From 3c8c20db769cd68e299b487035825e026b1a6ce9 Mon Sep 17 00:00:00 2001
From: Edwin Peer <edwin.peer@broadcom.com>
Date: Sun, 29 Aug 2021 03:34:58 -0400
Subject: bnxt_en: move HWRM API implementation into separate file

Move all firmware messaging functions and definitions to new
bnxt_hwrm.[ch].  The follow-on patches will make major modifications
to these APIs.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h | 100 +++++++++++++++++++++++++
 1 file changed, 100 insertions(+)
 create mode 100644 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h

(limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h')

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
new file mode 100644
index 000000000000..940c792b54c7
--- /dev/null
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
@@ -0,0 +1,100 @@
+/* Broadcom NetXtreme-C/E network driver.
+ *
+ * Copyright (c) 2020 Broadcom Limited
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation.
+ */
+
+#ifndef BNXT_HWRM_H
+#define BNXT_HWRM_H
+
+#include "bnxt_hsi.h"
+
+#define BNXT_HWRM_MAX_REQ_LEN		(bp->hwrm_max_req_len)
+#define BNXT_HWRM_SHORT_REQ_LEN		sizeof(struct hwrm_short_input)
+#define HWRM_CMD_MAX_TIMEOUT		40000
+#define SHORT_HWRM_CMD_TIMEOUT		20
+#define HWRM_CMD_TIMEOUT		(bp->hwrm_cmd_timeout)
+#define HWRM_RESET_TIMEOUT		((HWRM_CMD_TIMEOUT) * 4)
+#define HWRM_COREDUMP_TIMEOUT		((HWRM_CMD_TIMEOUT) * 12)
+#define BNXT_HWRM_REQ_MAX_SIZE		128
+#define BNXT_HWRM_REQS_PER_PAGE		(BNXT_PAGE_SIZE /	\
+					 BNXT_HWRM_REQ_MAX_SIZE)
+#define HWRM_SHORT_MIN_TIMEOUT		3
+#define HWRM_SHORT_MAX_TIMEOUT		10
+#define HWRM_SHORT_TIMEOUT_COUNTER	5
+
+#define HWRM_MIN_TIMEOUT		25
+#define HWRM_MAX_TIMEOUT		40
+
+#define HWRM_WAIT_MUST_ABORT(bp, req)					\
+	(le16_to_cpu((req)->req_type) != HWRM_VER_GET &&		\
+	 !bnxt_is_fw_healthy(bp))
+
+#define HWRM_TOTAL_TIMEOUT(n)	(((n) <= HWRM_SHORT_TIMEOUT_COUNTER) ?	\
+	((n) * HWRM_SHORT_MIN_TIMEOUT) :				\
+	(HWRM_SHORT_TIMEOUT_COUNTER * HWRM_SHORT_MIN_TIMEOUT +		\
+	 ((n) - HWRM_SHORT_TIMEOUT_COUNTER) * HWRM_MIN_TIMEOUT))
+
+#define HWRM_VALID_BIT_DELAY_USEC	150
+
+#define BNXT_HWRM_CHNL_CHIMP	0
+#define BNXT_HWRM_CHNL_KONG	1
+
+static inline bool bnxt_cfa_hwrm_message(u16 req_type)
+{
+	switch (req_type) {
+	case HWRM_CFA_ENCAP_RECORD_ALLOC:
+	case HWRM_CFA_ENCAP_RECORD_FREE:
+	case HWRM_CFA_DECAP_FILTER_ALLOC:
+	case HWRM_CFA_DECAP_FILTER_FREE:
+	case HWRM_CFA_EM_FLOW_ALLOC:
+	case HWRM_CFA_EM_FLOW_FREE:
+	case HWRM_CFA_EM_FLOW_CFG:
+	case HWRM_CFA_FLOW_ALLOC:
+	case HWRM_CFA_FLOW_FREE:
+	case HWRM_CFA_FLOW_INFO:
+	case HWRM_CFA_FLOW_FLUSH:
+	case HWRM_CFA_FLOW_STATS:
+	case HWRM_CFA_METER_PROFILE_ALLOC:
+	case HWRM_CFA_METER_PROFILE_FREE:
+	case HWRM_CFA_METER_PROFILE_CFG:
+	case HWRM_CFA_METER_INSTANCE_ALLOC:
+	case HWRM_CFA_METER_INSTANCE_FREE:
+		return true;
+	default:
+		return false;
+	}
+}
+
+static inline bool bnxt_kong_hwrm_message(struct bnxt *bp, struct input *req)
+{
+	return (bp->fw_cap & BNXT_FW_CAP_KONG_MB_CHNL &&
+		(bnxt_cfa_hwrm_message(le16_to_cpu(req->req_type)) ||
+		 le16_to_cpu(req->target_id) == HWRM_TARGET_ID_KONG));
+}
+
+static inline void *bnxt_get_hwrm_resp_addr(struct bnxt *bp, void *req)
+{
+	return bp->hwrm_cmd_resp_addr;
+}
+
+static inline u16 bnxt_get_hwrm_seq_id(struct bnxt *bp, u16 dst)
+{
+	u16 seq_id;
+
+	if (dst == BNXT_HWRM_CHNL_CHIMP)
+		seq_id = bp->hwrm_cmd_seq++;
+	else
+		seq_id = bp->hwrm_cmd_kong_seq++;
+	return seq_id;
+}
+
+void bnxt_hwrm_cmd_hdr_init(struct bnxt *, void *, u16, u16, u16);
+int _hwrm_send_message(struct bnxt *bp, void *msg, u32 len, int timeout);
+int _hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 len, int timeout);
+int hwrm_send_message(struct bnxt *bp, void *msg, u32 len, int timeout);
+int hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 len, int timeout);
+#endif
-- 
cgit v1.2.3


From f9ff578251dc2f1cf5b9b007e050033d8414829d Mon Sep 17 00:00:00 2001
From: Edwin Peer <edwin.peer@broadcom.com>
Date: Sun, 29 Aug 2021 03:34:59 -0400
Subject: bnxt_en: introduce new firmware message API based on DMA pools

This change constitutes a major step towards supporting multiple
firmware commands in flight by maintaining a separate response buffer
for the duration of each request. These firmware commands are also
known as Hardware Resource Manager (HWRM) commands.  Using separate
response buffers requires an API change in order for callers to be
able to free the buffer when done.

It is impossible to keep the existing APIs unchanged.  The existing
usage for a simple HWRM message request such as the following:

        struct input req = {0};
        bnxt_hwrm_cmd_hdr_init(bp, &req, REQ_TYPE, -1, -1);
        rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
        if (rc)
                /* error */

changes to:

         struct input *req;
         rc = hwrm_req_init(bp, req, REQ_TYPE);
         if (rc)
                 /* error */
         rc = hwrm_req_send(bp, req); /* consumes req */
         if (rc)
                 /* error */

The key changes are:

1. The req is no longer allocated on the stack.
2. The caller must call hwrm_req_init() to allocate a req buffer and
   check for a valid buffer.
3. The req buffer is automatically released when hwrm_req_send() returns.
4. If the caller wants to check the firmware response, the caller must
   call hwrm_req_hold() to take ownership of the response buffer and
   release it afterwards using hwrm_req_drop().  The caller is no longer
   required to explicitly hold the hwrm_cmd_lock mutex to read the
   response.
5. Because the firmware commands and responses all have different sizes,
   some safeguards are added to the code.

This patch maintains legacy API compatibiltiy, implementing the old
API in terms of the new.  The follow-on patches will convert all
callers to use the new APIs.

v2: Fix redefined writeq with parisc .config
    Fix "cast from pointer to integer of different size" warning in
hwrm_calc_sentinel()

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c      |  42 +--
 drivers/net/ethernet/broadcom/bnxt/bnxt.h      |   7 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c | 460 +++++++++++++++++++++----
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h |  54 ++-
 4 files changed, 442 insertions(+), 121 deletions(-)

(limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h')

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 10c39801ad5f..23486f382b91 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -3963,6 +3963,9 @@ static void bnxt_free_hwrm_resources(struct bnxt *bp)
 				  bp->hwrm_cmd_resp_dma_addr);
 		bp->hwrm_cmd_resp_addr = NULL;
 	}
+
+	dma_pool_destroy(bp->hwrm_dma_pool);
+	bp->hwrm_dma_pool = NULL;
 }
 
 static int bnxt_alloc_hwrm_resources(struct bnxt *bp)
@@ -3975,33 +3978,10 @@ static int bnxt_alloc_hwrm_resources(struct bnxt *bp)
 	if (!bp->hwrm_cmd_resp_addr)
 		return -ENOMEM;
 
-	return 0;
-}
-
-static void bnxt_free_hwrm_short_cmd_req(struct bnxt *bp)
-{
-	if (bp->hwrm_short_cmd_req_addr) {
-		struct pci_dev *pdev = bp->pdev;
-
-		dma_free_coherent(&pdev->dev, bp->hwrm_max_ext_req_len,
-				  bp->hwrm_short_cmd_req_addr,
-				  bp->hwrm_short_cmd_req_dma_addr);
-		bp->hwrm_short_cmd_req_addr = NULL;
-	}
-}
-
-static int bnxt_alloc_hwrm_short_cmd_req(struct bnxt *bp)
-{
-	struct pci_dev *pdev = bp->pdev;
-
-	if (bp->hwrm_short_cmd_req_addr)
-		return 0;
-
-	bp->hwrm_short_cmd_req_addr =
-		dma_alloc_coherent(&pdev->dev, bp->hwrm_max_ext_req_len,
-				   &bp->hwrm_short_cmd_req_dma_addr,
-				   GFP_KERNEL);
-	if (!bp->hwrm_short_cmd_req_addr)
+	bp->hwrm_dma_pool = dma_pool_create("bnxt_hwrm", &pdev->dev,
+					    BNXT_HWRM_DMA_SIZE,
+					    BNXT_HWRM_DMA_ALIGN, 0);
+	if (!bp->hwrm_dma_pool)
 		return -ENOMEM;
 
 	return 0;
@@ -11654,12 +11634,6 @@ static int bnxt_fw_init_one_p1(struct bnxt *bp)
 			return rc;
 	}
 
