Revert "crypto: tegra: Use separate buffer for each host1x command"

This reverts commit f75887b0c0e9340c3dd92b19dab99ed92f36bcb0.

Reason: To align with the upstream changes pushed as part of
https://lore.kernel.org/lkml/20250224091610.49683-1-akhilrajeev@nvidia.com/

Bug 4883011

Change-Id: If79728d00d3e8e3c8ad3e26a8170a7c1fa95414c
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3347660
Reviewed-by: Akhil R <akhilrajeev@nvidia.com>
Reviewed-by: Laxman Dewangan <ldewangan@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: Akhil R <akhilrajeev@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>

drivers/crypto/tegra/tegra-se-aes.c

@@ -44,7 +44,6 @@ struct tegra_aes_reqctx {
 	u32 *iv;
 	u32 key1_id;
 	u32 key2_id;
-	u32 *cmdbuf;
 };
 
 struct tegra_aead_ctx {
@@ -75,7 +74,6 @@ struct tegra_aead_reqctx {
 	u32 key_id;
 	u32 iv[4];
 	u8 authdata[16];
-	u32 *cmdbuf;
 };
 
 struct tegra_cmac_ctx {
@@ -102,7 +100,6 @@ struct tegra_cmac_reqctx {
 	u32 key_id;
 	u32 *iv;
 	u32 result[CMAC_RESULT_REG_COUNT];
-	u32 *cmdbuf;
 };
 
 /* increment counter (128-bit int) */
@@ -228,7 +225,7 @@ static int tegra234_aes_cfg(u32 alg, bool encrypt)
 static unsigned int tegra_aes_prep_cmd(struct tegra_se *se, struct tegra_aes_reqctx *rctx)
 {
 	unsigned int data_count, res_bits, i = 0, j;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 	dma_addr_t addr = rctx->datbuf.addr;
 
 	data_count = rctx->len / AES_BLOCK_SIZE;
@@ -332,14 +329,14 @@ static int tegra_aes_do_one_req(struct crypto_engine *engine, void *areq)
 
 	/* Prepare the command and submit for execution */
 	cmdlen = tegra_aes_prep_cmd(se, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	ret = tegra_se_host1x_submit(se, cmdlen);
 
 	/* Copy the result */
 	tegra_aes_update_iv(req, ctx);
 	scatterwalk_map_and_copy(rctx->datbuf.buf, req->dst, 0, req->cryptlen, 1);
 
 	/* Free the buffer */
-	dma_free_coherent(se->dev, rctx->datbuf.size,
+	dma_free_coherent(ctx->se->dev, rctx->datbuf.size,
 			  rctx->datbuf.buf, rctx->datbuf.addr);
 
 key2_free:
@@ -349,8 +346,6 @@ key1_free:
 	if (rctx->key1_id != ctx->key1_id)
 		tegra_key_invalidate(ctx->se, rctx->key1_id, ctx->alg);
 out:
-	kfree(rctx->cmdbuf);
-
 	crypto_finalize_skcipher_request(se->engine, req, ret);
 
 	return 0;
@@ -643,13 +638,6 @@ static int tegra_aes_crypt(struct skcipher_request *req, bool encrypt)
 	if (!req->cryptlen)
 		return 0;
 
-	if (ctx->alg == SE_ALG_ECB)
-		req->iv = NULL;
-
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		return -ENOMEM;
-
 	rctx->encrypt = encrypt;
 	rctx->config = se->regcfg->cfg(ctx->alg, encrypt);
 	rctx->crypto_config = se->regcfg->crypto_cfg(ctx->alg, encrypt);
@@ -803,7 +791,7 @@ static void tegra_aes_set_regcfg(struct tegra_se *se)
 static unsigned int tegra_gmac_prep_cmd(struct tegra_se *se, struct tegra_aead_reqctx *rctx)
 {
 	unsigned int data_count, res_bits, i = 0, j;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 
 	data_count = (rctx->assoclen / AES_BLOCK_SIZE);
 	res_bits = (rctx->assoclen % AES_BLOCK_SIZE) * 8;
@@ -849,7 +837,7 @@ static unsigned int tegra_gmac_prep_cmd(struct tegra_se *se, struct tegra_aead_r
 static unsigned int tegra_gcm_crypt_prep_cmd(struct tegra_se *se, struct tegra_aead_reqctx *rctx)
 {
 	unsigned int data_count, res_bits, i = 0, j;
-	u32 *cpuvaddr = rctx->cmdbuf, op;
+	u32 *cpuvaddr = se->cmdbuf->addr, op;
 
 	data_count = (rctx->cryptlen / AES_BLOCK_SIZE);
 	res_bits = (rctx->cryptlen % AES_BLOCK_SIZE) * 8;
@@ -971,7 +959,7 @@ static int tegra_gcm_do_gmac(struct tegra_aead_ctx *ctx, struct tegra_aead_reqct
 
 	cmdlen = tegra_gmac_prep_cmd(se, rctx);
 
