Commit b6c881b2 authored by Peter Zijlstra's avatar Peter Zijlstra Committed by Borislav Petkov (AMD)

x86/alternative: Complicate optimize_nops() some more

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: default avatarPeter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: default avatarBorislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230208171431.433132442@infradead.org
parent 6c480f22
......@@ -114,6 +114,8 @@ const unsigned char * const x86_nops[ASM_NOP_MAX+1] =
};
/*
* 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)
* for every single-byte NOP, try to generate the maximally available NOP of
* size <= ASM_NOP_MAX such that only a single CFI entry is generated (vs one for
......@@ -152,6 +154,9 @@ extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
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)
......@@ -165,6 +170,10 @@ static bool insn_is_nop(struct insn *insn)
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;
......@@ -180,12 +189,47 @@ static int skip_nops(u8 *instr, int offset, int len)
return offset;
}
/*
* Optimize a sequence of NOPs, possibly preceded by an 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;
......@@ -194,11 +238,7 @@ static void __init_or_module noinline optimize_nops(u8 *instr, size_t len)
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 +315,8 @@ 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)
{
int prev, target = 0;
for (int next, i = 0; i < len; i = next) {
struct insn insn;
......@@ -283,6 +325,9 @@ 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))
continue;
switch (insn.opcode.bytes[0]) {
case 0x0f:
if (insn.opcode.bytes[1] < 0x80 ||
......@@ -324,11 +369,6 @@ apply_relocation(u8 *buf, size_t len, u8 *dest, u8 *src, size_t src_len)
src - dest);
}
}
if (insn_is_nop(&insn)) {
next = skip_nops(buf, next, len);
add_nop(buf + i, next - i);
}
}
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment