Skip to contentSkip to conversation

bnxt_en: Implement new driver APIs to send FW messages

PATCH Applied Michael Chan 4 participants 16 messages Aug 29, 2021 – Sep 1, 2021

Subsystems (3)
lore
Share & tools
lore

Michael Chan

author maintainer
#0 lore raw

v2 cover bnxt_en: Implement new driver APIs to send FW messages

The current driver APIs to send messages to the firmware allow only one outstanding message in flight. There is only one buffer for the firmware response for each firmware channel. To send a firmware message, all callers must take a mutex and it is released after the firmware response has been read. This scheme does not allow multiple firmware messages in flight. Firmware may take a long time to respond to some messages (e.g. NVRAM related ones) and this causes the mutex to be held for a long time, blocking other callers.

This patchset intoduces the new driver APIs to address the above shortcomings. The new APIs are compatible with new and old firmware. But the new deferred firmware response mechanism will require newer firmware in order to allow multiple outstanding firmware commands.

All callers are updated to use the new APIs.

v2: Patch 4 and patch 9 updated to fix issues reported by test robot

Edwin Peer (11):
bnxt_en: remove DMA mapping for KONG response
bnxt_en: Refactor the HWRM_VER_GET firmware calls
bnxt_en: move HWRM API implementation into separate file
bnxt_en: introduce new firmware message API based on DMA pools
bnxt_en: discard out of sequence HWRM responses
bnxt_en: add HWRM request assignment API
bnxt_en: add support for HWRM request slices
bnxt_en: use link_lock instead of hwrm_cmd_lock to protect link_info
bnxt_en: update all firmware calls to use the new APIs
bnxt_en: remove legacy HWRM interface
bnxt_en: support multiple HWRM commands in flight

 drivers/net/ethernet/broadcom/bnxt/Makefile   |    2 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 2133 ++++++++---------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  105 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c |  185 +-
 .../net/ethernet/broadcom/bnxt/bnxt_devlink.c |   81 +-
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c |  548 +++--
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    |  763 ++++++
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    |  145 ++
 drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c |  130 +-
 .../net/ethernet/broadcom/bnxt/bnxt_sriov.c   |  455 ++--
 drivers/net/ethernet/broadcom/bnxt/bnxt_tc.c  |  264 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c |   31 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_vfr.c |   62 +-
 13 files changed, 2906 insertions(+), 1998 deletions(-)
 create mode 100644 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
 create mode 100644 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
Signature · 1 line
2.18.1
author maintainer
#1 lore raw

PATCH v2 1/11 bnxt_en: remove DMA mapping for KONG response

From: Edwin Peer <[email protected]>

The additional response buffer serves no useful purpose. There can be only one firmware command in flight due to the hwrm_cmd_lock mutex, which is taken for the entire duration of any command completion, KONG or otherwise. It is thus safe to share a single DMA buffer.

Removing the code associated with the additional mapping will simplify matters in the next patch, which allocates response buffers from DMA pools on a per request basis.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 42 +++--------------------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h | 16 ++-------
 2 files changed, 7 insertions(+), 51 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +4 -38
index ccf1e47d9e92..fb75fa9614c5 100644
@@ -3962,30 +3962,6 @@ static void bnxt_free_hwrm_resources(struct bnxt *bp)
 				  bp->hwrm_cmd_resp_dma_addr);
 		bp->hwrm_cmd_resp_addr = NULL;
 	}
-
-	if (bp->hwrm_cmd_kong_resp_addr) {
-		dma_free_coherent(&pdev->dev, PAGE_SIZE,
-				  bp->hwrm_cmd_kong_resp_addr,
-				  bp->hwrm_cmd_kong_resp_dma_addr);
-		bp->hwrm_cmd_kong_resp_addr = NULL;
-	}
-}
-
-static int bnxt_alloc_kong_hwrm_resources(struct bnxt *bp)
-{
-	struct pci_dev *pdev = bp->pdev;
-
-	if (bp->hwrm_cmd_kong_resp_addr)
-		return 0;
-
-	bp->hwrm_cmd_kong_resp_addr =
-		dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
-				   &bp->hwrm_cmd_kong_resp_dma_addr,
-				   GFP_KERNEL);
-	if (!bp->hwrm_cmd_kong_resp_addr)
-		return -ENOMEM;
-
-	return 0;
 }
 
 static int bnxt_alloc_hwrm_resources(struct bnxt *bp)
@@ -4581,10 +4557,7 @@ void bnxt_hwrm_cmd_hdr_init(struct bnxt *bp, void *request, u16 req_type,
 	req->req_type = cpu_to_le16(req_type);
 	req->cmpl_ring = cpu_to_le16(cmpl_ring);
 	req->target_id = cpu_to_le16(target_id);
-	if (bnxt_kong_hwrm_message(bp, req))
-		req->resp_addr = cpu_to_le64(bp->hwrm_cmd_kong_resp_dma_addr);
-	else
-		req->resp_addr = cpu_to_le64(bp->hwrm_cmd_resp_dma_addr);
+	req->resp_addr = cpu_to_le64(bp->hwrm_cmd_resp_dma_addr);
 }
 
 static int bnxt_hwrm_to_stderr(u32 hwrm_err)
@@ -4641,11 +4614,10 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 			return -EINVAL;
 	}
 
-	if (bnxt_hwrm_kong_chnl(bp, req)) {
+	if (bnxt_kong_hwrm_message(bp, req)) {
 		dst = BNXT_HWRM_CHNL_KONG;
 		bar_offset = BNXT_GRCPF_REG_KONG_COMM;
 		doorbell_offset = BNXT_GRCPF_REG_KONG_COMM_TRIGGER;
-		resp = bp->hwrm_cmd_kong_resp_addr;
 	}
 
 	memset(resp, 0, PAGE_SIZE);
@@ -11948,12 +11920,6 @@ static int bnxt_fw_init_one_p1(struct bnxt *bp)
 			return rc;
 	}
 
-	if (bp->fw_cap & BNXT_FW_CAP_KONG_MB_CHNL) {
-		rc = bnxt_alloc_kong_hwrm_resources(bp);
-		if (rc)
-			bp->fw_cap &= ~BNXT_FW_CAP_KONG_MB_CHNL;
-	}
-
 	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);
@@ -12136,8 +12102,8 @@ static void bnxt_reset_all(struct bnxt *bp)
 	} else if (fw_health->flags & ERROR_RECOVERY_QCFG_RESP_FLAGS_CO_CPU) {
 		struct hwrm_fw_reset_input req = {0};
 
-		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FW_RESET, -1, -1);
-		req.resp_addr = cpu_to_le64(bp->hwrm_cmd_kong_resp_dma_addr);
+		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FW_RESET, -1,
+				       HWRM_TARGET_ID_KONG);
 		req.embedded_proc_type = FW_RESET_REQ_EMBEDDED_PROC_TYPE_CHIP;
 		req.selfrst_status = FW_RESET_REQ_SELFRST_STATUS_SELFRSTASAP;
 		req.flags = FW_RESET_REQ_FLAGS_RESET_GRACEFUL;
drivers/net/ethernet/broadcom/bnxt/bnxt.h +3 -13
index dc96dd6957c9..a4fb1aa12b24 100644
@@ -1915,8 +1915,6 @@ struct bnxt {
 	dma_addr_t		hwrm_short_cmd_req_dma_addr;
 	void			*hwrm_cmd_resp_addr;
 	dma_addr_t		hwrm_cmd_resp_dma_addr;
-	void			*hwrm_cmd_kong_resp_addr;
-	dma_addr_t		hwrm_cmd_kong_resp_dma_addr;
 
 	struct rtnl_link_stats64	net_stats_prev;
 	struct bnxt_stats_mem	port_stats;
@@ -2216,21 +2214,13 @@ static inline bool bnxt_cfa_hwrm_message(u16 req_type)
 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)));
-}
-
-static inline bool bnxt_hwrm_kong_chnl(struct bnxt *bp, struct input *req)
-{
-	return (bp->fw_cap & BNXT_FW_CAP_KONG_MB_CHNL &&
-		req->resp_addr == cpu_to_le64(bp->hwrm_cmd_kong_resp_dma_addr));
+		(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)
 {
-	if (bnxt_hwrm_kong_chnl(bp, (struct input *)req))
-		return bp->hwrm_cmd_kong_resp_addr;
-	else
-		return bp->hwrm_cmd_resp_addr;
+	return bp->hwrm_cmd_resp_addr;
 }
 
 static inline u16 bnxt_get_hwrm_seq_id(struct bnxt *bp, u16 dst)
Signature · 1 line
2.18.1
author maintainer
#2 lore raw

PATCH v2 2/11 bnxt_en: Refactor the HWRM_VER_GET firmware calls

In reply to Michael Chan

From: Edwin Peer <[email protected]>

Refactor the code so that __bnxt_hwrm_ver_get() does not call bnxt_hwrm_do_send_msg() directly. The new APIs will not expose this internal call. Add a new bnxt_hwrm_poll() to poll the HWRM_VER_GET firmware call silently. The other bnxt_hwrm_ver_get() function will send the HWRM_VER_GET message directly with error logs enabled.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +11 -6
index fb75fa9614c5..dd2f80c394f5 100644
@@ -8036,7 +8036,7 @@ static int bnxt_hwrm_queue_qportcfg(struct bnxt *bp)
 	return rc;
 }
 
-static int __bnxt_hwrm_ver_get(struct bnxt *bp, bool silent)
+static int bnxt_hwrm_poll(struct bnxt *bp)
 {
 	struct hwrm_ver_get_input req = {0};
 	int rc;
@@ -8046,21 +8046,26 @@ static int __bnxt_hwrm_ver_get(struct bnxt *bp, bool silent)
 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
 
-	rc = bnxt_hwrm_do_send_msg(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT,
-				   silent);
+	rc = _hwrm_send_message_silent(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
 	return rc;
 }
 
 static int bnxt_hwrm_ver_get(struct bnxt *bp)
 {
 	struct hwrm_ver_get_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_ver_get_input req = {0};
 	u16 fw_maj, fw_min, fw_bld, fw_rsv;
 	u32 dev_caps_cfg, hwrm_ver;
 	int rc, len;
 
+	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VER_GET, -1, -1);
 	bp->hwrm_max_req_len = HWRM_MAX_REQ_LEN;
+	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
+	req.hwrm_intf_min = HWRM_VERSION_MINOR;
+	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
+
 	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = __bnxt_hwrm_ver_get(bp, false);
+	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
 	if (rc)
 		goto hwrm_ver_get_exit;
 
@@ -9791,7 +9796,7 @@ static int bnxt_try_recover_fw(struct bnxt *bp)
 		mutex_lock(&bp->hwrm_cmd_lock);
 		do {
 			sts = bnxt_fw_health_readl(bp, BNXT_FW_HEALTH_REG);
-			rc = __bnxt_hwrm_ver_get(bp, true);
+			rc = bnxt_hwrm_poll(bp);
 			if (!BNXT_FW_IS_BOOTING(sts) &&
 			    !BNXT_FW_IS_RECOVERING(sts))
 				break;
@@ -12234,7 +12239,7 @@ static void bnxt_fw_reset_task(struct work_struct *work)
 		fallthrough;
 	case BNXT_FW_RESET_STATE_POLL_FW:
 		bp->hwrm_cmd_timeout = SHORT_HWRM_CMD_TIMEOUT;
-		rc = __bnxt_hwrm_ver_get(bp, true);
+		rc = bnxt_hwrm_poll(bp);
 		if (rc) {
 			if (bnxt_fw_reset_timeout(bp)) {
 				netdev_err(bp->dev, "Firmware reset aborted\n");
Signature · 1 line
2.18.1
author maintainer
#3 lore raw

PATCH v2 3/11 bnxt_en: move HWRM API implementation into separate file

In reply to Michael Chan

From: Edwin Peer <[email protected]>

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

---
 drivers/net/ethernet/broadcom/bnxt/Makefile   |   2 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 273 +---------------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  84 -----
 drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c |   1 +
 .../net/ethernet/broadcom/bnxt/bnxt_devlink.c |   1 +
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c |   1 +
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 297 ++++++++++++++++++
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    | 100 ++++++
 drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c |   1 +
 .../net/ethernet/broadcom/bnxt/bnxt_sriov.c   |   1 +
 drivers/net/ethernet/broadcom/bnxt/bnxt_tc.c  |   1 +
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c |   1 +
 drivers/net/ethernet/broadcom/bnxt/bnxt_vfr.c |   1 +
 13 files changed, 407 insertions(+), 357 deletions(-)
 create mode 100644 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c
 create mode 100644 drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h
drivers/net/ethernet/broadcom/bnxt/Makefile +1 -1
index 2b8ae687b3c1..c6ef7ec2c115 100644
@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: GPL-2.0-only
 obj-$(CONFIG_BNXT) += bnxt_en.o
 
-bnxt_en-y := bnxt.o bnxt_sriov.o bnxt_ethtool.o bnxt_dcb.o bnxt_ulp.o bnxt_xdp.o bnxt_ptp.o bnxt_vfr.o bnxt_devlink.o bnxt_dim.o
+bnxt_en-y := bnxt.o bnxt_hwrm.o bnxt_sriov.o bnxt_ethtool.o bnxt_dcb.o bnxt_ulp.o bnxt_xdp.o bnxt_ptp.o bnxt_vfr.o bnxt_devlink.o bnxt_dim.o
 bnxt_en-$(CONFIG_BNXT_FLOWER_OFFLOAD) += bnxt_tc.o
 bnxt_en-$(CONFIG_DEBUG_FS) += bnxt_debugfs.o
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -272

285 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -84

103 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c +1 -0
index 8a68df4d9e59..df898665763a 100644
@@ -18,6 +18,7 @@
 #include <rdma/ib_verbs.h>
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_dcb.h"
 
 #ifdef CONFIG_BNXT_DCB
drivers/net/ethernet/broadcom/bnxt/bnxt_devlink.c +1 -0
index 2cd8bb37e641..00b284a028c5 100644
@@ -12,6 +12,7 @@
 #include <net/devlink.h>
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_vfr.h"
 #include "bnxt_devlink.h"
 #include "bnxt_ethtool.h"
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +1 -0
index 5852ae3b26a2..b6aaf14bd7fd 100644
@@ -24,6 +24,7 @@
 #include <linux/timecounter.h>
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_xdp.h"
 #include "bnxt_ptp.h"
 #include "bnxt_ethtool.h"
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c new file +297 -0

297 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h new file +100 -0

100 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c +1 -0
index 2fe3c9081f8d..4cc2379027cf 100644
@@ -18,6 +18,7 @@
 #include <linux/ptp_classify.h>
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_ptp.h"
 
 int bnxt_ptp_parse(struct sk_buff *skb, u16 *seq_id, u16 *hdr_off)
drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c +1 -0
index 7fa881e1cd80..7b0e308e44c2 100644
@@ -17,6 +17,7 @@
 #include <linux/etherdevice.h>
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_ulp.h"
 #include "bnxt_sriov.h"
 #include "bnxt_vfr.h"
drivers/net/ethernet/broadcom/bnxt/bnxt_tc.c +1 -0
index 5e4429b14b8c..c0c3cc426f7b 100644
@@ -22,6 +22,7 @@
 
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_sriov.h"
 #include "bnxt_tc.h"
 #include "bnxt_vfr.h"
drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c +1 -0
index 187ff643ad2a..f621cffccd3a 100644
@@ -22,6 +22,7 @@
 
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_ulp.h"
 
 static int bnxt_register_dev(struct bnxt_en_dev *edev, int ulp_id,
drivers/net/ethernet/broadcom/bnxt/bnxt_vfr.c +1 -0
index dd66302343a2..3ed712a08207 100644
@@ -15,6 +15,7 @@
 
 #include "bnxt_hsi.h"
 #include "bnxt.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_vfr.h"
 #include "bnxt_devlink.h"
 #include "bnxt_tc.h"
Signature · 1 line
2.18.1
author maintainer
#4 lore raw

PATCH v2 4/11 bnxt_en: introduce new firmware message API based on DMA pools

In reply to Michael Chan

From: Edwin Peer <[email protected]>

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()

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     |  42 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |   7 +-
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 460 +++++++++++++++---
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    |  54 +-
 4 files changed, 442 insertions(+), 121 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +7 -35
index 10c39801ad5f..23486f382b91 100644
@@ -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 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 
 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);
drivers/net/ethernet/broadcom/bnxt/bnxt.h +3 -4
index 5ff71eeffdd8..79a78a7468f3 100644
@@ -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);		\
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c +384 -76

567 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h +48 -6

79 lines not on this page · raw message

Signature · 1 line
2.18.1
#5 lore raw

Re: [PATCH net-next v2 04/11] bnxt_en: introduce new firmware message API based on DMA pools

Thank you for the patch! Yet something to improve:

Hi Michael,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on net-next/master]

url:    https://github.com/0day-ci/linux/commits/Michael-Chan/bnxt_en-Implement-new-driver-APIs-to-send-FW-messages/20210829-153728
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git c77225119daa0ca0a9c6c007945c0a87b3e4a2e8
config: parisc-allyesconfig (attached as .config)
compiler: hppa-linux-gcc (GCC) 11.2.0

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/cfca001de60095f2133fabd5dd01c092de1fd34d
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Michael-Chan/bnxt_en-Implement-new-driver-APIs-to-send-FW-messages/20210829-153728 git checkout cfca001de60095f2133fabd5dd01c092de1fd34d
# save the attached .config to linux build tree
mkdir build_dir
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross O=build_dir ARCH=parisc SHELL=/bin/bash

If you fix the issue, kindly add following tag as appropriate

All errors (new ones prefixed by >>):

In file included from include/linux/mm_types.h:9,
from include/linux/buildid.h:5,
from include/linux/module.h:14,
from drivers/net/ethernet/broadcom/bnxt/bnxt.c:11:
drivers/net/ethernet/broadcom/bnxt/bnxt.c: In function 'bnxt_init_board':

>> drivers/net/ethernet/broadcom/bnxt/bnxt.c:12062:27: error: 'struct bnxt' has no member named 'db_lock'
   12062 |         spin_lock_init(&bp->db_lock);
         |                           ^~
   include/linux/spinlock.h:338:45: note: in definition of macro 'spin_lock_init'
     338 |         __raw_spin_lock_init(spinlock_check(lock),              \
         |                                             ^~~~

