set_bits(loc32, 10, value, shift, 12);
}
-// Relaxable 12-bit low part of an absolute address. Pairs nicely with
-// patch_aarch64_21rx (below).
-void
-patch_aarch64_12x(unsigned char *location, uint64_t value)
-{
- // This can *only* be relaxed if it occurs immediately before a matching
- // patch_aarch64_21rx. If that happens, the JIT build step will replace both
- // calls with a single call to patch_aarch64_33rx. Otherwise, we end up
- // here, and the instruction is patched normally:
- patch_aarch64_12(location, value);
-}
-
// 16-bit low part of an absolute address.
void
patch_aarch64_16a(unsigned char *location, uint64_t value)
set_bits(loc32, 5, value, 2, 19);
}
-// Relaxable 21-bit count of pages between this page and an absolute address's
-// page. Pairs nicely with patch_aarch64_12x (above).
-void
-patch_aarch64_21rx(unsigned char *location, uint64_t value)
-{
- // This can *only* be relaxed if it occurs immediately before a matching
- // patch_aarch64_12x. If that happens, the JIT build step will replace both
- // calls with a single call to patch_aarch64_33rx. Otherwise, we end up
- // here, and the instruction is patched normally:
- patch_aarch64_21r(location, value);
-}
-
// 21-bit relative branch.
void
patch_aarch64_19r(unsigned char *location, uint64_t value)
set_bits(loc32, 0, value, 2, 26);
}
-// A pair of patch_aarch64_21rx and patch_aarch64_12x.
-void
-patch_aarch64_33rx(unsigned char *location, uint64_t value)
-{
- uint32_t *loc32 = (uint32_t *)location;
- // Try to relax the pair of GOT loads into an immediate value:
- assert(IS_AARCH64_ADRP(*loc32));
- unsigned char reg = get_bits(loc32[0], 0, 5);
- assert(IS_AARCH64_LDR_OR_STR(loc32[1]));
- // There should be only one register involved:
- assert(reg == get_bits(loc32[1], 0, 5)); // ldr's output register.
- assert(reg == get_bits(loc32[1], 5, 5)); // ldr's input register.
- uint64_t relaxed = *(uint64_t *)value;
- if (relaxed < (1UL << 16)) {
- // adrp reg, AAA; ldr reg, [reg + BBB] -> movz reg, XXX; nop
- loc32[0] = 0xD2800000 | (get_bits(relaxed, 0, 16) << 5) | reg;
- loc32[1] = 0xD503201F;
- return;
- }
- if (relaxed < (1ULL << 32)) {
- // adrp reg, AAA; ldr reg, [reg + BBB] -> movz reg, XXX; movk reg, YYY
- loc32[0] = 0xD2800000 | (get_bits(relaxed, 0, 16) << 5) | reg;
- loc32[1] = 0xF2A00000 | (get_bits(relaxed, 16, 16) << 5) | reg;
- return;
- }
- int64_t page_delta = (relaxed >> 12) - ((uintptr_t)location >> 12);
- if (page_delta >= -(1L << 20) &&
- page_delta < (1L << 20))
- {
- // adrp reg, AAA; ldr reg, [reg + BBB] -> adrp reg, AAA; add reg, reg, BBB
- patch_aarch64_21rx(location, relaxed);
- loc32[1] = 0x91000000 | get_bits(relaxed, 0, 12) << 10 | reg << 5 | reg;
- return;
- }
- relaxed = value - (uintptr_t)location;
- if ((relaxed & 0x3) == 0 &&
- (int64_t)relaxed >= -(1L << 19) &&
- (int64_t)relaxed < (1L << 19))
- {
- // adrp reg, AAA; ldr reg, [reg + BBB] -> ldr reg, XXX; nop
- loc32[0] = 0x58000000 | (get_bits(relaxed, 2, 19) << 5) | reg;
- loc32[1] = 0xD503201F;
- return;
- }
- // Couldn't do it. Just patch the two instructions normally:
- patch_aarch64_21rx(location, value);
- patch_aarch64_12x(location + 4, value);
-}
-
// Relaxable 32-bit relative address.
