From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- lib/crypto/powerpc/aes.h | 240 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 240 insertions(+) create mode 100644 lib/crypto/powerpc/aes.h (limited to 'lib/crypto/powerpc/aes.h') diff --git a/lib/crypto/powerpc/aes.h b/lib/crypto/powerpc/aes.h new file mode 100644 index 000000000..5a36b637e --- /dev/null +++ b/lib/crypto/powerpc/aes.h @@ -0,0 +1,240 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2015 Markus Stockhausen + * Copyright (C) 2015 International Business Machines Inc. + * Copyright 2026 Google LLC + */ +#include +#include +#include +#include +#include +#include + +#ifdef CONFIG_SPE + +EXPORT_SYMBOL_GPL(ppc_expand_key_128); +EXPORT_SYMBOL_GPL(ppc_expand_key_192); +EXPORT_SYMBOL_GPL(ppc_expand_key_256); +EXPORT_SYMBOL_GPL(ppc_generate_decrypt_key); +EXPORT_SYMBOL_GPL(ppc_encrypt_ecb); +EXPORT_SYMBOL_GPL(ppc_decrypt_ecb); +EXPORT_SYMBOL_GPL(ppc_encrypt_cbc); +EXPORT_SYMBOL_GPL(ppc_decrypt_cbc); +EXPORT_SYMBOL_GPL(ppc_crypt_ctr); +EXPORT_SYMBOL_GPL(ppc_encrypt_xts); +EXPORT_SYMBOL_GPL(ppc_decrypt_xts); + +void ppc_encrypt_aes(u8 *out, const u8 *in, const u32 *key_enc, u32 rounds); +void ppc_decrypt_aes(u8 *out, const u8 *in, const u32 *key_dec, u32 rounds); + +static void spe_begin(void) +{ + /* disable preemption and save users SPE registers if required */ + preempt_disable(); + enable_kernel_spe(); +} + +static void spe_end(void) +{ + disable_kernel_spe(); + /* reenable preemption */ + preempt_enable(); +} + +static void aes_preparekey_arch(union aes_enckey_arch *k, + union aes_invkey_arch *inv_k, + const u8 *in_key, int key_len, int nrounds) +{ + if (key_len == AES_KEYSIZE_128) + ppc_expand_key_128(k->spe_enc_key, in_key); + else if (key_len == AES_KEYSIZE_192) + ppc_expand_key_192(k->spe_enc_key, in_key); + else + ppc_expand_key_256(k->spe_enc_key, in_key); + + if (inv_k) + ppc_generate_decrypt_key(inv_k->spe_dec_key, k->spe_enc_key, + key_len); +} + +static void aes_encrypt_arch(const struct aes_enckey *key, + u8 out[AES_BLOCK_SIZE], + const u8 in[AES_BLOCK_SIZE]) +{ + spe_begin(); + ppc_encrypt_aes(out, in, key->k.spe_enc_key, key->nrounds / 2 - 1); + spe_end(); +} + +static void aes_decrypt_arch(const struct aes_key *key, + u8 out[AES_BLOCK_SIZE], + const u8 in[AES_BLOCK_SIZE]) +{ + spe_begin(); + ppc_decrypt_aes(out, in, key->inv_k.spe_dec_key, key->nrounds / 2 - 1); + spe_end(); +} + +#else /* CONFIG_SPE */ + +static __ro_after_init DEFINE_STATIC_KEY_FALSE(have_vec_crypto); + +EXPORT_SYMBOL_GPL(aes_p8_set_encrypt_key); +EXPORT_SYMBOL_GPL(aes_p8_set_decrypt_key); +EXPORT_SYMBOL_GPL(aes_p8_encrypt); +EXPORT_SYMBOL_GPL(aes_p8_decrypt); +EXPORT_SYMBOL_GPL(aes_p8_cbc_encrypt); +EXPORT_SYMBOL_GPL(aes_p8_ctr32_encrypt_blocks); +EXPORT_SYMBOL_GPL(aes_p8_xts_encrypt); +EXPORT_SYMBOL_GPL(aes_p8_xts_decrypt); + +static inline bool is_vsx_format(const struct p8_aes_key *key) +{ + return key->nrounds != 0; +} + +/* + * Convert a round key from VSX to generic format by reflecting all 16 bytes (if + * little endian) or reflecting the bytes in each 4-byte word (if big endian), + * and (if apply_inv_mix=true) applying InvMixColumn to each column. + * + * It would be nice if the VSX and generic key formats would be compatible. But + * that's very difficult to do, with the assembly code having been borrowed from + * OpenSSL and also targeted to POWER8 rather than POWER9. + * + * Fortunately, this conversion should only be needed in extremely rare cases, + * possibly not at all in practice. It's just included for full correctness. + */ +static void rndkey_from_vsx(u32 out[4], const u32 in[4], bool apply_inv_mix) +{ + const bool be = IS_ENABLED(CONFIG_CPU_BIG_ENDIAN); + u32 k0 = swab32(in[0]); + u32 k1 = swab32(in[1]); + u32 k2 = swab32(in[2]); + u32 k3 = swab32(in[3]); + + if (apply_inv_mix) { + k0 = inv_mix_columns(k0); + k1 = inv_mix_columns(k1); + k2 = inv_mix_columns(k2); + k3 = inv_mix_columns(k3); + } + out[0] = be ? k0 : k3; + out[1] = be ? k1 : k2; + out[2] = be ? k2 : k1; + out[3] = be ? k3 : k0; +} + +static void aes_preparekey_arch(union aes_enckey_arch *k, + union aes_invkey_arch *inv_k, + const u8 *in_key, int key_len, int nrounds) +{ + const int keybits = 8 * key_len; + int ret; + + if (static_branch_likely(&have_vec_crypto) && likely(may_use_simd())) { + preempt_disable(); + pagefault_disable(); + enable_kernel_vsx(); + ret = aes_p8_set_encrypt_key(in_key, keybits, &k->p8); + /* + * aes_p8_set_encrypt_key() should never fail here, since the + * key length was already validated. + */ + WARN_ON_ONCE(ret); + if (inv_k) { + ret = aes_p8_set_decrypt_key(in_key, keybits, + &inv_k->p8); + /* ... and likewise for aes_p8_set_decrypt_key(). */ + WARN_ON_ONCE(ret); + } + disable_kernel_vsx(); + pagefault_enable(); + preempt_enable(); + } else { + aes_expandkey_generic(k->rndkeys, + inv_k ? inv_k->inv_rndkeys : NULL, + in_key, key_len); + /* Mark the key as using the generic format. */ + k->p8.nrounds = 0; + if (inv_k) + inv_k->p8.nrounds = 0; + } +} + +static void aes_encrypt_arch(const struct aes_enckey *key, + u8 out[AES_BLOCK_SIZE], + const u8 in[AES_BLOCK_SIZE]) +{ + if (static_branch_likely(&have_vec_crypto) && + likely(is_vsx_format(&key->k.p8) && may_use_simd())) { + preempt_disable(); + pagefault_disable(); + enable_kernel_vsx(); + aes_p8_encrypt(in, out, &key->k.p8); + disable_kernel_vsx(); + pagefault_enable(); + preempt_enable(); + } else if (unlikely(is_vsx_format(&key->k.p8))) { + /* + * This handles (the hopefully extremely rare) case where a key + * was prepared using the VSX optimized format, then encryption + * is done in a context that cannot use VSX instructions. + */ + u32 rndkeys[AES_MAX_KEYLENGTH_U32]; + + for (int i = 0; i < 4 * (key->nrounds + 1); i += 4) + rndkey_from_vsx(&rndkeys[i], + &key->k.p8.rndkeys[i], false); + aes_encrypt_generic(rndkeys, key->nrounds, out, in); + } else { + aes_encrypt_generic(key->k.rndkeys, key->nrounds, out, in); + } +} + +static void aes_decrypt_arch(const struct aes_key *key, u8 out[AES_BLOCK_SIZE], + const u8 in[AES_BLOCK_SIZE]) +{ + if (static_branch_likely(&have_vec_crypto) && + likely(is_vsx_format(&key->inv_k.p8) && may_use_simd())) { + preempt_disable(); + pagefault_disable(); + enable_kernel_vsx(); + aes_p8_decrypt(in, out, &key->inv_k.p8); + disable_kernel_vsx(); + pagefault_enable(); + preempt_enable(); + } else if (unlikely(is_vsx_format(&key->inv_k.p8))) { + /* + * This handles (the hopefully extremely rare) case where a key + * was prepared using the VSX optimized format, then decryption + * is done in a context that cannot use VSX instructions. + */ + u32 inv_rndkeys[AES_MAX_KEYLENGTH_U32]; + int i; + + rndkey_from_vsx(&inv_rndkeys[0], + &key->inv_k.p8.rndkeys[0], false); + for (i = 4; i < 4 * key->nrounds; i += 4) { + rndkey_from_vsx(&inv_rndkeys[i], + &key->inv_k.p8.rndkeys[i], true); + } + rndkey_from_vsx(&inv_rndkeys[i], + &key->inv_k.p8.rndkeys[i], false); + aes_decrypt_generic(inv_rndkeys, key->nrounds, out, in); + } else { + aes_decrypt_generic(key->inv_k.inv_rndkeys, key->nrounds, + out, in); + } +} + +#define aes_mod_init_arch aes_mod_init_arch +static void aes_mod_init_arch(void) +{ + if (cpu_has_feature(CPU_FTR_ARCH_207S) && + (cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_VEC_CRYPTO)) + static_branch_enable(&have_vec_crypto); +} + +#endif /* !CONFIG_SPE */ -- cgit v1.3.1