-	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) ||
-	    bp->hwrm_max_ext_req_len > BNXT_HWRM_MAX_REQ_LEN) {
-		rc = bnxt_alloc_hwrm_short_cmd_req(bp);
-		if (rc)
-			return rc;
-	}
 	bnxt_nvm_cfg_ver_get(bp);
 
 	rc = bnxt_hwrm_func_reset(bp);
@@ -12588,7 +12562,6 @@ static void bnxt_remove_one(struct pci_dev *pdev)
 	bnxt_clear_int_mode(bp);
 	bnxt_hwrm_func_drv_unrgtr(bp);
 	bnxt_free_hwrm_resources(bp);
-	bnxt_free_hwrm_short_cmd_req(bp);
 	bnxt_ethtool_free(bp);
 	bnxt_dcb_free(bp);
 	kfree(bp->edev);
@@ -13188,7 +13161,6 @@ init_err_cleanup:
 
 init_err_pci_clean:
 	bnxt_hwrm_func_drv_unrgtr(bp);
-	bnxt_free_hwrm_short_cmd_req(bp);
 	bnxt_free_hwrm_resources(bp);
 	bnxt_ethtool_free(bp);
 	bnxt_ptp_clear(bp);
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index 5ff71eeffdd8..79a78a7468f3 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -1881,10 +1881,9 @@ struct bnxt {
 	u16			hwrm_cmd_seq;
 	u16                     hwrm_cmd_kong_seq;
 	u16			hwrm_intr_seq_id;
-	void			*hwrm_short_cmd_req_addr;
-	dma_addr_t		hwrm_short_cmd_req_dma_addr;
 	void			*hwrm_cmd_resp_addr;
 	dma_addr_t		hwrm_cmd_resp_dma_addr;
+	struct dma_pool		*hwrm_dma_pool;
 
 	struct rtnl_link_stats64	net_stats_prev;
 	struct bnxt_stats_mem	port_stats;
@@ -1984,7 +1983,7 @@ struct bnxt {
 	struct mutex		sriov_lock;
 #endif
 
-#if BITS_PER_LONG == 32
+#ifndef writeq
 	/* ensure atomic 64-bit doorbell writes on 32-bit systems. */
 	spinlock_t		db_lock;
 #endif
@@ -2113,7 +2112,7 @@ static inline u32 bnxt_tx_avail(struct bnxt *bp, struct bnxt_tx_ring_info *txr)
 		((txr->tx_prod - txr->tx_cons) & bp->tx_ring_mask);
 }
 
-#if BITS_PER_LONG == 32
+#ifndef writeq
 #define writeq(val64, db)			\
 do {						\
 	spin_lock(&bp->db_lock);		\
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
index b2a211b6cdd0..a45f2a619086 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
@@ -35,7 +35,220 @@ void bnxt_hwrm_cmd_hdr_init(struct bnxt *bp, void *request, u16 req_type,
 	req->resp_addr = cpu_to_le64(bp->hwrm_cmd_resp_dma_addr);
 }
 
-static int bnxt_hwrm_to_stderr(u32 hwrm_err)
+static u64 hwrm_calc_sentinel(struct bnxt_hwrm_ctx *ctx, u16 req_type)
+{
+	return (((uintptr_t)ctx) + req_type) ^ BNXT_HWRM_SENTINEL;
+}
+
+/**
+ * __hwrm_req_init() - Initialize an HWRM request.
+ * @bp: The driver context.
+ * @req: A pointer to the request pointer to initialize.
+ * @req_type: The request type. This will be converted to the little endian
+ *	before being written to the req_type field of the returned request.
+ * @req_len: The length of the request to be allocated.
+ *
+ * Allocate DMA resources and initialize a new HWRM request object of the
+ * given type. The response address field in the request is configured with
+ * the DMA bus address that has been mapped for the response and the passed
+ * request is pointed to kernel virtual memory mapped for the request (such
+ * that short_input indirection can be accomplished without copying). The
+ * request’s target and completion ring are initialized to default values and
+ * can be overridden by writing to the returned request object directly.
+ *
+ * The initialized request can be further customized by writing to its fields
+ * directly, taking care to covert such fields to little endian. The request
+ * object will be consumed (and all its associated resources release) upon
+ * passing it to hwrm_req_send() unless ownership of the request has been
+ * claimed by the caller via a call to hwrm_req_hold(). If the request is not
+ * consumed, either because it is never sent or because ownership has been
+ * claimed, then it must be released by a call to hwrm_req_drop().
+ *
+ * Return: zero on success, negative error code otherwise:
+ *	E2BIG: the type of request pointer is too large to fit.
+ *	ENOMEM: an allocation failure occurred.
+ */
+int __hwrm_req_init(struct bnxt *bp, void **req, u16 req_type, u32 req_len)
+{
+	struct bnxt_hwrm_ctx *ctx;
+	dma_addr_t dma_handle;
+	u8 *req_addr;
+
+	if (req_len > BNXT_HWRM_CTX_OFFSET)
+		return -E2BIG;
+
+	req_addr = dma_pool_alloc(bp->hwrm_dma_pool, GFP_KERNEL | __GFP_ZERO,
+				  &dma_handle);
+	if (!req_addr)
+		return -ENOMEM;
+
+	ctx = (struct bnxt_hwrm_ctx *)(req_addr + BNXT_HWRM_CTX_OFFSET);
+	/* safety first, sentinel used to check for invalid requests */
+	ctx->sentinel = hwrm_calc_sentinel(ctx, req_type);
+	ctx->req_len = req_len;
+	ctx->req = (struct input *)req_addr;
+	ctx->resp = (struct output *)(req_addr + BNXT_HWRM_RESP_OFFSET);
+	ctx->dma_handle = dma_handle;
+	ctx->flags = 0; /* __GFP_ZERO, but be explicit regarding ownership */
+	ctx->timeout = bp->hwrm_cmd_timeout ?: DFLT_HWRM_CMD_TIMEOUT;
+
+	/* initialize common request fields */
+	ctx->req->req_type = cpu_to_le16(req_type);
+	ctx->req->resp_addr = cpu_to_le64(dma_handle + BNXT_HWRM_RESP_OFFSET);
+	ctx->req->cmpl_ring = cpu_to_le16(BNXT_HWRM_NO_CMPL_RING);
+	ctx->req->target_id = cpu_to_le16(BNXT_HWRM_TARGET);
+	*req = ctx->req;
+
+	return 0;
+}
+
+static struct bnxt_hwrm_ctx *__hwrm_ctx(struct bnxt *bp, u8 *req_addr)
+{
+	void *ctx_addr = req_addr + BNXT_HWRM_CTX_OFFSET;
+	struct input *req = (struct input *)req_addr;
+	struct bnxt_hwrm_ctx *ctx = ctx_addr;
+	u64 sentinel;
+
+	if (!req) {
+		/* can only be due to software bug, be loud */
+		netdev_err(bp->dev, "null HWRM request");
+		dump_stack();
+		return NULL;
+	}
+
+	/* HWRM API has no type safety, verify sentinel to validate address */
+	sentinel = hwrm_calc_sentinel(ctx, le16_to_cpu(req->req_type));
+	if (ctx->sentinel != sentinel) {
+		/* can only be due to software bug, be loud */
+		netdev_err(bp->dev, "HWRM sentinel mismatch, req_type = %u\n",
+			   (u32)le16_to_cpu(req->req_type));
+		dump_stack();
+		return NULL;
+	}
+
+	return ctx;
+}
+
+/**
+ * hwrm_req_timeout() - Set the completion timeout for the request.
+ * @bp: The driver context.
+ * @req: The request to set the timeout.
+ * @timeout: The timeout in milliseconds.
+ *
+ * Set the timeout associated with the request for subsequent calls to
+ * hwrm_req_send(). Some requests are long running and require a different
+ * timeout than the default.
+ */
+void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+
+	if (ctx)
+		ctx->timeout = timeout;
+}
+
+/**
+ * hwrm_req_flags() - Set non internal flags of the ctx
+ * @bp: The driver context.
+ * @req: The request containing the HWRM command
+ * @flags: ctx flags that don't have BNXT_HWRM_INTERNAL_FLAG set
+ *
+ * ctx flags can be used by the callers to instruct how the subsequent
+ * hwrm_req_send() should behave. Example: callers can use hwrm_req_flags
+ * with BNXT_HWRM_CTX_SILENT to omit kernel prints of errors of hwrm_req_send()
+ * or with BNXT_HWRM_FULL_WAIT enforce hwrm_req_send() to wait for full timeout
+ * even if FW is not responding.
+ * This generic function can be used to set any flag that is not an internal flag
+ * of the HWRM module.
+ */
+void hwrm_req_flags(struct bnxt *bp, void *req, enum bnxt_hwrm_ctx_flags flags)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+
+	if (ctx)
+		ctx->flags |= (flags & HWRM_API_FLAGS);
+}
+
+/**
+ * hwrm_req_hold() - Claim ownership of the request's resources.
+ * @bp: The driver context.
+ * @req: A pointer to the request to own. The request will no longer be
+ *	consumed by calls to hwrm_req_send().
+ *
+ * Take ownership of the request. Ownership places responsibility on the
+ * caller to free the resources associated with the request via a call to
+ * hwrm_req_drop(). The caller taking ownership implies that a subsequent
+ * call to hwrm_req_send() will not consume the request (ie. sending will
+ * not free the associated resources if the request is owned by the caller).
+ * Taking ownership returns a reference to the response. Retaining and
+ * accessing the response data is the most common reason to take ownership
+ * of the request. Ownership can also be acquired in order to reuse the same
+ * request object across multiple invocations of hwrm_req_send().
+ *
+ * Return: A pointer to the response object.
+ *
+ * The resources associated with the response will remain available to the
+ * caller until ownership of the request is relinquished via a call to
+ * hwrm_req_drop(). It is not possible for hwrm_req_hold() to return NULL if
+ * a valid request is provided. A returned NULL value would imply a driver
+ * bug and the implementation will complain loudly in the logs to aid in
+ * detection. It should not be necessary to check the result for NULL.
+ */
+void *hwrm_req_hold(struct bnxt *bp, void *req)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+	struct input *input = (struct input *)req;
+
+	if (!ctx)
+		return NULL;
+
+	if (ctx->flags & BNXT_HWRM_INTERNAL_CTX_OWNED) {
+		/* can only be due to software bug, be loud */
+		netdev_err(bp->dev, "HWRM context already owned, req_type = %u\n",
+			   (u32)le16_to_cpu(input->req_type));
+		dump_stack();
+		return NULL;
+	}
+
+	ctx->flags |= BNXT_HWRM_INTERNAL_CTX_OWNED;
+	return ((u8 *)req) + BNXT_HWRM_RESP_OFFSET;
+}
+
+static void __hwrm_ctx_drop(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
+{
+	void *addr = ((u8 *)ctx) - BNXT_HWRM_CTX_OFFSET;
+	dma_addr_t dma_handle = ctx->dma_handle; /* save before invalidate */
+
+	/* invalidate, ensure ownership, sentinel and dma_handle are cleared */
+	memset(ctx, 0, sizeof(struct bnxt_hwrm_ctx));
+
+	/* return the buffer to the DMA pool */
+	if (dma_handle)
+		dma_pool_free(bp->hwrm_dma_pool, addr, dma_handle);
+}
+
+/**
+ * hwrm_req_drop() - Release all resources associated with the request.
+ * @bp: The driver context.
+ * @req: The request to consume, releasing the associated resources. The
+ *	request object and its associated response are no longer valid.
+ *
+ * It is legal to call hwrm_req_drop() on an unowned request, provided it
+ * has not already been consumed by hwrm_req_send() (for example, to release
+ * an aborted request). A given request should not be dropped more than once,
+ * nor should it be dropped after having been consumed by hwrm_req_send(). To
+ * do so is an error (the context will not be found and a stack trace will be
+ * rendered in the kernel log).
+ */
+void hwrm_req_drop(struct bnxt *bp, void *req)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+
+	if (ctx)
+		__hwrm_ctx_drop(bp, ctx);
+}
+
+static int __hwrm_to_stderr(u32 hwrm_err)
 {
 	switch (hwrm_err) {
 	case HWRM_ERR_CODE_SUCCESS:
@@ -64,78 +277,71 @@ static int bnxt_hwrm_to_stderr(u32 hwrm_err)
 	}
 }
 
