Re: [PATCH] x86/entry_32: Fix segment exceptions

From: Borislav Petkov <bp@alien8.de>
To: Peter Zijlstra <peterz@infradead.org>, Andy Lutomirski <luto@kernel.org>
Cc: Sean Christopherson <seanjc@google.com>,
	kernel test robot <oliver.sang@intel.com>,
	Josh Poimboeuf <jpoimboe@redhat.com>,
	LKML <linux-kernel@vger.kernel.org>,
	x86@kernel.org, lkp@lists.01.org, lkp@intel.com
Subject: Re: [PATCH] x86/entry_32: Fix segment exceptions
Date: Fri, 14 Jan 2022 12:24:52 +0100	[thread overview]
Message-ID: <YeFdeZsTWDt9WU8f@zn.tnic> (raw)
In-Reply-To: <YeCDudla868Ipf++@hirez.programming.kicks-ass.net>

On Thu, Jan 13, 2022 at 08:55:37PM +0100, Peter Zijlstra wrote:
> On IRC Andrew also noted that these EX_REG_* things should be __i386__
> only. Previosly when they lived as open-coded EX_DATA_REG() usage in
> entry_32.S that was implied, but now ...

I guess that then below.

amluto, I'd love it if we (and by that I mean you :-)) could document
the rules for GS on 32-bit so that it is clear what's going on there...

entry_32.S is only hinting at what's going on, we have comments here and
there but not all concentrated into a single location.

Thx.

---
From: Borislav Petkov <bp@suse.de>
Date: Fri, 14 Jan 2022 12:15:21 +0100
Subject: [PATCH] x86/entry_32: Remove GS from the pt_regs offsets and fixup
 regs

GS is special on 32-bit and segment exceptions fixup through it won't
work. Leave breadcrumbs for others not to walk into the same nasty.

Fixes: 9cdbeec40968 ("x86/entry_32: Fix segment exceptions")
Signed-off-by: Borislav Petkov <bp@suse.de>
---
 arch/x86/include/asm/extable_fixup_types.h | 5 +++--
 arch/x86/lib/insn-eval.c                   | 5 ++++-
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/arch/x86/include/asm/extable_fixup_types.h b/arch/x86/include/asm/extable_fixup_types.h
index 503622627400..0aa5f4d3234f 100644
--- a/arch/x86/include/asm/extable_fixup_types.h
+++ b/arch/x86/include/asm/extable_fixup_types.h
@@ -20,11 +20,12 @@
 #define EX_DATA_FLAG(flag)		((flag) << EX_DATA_FLAG_SHIFT)
 #define EX_DATA_IMM(imm)		((imm) << EX_DATA_IMM_SHIFT)
 
-/* segment regs */
+#ifdef CONFIG_X86_32
+/* segment regs, valid only for 32-bit code, see pt_regoff */
 #define EX_REG_DS			EX_DATA_REG(8)
 #define EX_REG_ES			EX_DATA_REG(9)
 #define EX_REG_FS			EX_DATA_REG(10)
-#define EX_REG_GS			EX_DATA_REG(11)
+#endif
 
 /* flags */
 #define EX_FLAG_CLEAR_AX		EX_DATA_FLAG(1)
diff --git a/arch/x86/lib/insn-eval.c b/arch/x86/lib/insn-eval.c
index b781d324211b..34001eee7482 100644
--- a/arch/x86/lib/insn-eval.c
+++ b/arch/x86/lib/insn-eval.c
@@ -432,7 +432,10 @@ static const int pt_regoff[] = {
 	offsetof(struct pt_regs, ds),
 	offsetof(struct pt_regs, es),
 	offsetof(struct pt_regs, fs),
-	offsetof(struct pt_regs, gs),
+	/*
+	 * Can't use that one - it is special - see entry_32.S
+	 * offsetof(struct pt_regs, gs),
+	 */
 #endif
 };
 
-- 
2.29.2


-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
