lib/crypto/x86/aes-aesni.S
Source file repositories/reference/linux-study-clean/lib/crypto/x86/aes-aesni.S
File Facts
- System
- Linux kernel
- Corpus path
lib/crypto/x86/aes-aesni.S- Extension
.S- Size
- 7126 bytes
- Lines
- 262
- Domain
- Kernel Services
- Bucket
- lib
- Inferred role
- Kernel Services: lib
- Status
- atlas-only
Why This File Exists
Shared kernel service surface used by multiple subsystems, including helpers, cryptography, virtualization support, and async I/O infrastructure.
- Shared kernel service surface used by multiple subsystems, including helpers, cryptography, virtualization support, and async I/O infrastructure.
Dependency Surface
linux/linkage.h
Detected Declarations
- No top-level syscall, struct, function, initcall, or export declaration detected by the generator.
Annotated Snippet
//
// AES block cipher using AES-NI instructions
//
// Copyright 2026 Google LLC
//
// The code in this file supports 32-bit and 64-bit CPUs, and it doesn't require
// AVX. It does use up to SSE4.1, which all CPUs with AES-NI have.
#include <linux/linkage.h>
.section .rodata
#ifdef __x86_64__
#define RODATA(label) label(%rip)
#else
#define RODATA(label) label
#endif
// A mask for pshufb that extracts the last dword, rotates it right by 8
// bits, and copies the result to all four dwords.
.p2align 4
.Lmask:
.byte 13, 14, 15, 12, 13, 14, 15, 12, 13, 14, 15, 12, 13, 14, 15, 12
// The AES round constants, used during key expansion
.Lrcon:
.long 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36
.text
// Transform four dwords [a0, a1, a2, a3] in \a into
// [a0, a0^a1, a0^a1^a2, a0^a1^a2^a3]. \tmp is a temporary xmm register.
//
// Note: this could be done in four instructions, shufps + pxor + shufps + pxor,
// if the temporary register were zero-initialized ahead of time. We instead do
// it in an easier-to-understand way that doesn't require zero-initialization
// and avoids the unusual shufps instruction. movdqa is usually "free" anyway.
.macro _prefix_sum a, tmp
movdqa \a, \tmp // [a0, a1, a2, a3]
pslldq $4, \a // [0, a0, a1, a2]
pxor \tmp, \a // [a0, a0^a1, a1^a2, a2^a3]
movdqa \a, \tmp
pslldq $8, \a // [0, 0, a0, a0^a1]
pxor \tmp, \a // [a0, a0^a1, a0^a1^a2, a0^a1^a2^a3]
.endm
.macro _gen_round_key a, b
// Compute four copies of rcon[i] ^ SubBytes(ror32(w, 8)), where w is
// the last dword of the previous round key (given in \b).
//
// 'aesenclast src, dst' does dst = src XOR SubBytes(ShiftRows(dst)).
// It is used here solely for the SubBytes and the XOR. The ShiftRows
// is a no-op because all four columns are the same here.
//
// Don't use the 'aeskeygenassist' instruction, since:
// - On most Intel CPUs it is microcoded, making it have a much higher
// latency and use more execution ports than 'aesenclast'.
// - It cannot be used in a loop, since it requires an immediate.
// - It doesn't do much more than 'aesenclast' in the first place.
movdqa \b, %xmm2
pshufb MASK, %xmm2
aesenclast RCON, %xmm2
// XOR in the prefix sum of the four dwords of \a, which is the
// previous round key (AES-128) or the first round key in the previous
// pair of round keys (AES-256). The result is the next round key.
_prefix_sum \a, tmp=%xmm3
pxor %xmm2, \a
// Store the next round key to memory. Also leave it in \a.
movdqu \a, (RNDKEYS)
.endm
Annotation
- Immediate include surface: `linux/linkage.h`.
- Atlas domain: Kernel Services / lib.
- Implementation status: atlas-only.
Implementation Notes
- This generated page is the file-by-file coverage layer; curated subsystem chapters should link here when they synthesize a multi-file control flow.
- Core OS pages should be promoted from atlas-only to deep-reviewed when they explain data structures, invariants, locking, lifecycle, and C implementation snippets.
- Driver-family pages are intentionally pattern-oriented unless they are part of the selected PCIe/NVMe representative device path.