1 files changed, 154 insertions, 251 deletions

diff --git a/drivers/crypto/caam/caamrng.c b/drivers/crypto/caam/caamrng.c
index e8baacaabe07..77d048dfe5d0 100644
--- a/drivers/crypto/caam/caamrng.c
+++ b/drivers/crypto/caam/caamrng.c
@@ -7,35 +7,12 @@
  *
  * Based on caamalg.c crypto API driver.
  *
- * relationship between job descriptors to shared descriptors:
- *
- * ---------------                     --------------
- * | JobDesc #0  |-------------------->| ShareDesc  |
- * | *(buffer 0) |      |------------->| (generate) |
- * ---------------      |              | (move)     |
- *                      |              | (store)    |
- * ---------------      |              --------------
- * | JobDesc #1  |------|
- * | *(buffer 1) |
- * ---------------
- *
- * A job desc looks like this:
- *
- * ---------------------
- * | Header            |
- * | ShareDesc Pointer |
- * | SEQ_OUT_PTR       |
- * | (output buffer)   |
- * ---------------------
- *
- * The SharedDesc never changes, and each job descriptor points to one of two
- * buffers for each device, from which the data will be copied into the
- * requested destination
  */
 
 #include <linux/hw_random.h>
 #include <linux/completion.h>
 #include <linux/atomic.h>
+#include <linux/kfifo.h>
 
 #include "compat.h"
 
@@ -45,278 +22,205 @@
 #include "jr.h"
 #include "error.h"
 
+#define CAAM_RNG_MAX_FIFO_STORE_SIZE	16
+
 /*
- * Maximum buffer size: maximum number of random, cache-aligned bytes that
- * will be generated and moved to seq out ptr (extlen not allowed)
+ * Length of used descriptors, see caam_init_desc()
  */
-#define RN_BUF_SIZE			(0xffff / L1_CACHE_BYTES * \
-					 L1_CACHE_BYTES)
-
-/* length of descriptors */
-#define DESC_JOB_O_LEN			(CAAM_CMD_SZ * 2 + CAAM_PTR_SZ_MAX * 2)
-#define DESC_RNG_LEN			(3 * CAAM_CMD_SZ)
-
-/* Buffer, its dma address and lock */
-struct buf_data {
-	u8 buf[RN_BUF_SIZE] ____cacheline_aligned;
-	dma_addr_t addr;
-	struct completion filled;
-	u32 hw_desc[DESC_JOB_O_LEN];
-#define BUF_NOT_EMPTY 0
-#define BUF_EMPTY 1
-#define BUF_PENDING 2  /* Empty, but with job pending --don't submit another */
-	atomic_t empty;
-};
+#define CAAM_RNG_DESC_LEN (CAAM_CMD_SZ +				\
+			   CAAM_CMD_SZ +				\
+			   CAAM_CMD_SZ + CAAM_PTR_SZ_MAX)
 
 /* rng per-device context */
 struct caam_rng_ctx {
+	struct hwrng rng;
 	struct device *jrdev;
-	dma_addr_t sh_desc_dma;
-	u32 sh_desc[DESC_RNG_LEN];
-	unsigned int cur_buf_idx;
-	int current_buf;
-	struct buf_data bufs[2];
+	struct device *ctrldev;
+	void *desc_async;
+	void *desc_sync;
+	struct work_struct worker;
+	struct kfifo fifo;
 };
 
-static struct caam_rng_ctx *rng_ctx;
-
-/*
- * Variable used to avoid double free of resources in case
- * algorithm registration was unsuccessful
- */
-static bool init_done;
-
-static inline void rng_unmap_buf(struct device *jrdev, struct buf_data *bd)
-{
-	if (bd->addr)
-		dma_unmap_single(jrdev, bd->addr, RN_BUF_SIZE,
-				 DMA_FROM_DEVICE);
-}
+struct caam_rng_job_ctx {
+	struct completion *done;
+	int *err;
+};
 