-	return tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	return tegra_se_host1x_submit(se, cmdlen);
 }
 
 static int tegra_gcm_do_crypt(struct tegra_aead_ctx *ctx, struct tegra_aead_reqctx *rctx)
@@ -988,7 +976,7 @@ static int tegra_gcm_do_crypt(struct tegra_aead_ctx *ctx, struct tegra_aead_reqc
 
 	/* Prepare command and submit */
 	cmdlen = tegra_gcm_crypt_prep_cmd(se, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	ret = tegra_se_host1x_submit(se, cmdlen);
 	if (ret)
 		return ret;
 
@@ -1002,7 +990,7 @@ static int tegra_gcm_do_crypt(struct tegra_aead_ctx *ctx, struct tegra_aead_reqc
 static int tegra_gcm_do_final(struct tegra_aead_ctx *ctx, struct tegra_aead_reqctx *rctx)
 {
 	struct tegra_se *se = ctx->se;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 	int cmdlen, ret, offset;
 
 	rctx->config = se->regcfg->cfg(ctx->final_alg, rctx->encrypt);
@@ -1011,7 +999,7 @@ static int tegra_gcm_do_final(struct tegra_aead_ctx *ctx, struct tegra_aead_reqc
 
 	/* Prepare command and submit */
 	cmdlen = tegra_gcm_prep_final_cmd(se, cpuvaddr, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	ret = tegra_se_host1x_submit(se, cmdlen);
 	if (ret)
 		return ret;
 
@@ -1028,7 +1016,7 @@ static int tegra_gcm_do_final(struct tegra_aead_ctx *ctx, struct tegra_aead_reqc
 static int tegra_gcm_hw_verify(struct tegra_aead_ctx *ctx, struct tegra_aead_reqctx *rctx, u8 *mac)
 {
 	struct tegra_se *se = ctx->se;
-	u32 result, *cpuvaddr = rctx->cmdbuf;
+	u32 result, *cpuvaddr = se->cmdbuf->addr;
 	int size, ret;
 
 	memcpy(rctx->inbuf.buf, mac, rctx->authsize);
@@ -1040,7 +1028,7 @@ static int tegra_gcm_hw_verify(struct tegra_aead_ctx *ctx, struct tegra_aead_req
 
 	/* Prepare command and submit */
 	size = tegra_gcm_prep_final_cmd(se, cpuvaddr, rctx);
-	ret = tegra_se_host1x_submit(se, cpuvaddr, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		return ret;
 
@@ -1089,7 +1077,7 @@ static inline int tegra_ccm_check_iv(const u8 *iv)
 static unsigned int tegra_cbcmac_prep_cmd(struct tegra_se *se, struct tegra_aead_reqctx *rctx)
 {
 	unsigned int data_count, i = 0;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 
 	data_count = (rctx->inbuf.size / AES_BLOCK_SIZE) - 1;
 
@@ -1123,7 +1111,7 @@ static unsigned int tegra_cbcmac_prep_cmd(struct tegra_se *se, struct tegra_aead
 static unsigned int tegra_ctr_prep_cmd(struct tegra_se *se, struct tegra_aead_reqctx *rctx)
 {
 	unsigned int i = 0, j;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 
 	cpuvaddr[i++] = host1x_opcode_setpayload(SE_CRYPTO_CTR_REG_COUNT);
 	cpuvaddr[i++] = se_host1x_opcode_incr_w(se->hw->regs->linear_ctr);
@@ -1171,7 +1159,7 @@ static int tegra_ccm_do_cbcmac(struct tegra_se *se, struct tegra_aead_reqctx *rc
 	/* Prepare command and submit */
 	cmdlen = tegra_cbcmac_prep_cmd(se, rctx);
 
