AES CBC, ECB mode input size are fixed with BLOCK_SIZE

[ljmblueday]

AES CBC and ECB mode input sizes are fixed as BLOCK_SIZE.
For GCM, CTR mode, use in_size as input->data_length. However, for CBC, ECB mode, in_size is fixed as BLOCK_SIZE. This results in poor performance when encrypting and decrypting.

TA_aes_update function in crypto_aes.c file
keymaster_error_t TA_aes_update(keymaster_operation_t *operation,
				keymaster_blob_t *input,
				keymaster_blob_t *output,
				uint32_t *out_size,
				const uint32_t input_provided,
				size_t *input_consumed,
				const keymaster_key_param_set_t *in_params,
				bool *is_input_ext)
{
	keymaster_error_t res = KM_ERROR_OK;
	uint32_t pos = 0U;
	uint32_t remainder = 0;
	**uint32_t in_size = BLOCK_SIZE;**
.
.
.

Performance test result (AES 256, data size 16MBytes)

CBC	6.7172
ECB	6.703
CTR	0.6988
GCM	0.705

In the case of CBC and ECB modes as follows, if in_size is put in as implemented in code, it shows the same performance as in GCM and CTR modes. Please give me a review opinion on the improved code.

} else {
		if (operation->mode == KM_MODE_CTR)
			/* CTR is a stream mode */
			in_size = input->data_length;
		else
			**/* KM_MODE_CBC, KM_MODE_ECB */
			in_size = input->data_length - ((input->data_length)% BLOCK_SIZE);**

		while (operation->mode == KM_MODE_CTR
			   || remainder / BLOCK_SIZE != 0) {
			/* calculate memory left.
			 * Add BLOCK_SIZE in case adding padding
			 */
	 		if(operation->purpose == KM_PURPOSE_DECRYPT && operation->padding == KM_PAD_PKCS7 && remainder == 16)
				break;

CBC	0.625
ECB	0.647
CTR	0.635
GCM	0.622