-static inline void rng_unmap_ctx(struct caam_rng_ctx *ctx)
+static struct caam_rng_ctx *to_caam_rng_ctx(struct hwrng *r)
 {
-	struct device *jrdev = ctx->jrdev;
-
-	if (ctx->sh_desc_dma)
-		dma_unmap_single(jrdev, ctx->sh_desc_dma,
-				 desc_bytes(ctx->sh_desc), DMA_TO_DEVICE);
-	rng_unmap_buf(jrdev, &ctx->bufs[0]);
-	rng_unmap_buf(jrdev, &ctx->bufs[1]);
+	return (struct caam_rng_ctx *)r->priv;
 }
 
-static void rng_done(struct device *jrdev, u32 *desc, u32 err, void *context)
+static void caam_rng_done(struct device *jrdev, u32 *desc, u32 err,
+			  void *context)
 {
-	struct buf_data *bd;
-
-	bd = container_of(desc, struct buf_data, hw_desc[0]);
+	struct caam_rng_job_ctx *jctx = context;
 
 	if (err)
-		caam_jr_strstatus(jrdev, err);
+		*jctx->err = caam_jr_strstatus(jrdev, err);
 
-	atomic_set(&bd->empty, BUF_NOT_EMPTY);
-	complete(&bd->filled);
-
-	/* Buffer refilled, invalidate cache */
-	dma_sync_single_for_cpu(jrdev, bd->addr, RN_BUF_SIZE, DMA_FROM_DEVICE);
-
-	print_hex_dump_debug("rng refreshed buf@: ", DUMP_PREFIX_ADDRESS, 16, 4,
-			     bd->buf, RN_BUF_SIZE, 1);
+	complete(jctx->done);
 }
 
-static inline int submit_job(struct caam_rng_ctx *ctx, int to_current)
+static u32 *caam_init_desc(u32 *desc, dma_addr_t dst_dma)
 {
-	struct buf_data *bd = &ctx->bufs[!(to_current ^ ctx->current_buf)];
-	struct device *jrdev = ctx->jrdev;
-	u32 *desc = bd->hw_desc;
-	int err;
-
-	dev_dbg(jrdev, "submitting job %d\n", !(to_current ^ ctx->current_buf));
-	init_completion(&bd->filled);
-	err = caam_jr_enqueue(jrdev, desc, rng_done, ctx);
-	if (err)
-		complete(&bd->filled); /* don't wait on failed job*/
-	else
-		atomic_inc(&bd->empty); /* note if pending */
-
-	return err;
+	init_job_desc(desc, 0);	/* + 1 cmd_sz */
+	/* Generate random bytes: + 1 cmd_sz */
+	append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG |
+			 OP_ALG_PR_ON);
+	/* Store bytes: + 1 cmd_sz + caam_ptr_sz  */
+	append_fifo_store(desc, dst_dma,
+			  CAAM_RNG_MAX_FIFO_STORE_SIZE, FIFOST_TYPE_RNGSTORE);
+
+	print_hex_dump_debug("rng job desc@: ", DUMP_PREFIX_ADDRESS,
+			     16, 4, desc, desc_bytes(desc), 1);
+
+	return desc;
 }
 
-static int caam_read(struct hwrng *rng, void *data, size_t max, bool wait)
+static int caam_rng_read_one(struct device *jrdev,
+			     void *dst, int len,
+			     void *desc,
+			     struct completion *done)
 {
-	struct caam_rng_ctx *ctx = rng_ctx;
-	struct buf_data *bd = &ctx->bufs[ctx->current_buf];
-	int next_buf_idx, copied_idx;
-	int err;
-
-	if (atomic_read(&bd->empty)) {
-		/* try to submit job if there wasn't one */
-		if (atomic_read(&bd->empty) == BUF_EMPTY) {
-			err = submit_job(ctx, 1);
-			/* if can't submit job, can't even wait */
-			if (err)
-				return 0;
-		}
-		/* no immediate data, so exit if not waiting */
-		if (!wait)
-			return 0;
-
-		/* waiting for pending job */
-		if (atomic_read(&bd->empty))
-			wait_for_completion(&bd->filled);
+	dma_addr_t dst_dma;
+	int err, ret = 0;
+	struct caam_rng_job_ctx jctx = {
+		.done = done,
+		.err  = &ret,
+	};
+
+	len = CAAM_RNG_MAX_FIFO_STORE_SIZE;
+
+	dst_dma = dma_map_single(jrdev, dst, len, DMA_FROM_DEVICE);
+	if (dma_mapping_error(jrdev, dst_dma)) {
+		dev_err(jrdev, "unable to map destination memory\n");
+		return -ENOMEM;
 	}
 