void
patch_x86_64_32rx(unsigned char *location, uint64_t value)
_PATCH_FUNCS = {
# aarch64-apple-darwin:
"ARM64_RELOC_BRANCH26": "patch_aarch64_26r",
- "ARM64_RELOC_GOT_LOAD_PAGE21": "patch_aarch64_21rx",
- "ARM64_RELOC_GOT_LOAD_PAGEOFF12": "patch_aarch64_12x",
+ "ARM64_RELOC_GOT_LOAD_PAGE21": "patch_aarch64_21r",
+ "ARM64_RELOC_GOT_LOAD_PAGEOFF12": "patch_aarch64_12",
"ARM64_RELOC_PAGE21": "patch_aarch64_21r",
"ARM64_RELOC_PAGEOFF12": "patch_aarch64_12",
"ARM64_RELOC_UNSIGNED": "patch_64",
# aarch64-pc-windows-msvc:
"IMAGE_REL_ARM64_BRANCH19": "patch_aarch64_19r",
"IMAGE_REL_ARM64_BRANCH26": "patch_aarch64_26r",
- "IMAGE_REL_ARM64_PAGEBASE_REL21": "patch_aarch64_21rx",
+ "IMAGE_REL_ARM64_PAGEBASE_REL21": "patch_aarch64_21r",
"IMAGE_REL_ARM64_PAGEOFFSET_12A": "patch_aarch64_12",
- "IMAGE_REL_ARM64_PAGEOFFSET_12L": "patch_aarch64_12x",
+ "IMAGE_REL_ARM64_PAGEOFFSET_12L": "patch_aarch64_12",
# i686-pc-windows-msvc:
"IMAGE_REL_I386_DIR32": "patch_32",
"IMAGE_REL_I386_REL32": "patch_x86_64_32rx",
# aarch64-unknown-linux-gnu:
"R_AARCH64_ABS64": "patch_64",
"R_AARCH64_ADD_ABS_LO12_NC": "patch_aarch64_12",
- "R_AARCH64_ADR_GOT_PAGE": "patch_aarch64_21rx",
+ "R_AARCH64_ADR_GOT_PAGE": "patch_aarch64_21r",
"R_AARCH64_ADR_PREL_PG_HI21": "patch_aarch64_21r",
"R_AARCH64_CALL26": "patch_aarch64_26r",
"R_AARCH64_CONDBR19": "patch_aarch64_19r",
"R_AARCH64_JUMP26": "patch_aarch64_26r",
- "R_AARCH64_LD64_GOT_LO12_NC": "patch_aarch64_12x",
+ "R_AARCH64_LD64_GOT_LO12_NC": "patch_aarch64_12",
"R_AARCH64_MOVW_UABS_G0_NC": "patch_aarch64_16a",
"R_AARCH64_MOVW_UABS_G1_NC": "patch_aarch64_16b",
"R_AARCH64_MOVW_UABS_G2_NC": "patch_aarch64_16c",
def __post_init__(self) -> None:
self.func = _PATCH_FUNCS[self.kind]
- def fold(self, other: typing.Self, body: bytearray) -> typing.Self | None:
- """Combine two holes into a single hole, if possible."""
- instruction_a = int.from_bytes(
- body[self.offset : self.offset + 4], byteorder=sys.byteorder
- )
- instruction_b = int.from_bytes(
- body[other.offset : other.offset + 4], byteorder=sys.byteorder
- )
- reg_a = instruction_a & 0b11111
- reg_b1 = instruction_b & 0b11111
- reg_b2 = (instruction_b >> 5) & 0b11111
-
- if (
- self.offset + 4 == other.offset
- and self.value == other.value
- and self.symbol == other.symbol
- and self.addend == other.addend
- and self.func == "patch_aarch64_21rx"
- and other.func == "patch_aarch64_12x"
- and reg_a == reg_b1 == reg_b2
- ):
- # These can *only* be properly relaxed when they appear together and
- # patch the same value:
- folded = self.replace()
- folded.func = "patch_aarch64_33rx"
- return folded
- return None
-
def as_c(self, where: str) -> str:
"""Dump this hole as a call to a patch_* function."""
if self.custom_location:
projects / thirdparty / Python / cpython.git / commitdiff
