instrumentation_begin();
        rcu_irq_enter_check_tick();
        trace_hardirqs_off_finish();
+       arch_save_aux_pt_regs(regs);
        instrumentation_end();

        return ret;