Hi Fan,
On 2017/11/28 0:47, Zhang, Roy Fan wrote:
Hi Jay,
Thanks for contributing to DPDK.
The code has been tested and works fine.
A few comments:
1. Could you split the patch into a patchset, as suggested in contribution
guide in http://dpdk.org/doc/guides/contributing/patches.html, section 5.4?
2. Please update doc/guides/cryptodevs for describing your virtio crypto PMD.
3. Please update doc/guides/rel_notes/release_18.02.
4. One more comment inline
For DPDK, I'm a newbie. Thanks for testing and pointing these steps
out, will fix them in V2.
-----Original Message-----
From: Jay Zhou [mailto:jianjay.z...@huawei.com]
Sent: Friday, November 17, 2017 5:10 PM
To: dev@dpdk.org
Cc: y...@fridaylinux.org; maxime.coque...@redhat.com;
arei.gong...@huawei.com; Zhang, Roy Fan <roy.fan.zh...@intel.com>; Zeng,
Xin <xin.z...@intel.com>; weidong.hu...@huawei.com;
wangxinxin.w...@huawei.com; longpe...@huawei.com;
jianjay.z...@huawei.com
Subject: [PATCH] virtio: add new driver for crypto devices
+ /*
+ * malloc memory to store indirect vring_desc entries, including
+ * ctrl request, cipher key, auth key, session input and desc vring
+ */
+ desc_offset = ctrl_req_length + cipher_keylen + auth_keylen
+ + input_length;
Instead of using rte_malloc() as below, you could pre-allocate a mempool and use
rte_mempool_get() or rte_mempool_get_bulk() to get these memory to store
descriptors.
You can use rte_mempool_virt2iova() to obtain the physical address of this
memory. This shall
have better performance.
I will have a try.
Regards,
Jay
+ virt_addr_started = rte_malloc(NULL,
+ desc_offset +
NUM_ENTRY_VIRTIO_CRYPTO_SYM_CREATE_SESSION
+ * sizeof(struct vring_desc), RTE_CACHE_LINE_SIZE);
+ if (virt_addr_started == NULL) {
+ PMD_SESSION_LOG(ERR, "not enough heap memory");
+ return -ENOSPC;
+ }
+ phys_addr_started = rte_malloc_virt2phy(virt_addr_started);
.
