summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/bng_re
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/infiniband/hw/bng_re')
-rw-r--r--drivers/infiniband/hw/bng_re/Kconfig10
-rw-r--r--drivers/infiniband/hw/bng_re/Makefile8
-rw-r--r--drivers/infiniband/hw/bng_re/bng_debugfs.c39
-rw-r--r--drivers/infiniband/hw/bng_re/bng_debugfs.h12
-rw-r--r--drivers/infiniband/hw/bng_re/bng_dev.c534
-rw-r--r--drivers/infiniband/hw/bng_re/bng_fw.c767
-rw-r--r--drivers/infiniband/hw/bng_re/bng_fw.h211
-rw-r--r--drivers/infiniband/hw/bng_re/bng_re.h85
-rw-r--r--drivers/infiniband/hw/bng_re/bng_res.c279
-rw-r--r--drivers/infiniband/hw/bng_re/bng_res.h215
-rw-r--r--drivers/infiniband/hw/bng_re/bng_sp.c131
-rw-r--r--drivers/infiniband/hw/bng_re/bng_sp.h47
-rw-r--r--drivers/infiniband/hw/bng_re/bng_tlv.h128
13 files changed, 2466 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/bng_re/Kconfig b/drivers/infiniband/hw/bng_re/Kconfig
new file mode 100644
index 000000000000..85845f72c64d
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/Kconfig
@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config INFINIBAND_BNG_RE
+ tristate "Broadcom Next generation RoCE HCA support"
+ depends on 64BIT
+ depends on INET && DCB && BNGE
+ help
+ This driver supports Broadcom Next generation
+ 50/100/200/400/800 gigabit RoCE HCAs. The module
+ will be called bng_re. To compile this driver
+ as a module, choose M here.
diff --git a/drivers/infiniband/hw/bng_re/Makefile b/drivers/infiniband/hw/bng_re/Makefile
new file mode 100644
index 000000000000..c6aaaf853c77
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/Makefile
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0
+ccflags-y := -I $(srctree)/drivers/net/ethernet/broadcom/bnge -I $(srctree)/drivers/infiniband/hw/bnxt_re
+
+obj-$(CONFIG_INFINIBAND_BNG_RE) += bng_re.o
+
+bng_re-y := bng_dev.o bng_fw.o \
+ bng_res.o bng_sp.o \
+ bng_debugfs.o
diff --git a/drivers/infiniband/hw/bng_re/bng_debugfs.c b/drivers/infiniband/hw/bng_re/bng_debugfs.c
new file mode 100644
index 000000000000..9ec5a8785250
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_debugfs.c
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2025 Broadcom.
+#include <linux/debugfs.h>
+#include <linux/pci.h>
+
+#include <rdma/ib_verbs.h>
+
+#include "bng_res.h"
+#include "bng_fw.h"
+#include "bnge.h"
+#include "bnge_auxr.h"
+#include "bng_re.h"
+#include "bng_debugfs.h"
+
+static struct dentry *bng_re_debugfs_root;
+
+void bng_re_debugfs_add_pdev(struct bng_re_dev *rdev)
+{
+ struct pci_dev *pdev = rdev->aux_dev->pdev;
+
+ rdev->dbg_root =
+ debugfs_create_dir(dev_name(&pdev->dev), bng_re_debugfs_root);
+}
+
+void bng_re_debugfs_rem_pdev(struct bng_re_dev *rdev)
+{
+ debugfs_remove_recursive(rdev->dbg_root);
+ rdev->dbg_root = NULL;
+}
+
+void bng_re_register_debugfs(void)
+{
+ bng_re_debugfs_root = debugfs_create_dir("bng_re", NULL);
+}
+
+void bng_re_unregister_debugfs(void)
+{
+ debugfs_remove(bng_re_debugfs_root);
+}
diff --git a/drivers/infiniband/hw/bng_re/bng_debugfs.h b/drivers/infiniband/hw/bng_re/bng_debugfs.h
new file mode 100644
index 000000000000..baef71df4242
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_debugfs.h
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+// Copyright (c) 2025 Broadcom.
+
+#ifndef __BNG_RE_DEBUGFS__
+#define __BNG_RE_DEBUGFS__
+
+void bng_re_debugfs_add_pdev(struct bng_re_dev *rdev);
+void bng_re_debugfs_rem_pdev(struct bng_re_dev *rdev);
+
+void bng_re_register_debugfs(void);
+void bng_re_unregister_debugfs(void);
+#endif
diff --git a/drivers/infiniband/hw/bng_re/bng_dev.c b/drivers/infiniband/hw/bng_re/bng_dev.c
new file mode 100644
index 000000000000..d8f8d7f7075f
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_dev.c
@@ -0,0 +1,534 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2025 Broadcom.
