[Acc] [PATCH 6/7] wd_comp: support nosva case

24 Jul 2025

Support mempcy, user pointer and sgl case
For flat memory:
If user call wd_xxx_setup_blkpool, will use user pointer mode.
uadk directly use pointer for app, assume it is continuous memory
and translate va to pa when configure register
Otherwise, alloc_sess will setup blkpool and use memcpy mode.
wd_comp alloc continuous memory for hardware and memcpy from
src pointer and memcpy results to dst pointer
For sgl memory:
App has to call wd_xxx_setup_blkpool.
The wd_datalist.data has to use continuous memory
Signed-off-by: Zhangfei Gao zhangfei.gao@linaro.org
---
 drv/hisi_comp.c           | 98 +++++++++++++++++++++++++++++----------
 drv/hisi_qm_udrv.c        | 14 ++++--
 drv/hisi_qm_udrv.h        |  3 +-
 drv/hisi_sec.c            |  8 ++--
 include/drv/wd_comp_drv.h |  7 +++
 wd_comp.c                 | 65 ++++++++++++++++++++++++++
 6 files changed, 162 insertions(+), 33 deletions(-)

diff --git a/drv/hisi_comp.c b/drv/hisi_comp.c
index cd558a8..c897efe 100644
--- a/drv/hisi_comp.c
+++ b/drv/hisi_comp.c
@@ -431,7 +431,14 @@ static int fill_buf_deflate_generic(struct hisi_zip_sqe *sqe,
    if (msg->ctx_buf)
    	ctx_buf = msg->ctx_buf + RSV_OFFSET;
-	fill_buf_addr_deflate(sqe, src, dst, ctx_buf);
+	if (msg->blkpool) {
+		fill_buf_addr_deflate(sqe,
+			wd_blkpool_phy(msg->blkpool, src),
+			wd_blkpool_phy(msg->blkpool, dst),
+			wd_blkpool_phy(msg->blkpool, ctx_buf));
+	} else {
+		fill_buf_addr_deflate(sqe, src, dst, ctx_buf);
+	}
return 0;
 }
@@ -464,32 +471,45 @@ static void fill_buf_type_sgl(struct hisi_zip_sqe *sqe)
 }
static int fill_buf_addr_deflate_sgl(handle_t h_qp, struct hisi_zip_sqe *sqe,
+				     struct wd_comp_msg *msg,
    			     struct wd_datalist	*list_src,
    			     struct wd_datalist *list_dst)
 {
    void *hw_sgl_in, *hw_sgl_out;
    handle_t h_sgl_pool;
-	h_sgl_pool = hisi_qm_get_sglpool(h_qp);
+	if (msg->h_sgl_pool)
+		h_sgl_pool = msg->h_sgl_pool;
+	else
+		h_sgl_pool = hisi_qm_get_sglpool(h_qp);
    if (unlikely(!h_sgl_pool)) {
    	WD_ERR("failed to get sglpool!\n");
    	return -WD_EINVAL;
    }
-	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, list_src);
+	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, list_src, msg->blkpool);
    if (unlikely(!hw_sgl_in)) {
    	WD_ERR("failed to get hw sgl in!\n");
    	return -WD_ENOMEM;
    }
-	hw_sgl_out = hisi_qm_get_hw_sgl(h_sgl_pool, list_dst);
+	hw_sgl_out = hisi_qm_get_hw_sgl(h_sgl_pool, list_dst, msg->blkpool);
    if (unlikely(!hw_sgl_out)) {
    	WD_ERR("failed to get hw sgl out!\n");
    	hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_in);
    	return -WD_ENOMEM;
    }
-	fill_buf_addr_deflate(sqe, hw_sgl_in, hw_sgl_out, NULL);
+	if (msg->h_sgl_pool) {
+		fill_buf_addr_deflate(sqe,
+			wd_blkpool_phy(msg->blkpool, hw_sgl_in),
+			wd_blkpool_phy(msg->blkpool, hw_sgl_out),
+			NULL);
+		msg->hw_sgl_in = hw_sgl_in;
+		msg->hw_sgl_out = hw_sgl_out;
+	} else {