-	return tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	return tegra_se_host1x_submit(se, cmdlen);
 }
 
 static int tegra_ccm_set_msg_len(u8 *block, unsigned int msglen, int csize)
@@ -1376,7 +1364,7 @@ static int tegra_ccm_do_ctr(struct tegra_se *se, struct tegra_aead_reqctx *rctx)
 
 	/* Prepare command and submit */
 	cmdlen = tegra_ctr_prep_cmd(se, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	ret = tegra_se_host1x_submit(se, cmdlen);
 	if (ret)
 		return ret;
 
@@ -1487,9 +1475,7 @@ key_free:
 	/* Free the keyslot if it is cloned for this request */
 	if (rctx->key_id != ctx->key_id)
 		tegra_key_invalidate(ctx->se, rctx->key_id, ctx->alg);
-
 out:
-	kfree(rctx->cmdbuf);
 	crypto_finalize_aead_request(ctx->se->engine, req, ret);
 
 	return 0;
@@ -1575,7 +1561,6 @@ outbuf_err:
 			  rctx->inbuf.buf, rctx->inbuf.addr);
 
 	/* Finalize the request if there are no errors */
-	kfree(rctx->cmdbuf);
 	crypto_finalize_aead_request(ctx->se->engine, req, ret);
 
 	return 0;
@@ -1699,10 +1684,6 @@ static int tegra_aead_crypt(struct aead_request *req, bool encrypt)
 	struct tegra_aead_ctx *ctx = crypto_aead_ctx(tfm);
 	struct tegra_aead_reqctx *rctx = aead_request_ctx(req);
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		return -ENOMEM;
-
 	rctx->encrypt = encrypt;
 
 	return crypto_transfer_aead_request_to_engine(ctx->se->engine, req);
@@ -1734,7 +1715,7 @@ static int tegra_aead_setkey(struct crypto_aead *tfm,
 static unsigned int tegra_cmac_prep_cmd(struct tegra_se *se, struct tegra_cmac_reqctx *rctx)
 {
 	unsigned int data_count, res_bits = 0, i = 0, j;
-	u32 *cpuvaddr = rctx->cmdbuf, op;
+	u32 *cpuvaddr = se->cmdbuf->addr, op;
 
 	data_count = (rctx->datbuf.size / AES_BLOCK_SIZE);
 
@@ -1881,7 +1862,7 @@ static int tegra_cmac_do_update(struct ahash_request *req)
 
 	cmdlen = tegra_cmac_prep_cmd(se, rctx);
 
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	ret = tegra_se_host1x_submit(se, cmdlen);
 	/*
 	 * If this is not the final update, copy the intermediate results
 	 * from the registers so that it can be used in the next 'update'
@@ -1925,7 +1906,7 @@ static int tegra_cmac_do_final(struct ahash_request *req)
 
 	/* Prepare command and submit */
 	cmdlen = tegra_cmac_prep_cmd(se, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, cmdlen);
+	ret = tegra_se_host1x_submit(se, cmdlen);
 	if (ret)
 		goto out;
 
@@ -1946,7 +1927,6 @@ out_free:
 			  rctx->residue.buf, rctx->residue.addr);
 	dma_free_coherent(se->dev, rctx->digest.size, rctx->digest.buf,
 			rctx->digest.addr);
-	kfree(rctx->cmdbuf);
 
 	return ret;
 }
@@ -2061,19 +2041,15 @@ static int tegra_cmac_init(struct ahash_request *req)
 			return -ENOMEM;
 	}
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		goto key_free;
-
 	rctx->digest.buf = dma_alloc_coherent(se->dev, rctx->digest.size,
 				&rctx->digest.addr, GFP_KERNEL);
 	if (!rctx->digest.buf)
-		goto cmdbuf_free;
+		goto digbuf_fail;
 
 	rctx->residue.buf = dma_alloc_coherent(se->dev, rctx->blk_size * 2,
 					       &rctx->residue.addr, GFP_KERNEL);
 	if (!rctx->residue.buf)
-		goto digbuf_free;
+		goto resbuf_fail;
 
 	rctx->residue.size = 0;
 	rctx->datbuf.size = 0;
@@ -2084,12 +2060,10 @@ static int tegra_cmac_init(struct ahash_request *req)
 
 	return 0;
 
