Received-SPF: pass (google.com: domain of linux-kernel+bounces-63816-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Date: Tue, 13 Feb 2024 15:36:00 -0000
From: "tip-bot2 for Borislav Petkov (AMD)" <tip-bot2@linutronix.de>
Sender: tip-bot2@linutronix.de
Reply-to: linux-kernel@vger.kernel.org
To: linux-tip-commits@vger.kernel.org
Subject: [tip: x86/alternatives] x86/alternatives: Use a temporary buffer when
 optimizing NOPs
Cc: "Borislav Petkov (AMD)" <bp@alien8.de>,
 Thomas Gleixner <tglx@linutronix.de>, x86@kernel.org,
 linux-kernel@vger.kernel.org
In-Reply-To: <20240130105941.19707-2-bp@alien8.de>
References: <20240130105941.19707-2-bp@alien8.de>
Precedence: bulk
MIME-Version: 1.0
Message-ID: <170783856041.398.6328711543597681720.tip-bot2@tip-bot2>
Robot-ID: <tip-bot2@linutronix.de>
Robot-Unsubscribe:
 Contact <mailto:tglx@linutronix.de> to get blacklisted from these emails
Precedence: bulk
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit

The following commit has been merged into the x86/alternatives branch of tip:

Commit-ID:     fca88864f9d5637381b71a0cd38b7f5e82eab01d
Gitweb:        https://git.kernel.org/tip/fca88864f9d5637381b71a0cd38b7f5e82eab01d
Author:        Borislav Petkov (AMD) <bp@alien8.de>
AuthorDate:    Tue, 30 Jan 2024 11:59:38 +01:00
Committer:     Borislav Petkov (AMD) <bp@alien8.de>
CommitterDate: Tue, 13 Feb 2024 16:16:02 +01:00

x86/alternatives: Use a temporary buffer when optimizing NOPs

Instead of optimizing NOPs in-place, use a temporary buffer like the
usual alternatives patching flow does. This obviates the need to grab
locks when patching, see

  6778977590da ("x86/alternatives: Disable interrupts and sync when optimizing NOPs in place")

While at it, add nomenclature definitions clarifying and simplifying the
naming of function-local variables in the alternatives code.

Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20240130105941.19707-2-bp@alien8.de
---
 arch/x86/kernel/alternative.c | 78 ++++++++++++++++++----------------
 1 file changed, 42 insertions(+), 36 deletions(-)

diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index 1d85cb7..835e343 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -125,6 +125,20 @@ const unsigned char * const x86_nops[ASM_NOP_MAX+1] =
 };
 
 /*
+ * Nomenclature for variable names to simplify and clarify this code and ease
+ * any potential staring at it:
+ *
+ * @instr: source address of the original instructions in the kernel text as
+ * generated by the compiler.
+ *
+ * @buf: temporary buffer on which the patching operates. This buffer is
+ * eventually text-poked into the kernel image.
+ *
+ * @replacement: pointer to the opcodes which are replacing @instr, located in
+ * the .altinstr_replacement section.
+ */
+
+/*
  * Fill the buffer with a single effective instruction of size @len.
  *
  * In order not to issue an ORC stack depth tracking CFI entry (Call Frame Info)
@@ -133,28 +147,28 @@ const unsigned char * const x86_nops[ASM_NOP_MAX+1] =
  * each single-byte NOPs). If @len to fill out is > ASM_NOP_MAX, pad with INT3 and
  * *jump* over instead of executing long and daft NOPs.
  */
-static void __init_or_module add_nop(u8 *instr, unsigned int len)
+static void __init_or_module add_nop(u8 *buf, unsigned int len)
 {
-	u8 *target = instr + len;
+	u8 *target = buf + len;
 
 	if (!len)
 		return;
 
 	if (len <= ASM_NOP_MAX) {
-		memcpy(instr, x86_nops[len], len);
+		memcpy(buf, x86_nops[len], len);
 		return;
 	}
 