-	next_buf_idx = ctx->cur_buf_idx + max;
-	dev_dbg(ctx->jrdev, "%s: start reading at buffer %d, idx %d\n",
-		 __func__, ctx->current_buf, ctx->cur_buf_idx);
-
-	/* if enough data in current buffer */
-	if (next_buf_idx < RN_BUF_SIZE) {
-		memcpy(data, bd->buf + ctx->cur_buf_idx, max);
-		ctx->cur_buf_idx = next_buf_idx;
-		return max;
+	init_completion(done);
+	err = caam_jr_enqueue(jrdev,
+			      caam_init_desc(desc, dst_dma),
+			      caam_rng_done, &jctx);
+	if (err == -EINPROGRESS) {
+		wait_for_completion(done);
+		err = 0;
 	}
 
-	/* else, copy what's left... */
-	copied_idx = RN_BUF_SIZE - ctx->cur_buf_idx;
-	memcpy(data, bd->buf + ctx->cur_buf_idx, copied_idx);
-	ctx->cur_buf_idx = 0;
-	atomic_set(&bd->empty, BUF_EMPTY);
-
-	/* ...refill... */
-	submit_job(ctx, 1);
+	dma_unmap_single(jrdev, dst_dma, len, DMA_FROM_DEVICE);
 
-	/* and use next buffer */
-	ctx->current_buf = !ctx->current_buf;
-	dev_dbg(ctx->jrdev, "switched to buffer %d\n", ctx->current_buf);
-
-	/* since there already is some data read, don't wait */
-	return copied_idx + caam_read(rng, data + copied_idx,
-				      max - copied_idx, false);
+	return err ?: (ret ?: len);
 }
 
-static inline int rng_create_sh_desc(struct caam_rng_ctx *ctx)
+static void caam_rng_fill_async(struct caam_rng_ctx *ctx)
 {
-	struct device *jrdev = ctx->jrdev;
-	u32 *desc = ctx->sh_desc;
-
-	init_sh_desc(desc, HDR_SHARE_SERIAL);
-
-	/* Generate random bytes */
-	append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG);
-
-	/* Store bytes */
-	append_seq_fifo_store(desc, RN_BUF_SIZE, FIFOST_TYPE_RNGSTORE);
+	struct scatterlist sg[1];
+	struct completion done;
+	int len, nents;
+
+	sg_init_table(sg, ARRAY_SIZE(sg));
+	nents = kfifo_dma_in_prepare(&ctx->fifo, sg, ARRAY_SIZE(sg),
+				     CAAM_RNG_MAX_FIFO_STORE_SIZE);
+	if (!nents)
+		return;
 
-	ctx->sh_desc_dma = dma_map_single(jrdev, desc, desc_bytes(desc),
-					  DMA_TO_DEVICE);
-	if (dma_mapping_error(jrdev, ctx->sh_desc_dma)) {
-		dev_err(jrdev, "unable to map shared descriptor\n");
-		return -ENOMEM;
-	}
+	len = caam_rng_read_one(ctx->jrdev, sg_virt(&sg[0]),
+				sg[0].length,
+				ctx->desc_async,
+				&done);
+	if (len < 0)
+		return;
 
