aesni: Push FPU sections down further

After commit 937b447 aesni_cipher_crypt() and aesni_cipher_mac()
execute in a FPU_KERN_NOCTX section, which means that they must run with
preemption disabled.  These functions handle discontiguous I/O buffers
by allocating a contiguous buffer and copying as necessary, but this
allocation cannot happen with preemption disabled.  Fix the problem by
pushing the FPU section down into aesni_cipher_crypt() and
aesni_cipher_mac().  In particular, encrypt-then-auth transforms need
not be handled with a single FPU section.

Reported by:	syzbot+78258dbb02eb92157357@syzkaller.appspotmail.com
Discussed with:	jhb
Fixes:		937b447 ("aesni: Switch to using FPU_KERN_NOCTX.")

sys/crypto/aesni/aesni.c

@@ -594,8 +594,6 @@ aesni_cipher_process(struct aesni_session *ses, struct cryptop *crp)
 		break;
 	}
 
-	fpu_kern_enter(curthread, NULL, FPU_KERN_NORMAL | FPU_KERN_NOCTX);
-
 	/* Do work */
 	if (csp->csp_mode == CSP_MODE_ETA) {
 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
@@ -612,7 +610,6 @@ aesni_cipher_process(struct aesni_session *ses, struct cryptop *crp)
 	else
 		error = aesni_cipher_crypt(ses, crp, csp);
 
-	fpu_kern_leave(curthread, NULL);
 	return (error);
 }
 
@@ -677,6 +674,8 @@ aesni_cipher_crypt(struct aesni_session *ses, struct cryptop *crp,
 		outcopy = allocated;
 	}
 
+	fpu_kern_enter(curthread, NULL, FPU_KERN_NORMAL | FPU_KERN_NOCTX);
+
 	error = 0;
 	encflag = CRYPTO_OP_IS_ENCRYPT(crp->crp_op);
 	if (crp->crp_cipher_key != NULL)
@@ -749,6 +748,9 @@ aesni_cipher_crypt(struct aesni_session *ses, struct cryptop *crp,
 		}
 		break;
 	}
+
+	fpu_kern_leave(curthread, NULL);
+
 	if (outcopy && error == 0)
 		crypto_copyback(crp, CRYPTO_HAS_OUTPUT_BUFFER(crp) ?
 		    crp->crp_payload_output_start : crp->crp_payload_start,
@@ -784,6 +786,8 @@ aesni_cipher_mac(struct aesni_session *ses, struct cryptop *crp,
 		key = csp->csp_auth_key;
 	keylen = csp->csp_auth_klen;
 
+	fpu_kern_enter(curthread, NULL, FPU_KERN_NORMAL | FPU_KERN_NOCTX);
+
 	if (ses->hmac) {
 		uint8_t hmac_key[SHA1_BLOCK_LEN] __aligned(16);
 
@@ -849,6 +853,8 @@ aesni_cipher_mac(struct aesni_session *ses, struct cryptop *crp,
 		ses->hash_finalize(res, &sctx);
 	}
 
+	fpu_kern_leave(curthread, NULL);
+
 	if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
 		uint32_t res2[SHA2_256_HASH_LEN / sizeof(uint32_t)];