 	if (len < 128) {
-		__text_gen_insn(instr, JMP8_INSN_OPCODE, instr, target, JMP8_INSN_SIZE);
-		instr += JMP8_INSN_SIZE;
+		__text_gen_insn(buf, JMP8_INSN_OPCODE, buf, target, JMP8_INSN_SIZE);
+		buf += JMP8_INSN_SIZE;
 	} else {
-		__text_gen_insn(instr, JMP32_INSN_OPCODE, instr, target, JMP32_INSN_SIZE);
-		instr += JMP32_INSN_SIZE;
+		__text_gen_insn(buf, JMP32_INSN_OPCODE, buf, target, JMP32_INSN_SIZE);
+		buf += JMP32_INSN_SIZE;
 	}
 
-	for (;instr < target; instr++)
-		*instr = INT3_INSN_OPCODE;
+	for (;buf < target; buf++)
+		*buf = INT3_INSN_OPCODE;
 }
 
 extern s32 __retpoline_sites[], __retpoline_sites_end[];
@@ -187,12 +201,12 @@ static bool insn_is_nop(struct insn *insn)
  * Find the offset of the first non-NOP instruction starting at @offset
  * but no further than @len.
  */
-static int skip_nops(u8 *instr, int offset, int len)
+static int skip_nops(u8 *buf, int offset, int len)
 {
 	struct insn insn;
 
 	for (; offset < len; offset += insn.length) {
-		if (insn_decode_kernel(&insn, &instr[offset]))
+		if (insn_decode_kernel(&insn, &buf[offset]))
 			break;
 
 		if (!insn_is_nop(&insn))
@@ -207,7 +221,7 @@ static int skip_nops(u8 *instr, int offset, int len)
  * to the end of the NOP sequence into a single NOP.
  */
 static bool __init_or_module
-__optimize_nops(u8 *instr, size_t len, struct insn *insn, int *next, int *prev, int *target)
+__optimize_nops(const u8 * const instr, u8 *buf, size_t len, struct insn *insn, int *next, int *prev, int *target)
 {
 	int i = *next - insn->length;
 
@@ -222,12 +236,12 @@ __optimize_nops(u8 *instr, size_t len, struct insn *insn, int *next, int *prev, 
 	if (insn_is_nop(insn)) {
 		int nop = i;
 
-		*next = skip_nops(instr, *next, len);
+		*next = skip_nops(buf, *next, len);
 		if (*target && *next == *target)
 			nop = *prev;
 
-		add_nop(instr + nop, *next - nop);
-		DUMP_BYTES(ALT, instr, len, "%px: [%d:%d) optimized NOPs: ", instr, nop, *next);
+		add_nop(buf + nop, *next - nop);
+		DUMP_BYTES(ALT, buf, len, "%px: [%d:%d) optimized NOPs: ", instr, nop, *next);
 		return true;
 	}
 
@@ -239,32 +253,22 @@ __optimize_nops(u8 *instr, size_t len, struct insn *insn, int *next, int *prev, 
  * "noinline" to cause control flow change and thus invalidate I$ and
  * cause refetch after modification.
  */
-static void __init_or_module noinline optimize_nops(u8 *instr, size_t len)
+static void __init_or_module noinline optimize_nops(const u8 * const instr, u8 *buf, size_t len)
 {
 	int prev, target = 0;
 
 	for (int next, i = 0; i < len; i = next) {
 		struct insn insn;
 
-		if (insn_decode_kernel(&insn, &instr[i]))
+		if (insn_decode_kernel(&insn, &buf[i]))
 			return;
 
 		next = i + insn.length;
 
-		__optimize_nops(instr, len, &insn, &next, &prev, &target);
+		__optimize_nops(instr, buf, len, &insn, &next, &prev, &target);
 	}
 }
 