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/auxiliary_bus.h>
+
+#include <rdma/ib_verbs.h>
+
+#include "bng_res.h"
+#include "bng_sp.h"
+#include "bng_fw.h"
+#include "bnge.h"
+#include "bnge_auxr.h"
+#include "bng_re.h"
+#include "bnge_hwrm.h"
+#include "bng_debugfs.h"
+
+MODULE_AUTHOR("Siva Reddy Kallam <siva.kallam@broadcom.com>");
+MODULE_DESCRIPTION(BNG_RE_DESC);
+MODULE_LICENSE("Dual BSD/GPL");
+
+static struct bng_re_dev *bng_re_dev_add(struct auxiliary_device *adev,
+ struct bnge_auxr_dev *aux_dev)
+{
+ struct bng_re_dev *rdev;
+
+ /* Allocate bng_re_dev instance */
+ rdev = ib_alloc_device(bng_re_dev, ibdev);
+ if (!rdev) {
+ pr_err("%s: bng_re_dev allocation failure!", KBUILD_MODNAME);
+ return NULL;
+ }
+
+ /* Assign auxiliary device specific data */
+ rdev->netdev = aux_dev->net;
+ rdev->aux_dev = aux_dev;
+ rdev->adev = adev;
+ rdev->fn_id = rdev->aux_dev->pdev->devfn;
+
+ return rdev;
+}
+
+
+static int bng_re_register_netdev(struct bng_re_dev *rdev)
+{
+ struct bnge_auxr_dev *aux_dev;
+
+ aux_dev = rdev->aux_dev;
+ return bnge_register_dev(aux_dev, rdev->adev);
+}
+
+static void bng_re_destroy_chip_ctx(struct bng_re_dev *rdev)
+{
+ struct bng_re_chip_ctx *chip_ctx;
+
+ if (!rdev->chip_ctx)
+ return;
+
+ kfree(rdev->dev_attr);
+ rdev->dev_attr = NULL;
+
+ chip_ctx = rdev->chip_ctx;
+ rdev->chip_ctx = NULL;
+ rdev->rcfw.res = NULL;
+ rdev->bng_res.cctx = NULL;
+ rdev->bng_res.pdev = NULL;
+ kfree(chip_ctx);
+}
+
+static int bng_re_setup_chip_ctx(struct bng_re_dev *rdev)
+{
+ struct bng_re_chip_ctx *chip_ctx;
+ struct bnge_auxr_dev *aux_dev;
+ int rc = -ENOMEM;
+
+ aux_dev = rdev->aux_dev;
+ rdev->bng_res.pdev = aux_dev->pdev;
+ rdev->rcfw.res = &rdev->bng_res;
+ chip_ctx = kzalloc(sizeof(*chip_ctx), GFP_KERNEL);
+ if (!chip_ctx)
+ return -ENOMEM;
+ chip_ctx->chip_num = aux_dev->chip_num;
+ chip_ctx->hw_stats_size = aux_dev->hw_ring_stats_size;
+
+ rdev->chip_ctx = chip_ctx;
+ rdev->bng_res.cctx = rdev->chip_ctx;
+ rdev->dev_attr = kzalloc(sizeof(*rdev->dev_attr), GFP_KERNEL);
+ if (!rdev->dev_attr)
+ goto free_chip_ctx;
+ rdev->bng_res.dattr = rdev->dev_attr;
+
+ return 0;
+free_chip_ctx:
+ kfree(rdev->chip_ctx);
+ rdev->chip_ctx = NULL;
+ return rc;
+}
+
+static void bng_re_init_hwrm_hdr(struct input *hdr, u16 opcd)
+{
+ hdr->req_type = cpu_to_le16(opcd);
+ hdr->cmpl_ring = cpu_to_le16(-1);
+ hdr->target_id = cpu_to_le16(-1);
+}
+
+static void bng_re_fill_fw_msg(struct bnge_fw_msg *fw_msg, void *msg,
+ int msg_len, void *resp, int resp_max_len,
+ int timeout)
+{
+ fw_msg->msg = msg;
+ fw_msg->msg_len = msg_len;
+ fw_msg->resp = resp;
+ fw_msg->resp_max_len = resp_max_len;
+ fw_msg->timeout = timeout;
+}
+
+static int bng_re_net_ring_free(struct bng_re_dev *rdev,
+ u16 fw_ring_id, int type)
+{
+ struct bnge_auxr_dev *aux_dev = rdev->aux_dev;
+ struct hwrm_ring_free_input req = {};
+ struct hwrm_ring_free_output resp;
+ struct bnge_fw_msg fw_msg = {};
+ int rc = -EINVAL;
+
+ if (!rdev)
+ return rc;
+
+ if (!aux_dev)
+ return rc;
+
+ bng_re_init_hwrm_hdr((void *)&req, HWRM_RING_FREE);
+ req.ring_type = type;
+ req.ring_id = cpu_to_le16(fw_ring_id);
+ bng_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
+ sizeof(resp), BNGE_DFLT_HWRM_CMD_TIMEOUT);
+ rc = bnge_send_msg(aux_dev, &fw_msg);
+ if (rc)
+ ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
+ req.ring_id, rc);
+ return rc;
+}
+
+static int bng_re_net_ring_alloc(struct bng_re_dev *rdev,
+ struct bng_re_ring_attr *ring_attr,
+ u16 *fw_ring_id)
+{
+ struct bnge_auxr_dev *aux_dev = rdev->aux_dev;
+ struct hwrm_ring_alloc_input req = {};
+ struct hwrm_ring_alloc_output resp;
+ struct bnge_fw_msg fw_msg = {};
+ int rc = -EINVAL;
+
+ if (!aux_dev)
+ return rc;
+
+ bng_re_init_hwrm_hdr((void *)&req, HWRM_RING_ALLOC);
+ req.enables = 0;
+ req.page_tbl_addr = cpu_to_le64(ring_attr->dma_arr[0]);
+ if (ring_attr->pages > 1) {
+ /* Page size is in log2 units */
+ req.page_size = BNGE_PAGE_SHIFT;
+ req.page_tbl_depth = 1;
+ }
+ req.fbo = 0;
+ /* Association of ring index with doorbell index and MSIX number */