vim +12062 drivers/net/ethernet/broadcom/bnxt/bnxt.c

230d1f0de754b4 Michael Chan       2019-08-29  11998  
c0c050c58d8409 Michael Chan       2015-10-22  11999  static int bnxt_init_board(struct pci_dev *pdev, struct net_device *dev)
c0c050c58d8409 Michael Chan       2015-10-22  12000  {
c0c050c58d8409 Michael Chan       2015-10-22  12001  	int rc;
c0c050c58d8409 Michael Chan       2015-10-22  12002  	struct bnxt *bp = netdev_priv(dev);
c0c050c58d8409 Michael Chan       2015-10-22  12003  
c0c050c58d8409 Michael Chan       2015-10-22  12004  	SET_NETDEV_DEV(dev, &pdev->dev);
c0c050c58d8409 Michael Chan       2015-10-22  12005  
c0c050c58d8409 Michael Chan       2015-10-22  12006  	/* enable device (incl. PCI PM wakeup), and bus-mastering */
c0c050c58d8409 Michael Chan       2015-10-22  12007  	rc = pci_enable_device(pdev);
c0c050c58d8409 Michael Chan       2015-10-22  12008  	if (rc) {
c0c050c58d8409 Michael Chan       2015-10-22  12009  		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
c0c050c58d8409 Michael Chan       2015-10-22  12010  		goto init_err;
c0c050c58d8409 Michael Chan       2015-10-22  12011  	}
c0c050c58d8409 Michael Chan       2015-10-22  12012  
c0c050c58d8409 Michael Chan       2015-10-22  12013  	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
c0c050c58d8409 Michael Chan       2015-10-22  12014  		dev_err(&pdev->dev,
c0c050c58d8409 Michael Chan       2015-10-22  12015  			"Cannot find PCI device base address, aborting\n");
c0c050c58d8409 Michael Chan       2015-10-22  12016  		rc = -ENODEV;
c0c050c58d8409 Michael Chan       2015-10-22  12017  		goto init_err_disable;
c0c050c58d8409 Michael Chan       2015-10-22  12018  	}
c0c050c58d8409 Michael Chan       2015-10-22  12019  
c0c050c58d8409 Michael Chan       2015-10-22  12020  	rc = pci_request_regions(pdev, DRV_MODULE_NAME);
c0c050c58d8409 Michael Chan       2015-10-22  12021  	if (rc) {
c0c050c58d8409 Michael Chan       2015-10-22  12022  		dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting\n");
c0c050c58d8409 Michael Chan       2015-10-22  12023  		goto init_err_disable;
c0c050c58d8409 Michael Chan       2015-10-22  12024  	}
c0c050c58d8409 Michael Chan       2015-10-22  12025  
c0c050c58d8409 Michael Chan       2015-10-22  12026  	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0 &&
c0c050c58d8409 Michael Chan       2015-10-22  12027  	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0) {
c0c050c58d8409 Michael Chan       2015-10-22  12028  		dev_err(&pdev->dev, "System does not support DMA, aborting\n");
3383176efc0fb0 Zhang Changzhong   2020-11-19  12029  		rc = -EIO;
c54bc3ced51066 Michael Chan       2020-11-20  12030  		goto init_err_release;
c0c050c58d8409 Michael Chan       2015-10-22  12031  	}
c0c050c58d8409 Michael Chan       2015-10-22  12032  
c0c050c58d8409 Michael Chan       2015-10-22  12033  	pci_set_master(pdev);
c0c050c58d8409 Michael Chan       2015-10-22  12034  
c0c050c58d8409 Michael Chan       2015-10-22  12035  	bp->dev = dev;
c0c050c58d8409 Michael Chan       2015-10-22  12036  	bp->pdev = pdev;
c0c050c58d8409 Michael Chan       2015-10-22  12037  
8ae2473842bdbb Michael Chan       2020-05-04  12038  	/* Doorbell BAR bp->bar1 is mapped after bnxt_fw_init_one_p2()
8ae2473842bdbb Michael Chan       2020-05-04  12039  	 * determines the BAR size.
8ae2473842bdbb Michael Chan       2020-05-04  12040  	 */
c0c050c58d8409 Michael Chan       2015-10-22  12041  	bp->bar0 = pci_ioremap_bar(pdev, 0);
c0c050c58d8409 Michael Chan       2015-10-22  12042  	if (!bp->bar0) {
c0c050c58d8409 Michael Chan       2015-10-22  12043  		dev_err(&pdev->dev, "Cannot map device registers, aborting\n");
c0c050c58d8409 Michael Chan       2015-10-22  12044  		rc = -ENOMEM;
c0c050c58d8409 Michael Chan       2015-10-22  12045  		goto init_err_release;
c0c050c58d8409 Michael Chan       2015-10-22  12046  	}
c0c050c58d8409 Michael Chan       2015-10-22  12047  
c0c050c58d8409 Michael Chan       2015-10-22  12048  	bp->bar2 = pci_ioremap_bar(pdev, 4);
c0c050c58d8409 Michael Chan       2015-10-22  12049  	if (!bp->bar2) {
c0c050c58d8409 Michael Chan       2015-10-22  12050  		dev_err(&pdev->dev, "Cannot map bar4 registers, aborting\n");
c0c050c58d8409 Michael Chan       2015-10-22  12051  		rc = -ENOMEM;
c0c050c58d8409 Michael Chan       2015-10-22  12052  		goto init_err_release;
c0c050c58d8409 Michael Chan       2015-10-22  12053  	}
c0c050c58d8409 Michael Chan       2015-10-22  12054  
6316ea6db93d87 Satish Baddipadige 2016-03-07  12055  	pci_enable_pcie_error_reporting(pdev);
6316ea6db93d87 Satish Baddipadige 2016-03-07  12056  
c0c050c58d8409 Michael Chan       2015-10-22  12057  	INIT_WORK(&bp->sp_task, bnxt_sp_task);
230d1f0de754b4 Michael Chan       2019-08-29  12058  	INIT_DELAYED_WORK(&bp->fw_reset_task, bnxt_fw_reset_task);
c0c050c58d8409 Michael Chan       2015-10-22  12059  
c0c050c58d8409 Michael Chan       2015-10-22  12060  	spin_lock_init(&bp->ntp_fltr_lock);
697197e5a17353 Michael Chan       2018-10-14  12061  #if BITS_PER_LONG == 32
697197e5a17353 Michael Chan       2018-10-14 @12062  	spin_lock_init(&bp->db_lock);
697197e5a17353 Michael Chan       2018-10-14  12063  #endif
c0c050c58d8409 Michael Chan       2015-10-22  12064  
c0c050c58d8409 Michael Chan       2015-10-22  12065  	bp->rx_ring_size = BNXT_DEFAULT_RX_RING_SIZE;
c0c050c58d8409 Michael Chan       2015-10-22  12066  	bp->tx_ring_size = BNXT_DEFAULT_TX_RING_SIZE;
c0c050c58d8409 Michael Chan       2015-10-22  12067  
18775aa8a91fcd Michael Chan       2017-10-26  12068  	bnxt_init_dflt_coal(bp);
51f307856b60e6 Michael Chan       2016-07-01  12069  
e99e88a9d2b067 Kees Cook          2017-10-16  12070  	timer_setup(&bp->timer, bnxt_timer, 0);
c0c050c58d8409 Michael Chan       2015-10-22  12071  	bp->current_interval = BNXT_TIMER_INTERVAL;
c0c050c58d8409 Michael Chan       2015-10-22  12072  
442a35a5a7aa72 Jakub Kicinski     2020-07-09  12073  	bp->vxlan_fw_dst_port_id = INVALID_HW_RING_ID;
442a35a5a7aa72 Jakub Kicinski     2020-07-09  12074  	bp->nge_fw_dst_port_id = INVALID_HW_RING_ID;
442a35a5a7aa72 Jakub Kicinski     2020-07-09  12075  
caefe526d7b5af Michael Chan       2015-12-09  12076  	clear_bit(BNXT_STATE_OPEN, &bp->state);
c0c050c58d8409 Michael Chan       2015-10-22  12077  	return 0;
c0c050c58d8409 Michael Chan       2015-10-22  12078  
c0c050c58d8409 Michael Chan       2015-10-22  12079  init_err_release:
17086399c113d9 Sathya Perla       2017-02-20  12080  	bnxt_unmap_bars(bp, pdev);
c0c050c58d8409 Michael Chan       2015-10-22  12081  	pci_release_regions(pdev);
c0c050c58d8409 Michael Chan       2015-10-22  12082  
c0c050c58d8409 Michael Chan       2015-10-22  12083  init_err_disable:
c0c050c58d8409 Michael Chan       2015-10-22  12084  	pci_disable_device(pdev);
c0c050c58d8409 Michael Chan       2015-10-22  12085  
c0c050c58d8409 Michael Chan       2015-10-22  12086  init_err:
c0c050c58d8409 Michael Chan       2015-10-22  12087  	return rc;
c0c050c58d8409 Michael Chan       2015-10-22  12088  }
c0c050c58d8409 Michael Chan       2015-10-22  12089  

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

#6 lore raw

Re: [PATCH net-next v2 04/11] bnxt_en: introduce new firmware message API based on DMA pools

In reply to Michael Chan
Quoted 9 lines · Michael Chan

From: Edwin Peer <[email protected]>

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.

parisc:allmodconfig, and probably others:

drivers/net/ethernet/broadcom/bnxt/bnxt.c: In function 'bnxt_init_board':
drivers/net/ethernet/broadcom/bnxt/bnxt.c:12280:27: error: 'struct bnxt' has no member named 'db_lock'

There is a difference between "#ifdef writeq" and "#if BITS_PER_LONG == 32".

Guenter

author maintainer
#7 lore raw

Re: [PATCH net-next v2 04/11] bnxt_en: introduce new firmware message API based on DMA pools

Quoted 12 lines · Guenter Roeck
Quoted 9 lines · Michael Chan

From: Edwin Peer <[email protected]>

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.

Quoted 6 lines

parisc:allmodconfig, and probably others:

drivers/net/ethernet/broadcom/bnxt/bnxt.c: In function 'bnxt_init_board':
drivers/net/ethernet/broadcom/bnxt/bnxt.c:12280:27: error: 'struct bnxt' has no member named 'db_lock'

There is a difference between "#ifdef writeq" and "#if BITS_PER_LONG == 32".

Yes. It is already fixed in this commit:

c6132f6f2e682c958f7022ecfd8bec35723a1a9d
bnxt_en: Fix 64-bit doorbell operation on 32-bit kernels

Thanks.

author maintainer
#8 lore raw

PATCH v2 5/11 bnxt_en: discard out of sequence HWRM responses

In reply to Michael Chan

From: Edwin Peer <[email protected]>

During firmware crash recovery, it is possible for firmware to respond to stale HWRM commands that have already timed out. Because response buffers may be reused, any out of sequence responses need to be ignored and only the matching seq_id should be accepted.

Also, READ_ONCE should be used for the reads from the DMA buffer to ensure that the necessary loads are scheduled.

---
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 21 +++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c +17 -4
index a45f2a619086..621daf687a00 100644
@@ -399,9 +399,10 @@ static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 					   le16_to_cpu(ctx->req->req_type));
 			goto exit;
 		}
-		len = le16_to_cpu(ctx->resp->resp_len);
+		len = le16_to_cpu(READ_ONCE(ctx->resp->resp_len));
 		valid = ((u8 *)ctx->resp) + len - 1;
 	} else {
+		__le16 seen_out_of_seq = ctx->req->seq_id; /* will never see */
 		int j;
 
 		/* Check if response len is updated */
@@ -411,9 +412,21 @@ static int __hwrm_send(struct bnxt *bp, struct bnxt_hwrm_ctx *ctx)
 			 */
 			if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state))
 				goto exit;
-			len = le16_to_cpu(ctx->resp->resp_len);
-			if (len)
-				break;
+			len = le16_to_cpu(READ_ONCE(ctx->resp->resp_len));
+			if (len) {
+				__le16 resp_seq = READ_ONCE(ctx->resp->seq_id);
+
+				if (resp_seq == ctx->req->seq_id)
+					break;
+				if (resp_seq != seen_out_of_seq) {
+					netdev_warn(bp->dev, "Discarding out of seq response: 0x%x for msg {0x%x 0x%x}\n",
+						    le16_to_cpu(resp_seq),
+						    le16_to_cpu(ctx->req->req_type),
+						    le16_to_cpu(ctx->req->seq_id));
+					seen_out_of_seq = resp_seq;
+				}
+			}
+
 			/* on first few passes, just barely sleep */
 			if (i < HWRM_SHORT_TIMEOUT_COUNTER) {
 				usleep_range(HWRM_SHORT_MIN_TIMEOUT,
Signature · 1 line
2.18.1
author maintainer
#9 lore raw

PATCH v2 6/11 bnxt_en: add HWRM request assignment API

In reply to Michael Chan

From: Edwin Peer <[email protected]>

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.

---
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 55 +++++++++++++++++++
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    |  1 +
 2 files changed, 56 insertions(+)
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c +55 -0
index 621daf687a00..39ef65025e17 100644
@@ -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.
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h +1 -0
index 199c646f5e71..c58d84cc692a 100644
@@ -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
Signature · 1 line
2.18.1
author maintainer
#10 lore raw

PATCH v2 7/11 bnxt_en: add support for HWRM request slices

In reply to Michael Chan

From: Edwin Peer <[email protected]>

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.

---
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 115 +++++++++++++++++-
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    |   7 ++
 2 files changed, 120 insertions(+), 2 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c +113 -2
index 39ef65025e17..6609a86d5226 100644
@@ -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;
+}
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h +7 -0
index c58d84cc692a..b3af7a88e2c7 100644
@@ -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
Signature · 1 line
2.18.1
author maintainer
#11 lore raw

PATCH v2 8/11 bnxt_en: use link_lock instead of hwrm_cmd_lock to protect link_info

In reply to Michael Chan

From: Edwin Peer <[email protected]>

We currently use the hwrm_cmd_lock to serialize the update of the firmware's link status response data and the copying of link status data to the VF. This won't work when we update the firmware message APIs, so we use the link_lock mutex instead. All link_info data should be updated under the link_lock mutex. Also add link_lock to functions that touch link_info in __bnxt_open_nic() and bnxt_probe_phy(). The locking is probably not strictly necessary during probe, but it's more consistent.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c       | 5 +++++
 drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c | 4 ++--
 2 files changed, 7 insertions(+), 2 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +5 -0
index 23486f382b91..b9aa56cc10d2 100644
@@ -10008,7 +10008,9 @@ static int __bnxt_open_nic(struct bnxt *bp, bool irq_re_init, bool link_re_init)
 	bnxt_tx_enable(bp);
 	mod_timer(&bp->timer, jiffies + bp->current_interval);
 	/* Poll link status and check for SFP+ module status */
+	mutex_lock(&bp->link_lock);
 	bnxt_get_port_module_status(bp);
+	mutex_unlock(&bp->link_lock);
 
 	/* VF-reps may need to be re-opened after the PF is re-opened */
 	if (BNXT_PF(bp))
@@ -12599,8 +12601,10 @@ static int bnxt_probe_phy(struct bnxt *bp, bool fw_dflt)
 	if (!fw_dflt)
 		return 0;
 
+	mutex_lock(&bp->link_lock);
 	rc = bnxt_update_link(bp, false);
 	if (rc) {
+		mutex_unlock(&bp->link_lock);
 		netdev_err(bp->dev, "Probe phy can't update link (rc: %x)\n",
 			   rc);
 		return rc;
@@ -12613,6 +12617,7 @@ static int bnxt_probe_phy(struct bnxt *bp, bool fw_dflt)
 		link_info->support_auto_speeds = link_info->support_speeds;
 
 	bnxt_init_ethtool_link_settings(bp);
+	mutex_unlock(&bp->link_lock);
 	return 0;
 }
 
drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c +2 -2
index 7b0e308e44c2..07e8e9f657e4 100644
@@ -1032,10 +1032,10 @@ static int bnxt_vf_set_link(struct bnxt *bp, struct bnxt_vf_info *vf)
 
 		phy_qcfg_req =
 		(struct hwrm_port_phy_qcfg_input *)vf->hwrm_cmd_req_addr;
-		mutex_lock(&bp->hwrm_cmd_lock);
+		mutex_lock(&bp->link_lock);
 		memcpy(&phy_qcfg_resp, &bp->link_info.phy_qcfg_resp,
 		       sizeof(phy_qcfg_resp));
-		mutex_unlock(&bp->hwrm_cmd_lock);
+		mutex_unlock(&bp->link_lock);
 		phy_qcfg_resp.resp_len = cpu_to_le16(sizeof(phy_qcfg_resp));
 		phy_qcfg_resp.seq_id = phy_qcfg_req->seq_id;
 		phy_qcfg_resp.valid = 1;
Signature · 1 line
2.18.1
author maintainer
#12 lore raw

PATCH v2 9/11 bnxt_en: update all firmware calls to use the new APIs

In reply to Michael Chan

From: Edwin Peer <[email protected]>

The conversion follows this general pattern for most of the calls:

1. The input message is changed from a stack variable initialized using bnxt_hwrm_cmd_hdr_init() to a pointer allocated and intialized using hwrm_req_init().

2. If we don't need to read the firmware response, the hwrm_send_message() call is replaced with hwrm_req_send().

3. If we need to read the firmware response, the mutex lock is replaced by hwrm_req_hold() to hold the response. When the response is read, the mutex unlock is replaced by hwrm_req_drop().

If additional DMA buffers are needed for firmware response data, the hwrm_req_dma_slice() is used instead of calling dma_alloc_coherent().

Some minor refactoring is also done while doing these conversions.

v2: Fix unintialized variable warnings in __bnxt_hwrm_get_tx_rings() and bnxt_approve_mac()

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 1747 ++++++++++-------
 drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c |  184 +-
 .../net/ethernet/broadcom/bnxt/bnxt_devlink.c |   80 +-
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c |  547 +++---
 drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c |  129 +-
 .../net/ethernet/broadcom/bnxt/bnxt_sriov.c   |  450 +++--
 drivers/net/ethernet/broadcom/bnxt/bnxt_tc.c  |  263 +--
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c |   30 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_vfr.c |   61 +-
 9 files changed, 1953 insertions(+), 1538 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +983 -764
index b9aa56cc10d2..6e9ca9b59e51 100644
@@ -4043,8 +4043,8 @@ static void bnxt_copy_hw_masks(u64 *mask_arr, __le64 *hw_mask_arr, int count)
 static int bnxt_hwrm_func_qstat_ext(struct bnxt *bp,
 				    struct bnxt_stats_mem *stats)
 {
-	struct hwrm_func_qstats_ext_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_func_qstats_ext_input req = {0};
+	struct hwrm_func_qstats_ext_output *resp;
+	struct hwrm_func_qstats_ext_input *req;
 	__le64 *hw_masks;
 	int rc;
 
@@ -4052,19 +4052,20 @@ static int bnxt_hwrm_func_qstat_ext(struct bnxt *bp,
 	    !(bp->flags & BNXT_FLAG_CHIP_P5))
 		return -EOPNOTSUPP;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QSTATS_EXT, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
-	req.flags = FUNC_QSTATS_EXT_REQ_FLAGS_COUNTER_MASK;
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_QSTATS_EXT);
 	if (rc)
-		goto qstat_exit;
+		return rc;
 
-	hw_masks = &resp->rx_ucast_pkts;
-	bnxt_copy_hw_masks(stats->hw_masks, hw_masks, stats->len / 8);
+	req->fid = cpu_to_le16(0xffff);
+	req->flags = FUNC_QSTATS_EXT_REQ_FLAGS_COUNTER_MASK;
 
-qstat_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
+	if (!rc) {
+		hw_masks = &resp->rx_ucast_pkts;
+		bnxt_copy_hw_masks(stats->hw_masks, hw_masks, stats->len / 8);
+	}
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -4533,34 +4534,35 @@ static void bnxt_enable_int(struct bnxt *bp)
 int bnxt_hwrm_func_drv_rgtr(struct bnxt *bp, unsigned long *bmap, int bmap_size,
 			    bool async_only)
 {
-	struct hwrm_func_drv_rgtr_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_func_drv_rgtr_input req = {0};
 	DECLARE_BITMAP(async_events_bmap, 256);
 	u32 *events = (u32 *)async_events_bmap;
+	struct hwrm_func_drv_rgtr_output *resp;
+	struct hwrm_func_drv_rgtr_input *req;
 	u32 flags;
 	int rc, i;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_DRV_RGTR, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_DRV_RGTR);
+	if (rc)
+		return rc;
 
-	req.enables =
-		cpu_to_le32(FUNC_DRV_RGTR_REQ_ENABLES_OS_TYPE |
-			    FUNC_DRV_RGTR_REQ_ENABLES_VER |
-			    FUNC_DRV_RGTR_REQ_ENABLES_ASYNC_EVENT_FWD);
+	req->enables = cpu_to_le32(FUNC_DRV_RGTR_REQ_ENABLES_OS_TYPE |
+				   FUNC_DRV_RGTR_REQ_ENABLES_VER |
+				   FUNC_DRV_RGTR_REQ_ENABLES_ASYNC_EVENT_FWD);
 
-	req.os_type = cpu_to_le16(FUNC_DRV_RGTR_REQ_OS_TYPE_LINUX);
+	req->os_type = cpu_to_le16(FUNC_DRV_RGTR_REQ_OS_TYPE_LINUX);
 	flags = FUNC_DRV_RGTR_REQ_FLAGS_16BIT_VER_MODE;
 	if (bp->fw_cap & BNXT_FW_CAP_HOT_RESET)
 		flags |= FUNC_DRV_RGTR_REQ_FLAGS_HOT_RESET_SUPPORT;
 	if (bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY)
 		flags |= FUNC_DRV_RGTR_REQ_FLAGS_ERROR_RECOVERY_SUPPORT |
 			 FUNC_DRV_RGTR_REQ_FLAGS_MASTER_SUPPORT;
-	req.flags = cpu_to_le32(flags);
-	req.ver_maj_8b = DRV_VER_MAJ;
-	req.ver_min_8b = DRV_VER_MIN;
-	req.ver_upd_8b = DRV_VER_UPD;
-	req.ver_maj = cpu_to_le16(DRV_VER_MAJ);
-	req.ver_min = cpu_to_le16(DRV_VER_MIN);
-	req.ver_upd = cpu_to_le16(DRV_VER_UPD);
+	req->flags = cpu_to_le32(flags);
+	req->ver_maj_8b = DRV_VER_MAJ;
+	req->ver_min_8b = DRV_VER_MIN;
+	req->ver_upd_8b = DRV_VER_UPD;
+	req->ver_maj = cpu_to_le16(DRV_VER_MAJ);
+	req->ver_min = cpu_to_le16(DRV_VER_MIN);
+	req->ver_upd = cpu_to_le16(DRV_VER_UPD);
 
 	if (BNXT_PF(bp)) {
 		u32 data[8];
@@ -4577,14 +4579,14 @@ int bnxt_hwrm_func_drv_rgtr(struct bnxt *bp, unsigned long *bmap, int bmap_size,
 		}
 
 		for (i = 0; i < 8; i++)
-			req.vf_req_fwd[i] = cpu_to_le32(data[i]);
+			req->vf_req_fwd[i] = cpu_to_le32(data[i]);
 
