linux-kernel - Re: linux-next: build failure after merge of the vfs tree

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]

Date:   Mon, 28 Sep 2020 23:10:56 -0500
From:   Josh Poimboeuf <jpoimboe@...hat.com>
To:     Al Viro <viro@...iv.linux.org.uk>
Cc:     Stephen Rothwell <sfr@...b.auug.org.au>,
        Linux Next Mailing List <linux-next@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        "Peter Zijlstra (Intel)" <peterz@...radead.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Borislav Petkov <bp@...e.de>
Subject: Re: linux-next: build failure after merge of the vfs tree

On Fri, Sep 25, 2020 at 02:38:20PM +0100, Al Viro wrote:
> On Fri, Sep 25, 2020 at 10:01:28PM +1000, Stephen Rothwell wrote:
> > $ x86_64-linux-gnu-gcc --version
> > x86_64-linux-gnu-gcc (Debian 10.2.0-9) 10.2.0
> > $ x86_64-linux-gnu-ld --version
> > GNU ld (GNU Binutils for Debian) 2.35
> > 
> > and the gcc plugins don't get built for the allnoconfig builds.
> 
> > I reverted my Revert commit after I finished linux-next today and built
> > the x86_64 allnoconfig verion of lib/iov_iter.s:
> > 
> > $ grep -A 1 '41 "/home/sfr/next/next/arch/x86/include/asm/barrier.h"' lib/iov_iter.s
> > # 41 "/home/sfr/next/next/arch/x86/include/asm/barrier.h" 1
> > 	cmp $140737488351232,%rdx; sbb %rcx,%rcx;	#, uaddr, mask
> 
> Wait a sec...
> static inline unsigned long array_index_mask_nospec(unsigned long index,
>                 unsigned long size)
> {
>         unsigned long mask;
> 
>         asm volatile ("cmp %1,%2; sbb %0,%0;"
>                         :"=r" (mask)
>                         :"g"(size),"r" (index)
>                         :"cc");
>         return mask;
> }  
> 
> used with large constant size will blow up - "g" is wrong, since cmp allows
> 64bit arguments to be register or memory ones; immediates can't go past
> 32bit.
> 
> Looks like on the configs where it builds we end up with not seeing it's
> a constant...
> 
> Josh, any ideas?  We could, of course, make it "r"(size), but that would
> be unpleasant in all existing callers...

Sorry, I've been traveling.  I'd just vote for making it "r".

array_index_nospec() is always called after a usercopy.  I don't think
anyone will notice the extra mov, for the cases where it would be
propagated as an immediate.  And the argument *is* an unsigned long
after all.

Stephen, can you confirm this fixes it?

diff --git a/arch/x86/include/asm/barrier.h b/arch/x86/include/asm/barrier.h
index d158ea1fa250..69045ac62f58 100644
--- a/arch/x86/include/asm/barrier.h
+++ b/arch/x86/include/asm/barrier.h
@@ -40,7 +40,7 @@ static inline unsigned long array_index_mask_nospec(unsigned long index,
 
 	asm volatile ("cmp %1,%2; sbb %0,%0;"
 			:"=r" (mask)
-			:"g"(size),"r" (index)
+			:"r"(size), "r"(index)
 			:"cc");
 	return mask;
 }