From 30bfa7b3488bfb1bb75c9f50a5fcac1832970c60 Mon Sep 17 00:00:00 2001 From: Andy Lutomirski Date: Wed, 16 Dec 2015 23:18:48 -0800 Subject: [PATCH] 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 Signed-off-by: Andy Lutomirski Reviewed-and-tested-by: Borislav Petkov Cc: Cc: Su Tao Cc: Denys Vlasenko Cc: Cc: Cc: Brian Gerst Cc: Mingwei Shi Cc: Linus Torvalds Signed-off-by: Thomas Gleixner --- arch/x86/entry/common.c | 6 +-- arch/x86/entry/entry_32.S | 2 +- arch/x86/entry/entry_64_compat.S | 10 ++--- arch/x86/entry/vdso/vdso32/system_call.S | 52 +++++++++++++++++++----- 4 files changed, 51 insertions(+), 19 deletions(-) diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c index a89fdbc1f0be..03663740c866 100644 --- a/arch/x86/entry/common.c +++ b/arch/x86/entry/common.c @@ -421,7 +421,7 @@ __visible long do_fast_syscall_32(struct pt_regs *regs) 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! */ @@ -432,10 +432,10 @@ __visible long do_fast_syscall_32(struct pt_regs *regs) * 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 ) { diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S index fcad8ac30a8e..f3b6d54e0042 100644 --- a/arch/x86/entry/entry_32.S +++ b/arch/x86/entry/entry_32.S @@ -292,7 +292,7 @@ ENTRY(entry_SYSENTER_32) 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 */ diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S index 402e34a21559..6a1ae3751e82 100644 --- a/arch/x86/entry/entry_64_compat.S +++ b/arch/x86/entry/entry_64_compat.S @@ -63,7 +63,7 @@ ENTRY(entry_SYSENTER_compat) /* 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 @@ -82,14 +82,14 @@ ENTRY(entry_SYSENTER_compat) 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 */ @@ -179,7 +179,7 @@ ENTRY(entry_SYSCALL_compat) 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 */ @@ -187,7 +187,7 @@ ENTRY(entry_SYSCALL_compat) 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 */ 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 -- 2.20.1