+ req.logical_id = cpu_to_le16(ring_attr->lrid);
+ req.length = cpu_to_le32(ring_attr->depth + 1);
+ req.ring_type = ring_attr->type;
+ req.int_mode = ring_attr->mode;
+ bng_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
+ sizeof(resp), BNGE_DFLT_HWRM_CMD_TIMEOUT);
+ rc = bnge_send_msg(aux_dev, &fw_msg);
+ if (!rc)
+ *fw_ring_id = le16_to_cpu(resp.ring_id);
+
+ return rc;
+}
+
+static int bng_re_stats_ctx_free(struct bng_re_dev *rdev)
+{
+ struct bnge_auxr_dev *aux_dev = rdev->aux_dev;
+ struct hwrm_stat_ctx_free_input req = {};
+ struct hwrm_stat_ctx_free_output resp = {};
+ struct bnge_fw_msg fw_msg = {};
+ int rc = -EINVAL;
+
+ if (!aux_dev)
+ return rc;
+
+ bng_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_FREE);
+ req.stat_ctx_id = cpu_to_le32(rdev->stats_ctx.fw_id);
+ bng_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
+ sizeof(resp), BNGE_DFLT_HWRM_CMD_TIMEOUT);
+ rc = bnge_send_msg(aux_dev, &fw_msg);
+ if (rc)
+ ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
+ rc);
+
+ return rc;
+}
+
+static int bng_re_stats_ctx_alloc(struct bng_re_dev *rdev)
+{
+ struct bnge_auxr_dev *aux_dev = rdev->aux_dev;
+ struct bng_re_stats *stats = &rdev->stats_ctx;
+ struct hwrm_stat_ctx_alloc_output resp = {};
+ struct hwrm_stat_ctx_alloc_input req = {};
+ struct bnge_fw_msg fw_msg = {};
+ int rc = -EINVAL;
+
+ stats->fw_id = BNGE_INVALID_STATS_CTX_ID;
+
+ if (!aux_dev)
+ return rc;
+
+ bng_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_ALLOC);
+ req.update_period_ms = cpu_to_le32(1000);
+ req.stats_dma_addr = cpu_to_le64(stats->dma_map);
+ req.stats_dma_length = cpu_to_le16(rdev->chip_ctx->hw_stats_size);
+ req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
+ bng_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
+ sizeof(resp), BNGE_DFLT_HWRM_CMD_TIMEOUT);
+ rc = bnge_send_msg(aux_dev, &fw_msg);
+ if (!rc)
+ stats->fw_id = le32_to_cpu(resp.stat_ctx_id);
+ return rc;
+}
+
+static void bng_re_query_hwrm_version(struct bng_re_dev *rdev)
+{
+ struct bnge_auxr_dev *aux_dev = rdev->aux_dev;
+ struct hwrm_ver_get_output ver_get_resp = {};
+ struct hwrm_ver_get_input ver_get_req = {};
+ struct bng_re_chip_ctx *cctx;
+ struct bnge_fw_msg fw_msg = {};
+ int rc;
+
+ bng_re_init_hwrm_hdr((void *)&ver_get_req, HWRM_VER_GET);
+ ver_get_req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
+ ver_get_req.hwrm_intf_min = HWRM_VERSION_MINOR;
+ ver_get_req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
+ bng_re_fill_fw_msg(&fw_msg, (void *)&ver_get_req, sizeof(ver_get_req),
+ (void *)&ver_get_resp, sizeof(ver_get_resp),
+ BNGE_DFLT_HWRM_CMD_TIMEOUT);
+ rc = bnge_send_msg(aux_dev, &fw_msg);
+ if (rc) {
+ ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
+ rc);
+ return;
+ }
+
+ cctx = rdev->chip_ctx;
+ cctx->hwrm_intf_ver =
+ (u64)le16_to_cpu(ver_get_resp.hwrm_intf_major) << 48 |
+ (u64)le16_to_cpu(ver_get_resp.hwrm_intf_minor) << 32 |
+ (u64)le16_to_cpu(ver_get_resp.hwrm_intf_build) << 16 |
+ le16_to_cpu(ver_get_resp.hwrm_intf_patch);
+
+ cctx->hwrm_cmd_max_timeout = le16_to_cpu(ver_get_resp.max_req_timeout);
+
+ if (!cctx->hwrm_cmd_max_timeout)
+ cctx->hwrm_cmd_max_timeout = BNG_ROCE_FW_MAX_TIMEOUT;
+}
+
+static void bng_re_dev_uninit(struct bng_re_dev *rdev)
+{
+ int rc;
+ bng_re_debugfs_rem_pdev(rdev);
+
+ if (test_and_clear_bit(BNG_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
+ rc = bng_re_deinit_rcfw(&rdev->rcfw);
+ if (rc)
+ ibdev_warn(&rdev->ibdev,
+ "Failed to deinitialize RCFW: %#x", rc);
+ bng_re_stats_ctx_free(rdev);
+ bng_re_free_stats_ctx_mem(rdev->bng_res.pdev, &rdev->stats_ctx);
+ bng_re_disable_rcfw_channel(&rdev->rcfw);
+ bng_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id,
+ RING_ALLOC_REQ_RING_TYPE_NQ);
+ bng_re_free_rcfw_channel(&rdev->rcfw);
+ }
+
+ kfree(rdev->nqr);
+ rdev->nqr = NULL;
+ bng_re_destroy_chip_ctx(rdev);
+ if (test_and_clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags))
+ bnge_unregister_dev(rdev->aux_dev);
+}
+