-static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
-				 int timeout, bool silent)
+static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 {
-	int i, intr_process, rc, tmo_count;
-	struct input *req = msg;
-	u32 *data = msg;
-	u8 *valid;
-	u16 cp_ring_id, len = 0;
-	struct hwrm_err_output *resp = bp->hwrm_cmd_resp_addr;
-	u16 max_req_len = BNXT_HWRM_MAX_REQ_LEN;
-	struct hwrm_short_input short_input = {0};
 	u32 doorbell_offset = BNXT_GRCPF_REG_CHIMP_COMM_TRIGGER;
 	u32 bar_offset = BNXT_GRCPF_REG_CHIMP_COMM;
+	struct hwrm_short_input short_input = {0};
+	u16 max_req_len = BNXT_HWRM_MAX_REQ_LEN;
+	unsigned int i, timeout, tmo_count;
 	u16 dst = BNXT_HWRM_CHNL_CHIMP;
+	int intr_process, rc = -EBUSY;
+	u32 *data = (u32 *)ctx->req;
+	u32 msg_len = ctx->req_len;
+	u16 cp_ring_id, len = 0;
+	u32 req_type;
+	u8 *valid;
 
-	if (BNXT_NO_FW_ACCESS(bp) &&
-	    le16_to_cpu(req->req_type) != HWRM_FUNC_RESET)
-		return -EBUSY;
+	if (ctx->flags & BNXT_HWRM_INTERNAL_RESP_DIRTY)
+		memset(ctx->resp, 0, PAGE_SIZE);
 
-	if (msg_len > BNXT_HWRM_MAX_REQ_LEN) {
-		if (msg_len > bp->hwrm_max_ext_req_len ||
-		    !bp->hwrm_short_cmd_req_addr)
-			return -EINVAL;
+	req_type = le16_to_cpu(ctx->req->req_type);
+	if (BNXT_NO_FW_ACCESS(bp) && req_type != HWRM_FUNC_RESET)
+		goto exit;
+
+	if (msg_len > BNXT_HWRM_MAX_REQ_LEN &&
+	    msg_len > bp->hwrm_max_ext_req_len) {
+		rc = -E2BIG;
+		goto exit;
 	}
 
-	if (bnxt_kong_hwrm_message(bp, req)) {
+	if (bnxt_kong_hwrm_message(bp, ctx->req)) {
 		dst = BNXT_HWRM_CHNL_KONG;
 		bar_offset = BNXT_GRCPF_REG_KONG_COMM;
 		doorbell_offset = BNXT_GRCPF_REG_KONG_COMM_TRIGGER;
+		if (le16_to_cpu(ctx->req->cmpl_ring) != INVALID_HW_RING_ID) {
+			netdev_err(bp->dev, "Ring completions not supported for KONG commands, req_type = %d\n",
+				   req_type);
+			rc = -EINVAL;
+			goto exit;
+		}
 	}
 
-	memset(resp, 0, PAGE_SIZE);
-	cp_ring_id = le16_to_cpu(req->cmpl_ring);
+	cp_ring_id = le16_to_cpu(ctx->req->cmpl_ring);
 	intr_process = (cp_ring_id == INVALID_HW_RING_ID) ? 0 : 1;
 
-	req->seq_id = cpu_to_le16(bnxt_get_hwrm_seq_id(bp, dst));
+	ctx->req->seq_id = cpu_to_le16(bnxt_get_hwrm_seq_id(bp, dst));
 	/* currently supports only one outstanding message */
 	if (intr_process)
-		bp->hwrm_intr_seq_id = le16_to_cpu(req->seq_id);
+		bp->hwrm_intr_seq_id = le16_to_cpu(ctx->req->seq_id);
 
 	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) ||
 	    msg_len > BNXT_HWRM_MAX_REQ_LEN) {
-		void *short_cmd_req = bp->hwrm_short_cmd_req_addr;
-		u16 max_msg_len;
-
-		/* Set boundary for maximum extended request length for short
-		 * cmd format. If passed up from device use the max supported
-		 * internal req length.
-		 */
-		max_msg_len = bp->hwrm_max_ext_req_len;
-
-		memcpy(short_cmd_req, req, msg_len);
-		if (msg_len < max_msg_len)
-			memset(short_cmd_req + msg_len, 0,
-			       max_msg_len - msg_len);
-
-		short_input.req_type = req->req_type;
+		short_input.req_type = ctx->req->req_type;
 		short_input.signature =
 				cpu_to_le16(SHORT_REQ_SIGNATURE_SHORT_CMD);
 		short_input.size = cpu_to_le16(msg_len);
-		short_input.req_addr =
-			cpu_to_le64(bp->hwrm_short_cmd_req_dma_addr);
+		short_input.req_addr = cpu_to_le64(ctx->dma_handle);
 
 		data = (u32 *)&short_input;
 		msg_len = sizeof(short_input);
 
-		/* Sync memory write before updating doorbell */
-		wmb();
-
 		max_req_len = BNXT_HWRM_SHORT_REQ_LEN;
 	}
 
+	/* Ensure any associated DMA buffers are written before doorbell */
+	wmb();
+
 	/* Write request msg to hwrm channel */
 	__iowrite32_copy(bp->bar0 + bar_offset, data, msg_len / 4);
 
@@ -145,13 +351,13 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 	/* Ring channel doorbell */
 	writel(1, bp->bar0 + doorbell_offset);
 
-	if (!pci_is_enabled(bp->pdev))
-		return -ENODEV;
+	if (!pci_is_enabled(bp->pdev)) {
+		rc = -ENODEV;
+		goto exit;
+	}
 
-	if (!timeout)
-		timeout = DFLT_HWRM_CMD_TIMEOUT;
 	/* Limit timeout to an upper limit */
-	timeout = min(timeout, HWRM_CMD_MAX_TIMEOUT);
+	timeout = min_t(uint, ctx->timeout, HWRM_CMD_MAX_TIMEOUT);
 	/* convert timeout to usec */
 	timeout *= 1000;
 
@@ -174,13 +380,13 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 			 * check has failed.
 			 */
 			if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state))
-				return -EBUSY;
+				goto exit;
 			/* on first few passes, just barely sleep */
 			if (i < HWRM_SHORT_TIMEOUT_COUNTER) {
 				usleep_range(HWRM_SHORT_MIN_TIMEOUT,
 					     HWRM_SHORT_MAX_TIMEOUT);
 			} else {
-				if (HWRM_WAIT_MUST_ABORT(bp, req))
+				if (HWRM_WAIT_MUST_ABORT(bp, ctx))
 					break;
 				usleep_range(HWRM_MIN_TIMEOUT,
 					     HWRM_MAX_TIMEOUT);
@@ -188,13 +394,13 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 		}
 
 		if (bp->hwrm_intr_seq_id != (u16)~seq_id) {
-			if (!silent)
+			if (!(ctx->flags & BNXT_HWRM_CTX_SILENT))
 				netdev_err(bp->dev, "Resp cmpl intr err msg: 0x%x\n",
-					   le16_to_cpu(req->req_type));
-			return -EBUSY;
+					   le16_to_cpu(ctx->req->req_type));
+			goto exit;
 		}
-		len = le16_to_cpu(resp->resp_len);
-		valid = ((u8 *)resp) + len - 1;
+		len = le16_to_cpu(ctx->resp->resp_len);
+		valid = ((u8 *)ctx->resp) + len - 1;
 	} else {
 		int j;
 