-static void __init_or_module noinline optimize_nops_inplace(u8 *instr, size_t len)
-{
-	unsigned long flags;
-
-	local_irq_save(flags);
-	optimize_nops(instr, len);
-	sync_core();
-	local_irq_restore(flags);
-}
-
 /*
  * In this context, "source" is where the instructions are placed in the
  * section .altinstr_replacement, for example during kernel build by the
@@ -336,7 +340,7 @@ bool need_reloc(unsigned long offset, u8 *src, size_t src_len)
 }
 
 static void __init_or_module noinline
-apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
+apply_relocation(const u8 * const instr, u8 *buf, size_t len, u8 *src, size_t src_len)
 {
 	int prev, target = 0;
 
@@ -348,7 +352,7 @@ apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
 
 		next = i + insn.length;
 
-		if (__optimize_nops(buf, len, &insn, &next, &prev, &target))
+		if (__optimize_nops(instr, buf, len, &insn, &next, &prev, &target))
 			continue;
 
 		switch (insn.opcode.bytes[0]) {
@@ -365,7 +369,7 @@ apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
 			if (need_reloc(next + insn.immediate.value, src, src_len)) {
 				apply_reloc(insn.immediate.nbytes,
 					    buf + i + insn_offset_immediate(&insn),
-					    src - dest);
+					    src - instr);
 			}
 
 			/*
@@ -373,7 +377,7 @@ apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
 			 */
 			if (insn.opcode.bytes[0] == JMP32_INSN_OPCODE) {
 				s32 imm = insn.immediate.value;
-				imm += src - dest;
+				imm += src - instr;
 				imm += JMP32_INSN_SIZE - JMP8_INSN_SIZE;
 				if ((imm >> 31) == (imm >> 7)) {
 					buf[i+0] = JMP8_INSN_OPCODE;
@@ -389,7 +393,7 @@ apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
 			if (need_reloc(next + insn.displacement.value, src, src_len)) {
 				apply_reloc(insn.displacement.nbytes,
 					    buf + i + insn_offset_displacement(&insn),
-					    src - dest);
+					    src - instr);
 			}
 		}
 	}
@@ -505,7 +509,9 @@ void __init_or_module noinline apply_alternatives(struct alt_instr *start,
 		 *   patch if feature is *NOT* present.
 		 */
 		if (!boot_cpu_has(a->cpuid) == !(a->flags & ALT_FLAG_NOT)) {
-			optimize_nops_inplace(instr, a->instrlen);
+			memcpy(insn_buff, instr, a->instrlen);
+			optimize_nops(instr, insn_buff, a->instrlen);
+			text_poke_early(instr, insn_buff, a->instrlen);
 			continue;
 		}
 
@@ -527,7 +533,7 @@ void __init_or_module noinline apply_alternatives(struct alt_instr *start,
 		for (; insn_buff_sz < a->instrlen; insn_buff_sz++)
 			insn_buff[insn_buff_sz] = 0x90;
 
-		apply_relocation(insn_buff, a->instrlen, instr, replacement, a->replacementlen);
+		apply_relocation(instr, insn_buff, a->instrlen, replacement, a->replacementlen);
 
 		DUMP_BYTES(ALT, instr, a->instrlen, "%px:   old_insn: ", instr);
 		DUMP_BYTES(ALT, replacement, a->replacementlen, "%px:   rpl_insn: ", replacement);
@@ -762,7 +768,7 @@ void __init_or_module noinline apply_retpolines(s32 *start, s32 *end)
 
 		len = patch_retpoline(addr, &insn, bytes);
 		if (len == insn.length) {
-			optimize_nops(bytes, len);
+			optimize_nops(addr, bytes, len);
 			DUMP_BYTES(RETPOLINE, ((u8*)addr),  len, "%px: orig: ", addr);
 			DUMP_BYTES(RETPOLINE, ((u8*)bytes), len, "%px: repl: ", addr);
 			text_poke_early(addr, bytes, len);