+static int bng_re_dev_init(struct bng_re_dev *rdev)
+{
+ struct bng_re_ring_attr rattr = {};
+ struct bng_re_creq_ctx *creq;
+ u32 db_offt;
+ int vid;
+ u8 type;
+ int rc;
+
+ /* Registered a new RoCE device instance to netdev */
+ rc = bng_re_register_netdev(rdev);
+ if (rc) {
+ ibdev_err(&rdev->ibdev,
+ "Failed to register with netedev: %#x\n", rc);
+ return -EINVAL;
+ }
+
+ set_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
+
+ if (rdev->aux_dev->auxr_info->msix_requested < BNG_RE_MIN_MSIX) {
+ ibdev_err(&rdev->ibdev,
+ "RoCE requires minimum 2 MSI-X vectors, but only %d reserved\n",
+ rdev->aux_dev->auxr_info->msix_requested);
+ bnge_unregister_dev(rdev->aux_dev);
+ clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
+ return -EINVAL;
+ }
+ ibdev_dbg(&rdev->ibdev, "Got %d MSI-X vectors\n",
+ rdev->aux_dev->auxr_info->msix_requested);
+
+ rc = bng_re_setup_chip_ctx(rdev);
+ if (rc) {
+ bnge_unregister_dev(rdev->aux_dev);
+ clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
+ ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
+ return -EINVAL;
+ }
+
+ bng_re_query_hwrm_version(rdev);
+
+ rc = bng_re_alloc_fw_channel(&rdev->bng_res, &rdev->rcfw);
+ if (rc) {
+ ibdev_err(&rdev->ibdev,
+ "Failed to allocate RCFW Channel: %#x\n", rc);
+ goto fail;
+ }
+
+ /* Allocate nq record memory */
+ rdev->nqr = kzalloc(sizeof(*rdev->nqr), GFP_KERNEL);
+ if (!rdev->nqr) {
+ bng_re_destroy_chip_ctx(rdev);
+ bnge_unregister_dev(rdev->aux_dev);
+ clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
+ return -ENOMEM;
+ }
+
+ rdev->nqr->num_msix = rdev->aux_dev->auxr_info->msix_requested;
+ memcpy(rdev->nqr->msix_entries, rdev->aux_dev->msix_info,
+ sizeof(struct bnge_msix_info) * rdev->nqr->num_msix);
+
+ type = RING_ALLOC_REQ_RING_TYPE_NQ;
+ creq = &rdev->rcfw.creq;
+ rattr.dma_arr = creq->hwq.pbl[BNG_PBL_LVL_0].pg_map_arr;
+ rattr.pages = creq->hwq.pbl[creq->hwq.level].pg_count;
+ rattr.type = type;
+ rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
+ rattr.depth = BNG_FW_CREQE_MAX_CNT - 1;
+ rattr.lrid = rdev->nqr->msix_entries[BNG_RE_CREQ_NQ_IDX].ring_idx;
+ rc = bng_re_net_ring_alloc(rdev, &rattr, &creq->ring_id);
+ if (rc) {
+ ibdev_err(&rdev->ibdev, "Failed to allocate CREQ: %#x\n", rc);
+ goto free_rcfw;
+ }
+ db_offt = rdev->nqr->msix_entries[BNG_RE_CREQ_NQ_IDX].db_offset;
+ vid = rdev->nqr->msix_entries[BNG_RE_CREQ_NQ_IDX].vector;
+
+ rc = bng_re_enable_fw_channel(&rdev->rcfw,
+ vid, db_offt);
+ if (rc) {
+ ibdev_err(&rdev->ibdev, "Failed to enable RCFW channel: %#x\n",
+ rc);
+ goto free_ring;
+ }
+
+ rc = bng_re_get_dev_attr(&rdev->rcfw);
+ if (rc)
+ goto disable_rcfw;
+
+ bng_re_debugfs_add_pdev(rdev);
+ rc = bng_re_alloc_stats_ctx_mem(rdev->bng_res.pdev, rdev->chip_ctx,
+ &rdev->stats_ctx);
+ if (rc) {
+ ibdev_err(&rdev->ibdev,
+ "Failed to allocate stats context: %#x\n", rc);
+ goto disable_rcfw;
+ }
+
+ rc = bng_re_stats_ctx_alloc(rdev);
+ if (rc) {
+ ibdev_err(&rdev->ibdev,
+ "Failed to allocate QPLIB context: %#x\n", rc);
+ goto free_stats_ctx;
+ }
+
+ rc = bng_re_init_rcfw(&rdev->rcfw, &rdev->stats_ctx);
+ if (rc) {
+ ibdev_err(&rdev->ibdev,
+ "Failed to initialize RCFW: %#x\n", rc);
+ goto free_sctx;
+ }
+ set_bit(BNG_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
+
+ return 0;
+free_sctx:
+ bng_re_stats_ctx_free(rdev);
+free_stats_ctx:
+ bng_re_free_stats_ctx_mem(rdev->bng_res.pdev, &rdev->stats_ctx);
+disable_rcfw:
+ bng_re_disable_rcfw_channel(&rdev->rcfw);
+free_ring:
+ bng_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
+free_rcfw:
+ bng_re_free_rcfw_channel(&rdev->rcfw);
+fail:
+ bng_re_dev_uninit(rdev);
+ return rc;
+}
+
+static int bng_re_add_device(struct auxiliary_device *adev)
+{
+ struct bnge_auxr_priv *auxr_priv =
+ container_of(adev, struct bnge_auxr_priv, aux_dev);
+ struct bng_re_en_dev_info *dev_info;
+ struct bng_re_dev *rdev;
+ int rc;
+
+ dev_info = auxiliary_get_drvdata(adev);
+
+ rdev = bng_re_dev_add(adev, auxr_priv->auxr_dev);
+ if (!rdev) {
+ rc = -ENOMEM;
+ goto exit;
+ }
+
+ dev_info->rdev = rdev;
+