@@ -204,8 +410,8 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 			 * check has failed.
 			 */
 			if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state))
-				return -EBUSY;
-			len = le16_to_cpu(resp->resp_len);
+				goto exit;
+			len = le16_to_cpu(ctx->resp->resp_len);
 			if (len)
 				break;
 			/* on first few passes, just barely sleep */
@@ -213,7 +419,7 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 				usleep_range(HWRM_SHORT_MIN_TIMEOUT,
 					     HWRM_SHORT_MAX_TIMEOUT);
 			} else {
-				if (HWRM_WAIT_MUST_ABORT(bp, req))
+				if (HWRM_WAIT_MUST_ABORT(bp, ctx))
 					goto timeout_abort;
 				usleep_range(HWRM_MIN_TIMEOUT,
 					     HWRM_MAX_TIMEOUT);
@@ -222,16 +428,16 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 
 		if (i >= tmo_count) {
 timeout_abort:
-			if (!silent)
-				netdev_err(bp->dev, "Error (timeout: %d) msg {0x%x 0x%x} len:%d\n",
-					   HWRM_TOTAL_TIMEOUT(i),
-					   le16_to_cpu(req->req_type),
-					   le16_to_cpu(req->seq_id), len);
-			return -EBUSY;
+			if (!(ctx->flags & BNXT_HWRM_CTX_SILENT))
+				netdev_err(bp->dev, "Error (timeout: %u) msg {0x%x 0x%x} len:%d\n",
+					   hwrm_total_timeout(i),
+					   le16_to_cpu(ctx->req->req_type),
+					   le16_to_cpu(ctx->req->seq_id), len);
+			goto exit;
 		}
 
 		/* Last byte of resp contains valid bit */
-		valid = ((u8 *)resp) + len - 1;
+		valid = ((u8 *)ctx->resp) + len - 1;
 		for (j = 0; j < HWRM_VALID_BIT_DELAY_USEC; j++) {
 			/* make sure we read from updated DMA memory */
 			dma_rmb();
@@ -241,13 +447,13 @@ timeout_abort:
 		}
 
 		if (j >= HWRM_VALID_BIT_DELAY_USEC) {
-			if (!silent)
-				netdev_err(bp->dev, "Error (timeout: %d) msg {0x%x 0x%x} len:%d v:%d\n",
-					   HWRM_TOTAL_TIMEOUT(i),
-					   le16_to_cpu(req->req_type),
-					   le16_to_cpu(req->seq_id), len,
+			if (!(ctx->flags & BNXT_HWRM_CTX_SILENT))
+				netdev_err(bp->dev, "Error (timeout: %u) msg {0x%x 0x%x} len:%d v:%d\n",
+					   hwrm_total_timeout(i),
+					   le16_to_cpu(ctx->req->req_type),
+					   le16_to_cpu(ctx->req->seq_id), len,
 					   *valid);
-			return -EBUSY;
+			goto exit;
 		}
 	}
 
@@ -256,12 +462,53 @@ timeout_abort:
 	 * a new field not implemented by old spec will read zero.
 	 */
 	*valid = 0;
-	rc = le16_to_cpu(resp->error_code);
-	if (rc && !silent)
+	rc = le16_to_cpu(ctx->resp->error_code);
+	if (rc && !(ctx->flags & BNXT_HWRM_CTX_SILENT)) {
 		netdev_err(bp->dev, "hwrm req_type 0x%x seq id 0x%x error 0x%x\n",
-			   le16_to_cpu(resp->req_type),
-			   le16_to_cpu(resp->seq_id), rc);
-	return bnxt_hwrm_to_stderr(rc);
+			   le16_to_cpu(ctx->resp->req_type),
+			   le16_to_cpu(ctx->resp->seq_id), rc);
+	}
+	rc = __hwrm_to_stderr(rc);
+exit:
+	if (ctx->flags & BNXT_HWRM_INTERNAL_CTX_OWNED)
+		ctx->flags |= BNXT_HWRM_INTERNAL_RESP_DIRTY;
+	else
+		__hwrm_ctx_drop(bp, ctx);
+	return rc;
+}
+
+static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
+				 int timeout, bool silent)
+{
+	struct bnxt_hwrm_ctx default_ctx = {0};
+	struct bnxt_hwrm_ctx *ctx = &default_ctx;
+	struct input *req = msg;
+	int rc;
+
+	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) ||
+	    msg_len > BNXT_HWRM_MAX_REQ_LEN) {
+		rc = __hwrm_req_init(bp, (void **)&req,
+				     le16_to_cpu(req->req_type), msg_len);
+		if (rc)
+			return rc;
+		memcpy(req, msg, msg_len); /* also copies resp_addr */
+		ctx = __hwrm_ctx(bp, (u8 *)req);
+		/* belts and brances, NULL ctx shouldn't be possible here */
+		if (!ctx)
+			return -ENOMEM;
+	}
+
+	ctx->req = req;
+	ctx->req_len = msg_len;
+	ctx->resp = bp->hwrm_cmd_resp_addr;
+	/* global response is not reallocated __GFP_ZERO between requests */
+	ctx->flags = BNXT_HWRM_INTERNAL_RESP_DIRTY;
+	ctx->timeout = timeout ?: DFLT_HWRM_CMD_TIMEOUT;
+	if (silent)
+		ctx->flags |= BNXT_HWRM_CTX_SILENT;
+
+	/* will consume req if allocated with __hwrm_req_init() */
+	return __hwrm_send(bp, ctx);
 }
 
 int _hwrm_send_message(struct bnxt *bp, void *msg, u32 msg_len, int timeout)
@@ -295,3 +542,64 @@ int hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 msg_len,
 	mutex_unlock(&bp->hwrm_cmd_lock);
 	return rc;
 }
+
+/**
+ * hwrm_req_send() - Execute an HWRM command.
+ * @bp: The driver context.
+ * @req: A pointer to the request to send. The DMA resources associated with
+ *	the request will be released (ie. the request will be consumed) unless
+ *	ownership of the request has been assumed by the caller via a call to
+ *	hwrm_req_hold().
+ *
+ * Send an HWRM request to the device and wait for a response. The request is
+ * consumed if it is not owned by the caller. This function will block until
+ * the request has either completed or times out due to an error.
+ *
+ * Return: A result code.
+ *
+ * The result is zero on success, otherwise the negative error code indicates
+ * one of the following errors:
+ *	E2BIG: The request was too large.
+ *	EBUSY: The firmware is in a fatal state or the request timed out
+ *	EACCESS: HWRM access denied.
+ *	ENOSPC: HWRM resource allocation error.
+ *	EINVAL: Request parameters are invalid.
+ *	ENOMEM: HWRM has no buffers.
+ *	EAGAIN: HWRM busy or reset in progress.
+ *	EOPNOTSUPP: Invalid request type.
+ *	EIO: Any other error.
+ * Error handling is orthogonal to request ownership. An unowned request will
+ * still be consumed on error. If the caller owns the request, then the caller
+ * is responsible for releasing the resources. Otherwise, hwrm_req_send() will
+ * always consume the request.
+ */
+int hwrm_req_send(struct bnxt *bp, void *req)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+	int rc;
+
+	if (!ctx)
+		return -EINVAL;
+
+	mutex_lock(&bp->hwrm_cmd_lock);
+	rc = __hwrm_send(bp, ctx);
+	mutex_unlock(&bp->hwrm_cmd_lock);
+	return rc;
+}
+
+/**
+ * hwrm_req_send_silent() - A silent version of hwrm_req_send().
+ * @bp: The driver context.
+ * @req: The request to send without logging.
+ *
+ * The same as hwrm_req_send(), except that the request is silenced using
+ * hwrm_req_silence() prior the call. This version of the function is
+ * provided solely to preserve the legacy API’s flavor for this functionality.
+ *
+ * Return: A result code, see hwrm_req_send().
+ */
+int hwrm_req_send_silent(struct bnxt *bp, void *req)
+{
+	hwrm_req_flags(bp, req, BNXT_HWRM_CTX_SILENT);
+	return hwrm_req_send(bp, req);
+}
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
index 940c792b54c7..199c646f5e71 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
@@ -12,6 +12,26 @@
 
 #include "bnxt_hsi.h"
 
