From: Peter Zijlstra <peterz@infradead.org>
To: x86@kernel.org
Cc: linux-kernel@vger.kernel.org, peterz@infradead.org,
mhiramat@kernel.org, kirill.shutemov@linux.intel.com,
Andrew.Cooper3@citrix.com, jpoimboe@redhat.com
Subject: [PATCH v3 4/4] x86/alternative: Complicate optimize_nops() some more
Date: Wed, 08 Feb 2023 18:10:54 +0100 [thread overview]
Message-ID: <20230208171431.433132442@infradead.org> (raw)
In-Reply-To: 20230208171050.490809180@infradead.org
Because:

SMP alternatives: ffffffff810026dc: [2:44) optimized NOPs: eb 2a eb 28 cc cc
cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc
cc cc cc cc cc cc cc cc cc cc cc cc cc
is quite daft, make things more complicated and have the NOP runlength
detection eat the preceding JMP if they both end at the same target.
SMP alternatives: ffffffff810026dc: [0:44) optimized NOPs: eb 2a cc cc cc cc
cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc
cc cc cc cc cc cc cc cc cc cc cc cc cc
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
arch/x86/kernel/alternative.c | 59 ++++++++++++++++++++++++++++++++++--------
1 file changed, 49 insertions(+), 10 deletions(-)
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -126,6 +126,7 @@ static void __init_or_module add_nops(vo
}
}
+/* Fill the buffer with a single effective instruction of size @len */
static void __init_or_module add_nop(u8 *instr, unsigned int len)
{
u8 *target = instr + len;
@@ -158,6 +159,9 @@ extern struct alt_instr __alt_instructio
extern s32 __smp_locks[], __smp_locks_end[];
void text_poke_early(void *addr, const void *opcode, size_t len);
+/*
+ * Matches NOP and NOPL, not any of the other possible NOPs.
+ */
static bool insn_is_nop(struct insn *insn)
{
if (insn->opcode.bytes[0] == 0x90)
@@ -171,6 +175,10 @@ static bool insn_is_nop(struct insn *ins
return false;
}
+/*
+ * 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)
{
struct insn insn;
@@ -187,11 +195,46 @@ static int skip_nops(u8 *instr, int offs
}
/*
+ * Optimize a sequence of NOPs, possibly preceded by a unconditional jump
+ * to the end of the NOP sequence into a single 'NOP'.
+ */
+static bool __optimize_nops(u8 *instr, size_t len, struct insn *insn,
+ int *next, int *prev, int *target)
+{
+ int i = *next - insn->length;
+
+ switch (insn->opcode.bytes[0]) {
+ case JMP8_INSN_OPCODE:
+ case JMP32_INSN_OPCODE:
+ *prev = i;
+ *target = *next + insn->immediate.value;
+ return false;
+ }
+
+ if (insn_is_nop(insn)) {
+ int nop = i;
+
+ *next = skip_nops(instr, *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);
+ return true;
+ }
+
+ *target = 0;
+ return false;
+}
+
+/*
* "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)
{
+ int prev, target = 0;
+
for (int next, i = 0; i < len; i = next) {
struct insn insn;
@@ -200,11 +243,7 @@ static void __init_or_module noinline op
next = i + insn.length;
- if (insn_is_nop(&insn)) {
- next = skip_nops(instr, next, len);
- add_nop(instr + i, next - i);
- DUMP_BYTES(ALT, instr, len, "%px: [%d:%d) optimized NOPs: ", instr, i, next);
- }
+ __optimize_nops(instr, len, &insn, &next, &prev, &target);
}
}
@@ -275,6 +314,8 @@ bool need_reloc(unsigned long offset, u8
static void __init_or_module noinline
apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
{
+ int prev, target = 0;
+
for (int next, i = 0; i < len; i = next) {
struct insn insn;
@@ -283,6 +324,9 @@ apply_relocation(u8 *buf, size_t len, u8
next = i + insn.length;
+ if (__optimize_nops(buf, len, &insn, &next, &prev, &target))
+ continue;
+
switch (insn.opcode.bytes[0]) {
case 0x0f:
if (insn.opcode.bytes[1] < 0x80 ||
@@ -324,11 +368,6 @@ apply_relocation(u8 *buf, size_t len, u8
src - dest);
}
}
-
- if (insn_is_nop(&insn)) {
- next = skip_nops(buf, next, len);
- add_nop(buf + i, next - i);
- }
}
}
next prev parent reply other threads:[~2023-05-11 12:58 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-02-08 17:10 [PATCH v3 0/4] x86: Fully relocatable alternatives and some NOPs Peter Zijlstra
2023-02-08 17:10 ` [PATCH v3 1/4] x86/alternative: Make debug-alternative selective Peter Zijlstra
2023-02-14 11:48 ` Borislav Petkov
2023-05-13 13:03 ` [tip: x86/alternatives] " tip-bot2 for Peter Zijlstra
2023-02-08 17:10 ` [PATCH v3 2/4] x86/alternative: Support relocations in alternatives Peter Zijlstra
2023-02-17 20:28 ` Borislav Petkov
2023-02-17 22:21 ` Borislav Petkov
2023-05-13 13:03 ` [tip: x86/alternatives] " tip-bot2 for Peter Zijlstra
2023-02-08 17:10 ` [PATCH v3 3/4] x86/alternative: Rewrite optimize_nops() some Peter Zijlstra
2023-02-08 19:52 ` Andrew.Cooper3
2023-02-08 20:29 ` Peter Zijlstra
2023-02-08 20:36 ` Peter Zijlstra
2023-02-08 20:44 ` Peter Zijlstra
2023-02-08 20:45 ` Peter Zijlstra
2023-02-08 21:01 ` Peter Zijlstra
2023-02-08 21:08 ` Peter Zijlstra
2023-02-08 21:21 ` Peter Zijlstra
2023-02-09 1:11 ` Andrew.Cooper3
2023-02-09 22:27 ` David Laight
2023-02-09 1:33 ` Andrew.Cooper3
2023-02-08 23:04 ` David Laight
2023-05-13 13:03 ` [tip: x86/alternatives] " tip-bot2 for Peter Zijlstra
2023-02-08 17:10 ` Peter Zijlstra [this message]
2023-05-13 13:03 ` [tip: x86/alternatives] x86/alternative: Complicate optimize_nops() some more tip-bot2 for Peter Zijlstra
2023-05-13 16:01 ` [PATCH] x86/alternatives: Fix section mismatch warnings Borislav Petkov
2023-05-13 16:10 ` [tip: x86/alternatives] " tip-bot2 for Borislav Petkov (AMD)
2023-02-27 10:49 ` [PATCH] x86/lib/memmove: Decouple ERMS from FSRM Borislav Petkov
2023-04-27 9:22 ` [PATCH TEST] " Yahu Gao
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230208171431.433132442@infradead.org \
--to=peterz@infradead.org \
--cc=Andrew.Cooper3@citrix.com \
--cc=jpoimboe@redhat.com \
--cc=kirill.shutemov@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mhiramat@kernel.org \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.