+		fill_buf_addr_deflate(sqe, hw_sgl_in, hw_sgl_out, NULL);
+	}
return 0;
 }
@@ -543,7 +563,7 @@ static int fill_buf_deflate_sgl_generic(handle_t h_qp, struct hisi_zip_sqe *sqe,
fill_buf_type_sgl(sqe);
-	ret = fill_buf_addr_deflate_sgl(h_qp, sqe, list_src, list_dst);
+	ret = fill_buf_addr_deflate_sgl(h_qp, sqe, msg, list_src, list_dst);
    if (unlikely(ret))
    	return ret;
@@ -738,34 +758,48 @@ static int fill_buf_lz77_zstd_sgl(handle_t h_qp, struct hisi_zip_sqe *sqe,
fill_buf_size_lz77_zstd(sqe, in_size, lits_size, out_size - lits_size);
-	h_sgl_pool = hisi_qm_get_sglpool(h_qp);
+	if (msg->h_sgl_pool)
+		h_sgl_pool = msg->h_sgl_pool;
+	else
+		h_sgl_pool = hisi_qm_get_sglpool(h_qp);
    if (unlikely(!h_sgl_pool)) {
    	WD_ERR("failed to get sglpool!\n");
    	return -WD_EINVAL;
    }
-	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, req->list_src);
+	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, req->list_src, msg->blkpool);
    if (unlikely(!hw_sgl_in)) {
    	WD_ERR("failed to get hw sgl in!\n");
    	return -WD_ENOMEM;
    }
-	hw_sgl_out_lit = hisi_qm_get_hw_sgl(h_sgl_pool, req->list_dst);
+	hw_sgl_out_lit = hisi_qm_get_hw_sgl(h_sgl_pool, req->list_dst, msg->blkpool);
    if (unlikely(!hw_sgl_out_lit)) {
    	WD_ERR("failed to get hw sgl out for literals!\n");
    	ret = -WD_ENOMEM;
    	goto err_free_sgl_in;
    }
-	hw_sgl_out_seq = hisi_qm_get_hw_sgl(h_sgl_pool, seq_start);
+	hw_sgl_out_seq = hisi_qm_get_hw_sgl(h_sgl_pool, seq_start, msg->blkpool);
    if (unlikely(!hw_sgl_out_seq)) {
    	WD_ERR("failed to get hw sgl out for sequences!\n");
    	ret = -WD_ENOMEM;
    	goto err_free_sgl_out_lit;
    }
-	fill_buf_addr_lz77_zstd(sqe, hw_sgl_in, hw_sgl_out_lit,
+	if (msg->h_sgl_pool) {
+		fill_buf_addr_lz77_zstd(sqe,
+			wd_blkpool_phy(msg->blkpool, hw_sgl_in),
+			wd_blkpool_phy(msg->blkpool, hw_sgl_out_lit),
+			wd_blkpool_phy(msg->blkpool, hw_sgl_out_seq),
+			NULL);
+		msg->hw_sgl_in = hw_sgl_in;
+		msg->hw_sgl_out = hw_sgl_out_lit;
+		msg->hw_sgl_out_seq = hw_sgl_out_seq;
+	} else {
+		fill_buf_addr_lz77_zstd(sqe, hw_sgl_in, hw_sgl_out_lit,
    			hw_sgl_out_seq, NULL);
+	}
return 0;
@@ -1116,27 +1150,41 @@ static int fill_zip_comp_sqe(struct hisi_qp *qp, struct wd_comp_msg *msg,
 }
static void free_hw_sgl(handle_t h_qp, struct hisi_zip_sqe *sqe,
+			struct wd_comp_msg *msg,
    		enum wd_comp_alg_type alg_type)
 {
    void *hw_sgl_in, *hw_sgl_out;
    handle_t h_sgl_pool;
-	h_sgl_pool = hisi_qm_get_sglpool(h_qp);
-	if (unlikely(!h_sgl_pool)) {
-		WD_ERR("failed to get sglpool to free hw sgl!\n");
-		return;
-	}
+	if (msg->h_sgl_pool) {
+		h_sgl_pool = msg->h_sgl_pool;
+		if (unlikely(!h_sgl_pool)) {
+			WD_ERR("failed to get sglpool to free hw sgl!\n");
+			return;
+		}
+		hisi_qm_put_hw_sgl(h_sgl_pool, msg->hw_sgl_in);
+		hisi_qm_put_hw_sgl(h_sgl_pool, msg->hw_sgl_out);
+		if (alg_type == WD_LZ77_ZSTD)
+			hisi_qm_put_hw_sgl(h_sgl_pool, msg->hw_sgl_out_seq);
+	} else {
-	hw_sgl_in = VA_ADDR(sqe->source_addr_h, sqe->source_addr_l);