-	print_hex_dump_debug("rng shdesc@: ", DUMP_PREFIX_ADDRESS, 16, 4,
-			     desc, desc_bytes(desc), 1);
+	kfifo_dma_in_finish(&ctx->fifo, len);
+}
 
-	return 0;
+static void caam_rng_worker(struct work_struct *work)
+{
+	struct caam_rng_ctx *ctx = container_of(work, struct caam_rng_ctx,
+						worker);
+	caam_rng_fill_async(ctx);
 }
 
-static inline int rng_create_job_desc(struct caam_rng_ctx *ctx, int buf_id)
+static int caam_read(struct hwrng *rng, void *dst, size_t max, bool wait)
 {
-	struct device *jrdev = ctx->jrdev;
-	struct buf_data *bd = &ctx->bufs[buf_id];
-	u32 *desc = bd->hw_desc;
-	int sh_len = desc_len(ctx->sh_desc);
+	struct caam_rng_ctx *ctx = to_caam_rng_ctx(rng);
+	int out;
 
-	init_job_desc_shared(desc, ctx->sh_desc_dma, sh_len, HDR_SHARE_DEFER |
-			     HDR_REVERSE);
+	if (wait) {
+		struct completion done;
 
-	bd->addr = dma_map_single(jrdev, bd->buf, RN_BUF_SIZE, DMA_FROM_DEVICE);
-	if (dma_mapping_error(jrdev, bd->addr)) {
-		dev_err(jrdev, "unable to map dst\n");
-		return -ENOMEM;
+		return caam_rng_read_one(ctx->jrdev, dst, max,
+					 ctx->desc_sync, &done);
 	}
 
-	append_seq_out_ptr_intlen(desc, bd->addr, RN_BUF_SIZE, 0);
-
-	print_hex_dump_debug("rng job desc@: ", DUMP_PREFIX_ADDRESS, 16, 4,
-			     desc, desc_bytes(desc), 1);
+	out = kfifo_out(&ctx->fifo, dst, max);
+	if (kfifo_is_empty(&ctx->fifo))
+		schedule_work(&ctx->worker);
 
-	return 0;
+	return out;
 }
 
 static void caam_cleanup(struct hwrng *rng)
 {
-	int i;
-	struct buf_data *bd;
-
-	for (i = 0; i < 2; i++) {
-		bd = &rng_ctx->bufs[i];
-		if (atomic_read(&bd->empty) == BUF_PENDING)
-			wait_for_completion(&bd->filled);
-	}
+	struct caam_rng_ctx *ctx = to_caam_rng_ctx(rng);
 
-	rng_unmap_ctx(rng_ctx);
+	flush_work(&ctx->worker);
+	caam_jr_free(ctx->jrdev);
+	kfifo_free(&ctx->fifo);
 }
 
-static int caam_init_buf(struct caam_rng_ctx *ctx, int buf_id)
+static int caam_init(struct hwrng *rng)
 {
-	struct buf_data *bd = &ctx->bufs[buf_id];
+	struct caam_rng_ctx *ctx = to_caam_rng_ctx(rng);
 	int err;
 
-	err = rng_create_job_desc(ctx, buf_id);
-	if (err)
-		return err;
-
-	atomic_set(&bd->empty, BUF_EMPTY);
-	submit_job(ctx, buf_id == ctx->current_buf);
-	wait_for_completion(&bd->filled);
+	ctx->desc_sync = devm_kzalloc(ctx->ctrldev, CAAM_RNG_DESC_LEN,
+				      GFP_DMA | GFP_KERNEL);
+	if (!ctx->desc_sync)
+		return -ENOMEM;
 
-	return 0;
-}
+	ctx->desc_async = devm_kzalloc(ctx->ctrldev, CAAM_RNG_DESC_LEN,
+				       GFP_DMA | GFP_KERNEL);
+	if (!ctx->desc_async)
+		return -ENOMEM;
 