-digbuf_free:
+resbuf_fail:
 	dma_free_coherent(se->dev, rctx->blk_size, rctx->digest.buf,
 				rctx->digest.addr);
-cmdbuf_free:
+digbuf_fail:
-	kfree(rctx->cmdbuf);
-key_free:
 	if (rctx->key_id != ctx->key_id)
 		tegra_key_invalidate(ctx->se, rctx->key_id, ctx->alg);
 

drivers/crypto/tegra/tegra-se-hash.c

@@ -45,7 +45,6 @@ struct tegra_sha_reqctx {
 	unsigned int blk_size;
 	unsigned int task;
 	u32 key_id;
-	u32 *cmdbuf;
 	u32 result[HASH_RESULT_REG_COUNT];
 	struct ahash_request fallback_req;
 };
@@ -312,7 +311,7 @@ static int tegra_sha_do_update(struct ahash_request *req)
 	struct tegra_sha_ctx *ctx = crypto_ahash_ctx(crypto_ahash_reqtfm(req));
 	struct tegra_sha_reqctx *rctx = ahash_request_ctx(req);
 	unsigned int nblks, nresidue, size, ret;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = ctx->se->cmdbuf->addr;
 
 	nresidue = (req->nbytes + rctx->residue.size) % rctx->blk_size;
 	nblks = (req->nbytes + rctx->residue.size) / rctx->blk_size;
@@ -373,7 +372,7 @@ static int tegra_sha_do_update(struct ahash_request *req)
 
 	size = tegra_sha_prep_cmd(ctx->se, cpuvaddr, rctx);
 
-	ret = tegra_se_host1x_submit(ctx->se, rctx->cmdbuf, size);
+	ret = tegra_se_host1x_submit(ctx->se, size);
 
 	/*
 	 * If this is not the final update, copy the intermediate results
@@ -395,7 +394,7 @@ static int tegra_sha_do_final(struct ahash_request *req)
 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
 	struct tegra_sha_ctx *ctx = crypto_ahash_ctx(tfm);
 	struct tegra_se *se = ctx->se;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 	int size, ret = 0;
 
 	rctx->datbuf.size = rctx->residue.size;
@@ -416,7 +415,7 @@ static int tegra_sha_do_final(struct ahash_request *req)
 	}
 
 	size = tegra_sha_prep_cmd(se, cpuvaddr, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		goto out;
 
@@ -432,8 +431,6 @@ out_free:
 			  rctx->residue.buf, rctx->residue.addr);
 	dma_free_coherent(se->dev, rctx->digest.size, rctx->digest.buf,
 			  rctx->digest.addr);
-	kfree(rctx->cmdbuf);
-
 	return ret;
 }
 
@@ -568,28 +565,22 @@ static int tegra_sha_init(struct ahash_request *req)
 			return -ENOMEM;
 	}
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		goto key_free;
-
 	rctx->digest.buf = dma_alloc_coherent(se->dev, rctx->digest.size,
 					      &rctx->digest.addr, GFP_KERNEL);
 	if (!rctx->digest.buf)
-		goto cmdbuf_free;
+		goto digbuf_fail;
 
 	rctx->residue.buf = dma_alloc_coherent(se->dev, rctx->blk_size,
 					       &rctx->residue.addr, GFP_KERNEL);
 	if (!rctx->residue.buf)
-		goto digbuf_free;
+		goto resbuf_fail;
 
 	return 0;
 
-digbuf_free:
+resbuf_fail:
 	dma_free_coherent(se->dev, rctx->digest.size, rctx->digest.buf,
 			  rctx->digest.addr);
-cmdbuf_free:
+digbuf_fail:
-	kfree(rctx->cmdbuf);
-key_free:
 	if (rctx->key_id != ctx->key_id)
 		tegra_key_invalidate(ctx->se, rctx->key_id, ctx->alg);
 

drivers/crypto/tegra/tegra-se-key.c

@@ -177,22 +177,22 @@ static int tegra_key_insert(struct tegra_se *se, const u8 *key,
 			    u32 keylen, u16 slot, u32 alg)
 {
 	const u32 *keyval = (u32 *)key;
-	u32 cmdbuf[100], size;
+	u32 *addr = se->cmdbuf->addr, size;
 
-	size = tegra_key_prep_ins_cmd(se, cmdbuf, keyval, keylen, slot, alg);
+	size = tegra_key_prep_ins_cmd(se, addr, keyval, keylen, slot, alg);
 