-	hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_in);
+		h_sgl_pool = hisi_qm_get_sglpool(h_qp);
+		if (unlikely(!h_sgl_pool)) {
+			WD_ERR("failed to get sglpool to free hw sgl!\n");
+			return;
+		}
-	hw_sgl_out = VA_ADDR(sqe->dest_addr_h, sqe->dest_addr_l);
-	hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_out);
+		hw_sgl_in = VA_ADDR(sqe->source_addr_h, sqe->source_addr_l);
+		hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_in);
-	if (alg_type == WD_LZ77_ZSTD) {
-		hw_sgl_out = VA_ADDR(sqe->literals_addr_h,
-				     sqe->literals_addr_l);
+		hw_sgl_out = VA_ADDR(sqe->dest_addr_h, sqe->dest_addr_l);
    	hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_out);
+
+		if (alg_type == WD_LZ77_ZSTD) {
+			hw_sgl_out = VA_ADDR(sqe->literals_addr_h,
+					sqe->literals_addr_l);
+			hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_out);
+		}
    }
 }
@@ -1163,7 +1211,7 @@ static int hisi_zip_comp_send(struct wd_alg_driver *drv, handle_t ctx, void *com
    ret = hisi_qm_send(h_qp, &sqe, 1, &count);
    if (unlikely(ret < 0)) {
    	if (msg->req.data_fmt == WD_SGL_BUF)
-			free_hw_sgl(h_qp, &sqe, msg->alg_type);
+			free_hw_sgl(h_qp, &sqe, msg, msg->alg_type);
    	if (ret != -WD_EBUSY)
    		WD_ERR("failed to send to hardware, ret = %d!\n", ret);
@@ -1304,7 +1352,7 @@ static int parse_zip_sqe(struct hisi_qp *qp, struct hisi_zip_sqe *sqe,
    recv_msg->alg_type = alg_type;
if (buf_type == WD_SGL_BUF)
-		free_hw_sgl((handle_t)qp, sqe, alg_type);
+		free_hw_sgl((handle_t)qp, sqe, msg, alg_type);
if (unlikely(recv_msg->req.status == WD_IN_EPARA))
    	dump_zip_msg(recv_msg);
diff --git a/drv/hisi_qm_udrv.c b/drv/hisi_qm_udrv.c
index 304764e..78f6583 100644
--- a/drv/hisi_qm_udrv.c
+++ b/drv/hisi_qm_udrv.c
@@ -9,6 +9,7 @@
#include "hisi_qm_udrv.h"
 #include "wd_util.h"
+#include "wd_bmm.h"
#define QM_DBELL_CMD_SQ		0
 #define QM_DBELL_CMD_CQ		1
@@ -842,7 +843,8 @@ static void hisi_qm_dump_sgl(void *sgl)
    }
 }
-void *hisi_qm_get_hw_sgl(handle_t sgl_pool, struct wd_datalist *sgl)
+void *hisi_qm_get_hw_sgl(handle_t sgl_pool, struct wd_datalist *sgl,
+			void *blkpool)
 {
    struct hisi_sgl_pool *pool = (struct hisi_sgl_pool *)sgl_pool;
    struct wd_datalist *tmp = sgl;
@@ -872,7 +874,10 @@ void *hisi_qm_get_hw_sgl(handle_t sgl_pool, struct wd_datalist *sgl)
    		goto err_out;
    	}
-		cur->sge_entries[i].buff = (uintptr_t)tmp->data;
+		if (blkpool)
+			cur->sge_entries[i].buff = (uintptr_t)wd_blkpool_phy(blkpool, tmp->data);
+		else
+			cur->sge_entries[i].buff = (uintptr_t)tmp->data;
    	cur->sge_entries[i].len = tmp->len;
    	cur->entry_sum_in_sgl++;
    	cur->entry_size_in_sgl += tmp->len;
@@ -890,7 +895,10 @@ void *hisi_qm_get_hw_sgl(handle_t sgl_pool, struct wd_datalist *sgl)
    			WD_ERR("invalid: the sgl pool is not enough!\n");
    			goto err_out;
    		}
-			cur->next_dma = (uintptr_t)next;
+			if (blkpool)
+				cur->next_dma = (uintptr_t)wd_blkpool_phy(blkpool, next);
+			else
+				cur->next_dma = (uintptr_t)next;
