On Sat, Mar 14, 2026 at 04:09:37PM +0800, Zorro Lang wrote: > [ 859.430320] KASAN: probably user-memory-access in range > [0x00000004e3534580-0x00000004e3534587] > [ 859.439808] crypto_transfer_skcipher_request_to_engine+0x1c/0x38 > [crypto_engine] > [ 859.439816] tegra_aes_crypt+0x17c/0x308 [tegra_se] > [ 859.443046] Mem abort info: > [ 859.450307] tegra_aes_encrypt+0x1c/0x28 [tegra_se] > [ 859.450316] crypto_skcipher_encrypt+0xe0/0x158 > [ 859.457490] ESR = 0x000000008600000f > [ 859.464835] fscrypt_crypt_data_unit+0x21c/0x2d0 > [ 859.472193] EC = 0x21: IABT (current EL), IL = 32 bits > [ 859.479359] fscrypt_encrypt_pagecache_blocks+0x1e8/0x350 > [ 859.479366] ext4_bio_write_folio+0x8e4/0x1128 [ext4]
Thanks. fscrypt isn't supposed to be using the tegra crypto driver. But that driver is actually so broken that it doesn't even set its flags correctly, causing the crypto_skcipher API to select it anyway. https://lore.kernel.org/linux-fscrypt/[email protected]/ fixes this by making the driver set the correct flags. I'm looking forward to lib/crypto/ offering all the algorithms that fscrypt needs so that we don't have to use crypto_skcipher. - Eric