-	return tegra_se_host1x_submit(se, cmdbuf, size);
+	return tegra_se_host1x_submit(se, size);
 }
 
 static int tegra_key_move_to_kds(struct tegra_se *se, u32 slot, u32 kds_id)
 {
-	u32 src_keyid, size, cmdbuf[100];
+	u32 src_keyid, size;
 	int ret;
 
 	src_keyid = SE_KSLT_REGION_ID_SYM | slot;
-	size = tegra_key_prep_mov_cmd(se, cmdbuf, src_keyid, kds_id);
+	size = tegra_key_prep_mov_cmd(se, se->cmdbuf->addr, src_keyid, kds_id);
 
-	ret = tegra_se_host1x_submit(se, cmdbuf, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		return ret;
 
@@ -201,13 +201,13 @@ static int tegra_key_move_to_kds(struct tegra_se *se, u32 slot, u32 kds_id)
 
 static unsigned int tegra_kac_get_from_kds(struct tegra_se *se, u32 keyid, u16 slot)
 {
-	u32 tgt_keyid, size, cmdbuf[100];
+	u32 tgt_keyid, size;
 	int ret;
 
 	tgt_keyid = SE_KSLT_REGION_ID_SYM | slot;
-	size = tegra_key_prep_mov_cmd(se, cmdbuf, keyid, tgt_keyid);
+	size = tegra_key_prep_mov_cmd(se, se->cmdbuf->addr, keyid, tgt_keyid);
 
-	ret = tegra_se_host1x_submit(se, cmdbuf, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		tegra_keyslot_free(slot);
 
@@ -216,10 +216,10 @@ static unsigned int tegra_kac_get_from_kds(struct tegra_se *se, u32 keyid, u16 s
 
 static void tegra_key_kds_invalidate(struct tegra_se *se, u32 keyid)
 {
-	unsigned int size, cmdbuf[100];
+	unsigned int size;
 
-	size = tegra_key_prep_invld_cmd(se, cmdbuf, keyid);
+	size = tegra_key_prep_invld_cmd(se, se->cmdbuf->addr, keyid);
-	tegra_se_host1x_submit(se, cmdbuf, size);
+	tegra_se_host1x_submit(se, size);
 	tegra_kds_free_id(keyid);
 }
 

drivers/crypto/tegra/tegra-se-main.c

@@ -143,10 +143,9 @@ static struct tegra_se_cmdbuf *tegra_se_host1x_bo_alloc(struct tegra_se *se, ssi
 	return cmdbuf;
 }
 
-int tegra_se_host1x_submit(struct tegra_se *se, u32 *cpuvaddr, u32 size)
+int tegra_se_host1x_submit(struct tegra_se *se, u32 size)
 {
 	struct host1x_job *job;
-	struct tegra_se_cmdbuf *cmdbuf;
 	int ret;
 
 	job = host1x_job_alloc(se->channel, 1, 0, true);
@@ -155,12 +154,6 @@ int tegra_se_host1x_submit(struct tegra_se *se, u32 *cpuvaddr, u32 size)
 		return -ENOMEM;
 	}
 
-	cmdbuf = tegra_se_host1x_bo_alloc(se, SZ_4K);
-	if (!cmdbuf)
-		goto job_put;
-
-	memcpy(cmdbuf->addr, cpuvaddr, size * 4);
-
 	job->syncpt = host1x_syncpt_get(se->syncpt);
 	job->syncpt_incrs = 1;
 	job->client = &se->client;
@@ -169,14 +162,14 @@ int tegra_se_host1x_submit(struct tegra_se *se, u32 *cpuvaddr, u32 size)
 	job->engine_fallback_streamid = se->stream_id;
 	job->engine_streamid_offset = SE_STREAM_ID;
 
-	cmdbuf->words = size;
+	se->cmdbuf->words = size;
 
-	host1x_job_add_gather(job, &cmdbuf->bo, size, 0);
+	host1x_job_add_gather(job, &se->cmdbuf->bo, size, 0);
 
 	ret = host1x_job_pin(job, se->dev);
 	if (ret) {
 		dev_err(se->dev, "failed to pin host1x job\n");
-		goto cmdbuf_put;
+		goto job_put;
 	}
 
 	ret = host1x_job_submit(job);
@@ -193,14 +186,10 @@ int tegra_se_host1x_submit(struct tegra_se *se, u32 *cpuvaddr, u32 size)
 	}
 
 	host1x_job_put(job);
-	tegra_se_cmdbuf_put(&cmdbuf->bo);
-
 	return 0;
 
 job_unpin:
 	host1x_job_unpin(job);
-cmdbuf_put:
-	tegra_se_cmdbuf_put(&cmdbuf->bo);
 job_put:
 	host1x_job_put(job);
 