-static int caam_init_rng(struct caam_rng_ctx *ctx, struct device *jrdev)
-{
-	int err;
+	if (kfifo_alloc(&ctx->fifo, CAAM_RNG_MAX_FIFO_STORE_SIZE,
+			GFP_DMA | GFP_KERNEL))
+		return -ENOMEM;
 
-	ctx->jrdev = jrdev;
+	INIT_WORK(&ctx->worker, caam_rng_worker);
 
-	err = rng_create_sh_desc(ctx);
-	if (err)
+	ctx->jrdev = caam_jr_alloc();
+	err = PTR_ERR_OR_ZERO(ctx->jrdev);
+	if (err) {
+		kfifo_free(&ctx->fifo);
+		pr_err("Job Ring Device allocation for transform failed\n");
 		return err;
+	}
 
-	ctx->current_buf = 0;
-	ctx->cur_buf_idx = 0;
+	/*
+	 * Fill async buffer to have early randomness data for
+	 * hw_random
+	 */
+	caam_rng_fill_async(ctx);
 
-	err = caam_init_buf(ctx, 0);
-	if (err)
-		return err;
-
-	return caam_init_buf(ctx, 1);
+	return 0;
 }
 
-static struct hwrng caam_rng = {
-	.name		= "rng-caam",
-	.cleanup	= caam_cleanup,
-	.read		= caam_read,
-};
+int caam_rng_init(struct device *ctrldev);
 
-void caam_rng_exit(void)
+void caam_rng_exit(struct device *ctrldev)
 {
-	if (!init_done)
-		return;
-
-	caam_jr_free(rng_ctx->jrdev);
-	hwrng_unregister(&caam_rng);
-	kfree(rng_ctx);
+	devres_release_group(ctrldev, caam_rng_init);
 }
 
 int caam_rng_init(struct device *ctrldev)
 {
-	struct device *dev;
+	struct caam_rng_ctx *ctx;
 	u32 rng_inst;
 	struct caam_drv_private *priv = dev_get_drvdata(ctrldev);
-	int err;
-	init_done = false;
+	int ret;
 
 	/* Check for an instantiated RNG before registration */
 	if (priv->era < 10)
@@ -328,31 +232,30 @@ int caam_rng_init(struct device *ctrldev)
 	if (!rng_inst)
 		return 0;
 
-	dev = caam_jr_alloc();
-	if (IS_ERR(dev)) {
-		pr_err("Job Ring Device allocation for transform failed\n");
-		return PTR_ERR(dev);
-	}
-	rng_ctx = kmalloc(sizeof(*rng_ctx), GFP_DMA | GFP_KERNEL);
-	if (!rng_ctx) {
-		err = -ENOMEM;
-		goto free_caam_alloc;
-	}
-	err = caam_init_rng(rng_ctx, dev);
-	if (err)
-		goto free_rng_ctx;
+	if (!devres_open_group(ctrldev, caam_rng_init, GFP_KERNEL))
+		return -ENOMEM;
 
-	dev_info(dev, "registering rng-caam\n");
+	ctx = devm_kzalloc(ctrldev, sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return -ENOMEM;
 
-	err = hwrng_register(&caam_rng);
-	if (!err) {
-		init_done = true;
-		return err;
+	ctx->ctrldev = ctrldev;
+
+	ctx->rng.name    = "rng-caam";
+	ctx->rng.init    = caam_init;
+	ctx->rng.cleanup = caam_cleanup;
+	ctx->rng.read    = caam_read;
+	ctx->rng.priv    = (unsigned long)ctx;
+	ctx->rng.quality = 1024;
+
+	dev_info(ctrldev, "registering rng-caam\n");
+
+	ret = devm_hwrng_register(ctrldev, &ctx->rng);
+	if (ret) {
+		caam_rng_exit(ctrldev);
+		return ret;
 	}
 
-free_rng_ctx:
-	kfree(rng_ctx);
-free_caam_alloc:
-	caam_jr_free(dev);
-	return err;
+	devres_close_group(ctrldev, caam_rng_init);
+	return 0;
 }