-		req.enables |=
+		req->enables |=
 			cpu_to_le32(FUNC_DRV_RGTR_REQ_ENABLES_VF_REQ_FWD);
 	}
 
 	if (bp->fw_cap & BNXT_FW_CAP_OVS_64BIT_HANDLE)
-		req.flags |= cpu_to_le32(
+		req->flags |= cpu_to_le32(
 			FUNC_DRV_RGTR_REQ_FLAGS_FLOW_HANDLE_64BIT_MODE);
 
 	memset(async_events_bmap, 0, sizeof(async_events_bmap));
@@ -4603,57 +4605,63 @@ int bnxt_hwrm_func_drv_rgtr(struct bnxt *bp, unsigned long *bmap, int bmap_size,
 		}
 	}
 	for (i = 0; i < 8; i++)
-		req.async_event_fwd[i] |= cpu_to_le32(events[i]);
+		req->async_event_fwd[i] |= cpu_to_le32(events[i]);
 
 	if (async_only)
-		req.enables =
+		req->enables =
 			cpu_to_le32(FUNC_DRV_RGTR_REQ_ENABLES_ASYNC_EVENT_FWD);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc) {
 		set_bit(BNXT_STATE_DRV_REGISTERED, &bp->state);
 		if (resp->flags &
 		    cpu_to_le32(FUNC_DRV_RGTR_RESP_FLAGS_IF_CHANGE_SUPPORTED))
 			bp->fw_cap |= BNXT_FW_CAP_IF_CHANGE;
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_func_drv_unrgtr(struct bnxt *bp)
 {
-	struct hwrm_func_drv_unrgtr_input req = {0};
+	struct hwrm_func_drv_unrgtr_input *req;
+	int rc;
 
 	if (!test_and_clear_bit(BNXT_STATE_DRV_REGISTERED, &bp->state))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_DRV_UNRGTR, -1, -1);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_DRV_UNRGTR);
+	if (rc)
+		return rc;
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_hwrm_tunnel_dst_port_free(struct bnxt *bp, u8 tunnel_type)
 {
-	u32 rc = 0;
-	struct hwrm_tunnel_dst_port_free_input req = {0};
+	struct hwrm_tunnel_dst_port_free_input *req;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_TUNNEL_DST_PORT_FREE, -1, -1);
-	req.tunnel_type = tunnel_type;
+	rc = hwrm_req_init(bp, req, HWRM_TUNNEL_DST_PORT_FREE);
+	if (rc)
+		return rc;
+
+	req->tunnel_type = tunnel_type;
 
 	switch (tunnel_type) {
 	case TUNNEL_DST_PORT_FREE_REQ_TUNNEL_TYPE_VXLAN:
-		req.tunnel_dst_port_id = cpu_to_le16(bp->vxlan_fw_dst_port_id);
+		req->tunnel_dst_port_id = cpu_to_le16(bp->vxlan_fw_dst_port_id);
 		bp->vxlan_fw_dst_port_id = INVALID_HW_RING_ID;
 		break;
 	case TUNNEL_DST_PORT_FREE_REQ_TUNNEL_TYPE_GENEVE:
-		req.tunnel_dst_port_id = cpu_to_le16(bp->nge_fw_dst_port_id);
+		req->tunnel_dst_port_id = cpu_to_le16(bp->nge_fw_dst_port_id);
 		bp->nge_fw_dst_port_id = INVALID_HW_RING_ID;
 		break;
 	default:
 		break;
 	}
 
-	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		netdev_err(bp->dev, "hwrm_tunnel_dst_port_free failed. rc:%d\n",
 			   rc);
@@ -4663,17 +4671,19 @@ static int bnxt_hwrm_tunnel_dst_port_free(struct bnxt *bp, u8 tunnel_type)
 static int bnxt_hwrm_tunnel_dst_port_alloc(struct bnxt *bp, __be16 port,
 					   u8 tunnel_type)
 {
-	u32 rc = 0;
-	struct hwrm_tunnel_dst_port_alloc_input req = {0};
-	struct hwrm_tunnel_dst_port_alloc_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_tunnel_dst_port_alloc_output *resp;
+	struct hwrm_tunnel_dst_port_alloc_input *req;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_TUNNEL_DST_PORT_ALLOC, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_TUNNEL_DST_PORT_ALLOC);
+	if (rc)
+		return rc;
 
-	req.tunnel_type = tunnel_type;
-	req.tunnel_dst_port_val = port;
+	req->tunnel_type = tunnel_type;
+	req->tunnel_dst_port_val = port;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc) {
 		netdev_err(bp->dev, "hwrm_tunnel_dst_port_alloc failed. rc:%d\n",
 			   rc);
@@ -4693,33 +4703,40 @@ static int bnxt_hwrm_tunnel_dst_port_alloc(struct bnxt *bp, __be16 port,
 	}
 
 err_out:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_cfa_l2_set_rx_mask(struct bnxt *bp, u16 vnic_id)
 {
-	struct hwrm_cfa_l2_set_rx_mask_input req = {0};
+	struct hwrm_cfa_l2_set_rx_mask_input *req;
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_L2_SET_RX_MASK, -1, -1);
-	req.vnic_id = cpu_to_le32(vnic->fw_vnic_id);
+	rc = hwrm_req_init(bp, req, HWRM_CFA_L2_SET_RX_MASK);
+	if (rc)
+		return rc;
 
-	req.num_mc_entries = cpu_to_le32(vnic->mc_list_count);
-	req.mc_tbl_addr = cpu_to_le64(vnic->mc_list_mapping);
-	req.mask = cpu_to_le32(vnic->rx_mask);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req->vnic_id = cpu_to_le32(vnic->fw_vnic_id);
+	req->num_mc_entries = cpu_to_le32(vnic->mc_list_count);
+	req->mc_tbl_addr = cpu_to_le64(vnic->mc_list_mapping);
+	req->mask = cpu_to_le32(vnic->rx_mask);
+	return hwrm_req_send_silent(bp, req);
 }
 
 #ifdef CONFIG_RFS_ACCEL
 static int bnxt_hwrm_cfa_ntuple_filter_free(struct bnxt *bp,
 					    struct bnxt_ntuple_filter *fltr)
 {
-	struct hwrm_cfa_ntuple_filter_free_input req = {0};
+	struct hwrm_cfa_ntuple_filter_free_input *req;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_NTUPLE_FILTER_FREE, -1, -1);
-	req.ntuple_filter_id = fltr->filter_id;
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_CFA_NTUPLE_FILTER_FREE);
+	if (rc)
+		return rc;
+
+	req->ntuple_filter_id = fltr->filter_id;
+	return hwrm_req_send(bp, req);
 }
 
 #define BNXT_NTP_FLTR_FLAGS					\
@@ -4744,69 +4761,70 @@ static int bnxt_hwrm_cfa_ntuple_filter_free(struct bnxt *bp,
 static int bnxt_hwrm_cfa_ntuple_filter_alloc(struct bnxt *bp,
 					     struct bnxt_ntuple_filter *fltr)
 {
-	struct hwrm_cfa_ntuple_filter_alloc_input req = {0};
 	struct hwrm_cfa_ntuple_filter_alloc_output *resp;
+	struct hwrm_cfa_ntuple_filter_alloc_input *req;
 	struct flow_keys *keys = &fltr->fkeys;
 	struct bnxt_vnic_info *vnic;
 	u32 flags = 0;
-	int rc = 0;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_NTUPLE_FILTER_ALLOC, -1, -1);
-	req.l2_filter_id = bp->vnic_info[0].fw_l2_filter_id[fltr->l2_fltr_idx];
+	rc = hwrm_req_init(bp, req, HWRM_CFA_NTUPLE_FILTER_ALLOC);
+	if (rc)
+		return rc;
+
+	req->l2_filter_id = bp->vnic_info[0].fw_l2_filter_id[fltr->l2_fltr_idx];
 
 	if (bp->fw_cap & BNXT_FW_CAP_CFA_RFS_RING_TBL_IDX_V2) {
 		flags = CFA_NTUPLE_FILTER_ALLOC_REQ_FLAGS_DEST_RFS_RING_IDX;
-		req.dst_id = cpu_to_le16(fltr->rxq);
+		req->dst_id = cpu_to_le16(fltr->rxq);
 	} else {
 		vnic = &bp->vnic_info[fltr->rxq + 1];
-		req.dst_id = cpu_to_le16(vnic->fw_vnic_id);
+		req->dst_id = cpu_to_le16(vnic->fw_vnic_id);
 	}
-	req.flags = cpu_to_le32(flags);
-	req.enables = cpu_to_le32(BNXT_NTP_FLTR_FLAGS);
+	req->flags = cpu_to_le32(flags);
+	req->enables = cpu_to_le32(BNXT_NTP_FLTR_FLAGS);
 
-	req.ethertype = htons(ETH_P_IP);
-	memcpy(req.src_macaddr, fltr->src_mac_addr, ETH_ALEN);
-	req.ip_addr_type = CFA_NTUPLE_FILTER_ALLOC_REQ_IP_ADDR_TYPE_IPV4;
-	req.ip_protocol = keys->basic.ip_proto;
+	req->ethertype = htons(ETH_P_IP);
+	memcpy(req->src_macaddr, fltr->src_mac_addr, ETH_ALEN);
+	req->ip_addr_type = CFA_NTUPLE_FILTER_ALLOC_REQ_IP_ADDR_TYPE_IPV4;
+	req->ip_protocol = keys->basic.ip_proto;
 
 	if (keys->basic.n_proto == htons(ETH_P_IPV6)) {
 		int i;
 
-		req.ethertype = htons(ETH_P_IPV6);
-		req.ip_addr_type =
+		req->ethertype = htons(ETH_P_IPV6);
+		req->ip_addr_type =
 			CFA_NTUPLE_FILTER_ALLOC_REQ_IP_ADDR_TYPE_IPV6;
-		*(struct in6_addr *)&req.src_ipaddr[0] =
+		*(struct in6_addr *)&req->src_ipaddr[0] =
 			keys->addrs.v6addrs.src;
-		*(struct in6_addr *)&req.dst_ipaddr[0] =
+		*(struct in6_addr *)&req->dst_ipaddr[0] =
 			keys->addrs.v6addrs.dst;
 		for (i = 0; i < 4; i++) {
-			req.src_ipaddr_mask[i] = cpu_to_be32(0xffffffff);
-			req.dst_ipaddr_mask[i] = cpu_to_be32(0xffffffff);
+			req->src_ipaddr_mask[i] = cpu_to_be32(0xffffffff);
+			req->dst_ipaddr_mask[i] = cpu_to_be32(0xffffffff);
 		}
 	} else {
-		req.src_ipaddr[0] = keys->addrs.v4addrs.src;
-		req.src_ipaddr_mask[0] = cpu_to_be32(0xffffffff);
-		req.dst_ipaddr[0] = keys->addrs.v4addrs.dst;
-		req.dst_ipaddr_mask[0] = cpu_to_be32(0xffffffff);
+		req->src_ipaddr[0] = keys->addrs.v4addrs.src;
+		req->src_ipaddr_mask[0] = cpu_to_be32(0xffffffff);
+		req->dst_ipaddr[0] = keys->addrs.v4addrs.dst;
+		req->dst_ipaddr_mask[0] = cpu_to_be32(0xffffffff);
 	}
 	if (keys->control.flags & FLOW_DIS_ENCAPSULATION) {
-		req.enables |= cpu_to_le32(BNXT_NTP_TUNNEL_FLTR_FLAG);
-		req.tunnel_type =
+		req->enables |= cpu_to_le32(BNXT_NTP_TUNNEL_FLTR_FLAG);
+		req->tunnel_type =
 			CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_ANYTUNNEL;
 	}
 
-	req.src_port = keys->ports.src;
-	req.src_port_mask = cpu_to_be16(0xffff);
-	req.dst_port = keys->ports.dst;
-	req.dst_port_mask = cpu_to_be16(0xffff);
+	req->src_port = keys->ports.src;
+	req->src_port_mask = cpu_to_be16(0xffff);
+	req->dst_port = keys->ports.dst;
+	req->dst_port_mask = cpu_to_be16(0xffff);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
-	if (!rc) {
-		resp = bnxt_get_hwrm_resp_addr(bp, &req);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
+	if (!rc)
 		fltr->filter_id = resp->ntuple_filter_id;
-	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 #endif
@@ -4814,62 +4832,62 @@ static int bnxt_hwrm_cfa_ntuple_filter_alloc(struct bnxt *bp,
 static int bnxt_hwrm_set_vnic_filter(struct bnxt *bp, u16 vnic_id, u16 idx,
 				     u8 *mac_addr)
 {
-	u32 rc = 0;
-	struct hwrm_cfa_l2_filter_alloc_input req = {0};
-	struct hwrm_cfa_l2_filter_alloc_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_cfa_l2_filter_alloc_output *resp;
+	struct hwrm_cfa_l2_filter_alloc_input *req;
+	int rc;
+
+	rc = hwrm_req_init(bp, req, HWRM_CFA_L2_FILTER_ALLOC);
+	if (rc)
+		return rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_L2_FILTER_ALLOC, -1, -1);
-	req.flags = cpu_to_le32(CFA_L2_FILTER_ALLOC_REQ_FLAGS_PATH_RX);
+	req->flags = cpu_to_le32(CFA_L2_FILTER_ALLOC_REQ_FLAGS_PATH_RX);
 	if (!BNXT_CHIP_TYPE_NITRO_A0(bp))
-		req.flags |=
+		req->flags |=
 			cpu_to_le32(CFA_L2_FILTER_ALLOC_REQ_FLAGS_OUTERMOST);
-	req.dst_id = cpu_to_le16(bp->vnic_info[vnic_id].fw_vnic_id);
-	req.enables =
+	req->dst_id = cpu_to_le16(bp->vnic_info[vnic_id].fw_vnic_id);
+	req->enables =
 		cpu_to_le32(CFA_L2_FILTER_ALLOC_REQ_ENABLES_L2_ADDR |
 			    CFA_L2_FILTER_ALLOC_REQ_ENABLES_DST_ID |
 			    CFA_L2_FILTER_ALLOC_REQ_ENABLES_L2_ADDR_MASK);
-	memcpy(req.l2_addr, mac_addr, ETH_ALEN);
-	req.l2_addr_mask[0] = 0xff;
-	req.l2_addr_mask[1] = 0xff;
-	req.l2_addr_mask[2] = 0xff;
-	req.l2_addr_mask[3] = 0xff;
-	req.l2_addr_mask[4] = 0xff;
-	req.l2_addr_mask[5] = 0xff;
-
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	memcpy(req->l2_addr, mac_addr, ETH_ALEN);
+	req->l2_addr_mask[0] = 0xff;
+	req->l2_addr_mask[1] = 0xff;
+	req->l2_addr_mask[2] = 0xff;
+	req->l2_addr_mask[3] = 0xff;
+	req->l2_addr_mask[4] = 0xff;
+	req->l2_addr_mask[5] = 0xff;
+
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		bp->vnic_info[vnic_id].fw_l2_filter_id[idx] =
 							resp->l2_filter_id;
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_clear_vnic_filter(struct bnxt *bp)
 {
+	struct hwrm_cfa_l2_filter_free_input *req;
 	u16 i, j, num_of_vnics = 1; /* only vnic 0 supported */
-	int rc = 0;
+	int rc;
 
 	/* Any associated ntuple filters will also be cleared by firmware. */
-	mutex_lock(&bp->hwrm_cmd_lock);
+	rc = hwrm_req_init(bp, req, HWRM_CFA_L2_FILTER_FREE);
+	if (rc)
+		return rc;
+	hwrm_req_hold(bp, req);
 	for (i = 0; i < num_of_vnics; i++) {
 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
 
 		for (j = 0; j < vnic->uc_filter_count; j++) {
-			struct hwrm_cfa_l2_filter_free_input req = {0};
+			req->l2_filter_id = vnic->fw_l2_filter_id[j];
 
-			bnxt_hwrm_cmd_hdr_init(bp, &req,
-					       HWRM_CFA_L2_FILTER_FREE, -1, -1);
-
-			req.l2_filter_id = vnic->fw_l2_filter_id[j];
-
-			rc = _hwrm_send_message(bp, &req, sizeof(req),
-						HWRM_CMD_TIMEOUT);
+			rc = hwrm_req_send(bp, req);
 		}
 		vnic->uc_filter_count = 0;
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
-
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -4877,12 +4895,15 @@ static int bnxt_hwrm_vnic_set_tpa(struct bnxt *bp, u16 vnic_id, u32 tpa_flags)
 {
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
 	u16 max_aggs = VNIC_TPA_CFG_REQ_MAX_AGGS_MAX;
-	struct hwrm_vnic_tpa_cfg_input req = {0};
+	struct hwrm_vnic_tpa_cfg_input *req;
+	int rc;
 
 	if (vnic->fw_vnic_id == INVALID_HW_RING_ID)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_TPA_CFG, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_TPA_CFG);
+	if (rc)
+		return rc;
 
 	if (tpa_flags) {
 		u16 mss = bp->dev->mtu - 40;
@@ -4896,9 +4917,9 @@ static int bnxt_hwrm_vnic_set_tpa(struct bnxt *bp, u16 vnic_id, u32 tpa_flags)
 		if (tpa_flags & BNXT_FLAG_GRO)
 			flags |= VNIC_TPA_CFG_REQ_FLAGS_GRO;
 
-		req.flags = cpu_to_le32(flags);
+		req->flags = cpu_to_le32(flags);
 
-		req.enables =
+		req->enables =
 			cpu_to_le32(VNIC_TPA_CFG_REQ_ENABLES_MAX_AGG_SEGS |
 				    VNIC_TPA_CFG_REQ_ENABLES_MAX_AGGS |
 				    VNIC_TPA_CFG_REQ_ENABLES_MIN_AGG_LEN);
@@ -4922,14 +4943,14 @@ static int bnxt_hwrm_vnic_set_tpa(struct bnxt *bp, u16 vnic_id, u32 tpa_flags)
 		} else {
 			segs = ilog2(nsegs);
 		}
-		req.max_agg_segs = cpu_to_le16(segs);
-		req.max_aggs = cpu_to_le16(max_aggs);
+		req->max_agg_segs = cpu_to_le16(segs);
+		req->max_aggs = cpu_to_le16(max_aggs);
 
-		req.min_agg_len = cpu_to_le32(512);
+		req->min_agg_len = cpu_to_le32(512);
 	}
-	req.vnic_id = cpu_to_le16(vnic->fw_vnic_id);
+	req->vnic_id = cpu_to_le16(vnic->fw_vnic_id);
 
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	return hwrm_req_send(bp, req);
 }
 
 static u16 bnxt_cp_ring_from_grp(struct bnxt *bp, struct bnxt_ring_struct *ring)
@@ -5073,86 +5094,102 @@ static void bnxt_fill_hw_rss_tbl(struct bnxt *bp, struct bnxt_vnic_info *vnic)
 static int bnxt_hwrm_vnic_set_rss(struct bnxt *bp, u16 vnic_id, bool set_rss)
 {
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
-	struct hwrm_vnic_rss_cfg_input req = {0};
+	struct hwrm_vnic_rss_cfg_input *req;
+	int rc;
 
 	if ((bp->flags & BNXT_FLAG_CHIP_P5) ||
 	    vnic->fw_rss_cos_lb_ctx[0] == INVALID_HW_RING_ID)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_RSS_CFG, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_RSS_CFG);
+	if (rc)
+		return rc;
+
 	if (set_rss) {
 		bnxt_fill_hw_rss_tbl(bp, vnic);
-		req.hash_type = cpu_to_le32(bp->rss_hash_cfg);
-		req.hash_mode_flags = VNIC_RSS_CFG_REQ_HASH_MODE_FLAGS_DEFAULT;
-		req.ring_grp_tbl_addr = cpu_to_le64(vnic->rss_table_dma_addr);
-		req.hash_key_tbl_addr =
+		req->hash_type = cpu_to_le32(bp->rss_hash_cfg);
+		req->hash_mode_flags = VNIC_RSS_CFG_REQ_HASH_MODE_FLAGS_DEFAULT;
+		req->ring_grp_tbl_addr = cpu_to_le64(vnic->rss_table_dma_addr);
+		req->hash_key_tbl_addr =
 			cpu_to_le64(vnic->rss_hash_key_dma_addr);
 	}
-	req.rss_ctx_idx = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[0]);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req->rss_ctx_idx = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[0]);
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_hwrm_vnic_set_rss_p5(struct bnxt *bp, u16 vnic_id, bool set_rss)
 {
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
-	struct hwrm_vnic_rss_cfg_input req = {0};
+	struct hwrm_vnic_rss_cfg_input *req;
 	dma_addr_t ring_tbl_map;
 	u32 i, nr_ctxs;
+	int rc;
+
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_RSS_CFG);
+	if (rc)
+		return rc;
+
+	req->vnic_id = cpu_to_le16(vnic->fw_vnic_id);
+	if (!set_rss)
+		return hwrm_req_send(bp, req);
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_RSS_CFG, -1, -1);
-	req.vnic_id = cpu_to_le16(vnic->fw_vnic_id);
-	if (!set_rss) {
-		hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
-		return 0;
-	}
 	bnxt_fill_hw_rss_tbl(bp, vnic);
-	req.hash_type = cpu_to_le32(bp->rss_hash_cfg);
-	req.hash_mode_flags = VNIC_RSS_CFG_REQ_HASH_MODE_FLAGS_DEFAULT;
-	req.hash_key_tbl_addr = cpu_to_le64(vnic->rss_hash_key_dma_addr);
+	req->hash_type = cpu_to_le32(bp->rss_hash_cfg);
+	req->hash_mode_flags = VNIC_RSS_CFG_REQ_HASH_MODE_FLAGS_DEFAULT;
+	req->hash_key_tbl_addr = cpu_to_le64(vnic->rss_hash_key_dma_addr);
 	ring_tbl_map = vnic->rss_table_dma_addr;
 	nr_ctxs = bnxt_get_nr_rss_ctxs(bp, bp->rx_nr_rings);
-	for (i = 0; i < nr_ctxs; ring_tbl_map += BNXT_RSS_TABLE_SIZE_P5, i++) {
-		int rc;
 