+enum bnxt_hwrm_ctx_flags {
+	/* Update the HWRM_API_FLAGS right below for any new non-internal bit added here */
+	BNXT_HWRM_INTERNAL_CTX_OWNED	= BIT(0), /* caller owns the context */
+	BNXT_HWRM_INTERNAL_RESP_DIRTY	= BIT(1), /* response contains data */
+	BNXT_HWRM_CTX_SILENT		= BIT(2), /* squelch firmware errors */
+	BNXT_HWRM_FULL_WAIT		= BIT(3), /* wait for full timeout of HWRM command */
+};
+
+#define HWRM_API_FLAGS (BNXT_HWRM_CTX_SILENT | BNXT_HWRM_FULL_WAIT)
+
+struct bnxt_hwrm_ctx {
+	u64 sentinel;
+	dma_addr_t dma_handle;
+	struct output *resp;
+	struct input *req;
+	u32 req_len;
+	enum bnxt_hwrm_ctx_flags flags;
+	unsigned int timeout;
+};
+
 #define BNXT_HWRM_MAX_REQ_LEN		(bp->hwrm_max_req_len)
 #define BNXT_HWRM_SHORT_REQ_LEN		sizeof(struct hwrm_short_input)
 #define HWRM_CMD_MAX_TIMEOUT		40000
@@ -19,7 +39,17 @@
 #define HWRM_CMD_TIMEOUT		(bp->hwrm_cmd_timeout)
 #define HWRM_RESET_TIMEOUT		((HWRM_CMD_TIMEOUT) * 4)
 #define HWRM_COREDUMP_TIMEOUT		((HWRM_CMD_TIMEOUT) * 12)
+#define BNXT_HWRM_TARGET		0xffff
+#define BNXT_HWRM_NO_CMPL_RING		-1
 #define BNXT_HWRM_REQ_MAX_SIZE		128
+#define BNXT_HWRM_DMA_SIZE		(2 * PAGE_SIZE) /* space for req+resp */
+#define BNXT_HWRM_RESP_RESERVED		PAGE_SIZE
+#define BNXT_HWRM_RESP_OFFSET		(BNXT_HWRM_DMA_SIZE -		\
+					 BNXT_HWRM_RESP_RESERVED)
+#define BNXT_HWRM_CTX_OFFSET		(BNXT_HWRM_RESP_OFFSET -	\
+					 sizeof(struct bnxt_hwrm_ctx))
+#define BNXT_HWRM_DMA_ALIGN		16
+#define BNXT_HWRM_SENTINEL		0xb6e1f68a12e9a7eb /* arbitrary value */
 #define BNXT_HWRM_REQS_PER_PAGE		(BNXT_PAGE_SIZE /	\
 					 BNXT_HWRM_REQ_MAX_SIZE)
 #define HWRM_SHORT_MIN_TIMEOUT		3
@@ -29,14 +59,17 @@
 #define HWRM_MIN_TIMEOUT		25
 #define HWRM_MAX_TIMEOUT		40
 
-#define HWRM_WAIT_MUST_ABORT(bp, req)					\
-	(le16_to_cpu((req)->req_type) != HWRM_VER_GET &&		\
+#define HWRM_WAIT_MUST_ABORT(bp, ctx)					\
+	(le16_to_cpu((ctx)->req->req_type) != HWRM_VER_GET &&		\
 	 !bnxt_is_fw_healthy(bp))
 
-#define HWRM_TOTAL_TIMEOUT(n)	(((n) <= HWRM_SHORT_TIMEOUT_COUNTER) ?	\
-	((n) * HWRM_SHORT_MIN_TIMEOUT) :				\
-	(HWRM_SHORT_TIMEOUT_COUNTER * HWRM_SHORT_MIN_TIMEOUT +		\
-	 ((n) - HWRM_SHORT_TIMEOUT_COUNTER) * HWRM_MIN_TIMEOUT))
+static inline unsigned int hwrm_total_timeout(unsigned int n)
+{
+	return n <= HWRM_SHORT_TIMEOUT_COUNTER ? n * HWRM_SHORT_MIN_TIMEOUT :
+		HWRM_SHORT_TIMEOUT_COUNTER * HWRM_SHORT_MIN_TIMEOUT +
+		(n - HWRM_SHORT_TIMEOUT_COUNTER) * HWRM_MIN_TIMEOUT;
+}
+
 
 #define HWRM_VALID_BIT_DELAY_USEC	150
 
@@ -97,4 +130,13 @@ int _hwrm_send_message(struct bnxt *bp, void *msg, u32 len, int timeout);
 int _hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 len, int timeout);
 int hwrm_send_message(struct bnxt *bp, void *msg, u32 len, int timeout);
 int hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 len, int timeout);
+int __hwrm_req_init(struct bnxt *bp, void **req, u16 req_type, u32 req_len);
+#define hwrm_req_init(bp, req, req_type) \
+	__hwrm_req_init((bp), (void **)&(req), (req_type), sizeof(*(req)))
+void *hwrm_req_hold(struct bnxt *bp, void *req);
+void hwrm_req_drop(struct bnxt *bp, void *req);
+void hwrm_req_flags(struct bnxt *bp, void *req, enum bnxt_hwrm_ctx_flags flags);
+void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout);
+int hwrm_req_send(struct bnxt *bp, void *req);
+int hwrm_req_send_silent(struct bnxt *bp, void *req);
 #endif
-- 
cgit v1.2.3


From ecddc29d928d0ecccbc8f339b59ed75e5c8e8ecf Mon Sep 17 00:00:00 2001
From: Edwin Peer <edwin.peer@broadcom.com>
Date: Sun, 29 Aug 2021 03:35:01 -0400
Subject: bnxt_en: add HWRM request assignment API

hwrm_req_replace() provides an assignment like operation to replace a
managed HWRM request object with data from a pre-built source. This is
useful for handling request data provided by higher layer HWRM clients.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c | 55 ++++++++++++++++++++++++++
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h |  1 +
 2 files changed, 56 insertions(+)

(limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h')

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
index 621daf687a00..39ef65025e17 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
@@ -147,6 +147,61 @@ void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout)
 		ctx->timeout = timeout;
 }
 
+/**
+ * hwrm_req_replace() - Replace request data.
+ * @bp: The driver context.
+ * @req: The request to modify. A call to hwrm_req_replace() is conceptually
+ *	an assignment of new_req to req. Subsequent calls to HWRM API functions,
+ *	such as hwrm_req_send(), should thus use req and not new_req (in fact,
+ *	calls to HWRM API functions will fail if non-managed request objects
+ *	are passed).
+ * @len: The length of new_req.
+ * @new_req: The pre-built request to copy or reference.
+ *
+ * Replaces the request data in req with that of new_req. This is useful in
+ * scenarios where a request object has already been constructed by a third
+ * party prior to creating a resource managed request using hwrm_req_init().
+ * Depending on the length, hwrm_req_replace() will either copy the new
+ * request data into the DMA memory allocated for req, or it will simply
+ * reference the new request and use it in lieu of req during subsequent
+ * calls to hwrm_req_send(). The resource management is associated with
+ * req and is independent of and does not apply to new_req. The caller must
+ * ensure that the lifetime of new_req is least as long as req.
+ *
+ * Return: zero on success, negative error code otherwise:
+ *     E2BIG: Request is too large.
+ *     EINVAL: Invalid request to modify.
+ */
+int hwrm_req_replace(struct bnxt *bp, void *req, void *new_req, u32 len)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+	struct input *internal_req = req;
+	u16 req_type;
+
+	if (!ctx)
+		return -EINVAL;
+
+	if (len > BNXT_HWRM_CTX_OFFSET)
+		return -E2BIG;
+
+	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) || len > BNXT_HWRM_MAX_REQ_LEN) {
+		memcpy(internal_req, new_req, len);
+	} else {
+		internal_req->req_type = ((struct input *)new_req)->req_type;
+		ctx->req = new_req;
+	}
+
+	ctx->req_len = len;
+	ctx->req->resp_addr = cpu_to_le64(ctx->dma_handle +
+					  BNXT_HWRM_RESP_OFFSET);
+
+	/* update sentinel for potentially new request type */
+	req_type = le16_to_cpu(internal_req->req_type);
+	ctx->sentinel = hwrm_calc_sentinel(ctx, req_type);
+
+	return 0;
+}
+
 /**
  * hwrm_req_flags() - Set non internal flags of the ctx
  * @bp: The driver context.
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
index 199c646f5e71..c58d84cc692a 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
@@ -139,4 +139,5 @@ void hwrm_req_flags(struct bnxt *bp, void *req, enum bnxt_hwrm_ctx_flags flags);
 void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout);
 int hwrm_req_send(struct bnxt *bp, void *req);
 int hwrm_req_send_silent(struct bnxt *bp, void *req);
+int hwrm_req_replace(struct bnxt *bp, void *req, void *new_req, u32 len);
 #endif
-- 
cgit v1.2.3


From 2138081708405fb9c16a76a9b6ef46c35d3f17a9 Mon Sep 17 00:00:00 2001
From: Edwin Peer <edwin.peer@broadcom.com>
Date: Sun, 29 Aug 2021 03:35:02 -0400
Subject: bnxt_en: add support for HWRM request slices

Slices are a mechanism for suballocating DMA mapped regions from the
request buffer. Such regions can be used for indirect command data
instead of creating new mappings with dma_alloc_coherent().

The advantage of using a slice is that the lifetime of the slice is
bound to the request and will be automatically unmapped when the
request is consumed.