+ rc = bng_re_dev_init(rdev);
+ if (rc)
+ goto re_dev_dealloc;
+
+ return 0;
+
+re_dev_dealloc:
+ ib_dealloc_device(&rdev->ibdev);
+exit:
+ return rc;
+}
+
+
+static void bng_re_remove_device(struct bng_re_dev *rdev,
+ struct auxiliary_device *aux_dev)
+{
+ bng_re_dev_uninit(rdev);
+ ib_dealloc_device(&rdev->ibdev);
+}
+
+
+static int bng_re_probe(struct auxiliary_device *adev,
+ const struct auxiliary_device_id *id)
+{
+ struct bnge_auxr_priv *aux_priv =
+ container_of(adev, struct bnge_auxr_priv, aux_dev);
+ struct bng_re_en_dev_info *en_info;
+ int rc;
+
+ en_info = kzalloc(sizeof(*en_info), GFP_KERNEL);
+ if (!en_info)
+ return -ENOMEM;
+
+ en_info->auxr_dev = aux_priv->auxr_dev;
+
+ auxiliary_set_drvdata(adev, en_info);
+
+ rc = bng_re_add_device(adev);
+ if (rc)
+ kfree(en_info);
+
+ return rc;
+}
+
+static void bng_re_remove(struct auxiliary_device *adev)
+{
+ struct bng_re_en_dev_info *dev_info = auxiliary_get_drvdata(adev);
+ struct bng_re_dev *rdev;
+
+ rdev = dev_info->rdev;
+
+ if (rdev)
+ bng_re_remove_device(rdev, adev);
+ kfree(dev_info);
+}
+
+static const struct auxiliary_device_id bng_re_id_table[] = {
+ { .name = BNG_RE_ADEV_NAME ".rdma", },
+ {},
+};
+
+MODULE_DEVICE_TABLE(auxiliary, bng_re_id_table);
+
+static struct auxiliary_driver bng_re_driver = {
+ .name = "rdma",
+ .probe = bng_re_probe,
+ .remove = bng_re_remove,
+ .id_table = bng_re_id_table,
+};
+
+static int __init bng_re_mod_init(void)
+{
+ int rc;
+
+
+ bng_re_register_debugfs();
+
+ rc = auxiliary_driver_register(&bng_re_driver);
+ if (rc) {
+ pr_err("%s: Failed to register auxiliary driver\n",
+ KBUILD_MODNAME);
+ goto unreg_debugfs;
+ }
+ return 0;
+unreg_debugfs:
+ bng_re_unregister_debugfs();
+ return rc;
+}
+
+static void __exit bng_re_mod_exit(void)
+{
+ auxiliary_driver_unregister(&bng_re_driver);
+ bng_re_unregister_debugfs();
+}
+
+module_init(bng_re_mod_init);
+module_exit(bng_re_mod_exit);
diff --git a/drivers/infiniband/hw/bng_re/bng_fw.c b/drivers/infiniband/hw/bng_re/bng_fw.c
new file mode 100644
index 000000000000..7d9539113cf5
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_fw.c
@@ -0,0 +1,767 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2025 Broadcom.
+#include <linux/pci.h>
+
+#include "roce_hsi.h"
+#include "bng_res.h"
+#include "bng_fw.h"
+#include "bng_sp.h"
+
+/**
+ * bng_re_map_rc - map return type based on opcode
+ * @opcode: roce slow path opcode
+ *
+ * case #1
+ * Firmware initiated error recovery is a safe state machine and
+ * driver can consider all the underlying rdma resources are free.
+ * In this state, it is safe to return success for opcodes related to
+ * destroying rdma resources (like destroy qp, destroy cq etc.).
+ *
+ * case #2
+ * If driver detect potential firmware stall, it is not safe state machine
+ * and the driver can not consider all the underlying rdma resources are
+ * freed.
+ * In this state, it is not safe to return success for opcodes related to
+ * destroying rdma resources (like destroy qp, destroy cq etc.).
+ *
+ * Scope of this helper function is only for case #1.
+ *
+ * Returns:
+ * 0 to communicate success to caller.
+ * Non zero error code to communicate failure to caller.
+ */
+static int bng_re_map_rc(u8 opcode)
+{
+ switch (opcode) {
+ case CMDQ_BASE_OPCODE_DESTROY_QP:
+ case CMDQ_BASE_OPCODE_DESTROY_SRQ:
+ case CMDQ_BASE_OPCODE_DESTROY_CQ:
+ case CMDQ_BASE_OPCODE_DEALLOCATE_KEY:
+ case CMDQ_BASE_OPCODE_DEREGISTER_MR:
+ case CMDQ_BASE_OPCODE_DELETE_GID:
+ case CMDQ_BASE_OPCODE_DESTROY_QP1:
+ case CMDQ_BASE_OPCODE_DESTROY_AH:
+ case CMDQ_BASE_OPCODE_DEINITIALIZE_FW:
+ case CMDQ_BASE_OPCODE_MODIFY_ROCE_CC:
+ case CMDQ_BASE_OPCODE_SET_LINK_AGGR_MODE:
+ return 0;
+ default:
+ return -ETIMEDOUT;
+ }
+}
+
+void bng_re_free_rcfw_channel(struct bng_re_rcfw *rcfw)
+{
+ kfree(rcfw->crsqe_tbl);
+ bng_re_free_hwq(rcfw->res, &rcfw->cmdq.hwq);
+ bng_re_free_hwq(rcfw->res, &rcfw->creq.hwq);
+ rcfw->pdev = NULL;
+}
+
+int bng_re_alloc_fw_channel(struct bng_re_res *res,
+ struct bng_re_rcfw *rcfw)
+{