-		req.ring_grp_tbl_addr = cpu_to_le64(ring_tbl_map);
-		req.ring_table_pair_index = i;
-		req.rss_ctx_idx = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[i]);
-		rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	hwrm_req_hold(bp, req);
+	for (i = 0; i < nr_ctxs; ring_tbl_map += BNXT_RSS_TABLE_SIZE_P5, i++) {
+		req->ring_grp_tbl_addr = cpu_to_le64(ring_tbl_map);
+		req->ring_table_pair_index = i;
+		req->rss_ctx_idx = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[i]);
+		rc = hwrm_req_send(bp, req);
 		if (rc)
-			return rc;
+			goto exit;
 	}
-	return 0;
+
+exit:
+	hwrm_req_drop(bp, req);
+	return rc;
 }
 
 static int bnxt_hwrm_vnic_set_hds(struct bnxt *bp, u16 vnic_id)
 {
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
-	struct hwrm_vnic_plcmodes_cfg_input req = {0};
+	struct hwrm_vnic_plcmodes_cfg_input *req;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_PLCMODES_CFG, -1, -1);
-	req.flags = cpu_to_le32(VNIC_PLCMODES_CFG_REQ_FLAGS_JUMBO_PLACEMENT |
-				VNIC_PLCMODES_CFG_REQ_FLAGS_HDS_IPV4 |
-				VNIC_PLCMODES_CFG_REQ_FLAGS_HDS_IPV6);
-	req.enables =
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_PLCMODES_CFG);
+	if (rc)
+		return rc;
+
+	req->flags = cpu_to_le32(VNIC_PLCMODES_CFG_REQ_FLAGS_JUMBO_PLACEMENT |
+				 VNIC_PLCMODES_CFG_REQ_FLAGS_HDS_IPV4 |
+				 VNIC_PLCMODES_CFG_REQ_FLAGS_HDS_IPV6);
+	req->enables =
 		cpu_to_le32(VNIC_PLCMODES_CFG_REQ_ENABLES_JUMBO_THRESH_VALID |
 			    VNIC_PLCMODES_CFG_REQ_ENABLES_HDS_THRESHOLD_VALID);
 	/* thresholds not implemented in firmware yet */
-	req.jumbo_thresh = cpu_to_le16(bp->rx_copy_thresh);
-	req.hds_threshold = cpu_to_le16(bp->rx_copy_thresh);
-	req.vnic_id = cpu_to_le32(vnic->fw_vnic_id);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req->jumbo_thresh = cpu_to_le16(bp->rx_copy_thresh);
+	req->hds_threshold = cpu_to_le16(bp->rx_copy_thresh);
+	req->vnic_id = cpu_to_le32(vnic->fw_vnic_id);
+	return hwrm_req_send(bp, req);
 }
 
 static void bnxt_hwrm_vnic_ctx_free_one(struct bnxt *bp, u16 vnic_id,
 					u16 ctx_idx)
 {
-	struct hwrm_vnic_rss_cos_lb_ctx_free_input req = {0};
+	struct hwrm_vnic_rss_cos_lb_ctx_free_input *req;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_RSS_COS_LB_CTX_FREE, -1, -1);
-	req.rss_cos_lb_ctx_id =
+	if (hwrm_req_init(bp, req, HWRM_VNIC_RSS_COS_LB_CTX_FREE))
+		return;
+
+	req->rss_cos_lb_ctx_id =
 		cpu_to_le16(bp->vnic_info[vnic_id].fw_rss_cos_lb_ctx[ctx_idx]);
 
-	hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	hwrm_req_send(bp, req);
 	bp->vnic_info[vnic_id].fw_rss_cos_lb_ctx[ctx_idx] = INVALID_HW_RING_ID;
 }
 
@@ -5173,20 +5210,20 @@ static void bnxt_hwrm_vnic_ctx_free(struct bnxt *bp)
 
 static int bnxt_hwrm_vnic_ctx_alloc(struct bnxt *bp, u16 vnic_id, u16 ctx_idx)
 {
+	struct hwrm_vnic_rss_cos_lb_ctx_alloc_output *resp;
+	struct hwrm_vnic_rss_cos_lb_ctx_alloc_input *req;
 	int rc;
-	struct hwrm_vnic_rss_cos_lb_ctx_alloc_input req = {0};
-	struct hwrm_vnic_rss_cos_lb_ctx_alloc_output *resp =
-						bp->hwrm_cmd_resp_addr;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_RSS_COS_LB_CTX_ALLOC, -1,
-			       -1);
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_RSS_COS_LB_CTX_ALLOC);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		bp->vnic_info[vnic_id].fw_rss_cos_lb_ctx[ctx_idx] =
 			le16_to_cpu(resp->rss_cos_lb_ctx_id);
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 
 	return rc;
 }
@@ -5200,47 +5237,50 @@ static u32 bnxt_get_roce_vnic_mode(struct bnxt *bp)
 
 int bnxt_hwrm_vnic_cfg(struct bnxt *bp, u16 vnic_id)
 {
-	unsigned int ring = 0, grp_idx;
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
-	struct hwrm_vnic_cfg_input req = {0};
+	struct hwrm_vnic_cfg_input *req;
+	unsigned int ring = 0, grp_idx;
 	u16 def_vlan = 0;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_CFG, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_CFG);
+	if (rc)
+		return rc;
 
 	if (bp->flags & BNXT_FLAG_CHIP_P5) {
 		struct bnxt_rx_ring_info *rxr = &bp->rx_ring[0];
 
-		req.default_rx_ring_id =
+		req->default_rx_ring_id =
 			cpu_to_le16(rxr->rx_ring_struct.fw_ring_id);
-		req.default_cmpl_ring_id =
+		req->default_cmpl_ring_id =
 			cpu_to_le16(bnxt_cp_ring_for_rx(bp, rxr));
-		req.enables =
+		req->enables =
 			cpu_to_le32(VNIC_CFG_REQ_ENABLES_DEFAULT_RX_RING_ID |
 				    VNIC_CFG_REQ_ENABLES_DEFAULT_CMPL_RING_ID);
 		goto vnic_mru;
 	}
-	req.enables = cpu_to_le32(VNIC_CFG_REQ_ENABLES_DFLT_RING_GRP);
+	req->enables = cpu_to_le32(VNIC_CFG_REQ_ENABLES_DFLT_RING_GRP);
 	/* Only RSS support for now TBD: COS & LB */
 	if (vnic->fw_rss_cos_lb_ctx[0] != INVALID_HW_RING_ID) {
-		req.rss_rule = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[0]);
-		req.enables |= cpu_to_le32(VNIC_CFG_REQ_ENABLES_RSS_RULE |
+		req->rss_rule = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[0]);
+		req->enables |= cpu_to_le32(VNIC_CFG_REQ_ENABLES_RSS_RULE |
 					   VNIC_CFG_REQ_ENABLES_MRU);
 	} else if (vnic->flags & BNXT_VNIC_RFS_NEW_RSS_FLAG) {
-		req.rss_rule =
+		req->rss_rule =
 			cpu_to_le16(bp->vnic_info[0].fw_rss_cos_lb_ctx[0]);
-		req.enables |= cpu_to_le32(VNIC_CFG_REQ_ENABLES_RSS_RULE |
+		req->enables |= cpu_to_le32(VNIC_CFG_REQ_ENABLES_RSS_RULE |
 					   VNIC_CFG_REQ_ENABLES_MRU);
-		req.flags |= cpu_to_le32(VNIC_CFG_REQ_FLAGS_RSS_DFLT_CR_MODE);
+		req->flags |= cpu_to_le32(VNIC_CFG_REQ_FLAGS_RSS_DFLT_CR_MODE);
 	} else {
-		req.rss_rule = cpu_to_le16(0xffff);
+		req->rss_rule = cpu_to_le16(0xffff);
 	}
 
 	if (BNXT_CHIP_TYPE_NITRO_A0(bp) &&
 	    (vnic->fw_rss_cos_lb_ctx[0] != INVALID_HW_RING_ID)) {
-		req.cos_rule = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[1]);
-		req.enables |= cpu_to_le32(VNIC_CFG_REQ_ENABLES_COS_RULE);
+		req->cos_rule = cpu_to_le16(vnic->fw_rss_cos_lb_ctx[1]);
+		req->enables |= cpu_to_le32(VNIC_CFG_REQ_ENABLES_COS_RULE);
 	} else {
-		req.cos_rule = cpu_to_le16(0xffff);
+		req->cos_rule = cpu_to_le16(0xffff);
 	}
 
 	if (vnic->flags & BNXT_VNIC_RSS_FLAG)
@@ -5251,34 +5291,36 @@ int bnxt_hwrm_vnic_cfg(struct bnxt *bp, u16 vnic_id)
 		ring = bp->rx_nr_rings - 1;
 
 	grp_idx = bp->rx_ring[ring].bnapi->index;
-	req.dflt_ring_grp = cpu_to_le16(bp->grp_info[grp_idx].fw_grp_id);
-	req.lb_rule = cpu_to_le16(0xffff);
+	req->dflt_ring_grp = cpu_to_le16(bp->grp_info[grp_idx].fw_grp_id);
+	req->lb_rule = cpu_to_le16(0xffff);
 vnic_mru:
-	req.mru = cpu_to_le16(bp->dev->mtu + ETH_HLEN + VLAN_HLEN);
+	req->mru = cpu_to_le16(bp->dev->mtu + ETH_HLEN + VLAN_HLEN);
 
-	req.vnic_id = cpu_to_le16(vnic->fw_vnic_id);
+	req->vnic_id = cpu_to_le16(vnic->fw_vnic_id);
 #ifdef CONFIG_BNXT_SRIOV
 	if (BNXT_VF(bp))
 		def_vlan = bp->vf.vlan;
 #endif
 	if ((bp->flags & BNXT_FLAG_STRIP_VLAN) || def_vlan)
-		req.flags |= cpu_to_le32(VNIC_CFG_REQ_FLAGS_VLAN_STRIP_MODE);
+		req->flags |= cpu_to_le32(VNIC_CFG_REQ_FLAGS_VLAN_STRIP_MODE);
 	if (!vnic_id && bnxt_ulp_registered(bp->edev, BNXT_ROCE_ULP))
-		req.flags |= cpu_to_le32(bnxt_get_roce_vnic_mode(bp));
+		req->flags |= cpu_to_le32(bnxt_get_roce_vnic_mode(bp));
 
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	return hwrm_req_send(bp, req);
 }
 
 static void bnxt_hwrm_vnic_free_one(struct bnxt *bp, u16 vnic_id)
 {
 	if (bp->vnic_info[vnic_id].fw_vnic_id != INVALID_HW_RING_ID) {
-		struct hwrm_vnic_free_input req = {0};
+		struct hwrm_vnic_free_input *req;
+
+		if (hwrm_req_init(bp, req, HWRM_VNIC_FREE))
+			return;
 
-		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_FREE, -1, -1);
-		req.vnic_id =
+		req->vnic_id =
 			cpu_to_le32(bp->vnic_info[vnic_id].fw_vnic_id);
 
-		hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+		hwrm_req_send(bp, req);
 		bp->vnic_info[vnic_id].fw_vnic_id = INVALID_HW_RING_ID;
 	}
 }
@@ -5295,11 +5337,15 @@ static int bnxt_hwrm_vnic_alloc(struct bnxt *bp, u16 vnic_id,
 				unsigned int start_rx_ring_idx,
 				unsigned int nr_rings)
 {
-	int rc = 0;
 	unsigned int i, j, grp_idx, end_idx = start_rx_ring_idx + nr_rings;
-	struct hwrm_vnic_alloc_input req = {0};
-	struct hwrm_vnic_alloc_output *resp = bp->hwrm_cmd_resp_addr;
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
+	struct hwrm_vnic_alloc_output *resp;
+	struct hwrm_vnic_alloc_input *req;
+	int rc;
+
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_ALLOC);
+	if (rc)
+		return rc;
 
 	if (bp->flags & BNXT_FLAG_CHIP_P5)
 		goto vnic_no_ring_grps;
@@ -5319,22 +5365,20 @@ static int bnxt_hwrm_vnic_alloc(struct bnxt *bp, u16 vnic_id,
 	for (i = 0; i < BNXT_MAX_CTX_PER_VNIC; i++)
 		vnic->fw_rss_cos_lb_ctx[i] = INVALID_HW_RING_ID;
 	if (vnic_id == 0)
-		req.flags = cpu_to_le32(VNIC_ALLOC_REQ_FLAGS_DEFAULT);
-
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_ALLOC, -1, -1);
+		req->flags = cpu_to_le32(VNIC_ALLOC_REQ_FLAGS_DEFAULT);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		vnic->fw_vnic_id = le32_to_cpu(resp->vnic_id);
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_vnic_qcaps(struct bnxt *bp)
 {
-	struct hwrm_vnic_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_vnic_qcaps_input req = {0};
+	struct hwrm_vnic_qcaps_output *resp;
+	struct hwrm_vnic_qcaps_input *req;
 	int rc;
 
 	bp->hw_ring_stats_size = sizeof(struct ctx_hw_stats);
@@ -5342,9 +5386,12 @@ static int bnxt_hwrm_vnic_qcaps(struct bnxt *bp)
 	if (bp->hwrm_spec_code < 0x10600)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VNIC_QCAPS, -1, -1);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_VNIC_QCAPS);
+	if (rc)
+		return rc;
+
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc) {
 		u32 flags = le32_to_cpu(resp->flags);
 
@@ -5370,92 +5417,96 @@ static int bnxt_hwrm_vnic_qcaps(struct bnxt *bp)
 				bp->hw_ring_stats_size = BNXT_RING_STATS_SIZE_P5_SR2;
 		}
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_ring_grp_alloc(struct bnxt *bp)
 {
+	struct hwrm_ring_grp_alloc_output *resp;
+	struct hwrm_ring_grp_alloc_input *req;
+	int rc;
 	u16 i;
-	u32 rc = 0;
 
 	if (bp->flags & BNXT_FLAG_CHIP_P5)
 		return 0;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
+	rc = hwrm_req_init(bp, req, HWRM_RING_GRP_ALLOC);
+	if (rc)
+		return rc;
+
+	resp = hwrm_req_hold(bp, req);
 	for (i = 0; i < bp->rx_nr_rings; i++) {
-		struct hwrm_ring_grp_alloc_input req = {0};
-		struct hwrm_ring_grp_alloc_output *resp =
-					bp->hwrm_cmd_resp_addr;
 		unsigned int grp_idx = bp->rx_ring[i].bnapi->index;
 
-		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_RING_GRP_ALLOC, -1, -1);
+		req->cr = cpu_to_le16(bp->grp_info[grp_idx].cp_fw_ring_id);
+		req->rr = cpu_to_le16(bp->grp_info[grp_idx].rx_fw_ring_id);
+		req->ar = cpu_to_le16(bp->grp_info[grp_idx].agg_fw_ring_id);
+		req->sc = cpu_to_le16(bp->grp_info[grp_idx].fw_stats_ctx);
 
-		req.cr = cpu_to_le16(bp->grp_info[grp_idx].cp_fw_ring_id);
-		req.rr = cpu_to_le16(bp->grp_info[grp_idx].rx_fw_ring_id);
-		req.ar = cpu_to_le16(bp->grp_info[grp_idx].agg_fw_ring_id);
-		req.sc = cpu_to_le16(bp->grp_info[grp_idx].fw_stats_ctx);
+		rc = hwrm_req_send(bp, req);
 
-		rc = _hwrm_send_message(bp, &req, sizeof(req),
-					HWRM_CMD_TIMEOUT);
 		if (rc)
 			break;
 
 		bp->grp_info[grp_idx].fw_grp_id =
 			le32_to_cpu(resp->ring_group_id);
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static void bnxt_hwrm_ring_grp_free(struct bnxt *bp)
 {
+	struct hwrm_ring_grp_free_input *req;
 	u16 i;
-	struct hwrm_ring_grp_free_input req = {0};
 
 	if (!bp->grp_info || (bp->flags & BNXT_FLAG_CHIP_P5))
 		return;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_RING_GRP_FREE, -1, -1);
+	if (hwrm_req_init(bp, req, HWRM_RING_GRP_FREE))
+		return;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
+	hwrm_req_hold(bp, req);
 	for (i = 0; i < bp->cp_nr_rings; i++) {
 		if (bp->grp_info[i].fw_grp_id == INVALID_HW_RING_ID)
 			continue;
-		req.ring_group_id =
+		req->ring_group_id =
 			cpu_to_le32(bp->grp_info[i].fw_grp_id);
 
-		_hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+		hwrm_req_send(bp, req);
 		bp->grp_info[i].fw_grp_id = INVALID_HW_RING_ID;
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 }
 
 static int hwrm_ring_alloc_send_msg(struct bnxt *bp,
 				    struct bnxt_ring_struct *ring,
 				    u32 ring_type, u32 map_index)
 {
-	int rc = 0, err = 0;
-	struct hwrm_ring_alloc_input req = {0};
-	struct hwrm_ring_alloc_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_ring_alloc_output *resp;
+	struct hwrm_ring_alloc_input *req;
 	struct bnxt_ring_mem_info *rmem = &ring->ring_mem;
 	struct bnxt_ring_grp_info *grp_info;
+	int rc, err = 0;
 	u16 ring_id;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_RING_ALLOC, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_RING_ALLOC);
+	if (rc)
+		goto exit;
 
-	req.enables = 0;
+	req->enables = 0;
 	if (rmem->nr_pages > 1) {
-		req.page_tbl_addr = cpu_to_le64(rmem->pg_tbl_map);
+		req->page_tbl_addr = cpu_to_le64(rmem->pg_tbl_map);
 		/* Page size is in log2 units */
-		req.page_size = BNXT_PAGE_SHIFT;
-		req.page_tbl_depth = 1;
+		req->page_size = BNXT_PAGE_SHIFT;
+		req->page_tbl_depth = 1;
 	} else {
-		req.page_tbl_addr =  cpu_to_le64(rmem->dma_arr[0]);
+		req->page_tbl_addr =  cpu_to_le64(rmem->dma_arr[0]);
 	}
-	req.fbo = 0;
+	req->fbo = 0;
 	/* Association of ring index with doorbell index and MSIX number */
-	req.logical_id = cpu_to_le16(map_index);
+	req->logical_id = cpu_to_le16(map_index);
 
 	switch (ring_type) {
 	case HWRM_RING_ALLOC_TX: {
@@ -5463,67 +5514,67 @@ static int hwrm_ring_alloc_send_msg(struct bnxt *bp,
 
 		txr = container_of(ring, struct bnxt_tx_ring_info,
 				   tx_ring_struct);
-		req.ring_type = RING_ALLOC_REQ_RING_TYPE_TX;
+		req->ring_type = RING_ALLOC_REQ_RING_TYPE_TX;
 		/* Association of transmit ring with completion ring */
 		grp_info = &bp->grp_info[ring->grp_idx];
-		req.cmpl_ring_id = cpu_to_le16(bnxt_cp_ring_for_tx(bp, txr));
-		req.length = cpu_to_le32(bp->tx_ring_mask + 1);
-		req.stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
-		req.queue_id = cpu_to_le16(ring->queue_id);
+		req->cmpl_ring_id = cpu_to_le16(bnxt_cp_ring_for_tx(bp, txr));
+		req->length = cpu_to_le32(bp->tx_ring_mask + 1);
+		req->stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
+		req->queue_id = cpu_to_le16(ring->queue_id);
 		break;
 	}
 	case HWRM_RING_ALLOC_RX:
-		req.ring_type = RING_ALLOC_REQ_RING_TYPE_RX;
-		req.length = cpu_to_le32(bp->rx_ring_mask + 1);
+		req->ring_type = RING_ALLOC_REQ_RING_TYPE_RX;
+		req->length = cpu_to_le32(bp->rx_ring_mask + 1);
 		if (bp->flags & BNXT_FLAG_CHIP_P5) {
 			u16 flags = 0;
 