    		cur = next;
    		head->entry_sum_in_chain += pool->sge_num;
    		/* In the new sgl chain, the subscript must be reset */
diff --git a/drv/hisi_qm_udrv.h b/drv/hisi_qm_udrv.h
index b02e8e7..ddb666e 100644
--- a/drv/hisi_qm_udrv.h
+++ b/drv/hisi_qm_udrv.h
@@ -162,7 +162,8 @@ void hisi_qm_destroy_sglpool(handle_t sgl_pool);
  *
  * Return the hw sgl addr which can fill into the sqe.
  */
-void *hisi_qm_get_hw_sgl(handle_t sgl_pool, struct wd_datalist *sgl);
+void *hisi_qm_get_hw_sgl(handle_t sgl_pool, struct wd_datalist *sgl,
+			 void *blkpool);
/**
  * hisi_qm_put_hw_sgl - Reback the hw sgl to the sgl pool.
diff --git a/drv/hisi_sec.c b/drv/hisi_sec.c
index 747d3a8..a305985 100644
--- a/drv/hisi_sec.c
+++ b/drv/hisi_sec.c
@@ -1050,7 +1050,7 @@ static int hisi_sec_fill_sgl(handle_t h_qp, __u8 **in, __u8 **out,
    	return -WD_EINVAL;
    }
-	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, (struct wd_datalist *)(*in));
+	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, (struct wd_datalist *)(*in), NULL);
    if (!hw_sgl_in) {
    	WD_ERR("failed to get sgl in for hw_v2!\n");
    	return -WD_EINVAL;
@@ -1060,7 +1060,7 @@ static int hisi_sec_fill_sgl(handle_t h_qp, __u8 **in, __u8 **out,
    	hw_sgl_out = *out;
    } else {
    	hw_sgl_out = hisi_qm_get_hw_sgl(h_sgl_pool,
-						(struct wd_datalist *)(*out));
+						(struct wd_datalist *)(*out), NULL);
    	if (!hw_sgl_out) {
    		WD_ERR("failed to get hw sgl out for hw_v2!\n");
    		hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_in);
@@ -1090,7 +1090,7 @@ static int hisi_sec_fill_sgl_v3(handle_t h_qp, __u8 **in, __u8 **out,
    	return -WD_EINVAL;
    }
-	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, (struct wd_datalist *)(*in));
+	hw_sgl_in = hisi_qm_get_hw_sgl(h_sgl_pool, (struct wd_datalist *)(*in), NULL);
    if (!hw_sgl_in) {
    	WD_ERR("failed to get sgl in for hw_v3!\n");
    	return -WD_EINVAL;
@@ -1101,7 +1101,7 @@ static int hisi_sec_fill_sgl_v3(handle_t h_qp, __u8 **in, __u8 **out,
    	sqe->bd_param |= SEC_PBUFF_MODE_MASK_V3;
    } else {
    	hw_sgl_out = hisi_qm_get_hw_sgl(h_sgl_pool,
-						(struct wd_datalist *)(*out));
+						(struct wd_datalist *)(*out), NULL);
    	if (!hw_sgl_out) {
    		WD_ERR("failed to get hw sgl out for hw_v3!\n");
    		hisi_qm_put_hw_sgl(h_sgl_pool, hw_sgl_in);
diff --git a/include/drv/wd_comp_drv.h b/include/drv/wd_comp_drv.h
index 213cf2d..1b8273c 100644
--- a/include/drv/wd_comp_drv.h
+++ b/include/drv/wd_comp_drv.h
@@ -53,6 +53,13 @@ struct wd_comp_msg {
    __u32 checksum;
    /* Request identifier */
    __u32 tag;
+	void *blkpool;
+	void *src;
+	void *dst;
+	handle_t h_sgl_pool;
+	void *hw_sgl_in;
+	void *hw_sgl_out;
+	void *hw_sgl_out_seq;
 };
struct wd_comp_msg *wd_comp_get_msg(__u32 idx, __u32 tag);
diff --git a/wd_comp.c b/wd_comp.c
index 21a57ac..f45621c 100644
--- a/wd_comp.c
+++ b/wd_comp.c
@@ -39,6 +39,10 @@ struct wd_comp_sess {
    __u32 checksum;
    __u8 *ctx_buf;
    void *sched_key;