+ struct bng_re_hwq_attr hwq_attr = {};
+ struct bng_re_sg_info sginfo = {};
+ struct bng_re_cmdq_ctx *cmdq;
+ struct bng_re_creq_ctx *creq;
+
+ rcfw->pdev = res->pdev;
+ cmdq = &rcfw->cmdq;
+ creq = &rcfw->creq;
+ rcfw->res = res;
+
+ sginfo.pgsize = PAGE_SIZE;
+ sginfo.pgshft = PAGE_SHIFT;
+
+ hwq_attr.sginfo = &sginfo;
+ hwq_attr.res = rcfw->res;
+ hwq_attr.depth = BNG_FW_CREQE_MAX_CNT;
+ hwq_attr.stride = BNG_FW_CREQE_UNITS;
+ hwq_attr.type = BNG_HWQ_TYPE_QUEUE;
+
+ if (bng_re_alloc_init_hwq(&creq->hwq, &hwq_attr)) {
+ dev_err(&rcfw->pdev->dev,
+ "HW channel CREQ allocation failed\n");
+ goto fail;
+ }
+
+ rcfw->cmdq_depth = BNG_FW_CMDQE_MAX_CNT;
+
+ sginfo.pgsize = bng_fw_cmdqe_page_size(rcfw->cmdq_depth);
+ hwq_attr.depth = rcfw->cmdq_depth & 0x7FFFFFFF;
+ hwq_attr.stride = BNG_FW_CMDQE_UNITS;
+ hwq_attr.type = BNG_HWQ_TYPE_CTX;
+ if (bng_re_alloc_init_hwq(&cmdq->hwq, &hwq_attr)) {
+ dev_err(&rcfw->pdev->dev,
+ "HW channel CMDQ allocation failed\n");
+ goto fail;
+ }
+
+ rcfw->crsqe_tbl = kcalloc(cmdq->hwq.max_elements,
+ sizeof(*rcfw->crsqe_tbl), GFP_KERNEL);
+ if (!rcfw->crsqe_tbl)
+ goto fail;
+
+ spin_lock_init(&rcfw->tbl_lock);
+
+ rcfw->max_timeout = res->cctx->hwrm_cmd_max_timeout;
+ return 0;
+
+fail:
+ bng_re_free_rcfw_channel(rcfw);
+ return -ENOMEM;
+}
+
+static int bng_re_process_qp_event(struct bng_re_rcfw *rcfw,
+ struct creq_qp_event *qp_event,
+ u32 *num_wait)
+{
+ struct bng_re_hwq *hwq = &rcfw->cmdq.hwq;
+ struct bng_re_crsqe *crsqe;
+ u32 req_size;
+ u16 cookie;
+ bool is_waiter_alive;
+ struct pci_dev *pdev;
+ u32 wait_cmds = 0;
+ int rc = 0;
+
+ pdev = rcfw->pdev;
+ switch (qp_event->event) {
+ case CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION:
+ dev_err(&pdev->dev, "Received QP error notification\n");
+ break;
+ default:
+ /*
+ * Command Response
+ * cmdq->lock needs to be acquired to synchronie
+ * the command send and completion reaping. This function
+ * is always called with creq->lock held. Using
+ * the nested variant of spin_lock.
+ *
+ */
+
+ spin_lock_nested(&hwq->lock, SINGLE_DEPTH_NESTING);
+ cookie = le16_to_cpu(qp_event->cookie);
+ cookie &= BNG_FW_MAX_COOKIE_VALUE;
+ crsqe = &rcfw->crsqe_tbl[cookie];
+
+ if (WARN_ONCE(test_bit(FIRMWARE_STALL_DETECTED,
+ &rcfw->cmdq.flags),
+ "Unreponsive rcfw channel detected.!!")) {
+ dev_info(&pdev->dev,
+ "rcfw timedout: cookie = %#x, free_slots = %d",
+ cookie, crsqe->free_slots);
+ spin_unlock(&hwq->lock);
+ return rc;
+ }
+
+ if (crsqe->is_waiter_alive) {
+ if (crsqe->resp) {
+ memcpy(crsqe->resp, qp_event, sizeof(*qp_event));
+ /* Insert write memory barrier to ensure that
+ * response data is copied before clearing the
+ * flags
+ */
+ smp_wmb();
+ }
+ }
+
+ wait_cmds++;
+
+ req_size = crsqe->req_size;
+ is_waiter_alive = crsqe->is_waiter_alive;
+
+ crsqe->req_size = 0;
+ if (!is_waiter_alive)
+ crsqe->resp = NULL;
+
+ crsqe->is_in_used = false;
+
+ hwq->cons += req_size;
+
+ spin_unlock(&hwq->lock);
+ }
+ *num_wait += wait_cmds;
+ return rc;
+}
+
+/* function events */
+static int bng_re_process_func_event(struct bng_re_rcfw *rcfw,
+ struct creq_func_event *func_event)
+{
+ switch (func_event->event) {
+ case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
+ case CREQ_FUNC_EVENT_EVENT_VF_COMM_REQUEST:
+ case CREQ_FUNC_EVENT_EVENT_RESOURCE_EXHAUSTED:
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/* CREQ Completion handlers */
+static void bng_re_service_creq(struct tasklet_struct *t)
+{
+ struct bng_re_rcfw *rcfw = from_tasklet(rcfw, t, creq.creq_tasklet);
+ struct bng_re_creq_ctx *creq = &rcfw->creq;
+ u32 type, budget = BNG_FW_CREQ_ENTRY_POLL_BUDGET;
+ struct bng_re_hwq *hwq = &creq->hwq;
+ struct creq_base *creqe;
+ u32 num_wakeup = 0;
+ u32 hw_polled = 0;
+
+ /* Service the CREQ until budget is over */
+ spin_lock_bh(&hwq->lock);
+ while (budget > 0) {
+ creqe = bng_re_get_qe(hwq, hwq->cons, NULL);
+ if (!BNG_FW_CREQ_CMP_VALID(creqe, creq->creq_db.dbinfo.flags))
+ break;
+ /* The valid test of the entry must be done first before
+ * reading any further.
