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>
regs->ip = landing_pad;
/*
- * Fetch ECX from where the vDSO stashed it.
+ * Fetch EBP from where the vDSO stashed it.
*
* WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
*/
* Micro-optimization: the pointer we're following is explicitly
* 32 bits, so it can't be out of range.
*/
- __get_user(*(u32 *)®s->cx,
+ __get_user(*(u32 *)®s->bp,
(u32 __user __force *)(unsigned long)(u32)regs->sp)
#else
- get_user(*(u32 *)®s->cx,
+ get_user(*(u32 *)®s->bp,
(u32 __user __force *)(unsigned long)(u32)regs->sp)
#endif
) {
movl TSS_sysenter_sp0(%esp), %esp
sysenter_past_esp:
pushl $__USER_DS /* pt_regs->ss */
- pushl %ecx /* pt_regs->sp (stashed in cx) */
+ pushl %ebp /* pt_regs->sp (stashed in bp) */
pushfl /* pt_regs->flags (except IF = 0) */
orl $X86_EFLAGS_IF, (%esp) /* Fix IF */
pushl $__USER_CS /* pt_regs->cs */
/* Construct struct pt_regs on stack */
pushq $__USER32_DS /* pt_regs->ss */
- pushq %rcx /* pt_regs->sp */
+ pushq %rbp /* pt_regs->sp (stashed in bp) */
/*
* Push flags. This is nasty. First, interrupts are currently
pushq %rdi /* pt_regs->di */
pushq %rsi /* pt_regs->si */
pushq %rdx /* pt_regs->dx */
- pushq %rcx /* pt_regs->cx (will be overwritten) */
+ pushq %rcx /* pt_regs->cx */
pushq $-ENOSYS /* pt_regs->ax */
pushq %r8 /* pt_regs->r8 = 0 */
pushq %r8 /* pt_regs->r9 = 0 */
pushq %r8 /* pt_regs->r10 = 0 */
pushq %r8 /* pt_regs->r11 = 0 */
pushq %rbx /* pt_regs->rbx */
- pushq %rbp /* pt_regs->rbp */
+ pushq %rbp /* pt_regs->rbp (will be overwritten) */
pushq %r8 /* pt_regs->r12 = 0 */
pushq %r8 /* pt_regs->r13 = 0 */
pushq %r8 /* pt_regs->r14 = 0 */
pushq %rdi /* pt_regs->di */
pushq %rsi /* pt_regs->si */
pushq %rdx /* pt_regs->dx */
- pushq %rcx /* pt_regs->cx (will be overwritten) */
+ pushq %rbp /* pt_regs->cx (stashed in bp) */
pushq $-ENOSYS /* pt_regs->ax */
xorq %r8,%r8
pushq %r8 /* pt_regs->r8 = 0 */
pushq %r8 /* pt_regs->r10 = 0 */
pushq %r8 /* pt_regs->r11 = 0 */
pushq %rbx /* pt_regs->rbx */
- pushq %rbp /* pt_regs->rbp */
+ pushq %rbp /* pt_regs->rbp (will be overwritten) */
pushq %r8 /* pt_regs->r12 = 0 */
pushq %r8 /* pt_regs->r13 = 0 */
pushq %r8 /* pt_regs->r14 = 0 */
/*
* 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