|
Jiri Slaby |
a43d3d |
From: Peter Zijlstra <peterz@infradead.org>
|
|
Jiri Slaby |
a43d3d |
Date: Thu, 26 Jan 2023 16:34:27 +0100
|
|
Jiri Slaby |
a43d3d |
Subject: [PATCH] x86/static_call: Add support for Jcc tail-calls
|
|
Jiri Slaby |
a43d3d |
References: bsc#1012628
|
|
Jiri Slaby |
a43d3d |
Patch-mainline: 6.2.1
|
|
Jiri Slaby |
a43d3d |
Git-commit: 923510c88d2b7d947c4217835fd9ca6bd65cc56c
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
commit 923510c88d2b7d947c4217835fd9ca6bd65cc56c upstream.
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
Clang likes to create conditional tail calls like:
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
0000000000000350 <amd_pmu_add_event>:
|
|
Jiri Slaby |
a43d3d |
350: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) 351: R_X86_64_NONE __fentry__-0x4
|
|
Jiri Slaby |
a43d3d |
355: 48 83 bf 20 01 00 00 00 cmpq $0x0,0x120(%rdi)
|
|
Jiri Slaby |
a43d3d |
35d: 0f 85 00 00 00 00 jne 363 <amd_pmu_add_event+0x13> 35f: R_X86_64_PLT32 __SCT__amd_pmu_branch_add-0x4
|
|
Jiri Slaby |
a43d3d |
363: e9 00 00 00 00 jmp 368 <amd_pmu_add_event+0x18> 364: R_X86_64_PLT32 __x86_return_thunk-0x4
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
Where 0x35d is a static call site that's turned into a conditional
|
|
Jiri Slaby |
a43d3d |
tail-call using the Jcc class of instructions.
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
Teach the in-line static call text patching about this.
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
Notably, since there is no conditional-ret, in that case patch the Jcc
|
|
Jiri Slaby |
a43d3d |
to point at an empty stub function that does the ret -- or the return
|
|
Jiri Slaby |
a43d3d |
thunk when needed.
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
Reported-by: "Erhard F." <erhard_f@mailbox.org>
|
|
Jiri Slaby |
a43d3d |
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
|
|
Jiri Slaby |
a43d3d |
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
Jiri Slaby |
a43d3d |
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
|
|
Jiri Slaby |
a43d3d |
Link: https://lore.kernel.org/r/Y9Kdg9QjHkr9G5b5@hirez.programming.kicks-ass.net
|
|
Jiri Slaby |
a43d3d |
Cc: Nathan Chancellor <nathan@kernel.org>
|
|
Jiri Slaby |
a43d3d |
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
Jiri Slaby |
a43d3d |
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
|
|
Jiri Slaby |
a43d3d |
---
|
|
Jiri Slaby |
a43d3d |
arch/x86/kernel/static_call.c | 50 ++++++++++++++++++++++++++++++++---
|
|
Jiri Slaby |
a43d3d |
1 file changed, 47 insertions(+), 3 deletions(-)
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
diff --git a/arch/x86/kernel/static_call.c b/arch/x86/kernel/static_call.c
|
|
Jiri Slaby |
a43d3d |
index 2ebc3389..b70670a9 100644
|
|
Jiri Slaby |
a43d3d |
--- a/arch/x86/kernel/static_call.c
|
|
Jiri Slaby |
a43d3d |
+++ b/arch/x86/kernel/static_call.c
|
|
Jiri Slaby |
a43d3d |
@@ -9,6 +9,7 @@ enum insn_type {
|
|
Jiri Slaby |
a43d3d |
NOP = 1, /* site cond-call */
|
|
Jiri Slaby |
a43d3d |
JMP = 2, /* tramp / site tail-call */
|
|
Jiri Slaby |
a43d3d |
RET = 3, /* tramp / site cond-tail-call */
|
|
Jiri Slaby |
a43d3d |
+ JCC = 4,
|
|
Jiri Slaby |
a43d3d |
};
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
/*
|
|
Jiri Slaby |
a43d3d |
@@ -25,12 +26,40 @@ static const u8 xor5rax[] = { 0x2e, 0x2e, 0x2e, 0x31, 0xc0 };
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
static const u8 retinsn[] = { RET_INSN_OPCODE, 0xcc, 0xcc, 0xcc, 0xcc };
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
+static u8 __is_Jcc(u8 *insn) /* Jcc.d32 */
|
|
Jiri Slaby |
a43d3d |
+{
|
|
Jiri Slaby |
a43d3d |
+ u8 ret = 0;
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+ if (insn[0] == 0x0f) {
|
|
Jiri Slaby |
a43d3d |