A single external region is also supported. This allows for regions
that will not fit inside the spare request buffer space such that
the same API can be used consistently even for larger mappings.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c | 115 ++++++++++++++++++++++++-
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h |   7 ++
 2 files changed, 120 insertions(+), 2 deletions(-)

(limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h')

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
index 39ef65025e17..6609a86d5226 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
@@ -91,6 +91,9 @@ int __hwrm_req_init(struct bnxt *bp, void **req, u16 req_type, u32 req_len)
 	ctx->dma_handle = dma_handle;
 	ctx->flags = 0; /* __GFP_ZERO, but be explicit regarding ownership */
 	ctx->timeout = bp->hwrm_cmd_timeout ?: DFLT_HWRM_CMD_TIMEOUT;
+	ctx->allocated = BNXT_HWRM_DMA_SIZE - BNXT_HWRM_CTX_OFFSET;
+	ctx->gfp = GFP_KERNEL;
+	ctx->slice_addr = NULL;
 
 	/* initialize common request fields */
 	ctx->req->req_type = cpu_to_le16(req_type);
@@ -147,6 +150,29 @@ void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout)
 		ctx->timeout = timeout;
 }
 
+/**
+ * hwrm_req_alloc_flags() - Sets GFP allocation flags for slices.
+ * @bp: The driver context.
+ * @req: The request for which calls to hwrm_req_dma_slice() will have altered
+ *	allocation flags.
+ * @flags: A bitmask of GFP flags. These flags are passed to
+ *	dma_alloc_coherent() whenever it is used to allocate backing memory
+ *	for slices. Note that calls to hwrm_req_dma_slice() will not always
+ *	result in new allocations, however, memory suballocated from the
+ *	request buffer is already __GFP_ZERO.
+ *
+ * Sets the GFP allocation flags associated with the request for subsequent
+ * calls to hwrm_req_dma_slice(). This can be useful for specifying __GFP_ZERO
+ * for slice allocations.
+ */
+void hwrm_req_alloc_flags(struct bnxt *bp, void *req, gfp_t gfp)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+
+	if (ctx)
+		ctx->gfp = gfp;
+}
+
 /**
  * hwrm_req_replace() - Replace request data.
  * @bp: The driver context.
@@ -166,7 +192,8 @@ void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout)
  * reference the new request and use it in lieu of req during subsequent
  * calls to hwrm_req_send(). The resource management is associated with
  * req and is independent of and does not apply to new_req. The caller must
- * ensure that the lifetime of new_req is least as long as req.
+ * ensure that the lifetime of new_req is least as long as req. Any slices
+ * that may have been associated with the original request are released.
  *
  * Return: zero on success, negative error code otherwise:
  *     E2BIG: Request is too large.
@@ -184,6 +211,15 @@ int hwrm_req_replace(struct bnxt *bp, void *req, void *new_req, u32 len)
 	if (len > BNXT_HWRM_CTX_OFFSET)
 		return -E2BIG;
 
+	/* free any existing slices */
+	ctx->allocated = BNXT_HWRM_DMA_SIZE - BNXT_HWRM_CTX_OFFSET;
+	if (ctx->slice_addr) {
+		dma_free_coherent(&bp->pdev->dev, ctx->slice_size,
+				  ctx->slice_addr, ctx->slice_handle);
+		ctx->slice_addr = NULL;
+	}
+	ctx->gfp = GFP_KERNEL;
+
 	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) || len > BNXT_HWRM_MAX_REQ_LEN) {
 		memcpy(internal_req, new_req, len);
 	} else {
@@ -274,6 +310,11 @@ static void __hwrm_ctx_drop(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 	void *addr = ((u8 *)ctx) - BNXT_HWRM_CTX_OFFSET;
 	dma_addr_t dma_handle = ctx->dma_handle; /* save before invalidate */
 
+	/* unmap any auxiliary DMA slice */
+	if (ctx->slice_addr)
+		dma_free_coherent(&bp->pdev->dev, ctx->slice_size,
+				  ctx->slice_addr, ctx->slice_handle);
+
 	/* invalidate, ensure ownership, sentinel and dma_handle are cleared */
 	memset(ctx, 0, sizeof(struct bnxt_hwrm_ctx));
 
@@ -286,7 +327,8 @@ static void __hwrm_ctx_drop(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
  * hwrm_req_drop() - Release all resources associated with the request.
  * @bp: The driver context.
  * @req: The request to consume, releasing the associated resources. The
- *	request object and its associated response are no longer valid.
+ *	request object, any slices, and its associated response are no
+ *	longer valid.
  *
  * It is legal to call hwrm_req_drop() on an unowned request, provided it
  * has not already been consumed by hwrm_req_send() (for example, to release
@@ -671,3 +713,72 @@ int hwrm_req_send_silent(struct bnxt *bp, void *req)
 	hwrm_req_flags(bp, req, BNXT_HWRM_CTX_SILENT);
 	return hwrm_req_send(bp, req);
 }
+
+/**
+ * hwrm_req_dma_slice() - Allocate a slice of DMA mapped memory.
+ * @bp: The driver context.
+ * @req: The request for which indirect data will be associated.
+ * @size: The size of the allocation.
+ * @dma: The bus address associated with the allocation. The HWRM API has no
+ *	knowledge about the type of the request and so cannot infer how the
+ *	caller intends to use the indirect data. Thus, the caller is
+ *	responsible for configuring the request object appropriately to
+ *	point to the associated indirect memory. Note, DMA handle has the
+ *	same definition as it does in dma_alloc_coherent(), the caller is
+ *	responsible for endian conversions via cpu_to_le64() before assigning
+ *	this address.
+ *
+ * Allocates DMA mapped memory for indirect data related to a request. The
+ * lifetime of the DMA resources will be bound to that of the request (ie.
+ * they will be automatically released when the request is either consumed by
+ * hwrm_req_send() or dropped by hwrm_req_drop()). Small allocations are
+ * efficiently suballocated out of the request buffer space, hence the name
+ * slice, while larger requests are satisfied via an underlying call to
+ * dma_alloc_coherent(). Multiple suballocations are supported, however, only
+ * one externally mapped region is.
+ *
+ * Return: The kernel virtual address of the DMA mapping.
+ */
+void *
+hwrm_req_dma_slice(struct bnxt *bp, void *req, u32 size, dma_addr_t *dma_handle)
+{
+	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
+	u8 *end = ((u8 *)req) + BNXT_HWRM_DMA_SIZE;
+	struct input *input = req;
+	u8 *addr, *req_addr = req;
+	u32 max_offset, offset;
+
+	if (!ctx)
+		return NULL;
+
+	max_offset = BNXT_HWRM_DMA_SIZE - ctx->allocated;
+	offset = max_offset - size;
+	offset = ALIGN_DOWN(offset, BNXT_HWRM_DMA_ALIGN);
+	addr = req_addr + offset;
+
+	if (addr < req_addr + max_offset && req_addr + ctx->req_len <= addr) {
+		ctx->allocated = end - addr;
+		*dma_handle = ctx->dma_handle + offset;
+		return addr;
+	}
+
+	/* could not suballocate from ctx buffer, try create a new mapping */
+	if (ctx->slice_addr) {
+		/* if one exists, can only be due to software bug, be loud */
+		netdev_err(bp->dev, "HWRM refusing to reallocate DMA slice, req_type = %u\n",
+			   (u32)le16_to_cpu(input->req_type));
+		dump_stack();
+		return NULL;
+	}
+
+	addr = dma_alloc_coherent(&bp->pdev->dev, size, dma_handle, ctx->gfp);
+
+	if (!addr)
+		return NULL;
+
+	ctx->slice_addr = addr;
+	ctx->slice_size = size;
+	ctx->slice_handle = *dma_handle;
+
+	return addr;
+}
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
index c58d84cc692a..b3af7a88e2c7 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
@@ -27,9 +27,14 @@ struct bnxt_hwrm_ctx {
 	dma_addr_t dma_handle;
 	struct output *resp;
 	struct input *req;
+	dma_addr_t slice_handle;
+	void *slice_addr;
+	u32 slice_size;
 	u32 req_len;
 	enum bnxt_hwrm_ctx_flags flags;
 	unsigned int timeout;
+	u32 allocated;
+	gfp_t gfp;
 };
 
 #define BNXT_HWRM_MAX_REQ_LEN		(bp->hwrm_max_req_len)
@@ -140,4 +145,6 @@ void hwrm_req_timeout(struct bnxt *bp, void *req, unsigned int timeout);
 int hwrm_req_send(struct bnxt *bp, void *req);
 int hwrm_req_send_silent(struct bnxt *bp, void *req);
 int hwrm_req_replace(struct bnxt *bp, void *req, void *new_req, u32 len);
+void hwrm_req_alloc_flags(struct bnxt *bp, void *req, gfp_t flags);
+void *hwrm_req_dma_slice(struct bnxt *bp, void *req, u32 size, dma_addr_t *dma);
 #endif
-- 
cgit v1.2.3


From b34695a894b88e50e16dd3dcb1098fe919023f14 Mon Sep 17 00:00:00 2001
From: Edwin Peer <edwin.peer@broadcom.com>
Date: Sun, 29 Aug 2021 03:35:05 -0400
Subject: bnxt_en: remove legacy HWRM interface

There are no longer any callers relying on the old API.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c      | 18 +-----
 drivers/net/ethernet/broadcom/bnxt/bnxt.h      |  2 -
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c | 77 --------------------------
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h | 10 ----
 4 files changed, 1 insertion(+), 106 deletions(-)

(limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h')

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 6e9ca9b59e51..ddec1163748d 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -3956,29 +3956,13 @@ out:
 
 static void bnxt_free_hwrm_resources(struct bnxt *bp)
 {
-	struct pci_dev *pdev = bp->pdev;
-
-	if (bp->hwrm_cmd_resp_addr) {
-		dma_free_coherent(&pdev->dev, PAGE_SIZE, bp->hwrm_cmd_resp_addr,
-				  bp->hwrm_cmd_resp_dma_addr);
-		bp->hwrm_cmd_resp_addr = NULL;
-	}
-
 	dma_pool_destroy(bp->hwrm_dma_pool);
 	bp->hwrm_dma_pool = NULL;
 }
 
 static int bnxt_alloc_hwrm_resources(struct bnxt *bp)
 {
-	struct pci_dev *pdev = bp->pdev;
-
-	bp->hwrm_cmd_resp_addr = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
-						   &bp->hwrm_cmd_resp_dma_addr,
-						   GFP_KERNEL);
-	if (!bp->hwrm_cmd_resp_addr)
-		return -ENOMEM;
-
-	bp->hwrm_dma_pool = dma_pool_create("bnxt_hwrm", &pdev->dev,
+	bp->hwrm_dma_pool = dma_pool_create("bnxt_hwrm", &bp->pdev->dev,
 					    BNXT_HWRM_DMA_SIZE,
 					    BNXT_HWRM_DMA_ALIGN, 0);
 	if (!bp->hwrm_dma_pool)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index 79a78a7468f3..f343e87bef0b 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -1881,8 +1881,6 @@ struct bnxt {
 	u16			hwrm_cmd_seq;
 	u16                     hwrm_cmd_kong_seq;
 	u16			hwrm_intr_seq_id;
-	void			*hwrm_cmd_resp_addr;
-	dma_addr_t		hwrm_cmd_resp_dma_addr;
 	struct dma_pool		*hwrm_dma_pool;
 
 	struct rtnl_link_stats64	net_stats_prev;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
index 6609a86d5226..60ec0caa5c56 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
@@ -24,17 +24,6 @@
 #include "bnxt.h"
 #include "bnxt_hwrm.h"
 