 			/* Association of rx ring with stats context */
 			grp_info = &bp->grp_info[ring->grp_idx];
-			req.rx_buf_size = cpu_to_le16(bp->rx_buf_use_size);
-			req.stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
-			req.enables |= cpu_to_le32(
+			req->rx_buf_size = cpu_to_le16(bp->rx_buf_use_size);
+			req->stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
+			req->enables |= cpu_to_le32(
 				RING_ALLOC_REQ_ENABLES_RX_BUF_SIZE_VALID);
 			if (NET_IP_ALIGN == 2)
 				flags = RING_ALLOC_REQ_FLAGS_RX_SOP_PAD;
-			req.flags = cpu_to_le16(flags);
+			req->flags = cpu_to_le16(flags);
 		}
 		break;
 	case HWRM_RING_ALLOC_AGG:
 		if (bp->flags & BNXT_FLAG_CHIP_P5) {
-			req.ring_type = RING_ALLOC_REQ_RING_TYPE_RX_AGG;
+			req->ring_type = RING_ALLOC_REQ_RING_TYPE_RX_AGG;
 			/* Association of agg ring with rx ring */
 			grp_info = &bp->grp_info[ring->grp_idx];
-			req.rx_ring_id = cpu_to_le16(grp_info->rx_fw_ring_id);
-			req.rx_buf_size = cpu_to_le16(BNXT_RX_PAGE_SIZE);
-			req.stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
-			req.enables |= cpu_to_le32(
+			req->rx_ring_id = cpu_to_le16(grp_info->rx_fw_ring_id);
+			req->rx_buf_size = cpu_to_le16(BNXT_RX_PAGE_SIZE);
+			req->stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
+			req->enables |= cpu_to_le32(
 				RING_ALLOC_REQ_ENABLES_RX_RING_ID_VALID |
 				RING_ALLOC_REQ_ENABLES_RX_BUF_SIZE_VALID);
 		} else {
-			req.ring_type = RING_ALLOC_REQ_RING_TYPE_RX;
+			req->ring_type = RING_ALLOC_REQ_RING_TYPE_RX;
 		}
-		req.length = cpu_to_le32(bp->rx_agg_ring_mask + 1);
+		req->length = cpu_to_le32(bp->rx_agg_ring_mask + 1);
 		break;
 	case HWRM_RING_ALLOC_CMPL:
-		req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
-		req.length = cpu_to_le32(bp->cp_ring_mask + 1);
+		req->ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
+		req->length = cpu_to_le32(bp->cp_ring_mask + 1);
 		if (bp->flags & BNXT_FLAG_CHIP_P5) {
 			/* Association of cp ring with nq */
 			grp_info = &bp->grp_info[map_index];
-			req.nq_ring_id = cpu_to_le16(grp_info->cp_fw_ring_id);
-			req.cq_handle = cpu_to_le64(ring->handle);
-			req.enables |= cpu_to_le32(
+			req->nq_ring_id = cpu_to_le16(grp_info->cp_fw_ring_id);
+			req->cq_handle = cpu_to_le64(ring->handle);
+			req->enables |= cpu_to_le32(
 				RING_ALLOC_REQ_ENABLES_NQ_RING_ID_VALID);
 		} else if (bp->flags & BNXT_FLAG_USING_MSIX) {
-			req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
+			req->int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
 		}
 		break;
 	case HWRM_RING_ALLOC_NQ:
-		req.ring_type = RING_ALLOC_REQ_RING_TYPE_NQ;
-		req.length = cpu_to_le32(bp->cp_ring_mask + 1);
+		req->ring_type = RING_ALLOC_REQ_RING_TYPE_NQ;
+		req->length = cpu_to_le32(bp->cp_ring_mask + 1);
 		if (bp->flags & BNXT_FLAG_USING_MSIX)
-			req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
+			req->int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
 		break;
 	default:
 		netdev_err(bp->dev, "hwrm alloc invalid ring type %d\n",
@@ -5531,12 +5582,13 @@ static int hwrm_ring_alloc_send_msg(struct bnxt *bp,
 		return -1;
 	}
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	err = le16_to_cpu(resp->error_code);
 	ring_id = le16_to_cpu(resp->ring_id);
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 
+exit:
 	if (rc || err) {
 		netdev_err(bp->dev, "hwrm_ring_alloc type %d failed. rc:%x err:%x\n",
 			   ring_type, rc, err);
@@ -5551,23 +5603,28 @@ static int bnxt_hwrm_set_async_event_cr(struct bnxt *bp, int idx)
 	int rc;
 
 	if (BNXT_PF(bp)) {
-		struct hwrm_func_cfg_input req = {0};
+		struct hwrm_func_cfg_input *req;
 
-		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
-		req.fid = cpu_to_le16(0xffff);
-		req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_ASYNC_EVENT_CR);
-		req.async_event_cr = cpu_to_le16(idx);
-		rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+		rc = hwrm_req_init(bp, req, HWRM_FUNC_CFG);
+		if (rc)
+			return rc;
+
+		req->fid = cpu_to_le16(0xffff);
+		req->enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_ASYNC_EVENT_CR);
+		req->async_event_cr = cpu_to_le16(idx);
+		return hwrm_req_send(bp, req);
 	} else {
-		struct hwrm_func_vf_cfg_input req = {0};
+		struct hwrm_func_vf_cfg_input *req;
+
+		rc = hwrm_req_init(bp, req, HWRM_FUNC_VF_CFG);
+		if (rc)
+			return rc;
 
-		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_VF_CFG, -1, -1);
-		req.enables =
+		req->enables =
 			cpu_to_le32(FUNC_VF_CFG_REQ_ENABLES_ASYNC_EVENT_CR);
-		req.async_event_cr = cpu_to_le16(idx);
-		rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+		req->async_event_cr = cpu_to_le16(idx);
+		return hwrm_req_send(bp, req);
 	}
-	return rc;
 }
 
 static void bnxt_set_db(struct bnxt *bp, struct bnxt_db_info *db, u32 ring_type,
@@ -5738,23 +5795,27 @@ static int hwrm_ring_free_send_msg(struct bnxt *bp,
 				   struct bnxt_ring_struct *ring,
 				   u32 ring_type, int cmpl_ring_id)
 {
+	struct hwrm_ring_free_output *resp;
+	struct hwrm_ring_free_input *req;
+	u16 error_code = 0;
 	int rc;
-	struct hwrm_ring_free_input req = {0};
-	struct hwrm_ring_free_output *resp = bp->hwrm_cmd_resp_addr;
-	u16 error_code;
 
 	if (BNXT_NO_FW_ACCESS(bp))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_RING_FREE, cmpl_ring_id, -1);
-	req.ring_type = ring_type;
-	req.ring_id = cpu_to_le16(ring->fw_ring_id);
+	rc = hwrm_req_init(bp, req, HWRM_RING_FREE);
+	if (rc)
+		goto exit;
+
+	req->cmpl_ring = cpu_to_le16(cmpl_ring_id);
+	req->ring_type = ring_type;
+	req->ring_id = cpu_to_le16(ring->fw_ring_id);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	error_code = le16_to_cpu(resp->error_code);
-	mutex_unlock(&bp->hwrm_cmd_lock);
-
+	hwrm_req_drop(bp, req);
+exit:
 	if (rc || error_code) {
 		netdev_err(bp->dev, "hwrm_ring_free type %d failed. rc:%x err:%x\n",
 			   ring_type, rc, error_code);
@@ -5869,20 +5930,23 @@ static int bnxt_trim_rings(struct bnxt *bp, int *rx, int *tx, int max,
 
 static int bnxt_hwrm_get_rings(struct bnxt *bp)
 {
-	struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
 	struct bnxt_hw_resc *hw_resc = &bp->hw_resc;
-	struct hwrm_func_qcfg_input req = {0};
+	struct hwrm_func_qcfg_output *resp;
+	struct hwrm_func_qcfg_input *req;
 	int rc;
 
 	if (bp->hwrm_spec_code < 0x10601)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCFG, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_QCFG);
+	if (rc)
+		return rc;
+
+	req->fid = cpu_to_le16(0xffff);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc) {
-		mutex_unlock(&bp->hwrm_cmd_lock);
+		hwrm_req_drop(bp, req);
 		return rc;
 	}
 
@@ -5916,39 +5980,45 @@ static int bnxt_hwrm_get_rings(struct bnxt *bp)
 		hw_resc->resv_cp_rings = cp;
 		hw_resc->resv_stat_ctxs = stats;
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return 0;
 }
 
-/* Caller must hold bp->hwrm_cmd_lock */
 int __bnxt_hwrm_get_tx_rings(struct bnxt *bp, u16 fid, int *tx_rings)
 {
-	struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_func_qcfg_input req = {0};
+	struct hwrm_func_qcfg_output *resp;
+	struct hwrm_func_qcfg_input *req;
 	int rc;
 
 	if (bp->hwrm_spec_code < 0x10601)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCFG, -1, -1);
-	req.fid = cpu_to_le16(fid);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_QCFG);
+	if (rc)
+		return rc;
+
+	req->fid = cpu_to_le16(fid);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		*tx_rings = le16_to_cpu(resp->alloc_tx_rings);
 
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static bool bnxt_rfs_supported(struct bnxt *bp);
 
-static void
-__bnxt_hwrm_reserve_pf_rings(struct bnxt *bp, struct hwrm_func_cfg_input *req,
-			     int tx_rings, int rx_rings, int ring_grps,
-			     int cp_rings, int stats, int vnics)
+static struct hwrm_func_cfg_input *
+__bnxt_hwrm_reserve_pf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
+			     int ring_grps, int cp_rings, int stats, int vnics)
 {
+	struct hwrm_func_cfg_input *req;
 	u32 enables = 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, req, HWRM_FUNC_CFG, -1, -1);
+	if (hwrm_req_init(bp, req, HWRM_FUNC_CFG))
+		return NULL;
+
 	req->fid = cpu_to_le16(0xffff);
 	enables |= tx_rings ? FUNC_CFG_REQ_ENABLES_NUM_TX_RINGS : 0;
 	req->num_tx_rings = cpu_to_le16(tx_rings);
@@ -5989,17 +6059,19 @@ __bnxt_hwrm_reserve_pf_rings(struct bnxt *bp, struct hwrm_func_cfg_input *req,
 		req->num_vnics = cpu_to_le16(vnics);
 	}
 	req->enables = cpu_to_le32(enables);
+	return req;
 }
 
-static void
-__bnxt_hwrm_reserve_vf_rings(struct bnxt *bp,
-			     struct hwrm_func_vf_cfg_input *req, int tx_rings,
-			     int rx_rings, int ring_grps, int cp_rings,
-			     int stats, int vnics)
+static struct hwrm_func_vf_cfg_input *
+__bnxt_hwrm_reserve_vf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
+			     int ring_grps, int cp_rings, int stats, int vnics)
 {
+	struct hwrm_func_vf_cfg_input *req;
 	u32 enables = 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, req, HWRM_FUNC_VF_CFG, -1, -1);
+	if (hwrm_req_init(bp, req, HWRM_FUNC_VF_CFG))
+		return NULL;
+
 	enables |= tx_rings ? FUNC_VF_CFG_REQ_ENABLES_NUM_TX_RINGS : 0;
 	enables |= rx_rings ? FUNC_VF_CFG_REQ_ENABLES_NUM_RX_RINGS |
 			      FUNC_VF_CFG_REQ_ENABLES_NUM_RSSCOS_CTXS : 0;
@@ -6031,21 +6103,27 @@ __bnxt_hwrm_reserve_vf_rings(struct bnxt *bp,
 	req->num_vnics = cpu_to_le16(vnics);
 
 	req->enables = cpu_to_le32(enables);
+	return req;
 }
 
 static int
 bnxt_hwrm_reserve_pf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 			   int ring_grps, int cp_rings, int stats, int vnics)
 {
-	struct hwrm_func_cfg_input req = {0};
+	struct hwrm_func_cfg_input *req;
 	int rc;
 
-	__bnxt_hwrm_reserve_pf_rings(bp, &req, tx_rings, rx_rings, ring_grps,
-				     cp_rings, stats, vnics);
-	if (!req.enables)
+	req = __bnxt_hwrm_reserve_pf_rings(bp, tx_rings, rx_rings, ring_grps,
+					   cp_rings, stats, vnics);
+	if (!req)
+		return -ENOMEM;
+
+	if (!req->enables) {
+		hwrm_req_drop(bp, req);
 		return 0;
+	}
 
-	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		return rc;
 
@@ -6059,7 +6137,7 @@ static int
 bnxt_hwrm_reserve_vf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 			   int ring_grps, int cp_rings, int stats, int vnics)
 {
-	struct hwrm_func_vf_cfg_input req = {0};
+	struct hwrm_func_vf_cfg_input *req;
 	int rc;
 
 	if (!BNXT_NEW_RM(bp)) {
@@ -6067,9 +6145,12 @@ bnxt_hwrm_reserve_vf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 		return 0;
 	}
 
-	__bnxt_hwrm_reserve_vf_rings(bp, &req, tx_rings, rx_rings, ring_grps,
-				     cp_rings, stats, vnics);
-	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req = __bnxt_hwrm_reserve_vf_rings(bp, tx_rings, rx_rings, ring_grps,
+					   cp_rings, stats, vnics);
+	if (!req)
+		return -ENOMEM;
+
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		return rc;
 
@@ -6270,14 +6351,14 @@ static int bnxt_hwrm_check_vf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 				    int ring_grps, int cp_rings, int stats,
 				    int vnics)
 {
-	struct hwrm_func_vf_cfg_input req = {0};
+	struct hwrm_func_vf_cfg_input *req;
 	u32 flags;
 
 	if (!BNXT_NEW_RM(bp))
 		return 0;
 
-	__bnxt_hwrm_reserve_vf_rings(bp, &req, tx_rings, rx_rings, ring_grps,
-				     cp_rings, stats, vnics);
+	req = __bnxt_hwrm_reserve_vf_rings(bp, tx_rings, rx_rings, ring_grps,
+					   cp_rings, stats, vnics);
 	flags = FUNC_VF_CFG_REQ_FLAGS_TX_ASSETS_TEST |
 		FUNC_VF_CFG_REQ_FLAGS_RX_ASSETS_TEST |
 		FUNC_VF_CFG_REQ_FLAGS_CMPL_ASSETS_TEST |
@@ -6287,20 +6368,19 @@ static int bnxt_hwrm_check_vf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 	if (!(bp->flags & BNXT_FLAG_CHIP_P5))
 		flags |= FUNC_VF_CFG_REQ_FLAGS_RING_GRP_ASSETS_TEST;
 
-	req.flags = cpu_to_le32(flags);
-	return hwrm_send_message_silent(bp, &req, sizeof(req),
-					HWRM_CMD_TIMEOUT);
+	req->flags = cpu_to_le32(flags);
+	return hwrm_req_send_silent(bp, req);
 }
 
 static int bnxt_hwrm_check_pf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 				    int ring_grps, int cp_rings, int stats,
 				    int vnics)
 {
-	struct hwrm_func_cfg_input req = {0};
+	struct hwrm_func_cfg_input *req;
 	u32 flags;
 
-	__bnxt_hwrm_reserve_pf_rings(bp, &req, tx_rings, rx_rings, ring_grps,
-				     cp_rings, stats, vnics);
+	req = __bnxt_hwrm_reserve_pf_rings(bp, tx_rings, rx_rings, ring_grps,
+					   cp_rings, stats, vnics);
 	flags = FUNC_CFG_REQ_FLAGS_TX_ASSETS_TEST;
 	if (BNXT_NEW_RM(bp)) {
 		flags |= FUNC_CFG_REQ_FLAGS_RX_ASSETS_TEST |
@@ -6314,9 +6394,8 @@ static int bnxt_hwrm_check_pf_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 			flags |= FUNC_CFG_REQ_FLAGS_RING_GRP_ASSETS_TEST;
 	}
 
-	req.flags = cpu_to_le32(flags);
-	return hwrm_send_message_silent(bp, &req, sizeof(req),
-					HWRM_CMD_TIMEOUT);
+	req->flags = cpu_to_le32(flags);
+	return hwrm_req_send_silent(bp, req);
 }
 
 static int bnxt_hwrm_check_rings(struct bnxt *bp, int tx_rings, int rx_rings,
@@ -6337,9 +6416,9 @@ static int bnxt_hwrm_check_rings(struct bnxt *bp, int tx_rings, int rx_rings,
 
 static void bnxt_hwrm_coal_params_qcaps(struct bnxt *bp)
 {
-	struct hwrm_ring_aggint_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
 	struct bnxt_coal_cap *coal_cap = &bp->coal_cap;
-	struct hwrm_ring_aggint_qcaps_input req = {0};
+	struct hwrm_ring_aggint_qcaps_output *resp;
+	struct hwrm_ring_aggint_qcaps_input *req;
 	int rc;
 
 	coal_cap->cmpl_params = BNXT_LEGACY_COAL_CMPL_PARAMS;
@@ -6355,9 +6434,11 @@ static void bnxt_hwrm_coal_params_qcaps(struct bnxt *bp)
 	if (bp->hwrm_spec_code < 0x10902)
 		return;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_RING_AGGINT_QCAPS, -1, -1);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message_silent(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	if (hwrm_req_init(bp, req, HWRM_RING_AGGINT_QCAPS))
+		return;
+
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send_silent(bp, req);
 	if (!rc) {
 		coal_cap->cmpl_params = le32_to_cpu(resp->cmpl_params);
 		coal_cap->nq_params = le32_to_cpu(resp->nq_params);
@@ -6377,7 +6458,7 @@ static void bnxt_hwrm_coal_params_qcaps(struct bnxt *bp)
 			le16_to_cpu(resp->num_cmpl_aggr_int_max);
 		coal_cap->timer_units = le16_to_cpu(resp->timer_units);
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 }
 
 static u16 bnxt_usec_to_coal_tmr(struct bnxt *bp, u16 usec)
@@ -6445,37 +6526,40 @@ static void bnxt_hwrm_set_coal_params(struct bnxt *bp,
 	req->enables |= cpu_to_le16(BNXT_COAL_CMPL_ENABLES);
 }
 
-/* Caller holds bp->hwrm_cmd_lock */
 static int __bnxt_hwrm_set_coal_nq(struct bnxt *bp, struct bnxt_napi *bnapi,
 				   struct bnxt_coal *hw_coal)
 {
-	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input req = {0};
+	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input *req;
 	struct bnxt_cp_ring_info *cpr = &bnapi->cp_ring;
 	struct bnxt_coal_cap *coal_cap = &bp->coal_cap;
 	u32 nq_params = coal_cap->nq_params;
 	u16 tmr;
+	int rc;
 
 	if (!(nq_params & RING_AGGINT_QCAPS_RESP_NQ_PARAMS_INT_LAT_TMR_MIN))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS,
-			       -1, -1);
-	req.ring_id = cpu_to_le16(cpr->cp_ring_struct.fw_ring_id);
-	req.flags =
+	rc = hwrm_req_init(bp, req, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS);
+	if (rc)
+		return rc;
+
+	req->ring_id = cpu_to_le16(cpr->cp_ring_struct.fw_ring_id);
+	req->flags =
 		cpu_to_le16(RING_CMPL_RING_CFG_AGGINT_PARAMS_REQ_FLAGS_IS_NQ);
 
 	tmr = bnxt_usec_to_coal_tmr(bp, hw_coal->coal_ticks) / 2;
 	tmr = clamp_t(u16, tmr, 1, coal_cap->int_lat_tmr_min_max);
-	req.int_lat_tmr_min = cpu_to_le16(tmr);
-	req.enables |= cpu_to_le16(BNXT_COAL_CMPL_MIN_TMR_ENABLE);
-	return _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req->int_lat_tmr_min = cpu_to_le16(tmr);
+	req->enables |= cpu_to_le16(BNXT_COAL_CMPL_MIN_TMR_ENABLE);
+	return hwrm_req_send(bp, req);
 }
 
 int bnxt_hwrm_set_ring_coal(struct bnxt *bp, struct bnxt_napi *bnapi)
 {
-	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input req_rx = {0};
+	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input *req_rx;
 	struct bnxt_cp_ring_info *cpr = &bnapi->cp_ring;
 	struct bnxt_coal coal;
+	int rc;
 
 	/* Tick values in micro seconds.
 	 * 1 coal_buf x bufs_per_record = 1 completion record.
@@ -6488,48 +6572,53 @@ int bnxt_hwrm_set_ring_coal(struct bnxt *bp, struct bnxt_napi *bnapi)
 	if (!bnapi->rx_ring)
 		return -ENODEV;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req_rx,
-			       HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS, -1, -1);
+	rc = hwrm_req_init(bp, req_rx, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS);
+	if (rc)
+		return rc;
 
-	bnxt_hwrm_set_coal_params(bp, &coal, &req_rx);
+	bnxt_hwrm_set_coal_params(bp, &coal, req_rx);
 
-	req_rx.ring_id = cpu_to_le16(bnxt_cp_ring_for_rx(bp, bnapi->rx_ring));
+	req_rx->ring_id = cpu_to_le16(bnxt_cp_ring_for_rx(bp, bnapi->rx_ring));
 
-	return hwrm_send_message(bp, &req_rx, sizeof(req_rx),
-				 HWRM_CMD_TIMEOUT);
+	return hwrm_req_send(bp, req_rx);
 }
 
 int bnxt_hwrm_set_coal(struct bnxt *bp)
 {
-	int i, rc = 0;
-	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input req_rx = {0},
-							   req_tx = {0}, *req;
+	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input *req_rx, *req_tx,
+							   *req;
+	int i, rc;
+
+	rc = hwrm_req_init(bp, req_rx, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS);
+	if (rc)
+		return rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req_rx,
-			       HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS, -1, -1);
-	bnxt_hwrm_cmd_hdr_init(bp, &req_tx,
-			       HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS, -1, -1);
+	rc = hwrm_req_init(bp, req_tx, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS);
+	if (rc) {
+		hwrm_req_drop(bp, req_rx);
+		return rc;
+	}
 
-	bnxt_hwrm_set_coal_params(bp, &bp->rx_coal, &req_rx);
-	bnxt_hwrm_set_coal_params(bp, &bp->tx_coal, &req_tx);
+	bnxt_hwrm_set_coal_params(bp, &bp->rx_coal, req_rx);
+	bnxt_hwrm_set_coal_params(bp, &bp->tx_coal, req_tx);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
+	hwrm_req_hold(bp, req_rx);
+	hwrm_req_hold(bp, req_tx);
 	for (i = 0; i < bp->cp_nr_rings; i++) {
 		struct bnxt_napi *bnapi = bp->bnapi[i];
 		struct bnxt_coal *hw_coal;
 		u16 ring_id;
 