+ */
+ dma_rmb();
+
+ type = creqe->type & CREQ_BASE_TYPE_MASK;
+ switch (type) {
+ case CREQ_BASE_TYPE_QP_EVENT:
+ bng_re_process_qp_event
+ (rcfw, (struct creq_qp_event *)creqe,
+ &num_wakeup);
+ creq->stats.creq_qp_event_processed++;
+ break;
+ case CREQ_BASE_TYPE_FUNC_EVENT:
+ if (!bng_re_process_func_event
+ (rcfw, (struct creq_func_event *)creqe))
+ creq->stats.creq_func_event_processed++;
+ else
+ dev_warn(&rcfw->pdev->dev,
+ "aeqe:%#x Not handled\n", type);
+ break;
+ default:
+ if (type != ASYNC_EVENT_CMPL_TYPE_HWRM_ASYNC_EVENT)
+ dev_warn(&rcfw->pdev->dev,
+ "creqe with event 0x%x not handled\n",
+ type);
+ break;
+ }
+ budget--;
+ hw_polled++;
+ bng_re_hwq_incr_cons(hwq->max_elements, &hwq->cons,
+ 1, &creq->creq_db.dbinfo.flags);
+ }
+
+ if (hw_polled)
+ bng_re_ring_nq_db(&creq->creq_db.dbinfo,
+ rcfw->res->cctx, true);
+ spin_unlock_bh(&hwq->lock);
+ if (num_wakeup)
+ wake_up_nr(&rcfw->cmdq.waitq, num_wakeup);
+}
+
+static int __send_message_basic_sanity(struct bng_re_rcfw *rcfw,
+ struct bng_re_cmdqmsg *msg,
+ u8 opcode)
+{
+ struct bng_re_cmdq_ctx *cmdq;
+
+ cmdq = &rcfw->cmdq;
+
+ if (test_bit(FIRMWARE_STALL_DETECTED, &cmdq->flags))
+ return -ETIMEDOUT;
+
+ if (test_bit(FIRMWARE_INITIALIZED_FLAG, &cmdq->flags) &&
+ opcode == CMDQ_BASE_OPCODE_INITIALIZE_FW) {
+ dev_err(&rcfw->pdev->dev, "RCFW already initialized!");
+ return -EINVAL;
+ }
+
+ if (!test_bit(FIRMWARE_INITIALIZED_FLAG, &cmdq->flags) &&
+ (opcode != CMDQ_BASE_OPCODE_QUERY_FUNC &&
+ opcode != CMDQ_BASE_OPCODE_INITIALIZE_FW &&
+ opcode != CMDQ_BASE_OPCODE_QUERY_VERSION)) {
+ dev_err(&rcfw->pdev->dev,
+ "RCFW not initialized, reject opcode 0x%x",
+ opcode);
+ return -EOPNOTSUPP;
+ }
+
+ return 0;
+}
+
+static int __send_message(struct bng_re_rcfw *rcfw,
+ struct bng_re_cmdqmsg *msg, u8 opcode)
+{
+ u32 bsize, free_slots, required_slots;
+ struct bng_re_cmdq_ctx *cmdq;
+ struct bng_re_crsqe *crsqe;
+ struct bng_fw_cmdqe *cmdqe;
+ struct bng_re_hwq *hwq;
+ u32 sw_prod, cmdq_prod;
+ struct pci_dev *pdev;
+ u16 cookie;
+ u8 *preq;
+
+ cmdq = &rcfw->cmdq;
+ hwq = &cmdq->hwq;
+ pdev = rcfw->pdev;
+
+ /* Cmdq are in 16-byte units, each request can consume 1 or more
+ * cmdqe
+ */
+ spin_lock_bh(&hwq->lock);
+ required_slots = bng_re_get_cmd_slots(msg->req);
+ free_slots = HWQ_FREE_SLOTS(hwq);
+ cookie = cmdq->seq_num & BNG_FW_MAX_COOKIE_VALUE;
+ crsqe = &rcfw->crsqe_tbl[cookie];
+
+ if (required_slots >= free_slots) {
+ dev_info_ratelimited(&pdev->dev,
+ "CMDQ is full req/free %d/%d!",
+ required_slots, free_slots);
+ spin_unlock_bh(&hwq->lock);
+ return -EAGAIN;
+ }
+ __set_cmdq_base_cookie(msg->req, msg->req_sz, cpu_to_le16(cookie));
+
+ bsize = bng_re_set_cmd_slots(msg->req);
+ crsqe->free_slots = free_slots;
+ crsqe->resp = (struct creq_qp_event *)msg->resp;
+ crsqe->is_waiter_alive = true;
+ crsqe->is_in_used = true;
+ crsqe->opcode = opcode;
+
+ crsqe->req_size = __get_cmdq_base_cmd_size(msg->req, msg->req_sz);
+ if (__get_cmdq_base_resp_size(msg->req, msg->req_sz) && msg->sb) {
+ struct bng_re_rcfw_sbuf *sbuf = msg->sb;
+
+ __set_cmdq_base_resp_addr(msg->req, msg->req_sz,
+ cpu_to_le64(sbuf->dma_addr));
+ __set_cmdq_base_resp_size(msg->req, msg->req_sz,
+ ALIGN(sbuf->size,
+ BNG_FW_CMDQE_UNITS) /
+ BNG_FW_CMDQE_UNITS);
+ }
+
+ preq = (u8 *)msg->req;
+ do {
+ /* Locate the next cmdq slot */
+ sw_prod = HWQ_CMP(hwq->prod, hwq);
+ cmdqe = bng_re_get_qe(hwq, sw_prod, NULL);
+ /* Copy a segment of the req cmd to the cmdq */
+ memset(cmdqe, 0, sizeof(*cmdqe));
+ memcpy(cmdqe, preq, min_t(u32, bsize, sizeof(*cmdqe)));
+ preq += min_t(u32, bsize, sizeof(*cmdqe));
+ bsize -= min_t(u32, bsize, sizeof(*cmdqe));
+ hwq->prod++;
+ } while (bsize > 0);
+ cmdq->seq_num++;
+
+ cmdq_prod = hwq->prod & 0xFFFF;
+ if (test_bit(FIRMWARE_FIRST_FLAG, &cmdq->flags)) {
+ /* The very first doorbell write
+ * is required to set this flag
+ * which prompts the FW to reset
+ * its internal pointers
+ */
+ cmdq_prod |= BIT(FIRMWARE_FIRST_FLAG);
+ clear_bit(FIRMWARE_FIRST_FLAG, &cmdq->flags);
+ }
+ /* ring CMDQ DB */
+ wmb();
+ writel(cmdq_prod, cmdq->cmdq_mbox.prod);
+ writel(BNG_FW_CMDQ_TRIG_VAL, cmdq->cmdq_mbox.db);
+ spin_unlock_bh(&hwq->lock);
+ /* Return the CREQ response pointer */
+ return 0;
+}
+
+/**
+ * __wait_for_resp - Don't hold the cpu context and wait for response
+ * @rcfw: rcfw channel instance of rdev
+ * @cookie: cookie to track the command
+ *
+ * Wait for command completion in sleepable context.
