Hi,
On 2019/12/3 20:01, Marco Elver wrote:
Avoid using __sync builtins and instead prefer the kernel's own
facilities:
See comments below for suggestions, preserving the assumed memory
ordering requirements (but please double-check). By using atomic_t
instead of __sync, you'd also avoid any data races due to plain
concurrent accesses.
Reported in: https://lore.kernel.org/linux-crypto/CANpmjNM2b26Oo6k-4EqfrJf1sBj3WoFf-NQnwsLr3EW9B=G8kw@xxxxxxxxxxxxxx/
Okay, I will check, and send out a patch to fixed them, thanks.
On Wed, 13 Nov 2019, Zaibo Xu wrote:
SEC driver provides PCIe hardware device initiation with
AES, SM4, and 3DES skcipher algorithms registered to Crypto.
It uses Hisilicon QM as interface to CPU.
Signed-off-by: Zaibo Xu <xuzaibo@xxxxxxxxxx>
Signed-off-by: Longfang Liu <liulongfang@xxxxxxxxxx>
---
drivers/crypto/hisilicon/Kconfig | 16 +
drivers/crypto/hisilicon/Makefile | 1 +
drivers/crypto/hisilicon/sec2/Makefile | 2 +
drivers/crypto/hisilicon/sec2/sec.h | 132 +++++
drivers/crypto/hisilicon/sec2/sec_crypto.c | 886 +++++++++++++++++++++++++++++
drivers/crypto/hisilicon/sec2/sec_crypto.h | 198 +++++++
drivers/crypto/hisilicon/sec2/sec_main.c | 640 +++++++++++++++++++++
7 files changed, 1875 insertions(+)
create mode 100644 drivers/crypto/hisilicon/sec2/Makefile
create mode 100644 drivers/crypto/hisilicon/sec2/sec.h
create mode 100644 drivers/crypto/hisilicon/sec2/sec_crypto.c
create mode 100644 drivers/crypto/hisilicon/sec2/sec_crypto.h
create mode 100644 drivers/crypto/hisilicon/sec2/sec_main.c
[...]
diff --git a/drivers/crypto/hisilicon/sec2/sec.h b/drivers/crypto/hisilicon/sec2/sec.h
new file mode 100644
index 0000000..443b6c5
--- /dev/null
+++ b/drivers/crypto/hisilicon/sec2/sec.h
@@ -0,0 +1,132 @@
[...]
+
+/* SEC request of Crypto */
+struct sec_req {
+ struct sec_sqe sec_sqe;
+ struct sec_ctx *ctx;
+ struct sec_qp_ctx *qp_ctx;
+
+ /* Cipher supported only at present */
+ struct sec_cipher_req c_req;
+ int err_type;
+ int req_id;
+
+ /* Status of the SEC request */
+ int fake_busy;
This could be
atomic_t fake_busy;
Yes, atomic_t is better.
+};
+
[...]
diff --git a/drivers/crypto/hisilicon/sec2/sec_crypto.c b/drivers/crypto/hisilicon/sec2/sec_crypto.c
new file mode 100644
index 0000000..23092a9
--- /dev/null
+++ b/drivers/crypto/hisilicon/sec2/sec_crypto.c
@@ -0,0 +1,886 @@
Add
#include <linux/atomic.h>
Okay.
[...]
+static int sec_bd_send(struct sec_ctx *ctx, struct sec_req *req)
+{
+ struct sec_qp_ctx *qp_ctx = req->qp_ctx;
+ int ret;
+
+ mutex_lock(&qp_ctx->req_lock);
+ ret = hisi_qp_send(qp_ctx->qp, &req->sec_sqe);
+ mutex_unlock(&qp_ctx->req_lock);
+
+ if (ret == -EBUSY)
+ return -ENOBUFS;
+
+ if (!ret) {
+ if (req->fake_busy)
This could be:
atomic_read(&req->fake_busy)
yes.
+ ret = -EBUSY;
+ else
+ ret = -EINPROGRESS;
+ }
+
+ return ret;
+}
[...]
+static void sec_skcipher_callback(struct sec_ctx *ctx, struct sec_req *req)
+{
+ struct skcipher_request *sk_req = req->c_req.sk_req;
+ struct sec_qp_ctx *qp_ctx = req->qp_ctx;
+
+ atomic_dec(&qp_ctx->pending_reqs);
+ sec_free_req_id(req);
+
+ /* IV output at encrypto of CBC mode */
+ if (ctx->c_ctx.c_mode == SEC_CMODE_CBC && req->c_req.encrypt)
+ sec_update_iv(req);
+
+ if (__sync_bool_compare_and_swap(&req->fake_busy, 1, 0))
This could be:
int expect_val = 1;
...
if (atomic_try_cmpxchg_relaxed(&req->fake_busy, &expect_val, 0))
okay
+ sk_req->base.complete(&sk_req->base, -EINPROGRESS);
+
+ sk_req->base.complete(&sk_req->base, req->err_type);
+}
+
+static void sec_request_uninit(struct sec_ctx *ctx, struct sec_req *req)
+{
+ struct sec_qp_ctx *qp_ctx = req->qp_ctx;
+
+ atomic_dec(&qp_ctx->pending_reqs);
+ sec_free_req_id(req);
+ sec_put_queue_id(ctx, req);
+}
+
+static int sec_request_init(struct sec_ctx *ctx, struct sec_req *req)
+{
+ struct sec_qp_ctx *qp_ctx;
+ int issue_id, ret;
+
+ /* To load balance */
+ issue_id = sec_get_queue_id(ctx, req);
+ qp_ctx = &ctx->qp_ctx[issue_id];
+
+ req->req_id = sec_alloc_req_id(req, qp_ctx);
+ if (req->req_id < 0) {
+ sec_put_queue_id(ctx, req);
+ return req->req_id;
+ }
+
+ if (ctx->fake_req_limit <= atomic_inc_return(&qp_ctx->pending_reqs))
+ req->fake_busy = 1;
+ else
+ req->fake_busy = 0;
These could be:
atomic_set(&req->fake_busy, ...)
Yes, thanks.
cheers,
Zaibo
.