-void bnxt_hwrm_cmd_hdr_init(struct bnxt *bp, void *request, u16 req_type,
-			    u16 cmpl_ring, u16 target_id)
-{
-	struct input *req = request;
-
-	req->req_type = cpu_to_le16(req_type);
-	req->cmpl_ring = cpu_to_le16(cmpl_ring);
-	req->target_id = cpu_to_le16(target_id);
-	req->resp_addr = cpu_to_le64(bp->hwrm_cmd_resp_dma_addr);
-}
-
 static u64 hwrm_calc_sentinel(struct bnxt_hwrm_ctx *ctx, u16 req_type)
 {
 	return (((uintptr_t)ctx) + req_type) ^ BNXT_HWRM_SENTINEL;
@@ -587,72 +576,6 @@ exit:
 	return rc;
 }
 
-static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
-				 int timeout, bool silent)
-{
-	struct bnxt_hwrm_ctx default_ctx = {0};
-	struct bnxt_hwrm_ctx *ctx = &default_ctx;
-	struct input *req = msg;
-	int rc;
-
-	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) ||
-	    msg_len > BNXT_HWRM_MAX_REQ_LEN) {
-		rc = __hwrm_req_init(bp, (void **)&req,
-				     le16_to_cpu(req->req_type), msg_len);
-		if (rc)
-			return rc;
-		memcpy(req, msg, msg_len); /* also copies resp_addr */
-		ctx = __hwrm_ctx(bp, (u8 *)req);
-		/* belts and brances, NULL ctx shouldn't be possible here */
-		if (!ctx)
-			return -ENOMEM;
-	}
-
-	ctx->req = req;
-	ctx->req_len = msg_len;
-	ctx->resp = bp->hwrm_cmd_resp_addr;
-	/* global response is not reallocated __GFP_ZERO between requests */
-	ctx->flags = BNXT_HWRM_INTERNAL_RESP_DIRTY;
-	ctx->timeout = timeout ?: DFLT_HWRM_CMD_TIMEOUT;
-	if (silent)
-		ctx->flags |= BNXT_HWRM_CTX_SILENT;
-
-	/* will consume req if allocated with __hwrm_req_init() */
-	return __hwrm_send(bp, ctx);
-}
-
-int _hwrm_send_message(struct bnxt *bp, void *msg, u32 msg_len, int timeout)
-{
-	return bnxt_hwrm_do_send_msg(bp, msg, msg_len, timeout, false);
-}
-
-int _hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 msg_len,
-			      int timeout)
-{
-	return bnxt_hwrm_do_send_msg(bp, msg, msg_len, timeout, true);
-}
-
-int hwrm_send_message(struct bnxt *bp, void *msg, u32 msg_len, int timeout)
-{
-	int rc;
-
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, msg, msg_len, timeout);
-	mutex_unlock(&bp->hwrm_cmd_lock);
-	return rc;
-}
-
-int hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 msg_len,
-			     int timeout)
-{
-	int rc;
-
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = bnxt_hwrm_do_send_msg(bp, msg, msg_len, timeout, true);
-	mutex_unlock(&bp->hwrm_cmd_lock);
-	return rc;
-}
-
 /**
  * hwrm_req_send() - Execute an HWRM command.
  * @bp: The driver context.
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
index b3af7a88e2c7..39032cf66258 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
@@ -114,11 +114,6 @@ static inline bool bnxt_kong_hwrm_message(struct bnxt *bp, struct input *req)
 		 le16_to_cpu(req->target_id) == HWRM_TARGET_ID_KONG));
 }
 
-static inline void *bnxt_get_hwrm_resp_addr(struct bnxt *bp, void *req)
-{
-	return bp->hwrm_cmd_resp_addr;
-}
-
 static inline u16 bnxt_get_hwrm_seq_id(struct bnxt *bp, u16 dst)
 {
 	u16 seq_id;
@@ -130,11 +125,6 @@ static inline u16 bnxt_get_hwrm_seq_id(struct bnxt *bp, u16 dst)
 	return seq_id;
 }
 
-void bnxt_hwrm_cmd_hdr_init(struct bnxt *, void *, u16, u16, u16);
-int _hwrm_send_message(struct bnxt *bp, void *msg, u32 len, int timeout);
-int _hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 len, int timeout);
-int hwrm_send_message(struct bnxt *bp, void *msg, u32 len, int timeout);
-int hwrm_send_message_silent(struct bnxt *bp, void *msg, u32 len, int timeout);
 int __hwrm_req_init(struct bnxt *bp, void **req, u16 req_type, u32 req_len);
 #define hwrm_req_init(bp, req, req_type) \
 	__hwrm_req_init((bp), (void **)&(req), (req_type), sizeof(*(req)))
-- 
cgit v1.2.3


From 68f684e257d7f3a6303b0e838bfa982c74f2c8da Mon Sep 17 00:00:00 2001
From: Edwin Peer <edwin.peer@broadcom.com>
Date: Sun, 29 Aug 2021 03:35:06 -0400
Subject: bnxt_en: support multiple HWRM commands in flight

Add infrastructure to maintain a pending list of HWRM commands awaiting
completion and reduce the scope of the hwrm_cmd_lock mutex so that it
protects only the request mailbox. The mailbox is free to use for one
or more concurrent commands after receiving deferred response events.

For uniformity and completeness, use the same pending list for
collecting completions for commands that respond via a completion ring.
These commands are only used for freeing rings and for IRQ test and
we only support one such command in flight.

Note deferred responses are also only supported on the main channel.
The secondary channel (KONG) does not support deferred responses.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c      | 21 ++++--
 drivers/net/ethernet/broadcom/bnxt/bnxt.h      |  2 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c | 96 ++++++++++++++++++++------
 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h | 33 +++++----
 4 files changed, 113 insertions(+), 39 deletions(-)

(limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h')

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index ddec1163748d..627f85ee3922 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -277,6 +277,7 @@ static const u16 bnxt_async_events_arr[] = {
 	ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY,
 	ASYNC_EVENT_CMPL_EVENT_ID_ERROR_RECOVERY,
 	ASYNC_EVENT_CMPL_EVENT_ID_DEBUG_NOTIFICATION,
+	ASYNC_EVENT_CMPL_EVENT_ID_DEFERRED_RESPONSE,
 	ASYNC_EVENT_CMPL_EVENT_ID_RING_MONITOR_MSG,
 	ASYNC_EVENT_CMPL_EVENT_ID_ECHO_REQUEST,
 	ASYNC_EVENT_CMPL_EVENT_ID_PPS_TIMESTAMP,
@@ -2269,6 +2270,12 @@ static int bnxt_async_event_process(struct bnxt *bp,
 		bnxt_event_error_report(bp, data1, data2);
 		goto async_event_process_exit;
 	}
+	case ASYNC_EVENT_CMPL_EVENT_ID_DEFERRED_RESPONSE: {
+		u16 seq_id = le32_to_cpu(cmpl->event_data2) & 0xffff;
+
+		hwrm_update_token(bp, seq_id, BNXT_HWRM_DEFERRED);
+		goto async_event_process_exit;
+	}
 	default:
 		goto async_event_process_exit;
 	}
@@ -2288,10 +2295,7 @@ static int bnxt_hwrm_handler(struct bnxt *bp, struct tx_cmp *txcmp)
 	switch (cmpl_type) {
 	case CMPL_BASE_TYPE_HWRM_DONE:
 		seq_id = le16_to_cpu(h_cmpl->sequence_id);
-		if (seq_id == bp->hwrm_intr_seq_id)
-			bp->hwrm_intr_seq_id = (u16)~bp->hwrm_intr_seq_id;
-		else
-			netdev_err(bp->dev, "Invalid hwrm seq id %d\n", seq_id);
+		hwrm_update_token(bp, seq_id, BNXT_HWRM_COMPLETE);
 		break;
 
 	case CMPL_BASE_TYPE_HWRM_FWD_REQ:
@@ -3956,8 +3960,15 @@ out:
 
 static void bnxt_free_hwrm_resources(struct bnxt *bp)
 {
+	struct bnxt_hwrm_wait_token *token;
+
 	dma_pool_destroy(bp->hwrm_dma_pool);
 	bp->hwrm_dma_pool = NULL;
+
+	rcu_read_lock();
+	hlist_for_each_entry_rcu(token, &bp->hwrm_pending_list, node)
+		WRITE_ONCE(token->state, BNXT_HWRM_CANCELLED);
+	rcu_read_unlock();
 }
 
 static int bnxt_alloc_hwrm_resources(struct bnxt *bp)
@@ -3968,6 +3979,8 @@ static int bnxt_alloc_hwrm_resources(struct bnxt *bp)
 	if (!bp->hwrm_dma_pool)
 		return -ENOMEM;
 