-		req = &req_rx;
+		req = req_rx;
 		if (!bnapi->rx_ring) {
 			ring_id = bnxt_cp_ring_for_tx(bp, bnapi->tx_ring);
-			req = &req_tx;
+			req = req_tx;
 		} else {
 			ring_id = bnxt_cp_ring_for_rx(bp, bnapi->rx_ring);
 		}
 		req->ring_id = cpu_to_le16(ring_id);
 
-		rc = _hwrm_send_message(bp, req, sizeof(*req),
-					HWRM_CMD_TIMEOUT);
+		rc = hwrm_req_send(bp, req);
 		if (rc)
 			break;
 
@@ -6537,11 +6626,10 @@ int bnxt_hwrm_set_coal(struct bnxt *bp)
 			continue;
 
 		if (bnapi->rx_ring && bnapi->tx_ring) {
-			req = &req_tx;
+			req = req_tx;
 			ring_id = bnxt_cp_ring_for_tx(bp, bnapi->tx_ring);
 			req->ring_id = cpu_to_le16(ring_id);
-			rc = _hwrm_send_message(bp, req, sizeof(*req),
-						HWRM_CMD_TIMEOUT);
+			rc = hwrm_req_send(bp, req);
 			if (rc)
 				break;
 		}
@@ -6551,14 +6639,15 @@ int bnxt_hwrm_set_coal(struct bnxt *bp)
 			hw_coal = &bp->tx_coal;
 		__bnxt_hwrm_set_coal_nq(bp, bnapi, hw_coal);
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req_rx);
+	hwrm_req_drop(bp, req_tx);
 	return rc;
 }
 
 static void bnxt_hwrm_stat_ctx_free(struct bnxt *bp)
 {
-	struct hwrm_stat_ctx_clr_stats_input req0 = {0};
-	struct hwrm_stat_ctx_free_input req = {0};
+	struct hwrm_stat_ctx_clr_stats_input *req0 = NULL;
+	struct hwrm_stat_ctx_free_input *req;
 	int i;
 
 	if (!bp->bnapi)
@@ -6567,53 +6656,60 @@ static void bnxt_hwrm_stat_ctx_free(struct bnxt *bp)
 	if (BNXT_CHIP_TYPE_NITRO_A0(bp))
 		return;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req0, HWRM_STAT_CTX_CLR_STATS, -1, -1);
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_STAT_CTX_FREE, -1, -1);
-
-	mutex_lock(&bp->hwrm_cmd_lock);
+	if (hwrm_req_init(bp, req, HWRM_STAT_CTX_FREE))
+		return;
+	if (BNXT_FW_MAJ(bp) <= 20) {
+		if (hwrm_req_init(bp, req0, HWRM_STAT_CTX_CLR_STATS)) {
+			hwrm_req_drop(bp, req);
+			return;
+		}
+		hwrm_req_hold(bp, req0);
+	}
+	hwrm_req_hold(bp, req);
 	for (i = 0; i < bp->cp_nr_rings; i++) {
 		struct bnxt_napi *bnapi = bp->bnapi[i];
 		struct bnxt_cp_ring_info *cpr = &bnapi->cp_ring;
 
 		if (cpr->hw_stats_ctx_id != INVALID_STATS_CTX_ID) {
-			req.stat_ctx_id = cpu_to_le32(cpr->hw_stats_ctx_id);
-			if (BNXT_FW_MAJ(bp) <= 20) {
-				req0.stat_ctx_id = req.stat_ctx_id;
-				_hwrm_send_message(bp, &req0, sizeof(req0),
-						   HWRM_CMD_TIMEOUT);
+			req->stat_ctx_id = cpu_to_le32(cpr->hw_stats_ctx_id);
+			if (req0) {
+				req0->stat_ctx_id = req->stat_ctx_id;
+				hwrm_req_send(bp, req0);
 			}
-			_hwrm_send_message(bp, &req, sizeof(req),
-					   HWRM_CMD_TIMEOUT);
+			hwrm_req_send(bp, req);
 
 			cpr->hw_stats_ctx_id = INVALID_STATS_CTX_ID;
 		}
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
+	if (req0)
+		hwrm_req_drop(bp, req0);
 }
 
 static int bnxt_hwrm_stat_ctx_alloc(struct bnxt *bp)
 {
-	int rc = 0, i;
-	struct hwrm_stat_ctx_alloc_input req = {0};
-	struct hwrm_stat_ctx_alloc_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_stat_ctx_alloc_output *resp;
+	struct hwrm_stat_ctx_alloc_input *req;
+	int rc, i;
 
 	if (BNXT_CHIP_TYPE_NITRO_A0(bp))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_STAT_CTX_ALLOC, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_STAT_CTX_ALLOC);
+	if (rc)
+		return rc;
 
-	req.stats_dma_length = cpu_to_le16(bp->hw_ring_stats_size);
-	req.update_period_ms = cpu_to_le32(bp->stats_coal_ticks / 1000);
+	req->stats_dma_length = cpu_to_le16(bp->hw_ring_stats_size);
+	req->update_period_ms = cpu_to_le32(bp->stats_coal_ticks / 1000);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
+	resp = hwrm_req_hold(bp, req);
 	for (i = 0; i < bp->cp_nr_rings; i++) {
 		struct bnxt_napi *bnapi = bp->bnapi[i];
 		struct bnxt_cp_ring_info *cpr = &bnapi->cp_ring;
 
-		req.stats_dma_addr = cpu_to_le64(cpr->stats.hw_stats_map);
+		req->stats_dma_addr = cpu_to_le64(cpr->stats.hw_stats_map);
 
-		rc = _hwrm_send_message(bp, &req, sizeof(req),
-					HWRM_CMD_TIMEOUT);
+		rc = hwrm_req_send(bp, req);
 		if (rc)
 			break;
 
@@ -6621,22 +6717,25 @@ static int bnxt_hwrm_stat_ctx_alloc(struct bnxt *bp)
 
 		bp->grp_info[i].fw_stats_ctx = cpr->hw_stats_ctx_id;
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_func_qcfg(struct bnxt *bp)
 {
-	struct hwrm_func_qcfg_input req = {0};
-	struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_func_qcfg_output *resp;
+	struct hwrm_func_qcfg_input *req;
 	u32 min_db_offset = 0;
 	u16 flags;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCFG, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_QCFG);
+	if (rc)
+		return rc;
+
+	req->fid = cpu_to_le16(0xffff);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto func_qcfg_exit;
 
@@ -6696,7 +6795,7 @@ static int bnxt_hwrm_func_qcfg(struct bnxt *bp)
 		bp->db_size = pci_resource_len(bp->pdev, 2);
 
 func_qcfg_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -6735,17 +6834,19 @@ static void bnxt_init_ctx_initializer(struct bnxt_ctx_mem_info *ctx,
 
 static int bnxt_hwrm_func_backing_store_qcaps(struct bnxt *bp)
 {
-	struct hwrm_func_backing_store_qcaps_input req = {0};
-	struct hwrm_func_backing_store_qcaps_output *resp =
-		bp->hwrm_cmd_resp_addr;
+	struct hwrm_func_backing_store_qcaps_output *resp;
+	struct hwrm_func_backing_store_qcaps_input *req;
 	int rc;
 
 	if (bp->hwrm_spec_code < 0x10902 || BNXT_VF(bp) || bp->ctx)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_BACKING_STORE_QCAPS, -1, -1);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message_silent(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_BACKING_STORE_QCAPS);
+	if (rc)
+		return rc;
+
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send_silent(bp, req);
 	if (!rc) {
 		struct bnxt_ctx_pg_info *ctx_pg;
 		struct bnxt_ctx_mem_info *ctx;
@@ -6810,7 +6911,7 @@ static int bnxt_hwrm_func_backing_store_qcaps(struct bnxt *bp)
 		rc = 0;
 	}
 ctx_err:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -6841,15 +6942,17 @@ static void bnxt_hwrm_set_pg_attr(struct bnxt_ring_mem_info *rmem, u8 *pg_attr,
 
 static int bnxt_hwrm_func_backing_store_cfg(struct bnxt *bp, u32 enables)
 {
-	struct hwrm_func_backing_store_cfg_input req = {0};
+	struct hwrm_func_backing_store_cfg_input *req;
 	struct bnxt_ctx_mem_info *ctx = bp->ctx;
 	struct bnxt_ctx_pg_info *ctx_pg;
-	u32 req_len = sizeof(req);
+	void **__req = (void **)&req;
+	u32 req_len = sizeof(*req);
 	__le32 *num_entries;
 	__le64 *pg_dir;
 	u32 flags = 0;
 	u8 *pg_attr;
 	u32 ena;
+	int rc;
 	int i;
 
 	if (!ctx)
@@ -6857,90 +6960,93 @@ static int bnxt_hwrm_func_backing_store_cfg(struct bnxt *bp, u32 enables)
 
 	if (req_len > bp->hwrm_max_ext_req_len)
 		req_len = BNXT_BACKING_STORE_CFG_LEGACY_LEN;
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_BACKING_STORE_CFG, -1, -1);
-	req.enables = cpu_to_le32(enables);
+	rc = __hwrm_req_init(bp, __req, HWRM_FUNC_BACKING_STORE_CFG, req_len);
+	if (rc)
+		return rc;
 
+	req->enables = cpu_to_le32(enables);
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_QP) {
 		ctx_pg = &ctx->qp_mem;
-		req.qp_num_entries = cpu_to_le32(ctx_pg->entries);
-		req.qp_num_qp1_entries = cpu_to_le16(ctx->qp_min_qp1_entries);
-		req.qp_num_l2_entries = cpu_to_le16(ctx->qp_max_l2_entries);
-		req.qp_entry_size = cpu_to_le16(ctx->qp_entry_size);
+		req->qp_num_entries = cpu_to_le32(ctx_pg->entries);
+		req->qp_num_qp1_entries = cpu_to_le16(ctx->qp_min_qp1_entries);
+		req->qp_num_l2_entries = cpu_to_le16(ctx->qp_max_l2_entries);
+		req->qp_entry_size = cpu_to_le16(ctx->qp_entry_size);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
-				      &req.qpc_pg_size_qpc_lvl,
-				      &req.qpc_page_dir);
+				      &req->qpc_pg_size_qpc_lvl,
+				      &req->qpc_page_dir);
 	}
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_SRQ) {
 		ctx_pg = &ctx->srq_mem;
-		req.srq_num_entries = cpu_to_le32(ctx_pg->entries);
-		req.srq_num_l2_entries = cpu_to_le16(ctx->srq_max_l2_entries);
-		req.srq_entry_size = cpu_to_le16(ctx->srq_entry_size);
+		req->srq_num_entries = cpu_to_le32(ctx_pg->entries);
+		req->srq_num_l2_entries = cpu_to_le16(ctx->srq_max_l2_entries);
+		req->srq_entry_size = cpu_to_le16(ctx->srq_entry_size);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
-				      &req.srq_pg_size_srq_lvl,
-				      &req.srq_page_dir);
+				      &req->srq_pg_size_srq_lvl,
+				      &req->srq_page_dir);
 	}
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_CQ) {
 		ctx_pg = &ctx->cq_mem;
-		req.cq_num_entries = cpu_to_le32(ctx_pg->entries);
-		req.cq_num_l2_entries = cpu_to_le16(ctx->cq_max_l2_entries);
-		req.cq_entry_size = cpu_to_le16(ctx->cq_entry_size);
-		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem, &req.cq_pg_size_cq_lvl,
-				      &req.cq_page_dir);
+		req->cq_num_entries = cpu_to_le32(ctx_pg->entries);
+		req->cq_num_l2_entries = cpu_to_le16(ctx->cq_max_l2_entries);
+		req->cq_entry_size = cpu_to_le16(ctx->cq_entry_size);
+		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
+				      &req->cq_pg_size_cq_lvl,
+				      &req->cq_page_dir);
 	}
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_VNIC) {
 		ctx_pg = &ctx->vnic_mem;
-		req.vnic_num_vnic_entries =
+		req->vnic_num_vnic_entries =
 			cpu_to_le16(ctx->vnic_max_vnic_entries);
-		req.vnic_num_ring_table_entries =
+		req->vnic_num_ring_table_entries =
 			cpu_to_le16(ctx->vnic_max_ring_table_entries);
-		req.vnic_entry_size = cpu_to_le16(ctx->vnic_entry_size);
+		req->vnic_entry_size = cpu_to_le16(ctx->vnic_entry_size);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
-				      &req.vnic_pg_size_vnic_lvl,
-				      &req.vnic_page_dir);
+				      &req->vnic_pg_size_vnic_lvl,
+				      &req->vnic_page_dir);
 	}
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_STAT) {
 		ctx_pg = &ctx->stat_mem;
-		req.stat_num_entries = cpu_to_le32(ctx->stat_max_entries);
-		req.stat_entry_size = cpu_to_le16(ctx->stat_entry_size);
+		req->stat_num_entries = cpu_to_le32(ctx->stat_max_entries);
+		req->stat_entry_size = cpu_to_le16(ctx->stat_entry_size);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
-				      &req.stat_pg_size_stat_lvl,
-				      &req.stat_page_dir);
+				      &req->stat_pg_size_stat_lvl,
+				      &req->stat_page_dir);
 	}
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_MRAV) {
 		ctx_pg = &ctx->mrav_mem;
-		req.mrav_num_entries = cpu_to_le32(ctx_pg->entries);
+		req->mrav_num_entries = cpu_to_le32(ctx_pg->entries);
 		if (ctx->mrav_num_entries_units)
 			flags |=
 			FUNC_BACKING_STORE_CFG_REQ_FLAGS_MRAV_RESERVATION_SPLIT;
-		req.mrav_entry_size = cpu_to_le16(ctx->mrav_entry_size);
+		req->mrav_entry_size = cpu_to_le16(ctx->mrav_entry_size);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
-				      &req.mrav_pg_size_mrav_lvl,
-				      &req.mrav_page_dir);
+				      &req->mrav_pg_size_mrav_lvl,
+				      &req->mrav_page_dir);
 	}
 	if (enables & FUNC_BACKING_STORE_CFG_REQ_ENABLES_TIM) {
 		ctx_pg = &ctx->tim_mem;
-		req.tim_num_entries = cpu_to_le32(ctx_pg->entries);
-		req.tim_entry_size = cpu_to_le16(ctx->tim_entry_size);
+		req->tim_num_entries = cpu_to_le32(ctx_pg->entries);
+		req->tim_entry_size = cpu_to_le16(ctx->tim_entry_size);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem,
-				      &req.tim_pg_size_tim_lvl,
-				      &req.tim_page_dir);
+				      &req->tim_pg_size_tim_lvl,
+				      &req->tim_page_dir);
 	}
-	for (i = 0, num_entries = &req.tqm_sp_num_entries,
-	     pg_attr = &req.tqm_sp_pg_size_tqm_sp_lvl,
-	     pg_dir = &req.tqm_sp_page_dir,
+	for (i = 0, num_entries = &req->tqm_sp_num_entries,
+	     pg_attr = &req->tqm_sp_pg_size_tqm_sp_lvl,
+	     pg_dir = &req->tqm_sp_page_dir,
 	     ena = FUNC_BACKING_STORE_CFG_REQ_ENABLES_TQM_SP;
 	     i < BNXT_MAX_TQM_RINGS;
 	     i++, num_entries++, pg_attr++, pg_dir++, ena <<= 1) {
 		if (!(enables & ena))
 			continue;
 
-		req.tqm_entry_size = cpu_to_le16(ctx->tqm_entry_size);
+		req->tqm_entry_size = cpu_to_le16(ctx->tqm_entry_size);
 		ctx_pg = ctx->tqm_mem[i];
 		*num_entries = cpu_to_le32(ctx_pg->entries);
 		bnxt_hwrm_set_pg_attr(&ctx_pg->ring_mem, pg_attr, pg_dir);
 	}
-	req.flags = cpu_to_le32(flags);
-	return hwrm_send_message(bp, &req, req_len, HWRM_CMD_TIMEOUT);
+	req->flags = cpu_to_le32(flags);
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
@@ -7220,17 +7326,18 @@ static int bnxt_alloc_ctx_mem(struct bnxt *bp)
 
 int bnxt_hwrm_func_resc_qcaps(struct bnxt *bp, bool all)
 {
-	struct hwrm_func_resource_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_func_resource_qcaps_input req = {0};
+	struct hwrm_func_resource_qcaps_output *resp;
+	struct hwrm_func_resource_qcaps_input *req;
 	struct bnxt_hw_resc *hw_resc = &bp->hw_resc;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_RESOURCE_QCAPS, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_RESOURCE_QCAPS);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message_silent(bp, &req, sizeof(req),
-				       HWRM_CMD_TIMEOUT);
+	req->fid = cpu_to_le16(0xffff);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send_silent(bp, req);
 	if (rc)
 		goto hwrm_func_resc_qcaps_exit;
 
@@ -7271,15 +7378,14 @@ int bnxt_hwrm_func_resc_qcaps(struct bnxt *bp, bool all)
 			pf->vf_resv_strategy = BNXT_VF_RESV_STRATEGY_MAXIMAL;
 	}
 hwrm_func_resc_qcaps_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
-/* bp->hwrm_cmd_lock already held. */
 static int __bnxt_hwrm_ptp_qcfg(struct bnxt *bp)
 {
-	struct hwrm_port_mac_ptp_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_port_mac_ptp_qcfg_input req = {0};
+	struct hwrm_port_mac_ptp_qcfg_output *resp;
+	struct hwrm_port_mac_ptp_qcfg_input *req;
 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
 	u8 flags;
 	int rc;
@@ -7289,21 +7395,27 @@ static int __bnxt_hwrm_ptp_qcfg(struct bnxt *bp)
 		goto no_ptp;
 	}
 
-	req.port_id = cpu_to_le16(bp->pf.port_id);
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_MAC_PTP_QCFG, -1, -1);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_MAC_PTP_QCFG);
 	if (rc)
 		goto no_ptp;
 
+	req->port_id = cpu_to_le16(bp->pf.port_id);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
+	if (rc)
+		goto exit;
+
 	flags = resp->flags;
 	if (!(flags & PORT_MAC_PTP_QCFG_RESP_FLAGS_HWRM_ACCESS)) {
 		rc = -ENODEV;
-		goto no_ptp;
+		goto exit;
 	}
 	if (!ptp) {
 		ptp = kzalloc(sizeof(*ptp), GFP_KERNEL);
-		if (!ptp)
-			return -ENOMEM;
+		if (!ptp) {
+			rc = -ENOMEM;
+			goto exit;
+		}
 		ptp->bp = bp;
 		bp->ptp_cfg = ptp;
 	}
@@ -7315,14 +7427,16 @@ static int __bnxt_hwrm_ptp_qcfg(struct bnxt *bp)
 		ptp->refclk_regs[1] = BNXT_TS_REG_TIMESYNC_TS0_UPPER;
 	} else {
 		rc = -ENODEV;
-		goto no_ptp;
+		goto exit;
 	}
 	rc = bnxt_ptp_init(bp);
+	if (rc)
+		netdev_warn(bp->dev, "PTP initialization failed.\n");
+exit:
+	hwrm_req_drop(bp, req);
 	if (!rc)
 		return 0;
 
-	netdev_warn(bp->dev, "PTP initialization failed.\n");
-
 no_ptp:
 	bnxt_ptp_clear(bp);
 	kfree(ptp);
@@ -7332,17 +7446,19 @@ static int __bnxt_hwrm_ptp_qcfg(struct bnxt *bp)
 
 static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
 {
-	int rc = 0;
-	struct hwrm_func_qcaps_input req = {0};
-	struct hwrm_func_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_func_qcaps_output *resp;
+	struct hwrm_func_qcaps_input *req;
 	struct bnxt_hw_resc *hw_resc = &bp->hw_resc;
 	u32 flags, flags_ext;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCAPS, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_QCAPS);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req->fid = cpu_to_le16(0xffff);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto hwrm_func_qcaps_exit;
 
@@ -7420,7 +7536,7 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
 	}
 
 hwrm_func_qcaps_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -7451,19 +7567,20 @@ static int bnxt_hwrm_func_qcaps(struct bnxt *bp)
 
 static int bnxt_hwrm_cfa_adv_flow_mgnt_qcaps(struct bnxt *bp)
 {
-	struct hwrm_cfa_adv_flow_mgnt_qcaps_input req = {0};
 	struct hwrm_cfa_adv_flow_mgnt_qcaps_output *resp;
-	int rc = 0;
+	struct hwrm_cfa_adv_flow_mgnt_qcaps_input *req;
 	u32 flags;
+	int rc;
 
 	if (!(bp->fw_cap & BNXT_FW_CAP_CFA_ADV_FLOW))
 		return 0;
 
-	resp = bp->hwrm_cmd_resp_addr;
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_ADV_FLOW_MGNT_QCAPS, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_CFA_ADV_FLOW_MGNT_QCAPS);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto hwrm_cfa_adv_qcaps_exit;
 