+ u8 tmp = insn[1];
|
|
Jiri Slaby |
a43d3d |
+ if ((tmp & 0xf0) == 0x80)
|
|
Jiri Slaby |
a43d3d |
+ ret = tmp;
|
|
Jiri Slaby |
a43d3d |
+ }
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+ return ret;
|
|
Jiri Slaby |
a43d3d |
+}
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+extern void __static_call_return(void);
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+asm (".global __static_call_return\n\t"
|
|
Jiri Slaby |
a43d3d |
+ ".type __static_call_return, @function\n\t"
|
|
Jiri Slaby |
a43d3d |
+ ASM_FUNC_ALIGN "\n\t"
|
|
Jiri Slaby |
a43d3d |
+ "__static_call_return:\n\t"
|
|
Jiri Slaby |
a43d3d |
+ ANNOTATE_NOENDBR
|
|
Jiri Slaby |
a43d3d |
+ ANNOTATE_RETPOLINE_SAFE
|
|
Jiri Slaby |
a43d3d |
+ "ret; int3\n\t"
|
|
Jiri Slaby |
a43d3d |
+ ".size __static_call_return, . - __static_call_return \n\t");
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
static void __ref __static_call_transform(void *insn, enum insn_type type,
|
|
Jiri Slaby |
a43d3d |
void *func, bool modinit)
|
|
Jiri Slaby |
a43d3d |
{
|
|
Jiri Slaby |
a43d3d |
const void *emulate = NULL;
|
|
Jiri Slaby |
a43d3d |
int size = CALL_INSN_SIZE;
|
|
Jiri Slaby |
a43d3d |
const void *code;
|
|
Jiri Slaby |
a43d3d |
+ u8 op, buf[6];
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+ if ((type == JMP || type == RET) && (op = __is_Jcc(insn)))
|
|
Jiri Slaby |
a43d3d |
+ type = JCC;
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
switch (type) {
|
|
Jiri Slaby |
a43d3d |
case CALL:
|
|
Jiri Slaby |
a43d3d |
@@ -57,6 +86,20 @@ static void __ref __static_call_transform(void *insn, enum insn_type type,
|
|
Jiri Slaby |
a43d3d |
else
|
|
Jiri Slaby |
a43d3d |
code = &retinsn;
|
|
Jiri Slaby |
a43d3d |
break;
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+ case JCC:
|
|
Jiri Slaby |
a43d3d |
+ if (!func) {
|
|
Jiri Slaby |
a43d3d |
+ func = __static_call_return;
|
|
Jiri Slaby |
a43d3d |
+ if (cpu_feature_enabled(X86_FEATURE_RETHUNK))
|
|
Jiri Slaby |
a43d3d |
+ func = x86_return_thunk;
|
|
Jiri Slaby |
a43d3d |
+ }
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+ buf[0] = 0x0f;
|
|
Jiri Slaby |
a43d3d |
+ __text_gen_insn(buf+1, op, insn+1, func, 5);
|
|
Jiri Slaby |
a43d3d |
+ code = buf;
|
|
Jiri Slaby |
a43d3d |
+ size = 6;
|
|
Jiri Slaby |
a43d3d |
+
|
|
Jiri Slaby |
a43d3d |
+ break;
|
|
Jiri Slaby |
a43d3d |
}
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
if (memcmp(insn, code, size) == 0)
|
|
Jiri Slaby |
a43d3d |
@@ -68,9 +111,9 @@ static void __ref __static_call_transform(void *insn, enum insn_type type,
|
|
Jiri Slaby |
a43d3d |
text_poke_bp(insn, code, size, emulate);
|
|
Jiri Slaby |
a43d3d |
}
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
-static void __static_call_validate(void *insn, bool tail, bool tramp)
|
|
Jiri Slaby |
a43d3d |
+static void __static_call_validate(u8 *insn, bool tail, bool tramp)
|
|
Jiri Slaby |
a43d3d |
{
|
|
Jiri Slaby |
a43d3d |
- u8 opcode = *(u8 *)insn;
|
|
Jiri Slaby |
a43d3d |
+ u8 opcode = insn[0];
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
if (tramp && memcmp(insn+5, tramp_ud, 3)) {
|
|
Jiri Slaby |
a43d3d |
pr_err("trampoline signature fail");
|
|
Jiri Slaby |
a43d3d |
@@ -79,7 +122,8 @@ static void __static_call_validate(void *insn, bool tail, bool tramp)
|
|
Jiri Slaby |
a43d3d |
|
|
Jiri Slaby |
a43d3d |
if (tail) {
|
|
Jiri Slaby |
a43d3d |
if (opcode == JMP32_INSN_OPCODE ||
|
|
Jiri Slaby |
a43d3d |
- opcode == RET_INSN_OPCODE)
|
|
Jiri Slaby |
a43d3d |
+ opcode == RET_INSN_OPCODE ||
|
|
Jiri Slaby |
a43d3d |
+ __is_Jcc(insn))
|
|
Jiri Slaby |
a43d3d |
return;
|
|
Jiri Slaby |
a43d3d |
} else {
|
|
Jiri Slaby |
a43d3d |
if (opcode == CALL_INSN_OPCODE ||
|
|
Jiri Slaby |
a43d3d |
--
|
|
Jiri Slaby |
a43d3d |
2.35.3
|
|
Jiri Slaby |
a43d3d |
|