@@ -227,14 +216,22 @@ static int tegra_se_client_init(struct host1x_client *client)
 
 	se->syncpt_id =  host1x_syncpt_id(se->syncpt);
 
+	se->cmdbuf = tegra_se_host1x_bo_alloc(se, SZ_4K);
+	if (!se->cmdbuf) {
+		ret = -ENOMEM;
+		goto syncpt_put;
+	}
+
 	ret = se->hw->init_alg(se);
 	if (ret) {
 		dev_err(se->dev, "failed to register algorithms\n");
-		goto syncpt_put;
+		goto cmdbuf_put;
 	}
 
 	return 0;
 
+cmdbuf_put:
+	tegra_se_cmdbuf_put(&se->cmdbuf->bo);
 syncpt_put:
 	host1x_syncpt_put(se->syncpt);
 channel_put:
@@ -248,6 +245,7 @@ static int tegra_se_client_deinit(struct host1x_client *client)
 	struct tegra_se *se = container_of(client, struct tegra_se, client);
 
 	se->hw->deinit_alg(se);
+	tegra_se_cmdbuf_put(&se->cmdbuf->bo);
 	host1x_syncpt_put(se->syncpt);
 	host1x_channel_put(se->channel);
 

drivers/crypto/tegra/tegra-se-sm4.c

@@ -2,7 +2,7 @@
 /*
  * Crypto driver for NVIDIA Security Engine for block cipher operations.
  *
- * Copyright (c) 2023-2024, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2023, NVIDIA CORPORATION.  All rights reserved.
  */
 
 #include <nvidia/conftest.h>
@@ -45,7 +45,6 @@ struct tegra_sm4_reqctx {
 	u32 crypto_cfg;
 	u32 key1_id;
 	u32 key2_id;
-	u32 *cmdbuf;
 };
 
 struct tegra_sm4_gcm_ctx {
@@ -60,7 +59,6 @@ struct tegra_sm4_gcm_ctx {
 	u32 verify_alg;
 	u32 keylen;
 	u32 key_id;
-	u32 *cmdbuf;
 };
 
 struct tegra_sm4_gcm_reqctx {
@@ -75,7 +73,6 @@ struct tegra_sm4_gcm_reqctx {
 	u32 config;
 	u32 crypto_config;
 	u32 key_id;
-	u32 *cmdbuf;
 	u32 iv[4];
 	u8 authdata[16];
 };
@@ -103,7 +100,6 @@ struct tegra_sm4_cmac_reqctx {
 	u32 crypto_config;
 	u32 key_id;
 	u32 result[CMAC_RESULT_REG_COUNT];
-	u32 *cmdbuf;
 	u32 *iv;
 };
 
@@ -229,7 +225,7 @@ static int tegra_sm4_do_one_req(struct crypto_engine *engine, void *areq)
 		goto key2_free;
 	}
 
-	cpuvaddr = rctx->cmdbuf;
+	cpuvaddr = se->cmdbuf->addr;
 	len = req->cryptlen;
 
 	/* Pad input to AES block size */
@@ -249,7 +245,7 @@ static int tegra_sm4_do_one_req(struct crypto_engine *engine, void *areq)
 	size = tegra_sm4_prep_cmd(se, cpuvaddr, iv, len, rctx->datbuf.addr,
 					 config, crypto_config);
 
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, size);
+	ret = tegra_se_host1x_submit(se, size);
 
 	/* Copy the result */
 	dst_nents = sg_nents(req->dst);
@@ -267,7 +263,6 @@ key1_free:
 	if (rctx->key1_id != ctx->key1_id)
 		tegra_key_invalidate(ctx->se, rctx->key1_id, ctx->alg);
 out:
-	kfree(rctx->cmdbuf);
 	crypto_finalize_skcipher_request(se->engine, req, ret);
 
 	return ret;
@@ -506,10 +501,6 @@ static int tegra_sm4_encrypt(struct skcipher_request *req)
 	rctx = skcipher_request_ctx(req);
 	rctx->encrypt = true;
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		return -ENOMEM;
-
 	return crypto_transfer_skcipher_request_to_engine(ctx->se->engine, req);
 }
 