@@ -7473,7 +7590,7 @@ static int bnxt_hwrm_cfa_adv_flow_mgnt_qcaps(struct bnxt *bp)
 		bp->fw_cap |= BNXT_FW_CAP_CFA_RFS_RING_TBL_IDX_V2;
 
 hwrm_cfa_adv_qcaps_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -7616,17 +7733,20 @@ static int bnxt_map_fw_health_regs(struct bnxt *bp)
 
 static int bnxt_hwrm_error_recovery_qcfg(struct bnxt *bp)
 {
-	struct hwrm_error_recovery_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
 	struct bnxt_fw_health *fw_health = bp->fw_health;
-	struct hwrm_error_recovery_qcfg_input req = {0};
+	struct hwrm_error_recovery_qcfg_output *resp;
+	struct hwrm_error_recovery_qcfg_input *req;
 	int rc, i;
 
 	if (!(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_ERROR_RECOVERY_QCFG, -1, -1);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_ERROR_RECOVERY_QCFG);
+	if (rc)
+		return rc;
+
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto err_recovery_out;
 	fw_health->flags = le32_to_cpu(resp->flags);
@@ -7668,7 +7788,7 @@ static int bnxt_hwrm_error_recovery_qcfg(struct bnxt *bp)
 			resp->delay_after_reset[i];
 	}
 err_recovery_out:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	if (!rc)
 		rc = bnxt_map_fw_health_regs(bp);
 	if (rc)
@@ -7678,12 +7798,16 @@ static int bnxt_hwrm_error_recovery_qcfg(struct bnxt *bp)
 
 static int bnxt_hwrm_func_reset(struct bnxt *bp)
 {
-	struct hwrm_func_reset_input req = {0};
+	struct hwrm_func_reset_input *req;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_RESET, -1, -1);
-	req.enables = 0;
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_RESET);
+	if (rc)
+		return rc;
 
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_RESET_TIMEOUT);
+	req->enables = 0;
+	hwrm_req_timeout(bp, req, HWRM_RESET_TIMEOUT);
+	return hwrm_req_send(bp, req);
 }
 
 static void bnxt_nvm_cfg_ver_get(struct bnxt *bp)
@@ -7698,16 +7822,18 @@ static void bnxt_nvm_cfg_ver_get(struct bnxt *bp)
 
 static int bnxt_hwrm_queue_qportcfg(struct bnxt *bp)
 {
-	int rc = 0;
-	struct hwrm_queue_qportcfg_input req = {0};
-	struct hwrm_queue_qportcfg_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_queue_qportcfg_output *resp;
+	struct hwrm_queue_qportcfg_input *req;
 	u8 i, j, *qptr;
 	bool no_rdma;
+	int rc = 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_QUEUE_QPORTCFG, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_QUEUE_QPORTCFG);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto qportcfg_exit;
 
@@ -7741,40 +7867,48 @@ static int bnxt_hwrm_queue_qportcfg(struct bnxt *bp)
 		bp->max_lltc = bp->max_tc;
 
 qportcfg_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_poll(struct bnxt *bp)
 {
-	struct hwrm_ver_get_input req = {0};
+	struct hwrm_ver_get_input *req;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VER_GET, -1, -1);
-	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
-	req.hwrm_intf_min = HWRM_VERSION_MINOR;
-	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
+	rc = hwrm_req_init(bp, req, HWRM_VER_GET);
+	if (rc)
+		return rc;
+
+	req->hwrm_intf_maj = HWRM_VERSION_MAJOR;
+	req->hwrm_intf_min = HWRM_VERSION_MINOR;
+	req->hwrm_intf_upd = HWRM_VERSION_UPDATE;
 
-	rc = _hwrm_send_message_silent(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	hwrm_req_flags(bp, req, BNXT_HWRM_CTX_SILENT | BNXT_HWRM_FULL_WAIT);
+	rc = hwrm_req_send(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_ver_get(struct bnxt *bp)
 {
-	struct hwrm_ver_get_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_ver_get_input req = {0};
+	struct hwrm_ver_get_output *resp;
+	struct hwrm_ver_get_input *req;
 	u16 fw_maj, fw_min, fw_bld, fw_rsv;
 	u32 dev_caps_cfg, hwrm_ver;
 	int rc, len;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_VER_GET, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_VER_GET);
+	if (rc)
+		return rc;
+
+	hwrm_req_flags(bp, req, BNXT_HWRM_FULL_WAIT);
 	bp->hwrm_max_req_len = HWRM_MAX_REQ_LEN;
-	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
-	req.hwrm_intf_min = HWRM_VERSION_MINOR;
-	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
+	req->hwrm_intf_maj = HWRM_VERSION_MAJOR;
+	req->hwrm_intf_min = HWRM_VERSION_MINOR;
+	req->hwrm_intf_upd = HWRM_VERSION_UPDATE;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto hwrm_ver_get_exit;
 
@@ -7866,29 +8000,33 @@ static int bnxt_hwrm_ver_get(struct bnxt *bp)
 		bp->fw_cap |= BNXT_FW_CAP_CFA_ADV_FLOW;
 
 hwrm_ver_get_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 int bnxt_hwrm_fw_set_time(struct bnxt *bp)
 {
-	struct hwrm_fw_set_time_input req = {0};
+	struct hwrm_fw_set_time_input *req;
 	struct tm tm;
 	time64_t now = ktime_get_real_seconds();
+	int rc;
 
 	if ((BNXT_VF(bp) && bp->hwrm_spec_code < 0x10901) ||
 	    bp->hwrm_spec_code < 0x10400)
 		return -EOPNOTSUPP;
 
 	time64_to_tm(now, 0, &tm);
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FW_SET_TIME, -1, -1);
-	req.year = cpu_to_le16(1900 + tm.tm_year);
-	req.month = 1 + tm.tm_mon;
-	req.day = tm.tm_mday;
-	req.hour = tm.tm_hour;
-	req.minute = tm.tm_min;
-	req.second = tm.tm_sec;
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_FW_SET_TIME);
+	if (rc)
+		return rc;
+
+	req->year = cpu_to_le16(1900 + tm.tm_year);
+	req->month = 1 + tm.tm_mon;
+	req->day = tm.tm_mday;
+	req->hour = tm.tm_hour;
+	req->minute = tm.tm_min;
+	req->second = tm.tm_sec;
+	return hwrm_req_send(bp, req);
 }
 
 static void bnxt_add_one_ctr(u64 hw, u64 *sw, u64 mask)
@@ -7976,8 +8114,9 @@ static void bnxt_accumulate_all_stats(struct bnxt *bp)
 
 static int bnxt_hwrm_port_qstats(struct bnxt *bp, u8 flags)
 {
+	struct hwrm_port_qstats_input *req;
 	struct bnxt_pf_info *pf = &bp->pf;
-	struct hwrm_port_qstats_input req = {0};
+	int rc;
 
 	if (!(bp->flags & BNXT_FLAG_PORT_STATS))
 		return 0;
@@ -7985,20 +8124,24 @@ static int bnxt_hwrm_port_qstats(struct bnxt *bp, u8 flags)
 	if (flags && !(bp->fw_cap & BNXT_FW_CAP_EXT_HW_STATS_SUPPORTED))
 		return -EOPNOTSUPP;
 
-	req.flags = flags;
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_QSTATS, -1, -1);
-	req.port_id = cpu_to_le16(pf->port_id);
-	req.tx_stat_host_addr = cpu_to_le64(bp->port_stats.hw_stats_map +
+	rc = hwrm_req_init(bp, req, HWRM_PORT_QSTATS);
+	if (rc)
+		return rc;
+
+	req->flags = flags;
+	req->port_id = cpu_to_le16(pf->port_id);
+	req->tx_stat_host_addr = cpu_to_le64(bp->port_stats.hw_stats_map +
 					    BNXT_TX_PORT_STATS_BYTE_OFFSET);
-	req.rx_stat_host_addr = cpu_to_le64(bp->port_stats.hw_stats_map);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req->rx_stat_host_addr = cpu_to_le64(bp->port_stats.hw_stats_map);
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_hwrm_port_qstats_ext(struct bnxt *bp, u8 flags)
 {
-	struct hwrm_port_qstats_ext_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_queue_pri2cos_qcfg_input req2 = {0};
-	struct hwrm_port_qstats_ext_input req = {0};
+	struct hwrm_queue_pri2cos_qcfg_output *resp_qc;
+	struct hwrm_queue_pri2cos_qcfg_input *req_qc;
+	struct hwrm_port_qstats_ext_output *resp_qs;
+	struct hwrm_port_qstats_ext_input *req_qs;
 	struct bnxt_pf_info *pf = &bp->pf;
 	u32 tx_stat_size;
 	int rc;
@@ -8009,46 +8152,53 @@ static int bnxt_hwrm_port_qstats_ext(struct bnxt *bp, u8 flags)
 	if (flags && !(bp->fw_cap & BNXT_FW_CAP_EXT_HW_STATS_SUPPORTED))
 		return -EOPNOTSUPP;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_QSTATS_EXT, -1, -1);
-	req.flags = flags;
-	req.port_id = cpu_to_le16(pf->port_id);
-	req.rx_stat_size = cpu_to_le16(sizeof(struct rx_port_stats_ext));
-	req.rx_stat_host_addr = cpu_to_le64(bp->rx_port_stats_ext.hw_stats_map);
+	rc = hwrm_req_init(bp, req_qs, HWRM_PORT_QSTATS_EXT);
+	if (rc)
+		return rc;
+
+	req_qs->flags = flags;
+	req_qs->port_id = cpu_to_le16(pf->port_id);
+	req_qs->rx_stat_size = cpu_to_le16(sizeof(struct rx_port_stats_ext));
+	req_qs->rx_stat_host_addr = cpu_to_le64(bp->rx_port_stats_ext.hw_stats_map);
 	tx_stat_size = bp->tx_port_stats_ext.hw_stats ?
 		       sizeof(struct tx_port_stats_ext) : 0;
-	req.tx_stat_size = cpu_to_le16(tx_stat_size);
-	req.tx_stat_host_addr = cpu_to_le64(bp->tx_port_stats_ext.hw_stats_map);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	req_qs->tx_stat_size = cpu_to_le16(tx_stat_size);
+	req_qs->tx_stat_host_addr = cpu_to_le64(bp->tx_port_stats_ext.hw_stats_map);
+	resp_qs = hwrm_req_hold(bp, req_qs);
+	rc = hwrm_req_send(bp, req_qs);
 	if (!rc) {
-		bp->fw_rx_stats_ext_size = le16_to_cpu(resp->rx_stat_size) / 8;
+		bp->fw_rx_stats_ext_size =
+			le16_to_cpu(resp_qs->rx_stat_size) / 8;
 		bp->fw_tx_stats_ext_size = tx_stat_size ?
-			le16_to_cpu(resp->tx_stat_size) / 8 : 0;
+			le16_to_cpu(resp_qs->tx_stat_size) / 8 : 0;
 	} else {
 		bp->fw_rx_stats_ext_size = 0;
 		bp->fw_tx_stats_ext_size = 0;
 	}
+	hwrm_req_drop(bp, req_qs);
+
 	if (flags)
-		goto qstats_done;
+		return rc;
 
 	if (bp->fw_tx_stats_ext_size <=
 	    offsetof(struct tx_port_stats_ext, pfc_pri0_tx_duration_us) / 8) {
-		mutex_unlock(&bp->hwrm_cmd_lock);
 		bp->pri2cos_valid = 0;
 		return rc;
 	}
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req2, HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
-	req2.flags = cpu_to_le32(QUEUE_PRI2COS_QCFG_REQ_FLAGS_IVLAN);
+	rc = hwrm_req_init(bp, req_qc, HWRM_QUEUE_PRI2COS_QCFG);
+	if (rc)
+		return rc;
+
+	req_qc->flags = cpu_to_le32(QUEUE_PRI2COS_QCFG_REQ_FLAGS_IVLAN);
 
-	rc = _hwrm_send_message(bp, &req2, sizeof(req2), HWRM_CMD_TIMEOUT);
+	resp_qc = hwrm_req_hold(bp, req_qc);
+	rc = hwrm_req_send(bp, req_qc);
 	if (!rc) {
-		struct hwrm_queue_pri2cos_qcfg_output *resp2;
 		u8 *pri2cos;
 		int i, j;
 
-		resp2 = bp->hwrm_cmd_resp_addr;
-		pri2cos = &resp2->pri0_cos_queue_id;
+		pri2cos = &resp_qc->pri0_cos_queue_id;
 		for (i = 0; i < 8; i++) {
 			u8 queue_id = pri2cos[i];
 			u8 queue_idx;
@@ -8057,17 +8207,18 @@ static int bnxt_hwrm_port_qstats_ext(struct bnxt *bp, u8 flags)
 			queue_idx = queue_id % 10;
 			if (queue_idx > BNXT_MAX_QUEUE) {
 				bp->pri2cos_valid = false;
-				goto qstats_done;
+				hwrm_req_drop(bp, req_qc);
+				return rc;
 			}
 			for (j = 0; j < bp->max_q; j++) {
 				if (bp->q_ids[j] == queue_id)
 					bp->pri2cos_idx[i] = queue_idx;
 			}
 		}
-		bp->pri2cos_valid = 1;
+		bp->pri2cos_valid = true;
 	}
-qstats_done:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req_qc);
+
 	return rc;
 }
 
@@ -8142,35 +8293,46 @@ static void bnxt_hwrm_resource_free(struct bnxt *bp, bool close_path,
 
 static int bnxt_hwrm_set_br_mode(struct bnxt *bp, u16 br_mode)
 {
-	struct hwrm_func_cfg_input req = {0};
+	struct hwrm_func_cfg_input *req;
+	u8 evb_mode;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
-	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_EVB_MODE);
 	if (br_mode == BRIDGE_MODE_VEB)
-		req.evb_mode = FUNC_CFG_REQ_EVB_MODE_VEB;
+		evb_mode = FUNC_CFG_REQ_EVB_MODE_VEB;
 	else if (br_mode == BRIDGE_MODE_VEPA)
-		req.evb_mode = FUNC_CFG_REQ_EVB_MODE_VEPA;
+		evb_mode = FUNC_CFG_REQ_EVB_MODE_VEPA;
 	else
 		return -EINVAL;
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_CFG);
+	if (rc)
+		return rc;
+
+	req->fid = cpu_to_le16(0xffff);
+	req->enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_EVB_MODE);
+	req->evb_mode = evb_mode;
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_hwrm_set_cache_line_size(struct bnxt *bp, int size)
 {
-	struct hwrm_func_cfg_input req = {0};
+	struct hwrm_func_cfg_input *req;
+	int rc;
 
 	if (BNXT_VF(bp) || bp->hwrm_spec_code < 0x10803)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
-	req.fid = cpu_to_le16(0xffff);
-	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_CACHE_LINESIZE);
-	req.options = FUNC_CFG_REQ_OPTIONS_CACHE_LINESIZE_SIZE_64;
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_CFG);
+	if (rc)
+		return rc;
+
+	req->fid = cpu_to_le16(0xffff);
+	req->enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_CACHE_LINESIZE);
+	req->options = FUNC_CFG_REQ_OPTIONS_CACHE_LINESIZE_SIZE_64;
 	if (size == 128)
-		req.options = FUNC_CFG_REQ_OPTIONS_CACHE_LINESIZE_SIZE_128;
+		req->options = FUNC_CFG_REQ_OPTIONS_CACHE_LINESIZE_SIZE_128;
 
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	return hwrm_req_send(bp, req);
 }
 
 static int __bnxt_setup_vnic(struct bnxt *bp, u16 vnic_id)
@@ -9118,18 +9280,20 @@ static bool bnxt_phy_qcaps_no_speed(struct hwrm_port_phy_qcaps_output *resp)
 
 static int bnxt_hwrm_phy_qcaps(struct bnxt *bp)
 {
-	int rc = 0;
-	struct hwrm_port_phy_qcaps_input req = {0};
-	struct hwrm_port_phy_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
 	struct bnxt_link_info *link_info = &bp->link_info;
+	struct hwrm_port_phy_qcaps_output *resp;
+	struct hwrm_port_phy_qcaps_input *req;
+	int rc = 0;
 
 	if (bp->hwrm_spec_code < 0x10201)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_QCAPS, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_QCAPS);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc)
 		goto hwrm_phy_qcaps_exit;
 
@@ -9167,7 +9331,7 @@ static int bnxt_hwrm_phy_qcaps(struct bnxt *bp)
 	bp->port_count = resp->port_cnt;
 
 hwrm_phy_qcaps_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
@@ -9180,19 +9344,21 @@ static bool bnxt_support_dropped(u16 advertising, u16 supported)
 
 int bnxt_update_link(struct bnxt *bp, bool chng_link_state)
 {
-	int rc = 0;
 	struct bnxt_link_info *link_info = &bp->link_info;
-	struct hwrm_port_phy_qcfg_input req = {0};
-	struct hwrm_port_phy_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_port_phy_qcfg_output *resp;
+	struct hwrm_port_phy_qcfg_input *req;
 	u8 link_up = link_info->link_up;
 	bool support_changed = false;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_QCFG, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_QCFG);
+	if (rc)
+		return rc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc) {
-		mutex_unlock(&bp->hwrm_cmd_lock);
+		hwrm_req_drop(bp, req);
 		return rc;
 	}
 
@@ -9287,7 +9453,7 @@ int bnxt_update_link(struct bnxt *bp, bool chng_link_state)
 		/* alwasy link down if not require to update link state */
 		link_info->link_up = 0;
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 
 	if (!BNXT_PHY_CFG_ABLE(bp))
 		return 0;
@@ -9397,18 +9563,20 @@ static void bnxt_hwrm_set_link_common(struct bnxt *bp, struct hwrm_port_phy_cfg_
 
 int bnxt_hwrm_set_pause(struct bnxt *bp)
 {
-	struct hwrm_port_phy_cfg_input req = {0};
+	struct hwrm_port_phy_cfg_input *req;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_CFG, -1, -1);
-	bnxt_hwrm_set_pause_common(bp, &req);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_CFG);
+	if (rc)
+		return rc;
+
+	bnxt_hwrm_set_pause_common(bp, req);
 
 	if ((bp->link_info.autoneg & BNXT_AUTONEG_FLOW_CTRL) ||
 	    bp->link_info.force_link_chng)
-		bnxt_hwrm_set_link_common(bp, &req);
+		bnxt_hwrm_set_link_common(bp, req);
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_send(bp, req);
 	if (!rc && !(bp->link_info.autoneg & BNXT_AUTONEG_FLOW_CTRL)) {
 		/* since changing of pause setting doesn't trigger any link
 		 * change event, the driver needs to update the current pause
@@ -9421,7 +9589,6 @@ int bnxt_hwrm_set_pause(struct bnxt *bp)
 			bnxt_report_link(bp);
 	}
 	bp->link_info.force_link_chng = false;
-	mutex_unlock(&bp->hwrm_cmd_lock);
 	return rc;
 }
 
@@ -9450,22 +9617,27 @@ static void bnxt_hwrm_set_eee(struct bnxt *bp,
 
 int bnxt_hwrm_set_link_setting(struct bnxt *bp, bool set_pause, bool set_eee)
 {
-	struct hwrm_port_phy_cfg_input req = {0};
+	struct hwrm_port_phy_cfg_input *req;
+	int rc;
+
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_CFG);
+	if (rc)
+		return rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_CFG, -1, -1);
 	if (set_pause)
-		bnxt_hwrm_set_pause_common(bp, &req);
+		bnxt_hwrm_set_pause_common(bp, req);
 
-	bnxt_hwrm_set_link_common(bp, &req);
+	bnxt_hwrm_set_link_common(bp, req);
 
 	if (set_eee)
-		bnxt_hwrm_set_eee(bp, &req);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+		bnxt_hwrm_set_eee(bp, req);
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_hwrm_shutdown_link(struct bnxt *bp)
 {
-	struct hwrm_port_phy_cfg_input req = {0};
+	struct hwrm_port_phy_cfg_input *req;
+	int rc;
 
 	if (!BNXT_SINGLE_PF(bp))
 		return 0;
@@ -9474,9 +9646,12 @@ static int bnxt_hwrm_shutdown_link(struct bnxt *bp)
 	    !(bp->phy_flags & BNXT_PHY_FL_FW_MANAGED_LKDN))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_CFG, -1, -1);
-	req.flags = cpu_to_le32(PORT_PHY_CFG_REQ_FLAGS_FORCE_LINK_DWN);
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_CFG);
+	if (rc)
+		return rc;
+
+	req->flags = cpu_to_le32(PORT_PHY_CFG_REQ_FLAGS_FORCE_LINK_DWN);
+	return hwrm_req_send(bp, req);
 }
 
 static int bnxt_fw_init_one(struct bnxt *bp);
@@ -9502,7 +9677,6 @@ static int bnxt_try_recover_fw(struct bnxt *bp)
 		int retry = 0, rc;
 		u32 sts;
 