+	INIT_HLIST_HEAD(&bp->hwrm_pending_list);
+
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index f343e87bef0b..a8212dcdad5f 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -1880,8 +1880,8 @@ struct bnxt {
 	u32			hwrm_spec_code;
 	u16			hwrm_cmd_seq;
 	u16                     hwrm_cmd_kong_seq;
-	u16			hwrm_intr_seq_id;
 	struct dma_pool		*hwrm_dma_pool;
+	struct hlist_head	hwrm_pending_list;
 
 	struct rtnl_link_stats64	net_stats_prev;
 	struct bnxt_stats_mem	port_stats;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
index 60ec0caa5c56..acef61abe35d 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
@@ -16,6 +16,7 @@
 #include <linux/io.h>
 #include <linux/irq.h>
 #include <linux/kernel.h>
+#include <linux/list.h>
 #include <linux/netdevice.h>
 #include <linux/pci.h>
 #include <linux/skbuff.h>
@@ -363,19 +364,72 @@ static int __hwrm_to_stderr(u32 hwrm_err)
 	}
 }
 
+static struct bnxt_hwrm_wait_token *
+__hwrm_acquire_token(struct bnxt *bp, enum bnxt_hwrm_chnl dst)
+{
+	struct bnxt_hwrm_wait_token *token;
+
+	token = kzalloc(sizeof(*token), GFP_KERNEL);
+	if (!token)
+		return NULL;
+
+	mutex_lock(&bp->hwrm_cmd_lock);
+
+	token->dst = dst;
+	token->state = BNXT_HWRM_PENDING;
+	if (dst == BNXT_HWRM_CHNL_CHIMP) {
+		token->seq_id = bp->hwrm_cmd_seq++;
+		hlist_add_head_rcu(&token->node, &bp->hwrm_pending_list);
+	} else {
+		token->seq_id = bp->hwrm_cmd_kong_seq++;
+	}
+
+	return token;
+}
+
+static void
+__hwrm_release_token(struct bnxt *bp, struct bnxt_hwrm_wait_token *token)
+{
+	if (token->dst == BNXT_HWRM_CHNL_CHIMP) {
+		hlist_del_rcu(&token->node);
+		kfree_rcu(token, rcu);
+	} else {
+		kfree(token);
+	}
+	mutex_unlock(&bp->hwrm_cmd_lock);
+}
+
+void
+hwrm_update_token(struct bnxt *bp, u16 seq_id, enum bnxt_hwrm_wait_state state)
+{
+	struct bnxt_hwrm_wait_token *token;
+
+	rcu_read_lock();
+	hlist_for_each_entry_rcu(token, &bp->hwrm_pending_list, node) {
+		if (token->seq_id == seq_id) {
+			WRITE_ONCE(token->state, state);
+			rcu_read_unlock();
+			return;
+		}
+	}
+	rcu_read_unlock();
+	netdev_err(bp->dev, "Invalid hwrm seq id %d\n", seq_id);
+}
+
 static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 {
 	u32 doorbell_offset = BNXT_GRCPF_REG_CHIMP_COMM_TRIGGER;
+	enum bnxt_hwrm_chnl dst = BNXT_HWRM_CHNL_CHIMP;
 	u32 bar_offset = BNXT_GRCPF_REG_CHIMP_COMM;
+	struct bnxt_hwrm_wait_token *token = NULL;
 	struct hwrm_short_input short_input = {0};
 	u16 max_req_len = BNXT_HWRM_MAX_REQ_LEN;
 	unsigned int i, timeout, tmo_count;
-	u16 dst = BNXT_HWRM_CHNL_CHIMP;
-	int intr_process, rc = -EBUSY;
 	u32 *data = (u32 *)ctx->req;
 	u32 msg_len = ctx->req_len;
-	u16 cp_ring_id, len = 0;
+	int rc = -EBUSY;
 	u32 req_type;
+	u16 len = 0;
 	u8 *valid;
 
 	if (ctx->flags & BNXT_HWRM_INTERNAL_RESP_DIRTY)
@@ -403,13 +457,12 @@ static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 		}
 	}
 
-	cp_ring_id = le16_to_cpu(ctx->req->cmpl_ring);
-	intr_process = (cp_ring_id == INVALID_HW_RING_ID) ? 0 : 1;
-
-	ctx->req->seq_id = cpu_to_le16(bnxt_get_hwrm_seq_id(bp, dst));
-	/* currently supports only one outstanding message */
-	if (intr_process)
-		bp->hwrm_intr_seq_id = le16_to_cpu(ctx->req->seq_id);
+	token = __hwrm_acquire_token(bp, dst);
+	if (!token) {
+		rc = -ENOMEM;
+		goto exit;
+	}
+	ctx->req->seq_id = cpu_to_le16(token->seq_id);
 
 	if ((bp->fw_cap & BNXT_FW_CAP_SHORT_CMD) ||
 	    msg_len > BNXT_HWRM_MAX_REQ_LEN) {
@@ -456,11 +509,9 @@ static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 	timeout = timeout - HWRM_SHORT_MIN_TIMEOUT * HWRM_SHORT_TIMEOUT_COUNTER;
 	tmo_count += DIV_ROUND_UP(timeout, HWRM_MIN_TIMEOUT);
 
-	if (intr_process) {
-		u16 seq_id = bp->hwrm_intr_seq_id;
-
+	if (le16_to_cpu(ctx->req->cmpl_ring) != INVALID_HW_RING_ID) {
 		/* Wait until hwrm response cmpl interrupt is processed */
-		while (bp->hwrm_intr_seq_id != (u16)~seq_id &&
+		while (READ_ONCE(token->state) < BNXT_HWRM_COMPLETE &&
 		       i++ < tmo_count) {
 			/* Abort the wait for completion if the FW health
 			 * check has failed.
@@ -479,7 +530,7 @@ static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 			}
 		}
 
-		if (bp->hwrm_intr_seq_id != (u16)~seq_id) {
+		if (READ_ONCE(token->state) != BNXT_HWRM_COMPLETE) {
 			if (!(ctx->flags & BNXT_HWRM_CTX_SILENT))
 				netdev_err(bp->dev, "Resp cmpl intr err msg: 0x%x\n",
 					   le16_to_cpu(ctx->req->req_type));
@@ -498,6 +549,13 @@ static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 			 */
 			if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state))
 				goto exit;
+
+			if (token &&
+			    READ_ONCE(token->state) == BNXT_HWRM_DEFERRED) {
+				__hwrm_release_token(bp, token);
+				token = NULL;
+			}
+
 			len = le16_to_cpu(READ_ONCE(ctx->resp->resp_len));
 			if (len) {
 				__le16 resp_seq = READ_ONCE(ctx->resp->seq_id);
@@ -569,6 +627,8 @@ timeout_abort:
 	}
 	rc = __hwrm_to_stderr(rc);
 exit:
+	if (token)
+		__hwrm_release_token(bp, token);
 	if (ctx->flags & BNXT_HWRM_INTERNAL_CTX_OWNED)
 		ctx->flags |= BNXT_HWRM_INTERNAL_RESP_DIRTY;
 	else
@@ -609,15 +669,11 @@ exit:
 int hwrm_req_send(struct bnxt *bp, void *req)
 {
 	struct bnxt_hwrm_ctx *ctx = __hwrm_ctx(bp, req);
-	int rc;
 
 	if (!ctx)
 		return -EINVAL;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = __hwrm_send(bp, ctx);
-	mutex_unlock(&bp->hwrm_cmd_lock);
-	return rc;
+	return __hwrm_send(bp, ctx);
 }
 
 /**
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
index 39032cf66258..4d17f0d5363b 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
@@ -37,6 +37,25 @@ struct bnxt_hwrm_ctx {
 	gfp_t gfp;
 };
 
+enum bnxt_hwrm_wait_state {
+	BNXT_HWRM_PENDING,
+	BNXT_HWRM_DEFERRED,
+	BNXT_HWRM_COMPLETE,
+	BNXT_HWRM_CANCELLED,
+};
+
+enum bnxt_hwrm_chnl { BNXT_HWRM_CHNL_CHIMP, BNXT_HWRM_CHNL_KONG };
+
+struct bnxt_hwrm_wait_token {
+	struct rcu_head rcu;
+	struct hlist_node node;
+	enum bnxt_hwrm_wait_state state;
+	enum bnxt_hwrm_chnl dst;
+	u16 seq_id;
+};
+
+void hwrm_update_token(struct bnxt *bp, u16 seq, enum bnxt_hwrm_wait_state s);
+
 #define BNXT_HWRM_MAX_REQ_LEN		(bp->hwrm_max_req_len)
 #define BNXT_HWRM_SHORT_REQ_LEN		sizeof(struct hwrm_short_input)
 #define HWRM_CMD_MAX_TIMEOUT		40000
@@ -78,9 +97,6 @@ static inline unsigned int hwrm_total_timeout(unsigned int n)
 
 #define HWRM_VALID_BIT_DELAY_USEC	150
 
-#define BNXT_HWRM_CHNL_CHIMP	0
-#define BNXT_HWRM_CHNL_KONG	1
-
 static inline bool bnxt_cfa_hwrm_message(u16 req_type)
 {
 	switch (req_type) {
@@ -114,17 +130,6 @@ static inline bool bnxt_kong_hwrm_message(struct bnxt *bp, struct input *req)
 		 le16_to_cpu(req->target_id) == HWRM_TARGET_ID_KONG));
 }
 
-static inline u16 bnxt_get_hwrm_seq_id(struct bnxt *bp, u16 dst)
-{
-	u16 seq_id;
-
-	if (dst == BNXT_HWRM_CHNL_CHIMP)
-		seq_id = bp->hwrm_cmd_seq++;
-	else
-		seq_id = bp->hwrm_cmd_kong_seq++;
-	return seq_id;
-}
-
 int __hwrm_req_init(struct bnxt *bp, void **req, u16 req_type, u32 req_len);
 #define hwrm_req_init(bp, req, req_type) \
 	__hwrm_req_init((bp), (void **)&(req), (req_type), sizeof(*(req)))
-- 
cgit v1.2.3