@@ -522,10 +513,6 @@ static int tegra_sm4_decrypt(struct skcipher_request *req)
 	rctx = skcipher_request_ctx(req);
 	rctx->encrypt = false;
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		return -ENOMEM;
-
 	return crypto_transfer_skcipher_request_to_engine(ctx->se->engine, req);
 }
 
@@ -842,7 +829,7 @@ static int tegra_sm4_gcm_prep_final_cmd(struct tegra_se *se, u32 *cpuvaddr,
 static int tegra_sm4_gcm_do_gmac(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm4_gcm_reqctx *rctx)
 {
 	struct tegra_se *se = ctx->se;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 	unsigned int nents, size;
 
 	nents = sg_nents(rctx->src_sg);
@@ -855,13 +842,13 @@ static int tegra_sm4_gcm_do_gmac(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm4
 
 	size = tegra_sm4_gmac_prep_cmd(se, cpuvaddr, rctx);
 
-	return tegra_se_host1x_submit(se, cpuvaddr, size);
+	return tegra_se_host1x_submit(se, size);
 }
 
 static int tegra_sm4_gcm_do_crypt(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm4_gcm_reqctx *rctx)
 {
 	struct tegra_se *se = ctx->se;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 	int size, ret;
 
 	scatterwalk_map_and_copy(rctx->inbuf.buf, rctx->src_sg,
@@ -873,7 +860,7 @@ static int tegra_sm4_gcm_do_crypt(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm
 
 	/* Prepare command and submit */
 	size = tegra_sm4_gcm_crypt_prep_cmd(se, cpuvaddr, rctx);
-	ret = tegra_se_host1x_submit(se, cpuvaddr, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		return ret;
 
@@ -887,7 +874,7 @@ static int tegra_sm4_gcm_do_crypt(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm
 static int tegra_sm4_gcm_do_final(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm4_gcm_reqctx *rctx)
 {
 	struct tegra_se *se = ctx->se;
-	u32 *cpuvaddr = rctx->cmdbuf;
+	u32 *cpuvaddr = se->cmdbuf->addr;
 	int size, ret, off;
 
 	rctx->config = se->regcfg->cfg(ctx->final_alg, rctx->encrypt);
@@ -896,7 +883,7 @@ static int tegra_sm4_gcm_do_final(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm
 
 	/* Prepare command and submit */
 	size = tegra_sm4_gcm_prep_final_cmd(se, cpuvaddr, rctx);
-	ret = tegra_se_host1x_submit(se, cpuvaddr, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		return ret;
 
@@ -913,7 +900,7 @@ static int tegra_sm4_gcm_do_final(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm
 static int tegra_sm4_gcm_hw_verify(struct tegra_sm4_gcm_ctx *ctx, struct tegra_sm4_gcm_reqctx *rctx, u8 *mac)
 {
 	struct tegra_se *se = ctx->se;
-	u32 result, *cpuvaddr = rctx->cmdbuf;
+	u32 result, *cpuvaddr = se->cmdbuf->addr;
 	int size, ret;
 
 	memcpy(rctx->inbuf.buf, mac, rctx->authsize);
@@ -925,7 +912,7 @@ static int tegra_sm4_gcm_hw_verify(struct tegra_sm4_gcm_ctx *ctx, struct tegra_s
 
 	/* Prepare command and submit */
 	size = tegra_sm4_gcm_prep_final_cmd(se, cpuvaddr, rctx);
-	ret = tegra_se_host1x_submit(se, cpuvaddr, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		return ret;
 
@@ -1035,7 +1022,6 @@ key_free:
 		tegra_key_invalidate(ctx->se, rctx->key_id, ctx->alg);
 
 out:
-	kfree(rctx->cmdbuf);
 	crypto_finalize_aead_request(se->engine, req, ret);
 
 	return 0;
@@ -1099,10 +1085,6 @@ static int tegra_sm4_gcm_crypt(struct aead_request *req, bool encrypt)
 
 	rctx->encrypt = encrypt;
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		return -ENOMEM;
-
 	return crypto_transfer_aead_request_to_engine(ctx->se->engine, req);
 }
 
