x86/entry: Restore traditional SYSENTER calling convention

It turns out that some Android versions hardcode the SYSENTER
calling convention.  This is buggy and will cause problems no
matter what the kernel does.  Nonetheless, we should try to
support it.

Credit goes to Linus for pointing out a clean way to handle
the SYSENTER/SYSCALL clobber differences while preserving
straightforward DWARF annotations.

I believe that the original offending Android commit was:

https://android.googlesource.com/platform%2Fbionic/+/7dc3684d7a2587e43e6d2a8e0e3f39bf759bd535

Reported-by: Qiuxu Zhuo <qiuxu.zhuo@intel.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Reviewed-and-tested-by: Borislav Petkov <bp@alien8.de>
Cc: <mark.gross@intel.com>
Cc: Su Tao <tao.su@intel.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: <frank.wang@intel.com>
Cc: <borun.fu@intel.com>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Mingwei Shi <mingwei.shi@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

1 files changed, 42 insertions, 10 deletions

diff --git a/arch/x86/entry/vdso/vdso32/system_call.S b/arch/x86/entry/vdso/vdso32/system_call.S
index 8f42b1b9e8df..3a1d9297074b 100644
--- a/arch/x86/entry/vdso/vdso32/system_call.S
+++ b/arch/x86/entry/vdso/vdso32/system_call.S
@@ -21,35 +21,67 @@ __kernel_vsyscall:
 	/*
 	 * Reshuffle regs so that all of any of the entry instructions
 	 * will preserve enough state.
+	 *
+	 * A really nice entry sequence would be:
+	 *  pushl %edx
+	 *  pushl %ecx
+	 *  movl  %esp, %ecx
+	 *
+	 * Unfortunately, naughty Android versions between July and December
+	 * 2015 actually hardcode the traditional Linux SYSENTER entry
+	 * sequence.  That is severely broken for a number of reasons (ask
+	 * anyone with an AMD CPU, for example).  Nonetheless, we try to keep
+	 * it working approximately as well as it ever worked.
+	 *
+	 * This link may eludicate some of the history:
+	 *   https://android-review.googlesource.com/#/q/Iac3295376d61ef83e713ac9b528f3b50aa780cd7
+	 * personally, I find it hard to understand what's going on there.
+	 *
+	 * Note to future user developers: DO NOT USE SYSENTER IN YOUR CODE.
+	 * Execute an indirect call to the address in the AT_SYSINFO auxv
+	 * entry.  That is the ONLY correct way to make a fast 32-bit system
+	 * call on Linux.  (Open-coding int $0x80 is also fine, but it's
+	 * slow.)
 	 */
+	pushl	%ecx
+	CFI_ADJUST_CFA_OFFSET	4
+	CFI_REL_OFFSET		ecx, 0
 	pushl	%edx
 	CFI_ADJUST_CFA_OFFSET	4
 	CFI_REL_OFFSET		edx, 0
-	pushl	%ecx
+	pushl	%ebp
 	CFI_ADJUST_CFA_OFFSET	4
-	CFI_REL_OFFSET		ecx, 0
-	movl	%esp, %ecx
+	CFI_REL_OFFSET		ebp, 0
+
+	#define SYSENTER_SEQUENCE	"movl %esp, %ebp; sysenter"
+	#define SYSCALL_SEQUENCE	"movl %ecx, %ebp; syscall"
 
 #ifdef CONFIG_X86_64
 	/* If SYSENTER (Intel) or SYSCALL32 (AMD) is available, use it. */
-	ALTERNATIVE_2 "", "sysenter", X86_FEATURE_SYSENTER32, \
-	                  "syscall",  X86_FEATURE_SYSCALL32
+	ALTERNATIVE_2 "", SYSENTER_SEQUENCE, X86_FEATURE_SYSENTER32, \
+	                  SYSCALL_SEQUENCE,  X86_FEATURE_SYSCALL32
 #else
-	ALTERNATIVE "", "sysenter", X86_FEATURE_SEP
+	ALTERNATIVE "", SYSENTER_SEQUENCE, X86_FEATURE_SEP
 #endif
 
 	/* Enter using int $0x80 */
-	movl	(%esp), %ecx
 	int	$0x80
 GLOBAL(int80_landing_pad)
 
-	/* Restore ECX and EDX in case they were clobbered. */
-	popl	%ecx
-	CFI_RESTORE		ecx
+	/*
+	 * Restore EDX and ECX in case they were clobbered.  EBP is not
+	 * clobbered (the kernel restores it), but it's cleaner and
+	 * probably faster to pop it than to adjust ESP using addl.
+	 */
+	popl	%ebp
+	CFI_RESTORE		ebp
 	CFI_ADJUST_CFA_OFFSET	-4
 	popl	%edx
 	CFI_RESTORE		edx
 	CFI_ADJUST_CFA_OFFSET	-4
+	popl	%ecx
+	CFI_RESTORE		ecx
+	CFI_ADJUST_CFA_OFFSET	-4
 	ret
 	CFI_ENDPROC