-		mutex_lock(&bp->hwrm_cmd_lock);
 		do {
 			sts = bnxt_fw_health_readl(bp, BNXT_FW_HEALTH_REG);
 			rc = bnxt_hwrm_poll(bp);
@@ -9511,7 +9685,6 @@ static int bnxt_try_recover_fw(struct bnxt *bp)
 				break;
 			retry++;
 		} while (rc == -EBUSY && retry < BNXT_FW_RETRY);
-		mutex_unlock(&bp->hwrm_cmd_lock);
 
 		if (!BNXT_FW_IS_HEALTHY(sts)) {
 			netdev_err(bp->dev,
@@ -9531,8 +9704,8 @@ static int bnxt_try_recover_fw(struct bnxt *bp)
 
 static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 {
-	struct hwrm_func_drv_if_change_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_func_drv_if_change_input req = {0};
+	struct hwrm_func_drv_if_change_output *resp;
+	struct hwrm_func_drv_if_change_input *req;
 	bool fw_reset = !bp->irq_tbl;
 	bool resc_reinit = false;
 	int rc, retry = 0;
@@ -9541,29 +9714,34 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 	if (!(bp->fw_cap & BNXT_FW_CAP_IF_CHANGE))
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_DRV_IF_CHANGE, -1, -1);
+	rc = hwrm_req_init(bp, req, HWRM_FUNC_DRV_IF_CHANGE);
+	if (rc)
+		return rc;
+
 	if (up)
-		req.flags = cpu_to_le32(FUNC_DRV_IF_CHANGE_REQ_FLAGS_UP);
-	mutex_lock(&bp->hwrm_cmd_lock);
+		req->flags = cpu_to_le32(FUNC_DRV_IF_CHANGE_REQ_FLAGS_UP);
+	resp = hwrm_req_hold(bp, req);
+
+	hwrm_req_flags(bp, req, BNXT_HWRM_FULL_WAIT);
 	while (retry < BNXT_FW_IF_RETRY) {
-		rc = _hwrm_send_message(bp, &req, sizeof(req),
-					HWRM_CMD_TIMEOUT);
+		rc = hwrm_req_send(bp, req);
 		if (rc != -EAGAIN)
 			break;
 
 		msleep(50);
 		retry++;
 	}
-	if (!rc)
-		flags = le32_to_cpu(resp->flags);
-	mutex_unlock(&bp->hwrm_cmd_lock);
 
-	if (rc == -EAGAIN)
+	if (rc == -EAGAIN) {
+		hwrm_req_drop(bp, req);
 		return rc;
-	if (rc && up) {
+	} else if (!rc) {
+		flags = le32_to_cpu(resp->flags);
+	} else if (up) {
 		rc = bnxt_try_recover_fw(bp);
 		fw_reset = true;
 	}
+	hwrm_req_drop(bp, req);
 	if (rc)
 		return rc;
 
@@ -9632,8 +9810,8 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 
 static int bnxt_hwrm_port_led_qcaps(struct bnxt *bp)
 {
-	struct hwrm_port_led_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_port_led_qcaps_input req = {0};
+	struct hwrm_port_led_qcaps_output *resp;
+	struct hwrm_port_led_qcaps_input *req;
 	struct bnxt_pf_info *pf = &bp->pf;
 	int rc;
 
@@ -9641,12 +9819,15 @@ static int bnxt_hwrm_port_led_qcaps(struct bnxt *bp)
 	if (BNXT_VF(bp) || bp->hwrm_spec_code < 0x10601)
 		return 0;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_LED_QCAPS, -1, -1);
-	req.port_id = cpu_to_le16(pf->port_id);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_LED_QCAPS);
+	if (rc)
+		return rc;
+
+	req->port_id = cpu_to_le16(pf->port_id);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (rc) {
-		mutex_unlock(&bp->hwrm_cmd_lock);
+		hwrm_req_drop(bp, req);
 		return rc;
 	}
 	if (resp->num_leds > 0 && resp->num_leds < BNXT_MAX_LED) {
@@ -9666,52 +9847,64 @@ static int bnxt_hwrm_port_led_qcaps(struct bnxt *bp)
 			}
 		}
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return 0;
 }
 
 int bnxt_hwrm_alloc_wol_fltr(struct bnxt *bp)
 {
-	struct hwrm_wol_filter_alloc_input req = {0};
-	struct hwrm_wol_filter_alloc_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_wol_filter_alloc_output *resp;
+	struct hwrm_wol_filter_alloc_input *req;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_WOL_FILTER_ALLOC, -1, -1);
-	req.port_id = cpu_to_le16(bp->pf.port_id);
-	req.wol_type = WOL_FILTER_ALLOC_REQ_WOL_TYPE_MAGICPKT;
-	req.enables = cpu_to_le32(WOL_FILTER_ALLOC_REQ_ENABLES_MAC_ADDRESS);
-	memcpy(req.mac_address, bp->dev->dev_addr, ETH_ALEN);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_WOL_FILTER_ALLOC);
+	if (rc)
+		return rc;
+
+	req->port_id = cpu_to_le16(bp->pf.port_id);
+	req->wol_type = WOL_FILTER_ALLOC_REQ_WOL_TYPE_MAGICPKT;
+	req->enables = cpu_to_le32(WOL_FILTER_ALLOC_REQ_ENABLES_MAC_ADDRESS);
+	memcpy(req->mac_address, bp->dev->dev_addr, ETH_ALEN);
+
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		bp->wol_filter_id = resp->wol_filter_id;
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 int bnxt_hwrm_free_wol_fltr(struct bnxt *bp)
 {
-	struct hwrm_wol_filter_free_input req = {0};
+	struct hwrm_wol_filter_free_input *req;
+	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_WOL_FILTER_FREE, -1, -1);
-	req.port_id = cpu_to_le16(bp->pf.port_id);
-	req.enables = cpu_to_le32(WOL_FILTER_FREE_REQ_ENABLES_WOL_FILTER_ID);
-	req.wol_filter_id = bp->wol_filter_id;
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_WOL_FILTER_FREE);
+	if (rc)
+		return rc;
+
+	req->port_id = cpu_to_le16(bp->pf.port_id);
+	req->enables = cpu_to_le32(WOL_FILTER_FREE_REQ_ENABLES_WOL_FILTER_ID);
+	req->wol_filter_id = bp->wol_filter_id;
+
+	return hwrm_req_send(bp, req);
 }
 
 static u16 bnxt_hwrm_get_wol_fltrs(struct bnxt *bp, u16 handle)
 {
-	struct hwrm_wol_filter_qcfg_input req = {0};
-	struct hwrm_wol_filter_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_wol_filter_qcfg_output *resp;
+	struct hwrm_wol_filter_qcfg_input *req;
 	u16 next_handle = 0;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_WOL_FILTER_QCFG, -1, -1);
-	req.port_id = cpu_to_le16(bp->pf.port_id);
-	req.handle = cpu_to_le16(handle);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_WOL_FILTER_QCFG);
+	if (rc)
+		return rc;
+
+	req->port_id = cpu_to_le16(bp->pf.port_id);
+	req->handle = cpu_to_le16(handle);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc) {
 		next_handle = le16_to_cpu(resp->next_handle);
 		if (next_handle != 0) {
@@ -9722,7 +9915,7 @@ static u16 bnxt_hwrm_get_wol_fltrs(struct bnxt *bp, u16 handle)
 			}
 		}
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return next_handle;
 }
 
@@ -9743,19 +9936,20 @@ static void bnxt_get_wol_settings(struct bnxt *bp)
 static ssize_t bnxt_show_temp(struct device *dev,
 			      struct device_attribute *devattr, char *buf)
 {
-	struct hwrm_temp_monitor_query_input req = {0};
 	struct hwrm_temp_monitor_query_output *resp;
+	struct hwrm_temp_monitor_query_input *req;
 	struct bnxt *bp = dev_get_drvdata(dev);
 	u32 len = 0;
 	int rc;
 
-	resp = bp->hwrm_cmd_resp_addr;
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_TEMP_MONITOR_QUERY, -1, -1);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_TEMP_MONITOR_QUERY);
+	if (rc)
+		return rc;
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		len = sprintf(buf, "%u\n", resp->temp * 1000); /* display millidegree */
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	if (rc)
 		return rc;
 	return len;
@@ -9778,12 +9972,13 @@ static void bnxt_hwmon_close(struct bnxt *bp)
 
 static void bnxt_hwmon_open(struct bnxt *bp)
 {
-	struct hwrm_temp_monitor_query_input req = {0};
+	struct hwrm_temp_monitor_query_input *req;
 	struct pci_dev *pdev = bp->pdev;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_TEMP_MONITOR_QUERY, -1, -1);
-	rc = hwrm_send_message_silent(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_TEMP_MONITOR_QUERY);
+	if (!rc)
+		rc = hwrm_req_send_silent(bp, req);
 	if (rc == -EACCES || rc == -EOPNOTSUPP) {
 		bnxt_hwmon_close(bp);
 		return;
@@ -10220,53 +10415,60 @@ static int bnxt_close(struct net_device *dev)
 static int bnxt_hwrm_port_phy_read(struct bnxt *bp, u16 phy_addr, u16 reg,
 				   u16 *val)
 {
-	struct hwrm_port_phy_mdio_read_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_port_phy_mdio_read_input req = {0};
+	struct hwrm_port_phy_mdio_read_output *resp;
+	struct hwrm_port_phy_mdio_read_input *req;
 	int rc;
 
 	if (bp->hwrm_spec_code < 0x10a00)
 		return -EOPNOTSUPP;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_MDIO_READ, -1, -1);
-	req.port_id = cpu_to_le16(bp->pf.port_id);
-	req.phy_addr = phy_addr;
-	req.reg_addr = cpu_to_le16(reg & 0x1f);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_MDIO_READ);
+	if (rc)
+		return rc;
+
+	req->port_id = cpu_to_le16(bp->pf.port_id);
+	req->phy_addr = phy_addr;
+	req->reg_addr = cpu_to_le16(reg & 0x1f);
 	if (mdio_phy_id_is_c45(phy_addr)) {
-		req.cl45_mdio = 1;
-		req.phy_addr = mdio_phy_id_prtad(phy_addr);
-		req.dev_addr = mdio_phy_id_devad(phy_addr);
-		req.reg_addr = cpu_to_le16(reg);
+		req->cl45_mdio = 1;
+		req->phy_addr = mdio_phy_id_prtad(phy_addr);
+		req->dev_addr = mdio_phy_id_devad(phy_addr);
+		req->reg_addr = cpu_to_le16(reg);
 	}
 
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	resp = hwrm_req_hold(bp, req);
+	rc = hwrm_req_send(bp, req);
 	if (!rc)
 		*val = le16_to_cpu(resp->reg_data);
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_hwrm_port_phy_write(struct bnxt *bp, u16 phy_addr, u16 reg,
 				    u16 val)
 {
-	struct hwrm_port_phy_mdio_write_input req = {0};
+	struct hwrm_port_phy_mdio_write_input *req;
+	int rc;
 
 	if (bp->hwrm_spec_code < 0x10a00)
 		return -EOPNOTSUPP;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_PHY_MDIO_WRITE, -1, -1);
-	req.port_id = cpu_to_le16(bp->pf.port_id);
-	req.phy_addr = phy_addr;
-	req.reg_addr = cpu_to_le16(reg & 0x1f);
+	rc = hwrm_req_init(bp, req, HWRM_PORT_PHY_MDIO_WRITE);
+	if (rc)
+		return rc;
+
+	req->port_id = cpu_to_le16(bp->pf.port_id);
+	req->phy_addr = phy_addr;
+	req->reg_addr = cpu_to_le16(reg & 0x1f);
 	if (mdio_phy_id_is_c45(phy_addr)) {
-		req.cl45_mdio = 1;
-		req.phy_addr = mdio_phy_id_prtad(phy_addr);
-		req.dev_addr = mdio_phy_id_devad(phy_addr);
-		req.reg_addr = cpu_to_le16(reg);
+		req->cl45_mdio = 1;
+		req->phy_addr = mdio_phy_id_prtad(phy_addr);
+		req->dev_addr = mdio_phy_id_devad(phy_addr);
+		req->reg_addr = cpu_to_le16(reg);
 	}
-	req.reg_data = cpu_to_le16(val);
+	req->reg_data = cpu_to_le16(val);
 
-	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	return hwrm_req_send(bp, req);
 }
 
 /* rtnl_lock held */
@@ -10508,6 +10710,7 @@ static int bnxt_cfg_rx_mode(struct bnxt *bp)
 {
 	struct net_device *dev = bp->dev;
 	struct bnxt_vnic_info *vnic = &bp->vnic_info[0];
+	struct hwrm_cfa_l2_filter_free_input *req;
 	struct netdev_hw_addr *ha;
 	int i, off = 0, rc;
 	bool uc_update;
@@ -10519,19 +10722,16 @@ static int bnxt_cfg_rx_mode(struct bnxt *bp)
 	if (!uc_update)
 		goto skip_uc;
 
-	mutex_lock(&bp->hwrm_cmd_lock);
+	rc = hwrm_req_init(bp, req, HWRM_CFA_L2_FILTER_FREE);
+	if (rc)
+		return rc;
+	hwrm_req_hold(bp, req);
 	for (i = 1; i < vnic->uc_filter_count; i++) {
-		struct hwrm_cfa_l2_filter_free_input req = {0};
-
-		bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_L2_FILTER_FREE, -1,
-				       -1);
+		req->l2_filter_id = vnic->fw_l2_filter_id[i];
 
-		req.l2_filter_id = vnic->fw_l2_filter_id[i];
-
-		rc = _hwrm_send_message(bp, &req, sizeof(req),
-					HWRM_CMD_TIMEOUT);
+		rc = hwrm_req_send(bp, req);
 	}
-	mutex_unlock(&bp->hwrm_cmd_lock);
+	hwrm_req_drop(bp, req);
 
 	vnic->uc_filter_count = 1;
 
@@ -10883,22 +11083,30 @@ static netdev_features_t bnxt_features_check(struct sk_buff *skb,
 int bnxt_dbg_hwrm_rd_reg(struct bnxt *bp, u32 reg_off, u16 num_words,
 			 u32 *reg_buf)
 {
-	struct hwrm_dbg_read_direct_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_dbg_read_direct_input req = {0};
+	struct hwrm_dbg_read_direct_output *resp;
+	struct hwrm_dbg_read_direct_input *req;
 	__le32 *dbg_reg_buf;
 	dma_addr_t mapping;
 	int rc, i;
 
-	dbg_reg_buf = dma_alloc_coherent(&bp->pdev->dev, num_words * 4,
-					 &mapping, GFP_KERNEL);
-	if (!dbg_reg_buf)
-		return -ENOMEM;
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_DBG_READ_DIRECT, -1, -1);
-	req.host_dest_addr = cpu_to_le64(mapping);
-	req.read_addr = cpu_to_le32(reg_off + CHIMP_REG_VIEW_ADDR);
-	req.read_len32 = cpu_to_le32(num_words);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_DBG_READ_DIRECT);
+	if (rc)
+		return rc;
+
+	dbg_reg_buf = hwrm_req_dma_slice(bp, req, num_words * 4,
+					 &mapping);
+	if (!dbg_reg_buf) {
+		rc = -ENOMEM;
+		goto dbg_rd_reg_exit;
+	}
+
+	req->host_dest_addr = cpu_to_le64(mapping);
+
+	resp = hwrm_req_hold(bp, req);
+	req->read_addr = cpu_to_le32(reg_off + CHIMP_REG_VIEW_ADDR);
+	req->read_len32 = cpu_to_le32(num_words);
+
+	rc = hwrm_req_send(bp, req);
 	if (rc || resp->error_code) {
 		rc = -EIO;
 		goto dbg_rd_reg_exit;
@@ -10907,28 +11115,30 @@ int bnxt_dbg_hwrm_rd_reg(struct bnxt *bp, u32 reg_off, u16 num_words,
 		reg_buf[i] = le32_to_cpu(dbg_reg_buf[i]);
 
 dbg_rd_reg_exit:
-	mutex_unlock(&bp->hwrm_cmd_lock);
-	dma_free_coherent(&bp->pdev->dev, num_words * 4, dbg_reg_buf, mapping);
+	hwrm_req_drop(bp, req);
 	return rc;
 }
 
 static int bnxt_dbg_hwrm_ring_info_get(struct bnxt *bp, u8 ring_type,
 				       u32 ring_id, u32 *prod, u32 *cons)
 {
-	struct hwrm_dbg_ring_info_get_output *resp = bp->hwrm_cmd_resp_addr;
-	struct hwrm_dbg_ring_info_get_input req = {0};
+	struct hwrm_dbg_ring_info_get_output *resp;
+	struct hwrm_dbg_ring_info_get_input *req;
 	int rc;
 
-	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_DBG_RING_INFO_GET, -1, -1);
-	req.ring_type = ring_type;
-	req.fw_ring_id = cpu_to_le32(ring_id);
-	mutex_lock(&bp->hwrm_cmd_lock);
-	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	rc = hwrm_req_init(bp, req, HWRM_DBG_RING_INFO_GET);
+	if (rc)
+		return rc;
+
+	req->ring_type = ring_type;
+	req->fw_ring_id = cpu_to_le32(ring_id);
[… more lines in the raw message]

Lines of this file are not shown. View the raw message.

drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c +107 -77

184 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_devlink.c +48 -32

80 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +303 -244

547 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c +78 -51

129 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c +246 -204

450 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_tc.c +138 -125

263 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c +18 -12
index f621cffccd3a..fde0c3e8ac57 100644

Lines of this file are not shown. View the raw message.

drivers/net/ethernet/broadcom/bnxt/bnxt_vfr.c +32 -29

61 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Signature · 1 line
2.18.1

Part of the diff is not shown here. View the raw message.

author maintainer
#13 lore raw

PATCH v2 10/11 bnxt_en: remove legacy HWRM interface

In reply to Michael Chan

From: Edwin Peer <[email protected]>

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

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 18 +----
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  2 -
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 77 -------------------
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    | 10 ---
 4 files changed, 1 insertion(+), 106 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -17
index 6e9ca9b59e51..ddec1163748d 100644
@@ -3956,29 +3956,13 @@ static int bnxt_alloc_vnic_attributes(struct bnxt *bp)
 
 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)
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -2
index 79a78a7468f3..f343e87bef0b 100644
@@ -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;
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c +0 -77

89 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h +0 -10
index b3af7a88e2c7..39032cf66258 100644
@@ -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)))
Signature · 1 line
2.18.1
author maintainer
#14 lore raw

PATCH v2 11/11 bnxt_en: support multiple HWRM commands in flight

In reply to Michael Chan

From: Edwin Peer <[email protected]>

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.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 21 +++-
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  2 +-
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.c    | 96 +++++++++++++++----
 .../net/ethernet/broadcom/bnxt/bnxt_hwrm.h    | 33 ++++---
 4 files changed, 113 insertions(+), 39 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +17 -4
index ddec1163748d..627f85ee3922 100644
@@ -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 @@ static int bnxt_alloc_vnic_attributes(struct bnxt *bp)
 
 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;
 }
 
drivers/net/ethernet/broadcom/bnxt/bnxt.h +1 -1
index f343e87bef0b..a8212dcdad5f 100644
@@ -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;
drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.c +76 -20

159 lines not on this page · raw message

drivers/net/ethernet/broadcom/bnxt/bnxt_hwrm.h +19 -14
index 39032cf66258..4d17f0d5363b 100644
@@ -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)))
Signature · 1 line
2.18.1
#15 lore raw
In reply to Michael Chan
This series was applied to netdev/net-next.git (refs/heads/master):

Hello:

This series was applied to netdev/net-next.git (refs/heads/master):

Quoted 10 lines · On Sun, 29 Aug 2021 03:34:55 -0400 you wrote:

The current driver APIs to send messages to the firmware allow only one outstanding message in flight. There is only one buffer for the firmware response for each firmware channel. To send a firmware message, all callers must take a mutex and it is released after the firmware response has been read. This scheme does not allow multiple firmware messages in flight. Firmware may take a long time to respond to some messages (e.g. NVRAM related ones) and this causes the mutex to be held for a long time, blocking other callers.

[...]

Here is the summary with links:
- [net-next,v2,01/11] bnxt_en: remove DMA mapping for KONG response

    https://git.kernel.org/netdev/net-next/c/6c172d59ad79
  - [net-next,v2,02/11] bnxt_en: Refactor the HWRM_VER_GET firmware calls
    https://git.kernel.org/netdev/net-next/c/7b370ad77392
  - [net-next,v2,03/11] bnxt_en: move HWRM API implementation into separate file
    https://git.kernel.org/netdev/net-next/c/3c8c20db769c
  - [net-next,v2,04/11] bnxt_en: introduce new firmware message API based on DMA pools
    https://git.kernel.org/netdev/net-next/c/f9ff578251dc
  - [net-next,v2,05/11] bnxt_en: discard out of sequence HWRM responses
    https://git.kernel.org/netdev/net-next/c/02b9aa106868
  - [net-next,v2,06/11] bnxt_en: add HWRM request assignment API
    https://git.kernel.org/netdev/net-next/c/ecddc29d928d
  - [net-next,v2,07/11] bnxt_en: add support for HWRM request slices
    https://git.kernel.org/netdev/net-next/c/213808170840
  - [net-next,v2,08/11] bnxt_en: use link_lock instead of hwrm_cmd_lock to protect link_info
    https://git.kernel.org/netdev/net-next/c/3c10ed497fa8
  - [net-next,v2,09/11] bnxt_en: update all firmware calls to use the new APIs
    https://git.kernel.org/netdev/net-next/c/bbf33d1d9805
  - [net-next,v2,10/11] bnxt_en: remove legacy HWRM interface
    https://git.kernel.org/netdev/net-next/c/b34695a894b8
  - [net-next,v2,11/11] bnxt_en: support multiple HWRM commands in flight
    https://git.kernel.org/netdev/net-next/c/68f684e257d7

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html