@@ -1244,9 +1226,9 @@ static int tegra_sm4_cmac_do_update(struct ahash_request *req)
 	/* Update residue value with the residue after current block */
 	rctx->residue.size = nresidue;
 
-	size = tegra_sm4_cmac_prep_cmd(se, rctx->cmdbuf, rctx);
+	size = tegra_sm4_cmac_prep_cmd(se, se->cmdbuf->addr, rctx);
 
-	return tegra_se_host1x_submit(se, rctx->cmdbuf, size);
+	return tegra_se_host1x_submit(se, size);
 }
 
 static int tegra_sm4_cmac_do_final(struct ahash_request *req)
@@ -1273,8 +1255,8 @@ static int tegra_sm4_cmac_do_final(struct ahash_request *req)
 	}
 
 	/* Prepare command and submit */
-	size = tegra_sm4_cmac_prep_cmd(se, rctx->cmdbuf, rctx);
+	size = tegra_sm4_cmac_prep_cmd(se, se->cmdbuf->addr, rctx);
-	ret = tegra_se_host1x_submit(se, rctx->cmdbuf, size);
+	ret = tegra_se_host1x_submit(se, size);
 	if (ret)
 		goto out;
 
@@ -1297,8 +1279,6 @@ out_free:
 	dma_free_coherent(se->dev, rctx->digest.size, rctx->digest.buf,
 			rctx->digest.addr);
 
-	kfree(rctx->cmdbuf);
-
 	return ret;
 }
 
@@ -1386,19 +1366,15 @@ static int tegra_sm4_cmac_init(struct ahash_request *req)
 			return -ENOMEM;
 	}
 
-	rctx->cmdbuf = kzalloc(SE_MAX_CMDLEN, GFP_KERNEL);
-	if (!rctx->cmdbuf)
-		goto key_free;
-
 	rctx->digest.buf = dma_alloc_coherent(se->dev, rctx->digest.size,
 				&rctx->digest.addr, GFP_KERNEL);
 	if (!rctx->digest.buf)
-		goto cmdbuf_free;
+		goto digbuf_fail;
 
 	rctx->residue.buf = dma_alloc_coherent(se->dev, rctx->blk_size * 2,
 					&rctx->residue.addr, GFP_KERNEL);
 	if (!rctx->residue.buf)
-		goto digbuf_free;
+		goto resbuf_fail;
 
 	rctx->residue.size = 0;
 	rctx->datbuf.size = 0;
@@ -1409,12 +1385,10 @@ static int tegra_sm4_cmac_init(struct ahash_request *req)
 
 	return 0;
 
-digbuf_free:
+resbuf_fail:
 	dma_free_coherent(se->dev, rctx->blk_size, rctx->digest.buf,
 				rctx->digest.addr);
-cmdbuf_free:
+digbuf_fail:
-	kfree(rctx->cmdbuf);
-key_free:
 	if (rctx->key_id != ctx->key_id)
 		tegra_key_invalidate(ctx->se, rctx->key_id, ctx->alg);
 

drivers/crypto/tegra/tegra-se.h

@@ -23,7 +23,6 @@
 #define SE_OWNERSHIP_UID(x)				FIELD_GET(GENMASK(7, 0), x)
 #define TEGRA_GPSE_ID					3
 
-#define SE_MAX_CMDLEN					(100 * 4) /* max 100 commands of 4 bytes each */
 #define SE_STREAM_ID					0x90
 
 #define SE_SHA_CFG					0x4004
@@ -656,6 +655,7 @@ struct tegra_se {
 	struct host1x_client client;
 	struct host1x_channel *channel;
 	struct tegra_se_regcfg *regcfg;
+	struct tegra_se_cmdbuf *cmdbuf;
 	struct crypto_engine *engine;
 	struct host1x_syncpt *syncpt;
 	struct device *dev;
@@ -770,7 +770,7 @@ int tegra_key_submit(struct tegra_se *se, const u8 *key,
 		     u32 keylen, u32 alg, u32 *keyid);
 void tegra_key_invalidate(struct tegra_se *se, u32 keyid, u32 alg);
 unsigned int tegra_key_get_idx(struct tegra_se *se, u32 keyid);
-int tegra_se_host1x_submit(struct tegra_se *se, u32 *cpuvaddr, u32 size);
+int tegra_se_host1x_submit(struct tegra_se *se, u32 size);
 
 u32 tegra_kds_get_id(void);
 void tegra_kds_free_id(u32 keyid);