+	void *blkpool;
+	__u8 *blkpool_ctxbuf;
+	handle_t h_sgl_pool;
+	__u8 blkpool_mode;
 };
struct wd_comp_setting {
@@ -436,6 +440,8 @@ static int wd_comp_check_sess_params(struct wd_comp_sess_setup *setup)
handle_t wd_comp_alloc_sess(struct wd_comp_sess_setup *setup)
 {
+	struct wd_ctx_config_internal *config = &wd_comp_setting.config;
+	struct wd_ctx_internal *ctx = config->ctxs;
    struct wd_comp_sess *sess;
    int ret;
@@ -467,6 +473,28 @@ handle_t wd_comp_alloc_sess(struct wd_comp_sess_setup *setup)
    	goto sched_err;
    }
+	if (ctx->blkpool) {
+		sess->blkpool = ctx->blkpool;
+		sess->h_sgl_pool = ctx->h_sgl_pool;
+		sess->blkpool_mode = ctx->blkpool_mode;
+
+		if (ctx->blkpool_mode == BLKPOOL_MODE_MEMCPY) {
+			struct wd_blkpool_setup blksetup;
+
+			memset(&blksetup, 0, sizeof(blksetup));
+			blksetup.block_size = HW_CTX_SIZE;
+			blksetup.block_num = DEFAULT_BLOCK_NM;
+			blksetup.align_size = DEFAULT_ALIGN_SIZE;
+			ret = wd_blkpool_setup(sess->blkpool, &blksetup);
+			if (ret)
+				goto sched_err;
+		}
+
+		sess->blkpool_ctxbuf = wd_blkpool_alloc(sess->blkpool, HW_CTX_SIZE);
+		if (!sess->blkpool_ctxbuf)
+			goto sched_err;
+	}
+
    return (handle_t)sess;
sched_err:
@@ -486,6 +514,9 @@ void wd_comp_free_sess(handle_t h_sess)
    if (sess->ctx_buf)
    	free(sess->ctx_buf);
+	if (sess->blkpool_ctxbuf)
+		wd_blkpool_free(sess->blkpool, sess->blkpool_ctxbuf);
+
    if (sess->sched_key)
    	free(sess->sched_key);
@@ -506,6 +537,9 @@ int wd_comp_reset_sess(handle_t h_sess)
    if (sess->ctx_buf)
    	memset(sess->ctx_buf, 0, HW_CTX_SIZE);
+	if (sess->blkpool_ctxbuf)
+		memset(sess->blkpool_ctxbuf, 0, HW_CTX_SIZE);
+
    return 0;
 }
@@ -628,6 +662,29 @@ static int wd_comp_sync_job(struct wd_comp_sess *sess,
    wd_dfx_msg_cnt(config, WD_CTX_CNT_NUM, idx);
    ctx = config->ctxs + idx;
+	if (sess->blkpool) {
+		msg->ctx_buf = sess->blkpool_ctxbuf;
+		msg->blkpool = sess->blkpool;
+		msg->h_sgl_pool = sess->h_sgl_pool;
+
+		if (sess->blkpool_mode == BLKPOOL_MODE_MEMCPY) {
+			void *src = wd_blkpool_alloc(sess->blkpool, req->src_len);
+			void *dst = wd_blkpool_alloc(sess->blkpool, req->dst_len);
+
+			if (!src || !dst)
+				return -ENOMEM;
+
+			/* save */
+			msg->src = msg->req.src;
+			msg->dst = msg->req.dst;
+			/* replace */
+			msg->req.src = src;
+			msg->req.dst = dst;
+
+			memcpy(msg->req.src, msg->src, req->src_len);
+		}
+	}
+
    msg_handle.send = wd_comp_setting.driver->send;
    msg_handle.recv = wd_comp_setting.driver->recv;
@@ -636,6 +693,14 @@ static int wd_comp_sync_job(struct wd_comp_sess *sess,
    			 msg, NULL, config->epoll_en);
    pthread_spin_unlock(&ctx->lock);
+	if (sess->blkpool) {
+		if (sess->blkpool_mode == BLKPOOL_MODE_MEMCPY) {
+			memcpy(msg->dst, req->dst, req->dst_len);
+			wd_blkpool_free(sess->blkpool, msg->req.src);
+			wd_blkpool_free(sess->blkpool, msg->req.dst);
+		}
+	}
+
    return ret;
 }
-- 
2.25.1


    

2025

2024

2023

2022

[Acc] [PATCH 6/7] wd_comp: support nosva case