+ *
+ * Returns:
+ * 0 if command is completed by firmware.
+ * Non zero error code for rest of the case.
+ */
+static int __wait_for_resp(struct bng_re_rcfw *rcfw, u16 cookie)
+{
+ struct bng_re_cmdq_ctx *cmdq;
+ struct bng_re_crsqe *crsqe;
+
+ cmdq = &rcfw->cmdq;
+ crsqe = &rcfw->crsqe_tbl[cookie];
+
+ do {
+ wait_event_timeout(cmdq->waitq,
+ !crsqe->is_in_used,
+ secs_to_jiffies(rcfw->max_timeout));
+
+ if (!crsqe->is_in_used)
+ return 0;
+
+ bng_re_service_creq(&rcfw->creq.creq_tasklet);
+
+ if (!crsqe->is_in_used)
+ return 0;
+ } while (true);
+};
+
+/**
+ * bng_re_rcfw_send_message - interface to send
+ * and complete rcfw command.
+ * @rcfw: rcfw channel instance of rdev
+ * @msg: message to send
+ *
+ * This function does not account shadow queue depth. It will send
+ * all the command unconditionally as long as send queue is not full.
+ *
+ * Returns:
+ * 0 if command completed by firmware.
+ * Non zero if the command is not completed by firmware.
+ */
+int bng_re_rcfw_send_message(struct bng_re_rcfw *rcfw,
+ struct bng_re_cmdqmsg *msg)
+{
+ struct creq_qp_event *evnt = (struct creq_qp_event *)msg->resp;
+ struct bng_re_crsqe *crsqe;
+ u16 cookie;
+ int rc;
+ u8 opcode;
+
+ opcode = __get_cmdq_base_opcode(msg->req, msg->req_sz);
+
+ rc = __send_message_basic_sanity(rcfw, msg, opcode);
+ if (rc)
+ return rc == -ENXIO ? bng_re_map_rc(opcode) : rc;
+
+ rc = __send_message(rcfw, msg, opcode);
+ if (rc)
+ return rc;
+
+ cookie = le16_to_cpu(__get_cmdq_base_cookie(msg->req, msg->req_sz))
+ & BNG_FW_MAX_COOKIE_VALUE;
+
+ rc = __wait_for_resp(rcfw, cookie);
+
+ if (rc) {
+ spin_lock_bh(&rcfw->cmdq.hwq.lock);
+ crsqe = &rcfw->crsqe_tbl[cookie];
+ crsqe->is_waiter_alive = false;
+ if (rc == -ENODEV)
+ set_bit(FIRMWARE_STALL_DETECTED, &rcfw->cmdq.flags);
+ spin_unlock_bh(&rcfw->cmdq.hwq.lock);
+ return -ETIMEDOUT;
+ }
+
+ if (evnt->status) {
+ /* failed with status */
+ dev_err(&rcfw->pdev->dev, "cmdq[%#x]=%#x status %#x\n",
+ cookie, opcode, evnt->status);
+ rc = -EIO;
+ }
+
+ return rc;
+}
+
+static int bng_re_map_cmdq_mbox(struct bng_re_rcfw *rcfw)
+{
+ struct bng_re_cmdq_mbox *mbox;
+ resource_size_t bar_reg;
+ struct pci_dev *pdev;
+
+ pdev = rcfw->pdev;
+ mbox = &rcfw->cmdq.cmdq_mbox;
+
+ mbox->reg.bar_id = BNG_FW_COMM_PCI_BAR_REGION;
+ mbox->reg.len = BNG_FW_COMM_SIZE;
+ mbox->reg.bar_base = pci_resource_start(pdev, mbox->reg.bar_id);
+ if (!mbox->reg.bar_base) {
+ dev_err(&pdev->dev,
+ "CMDQ BAR region %d resc start is 0!\n",
+ mbox->reg.bar_id);
+ return -ENOMEM;
+ }
+
+ bar_reg = mbox->reg.bar_base + BNG_FW_COMM_BASE_OFFSET;
+ mbox->reg.len = BNG_FW_COMM_SIZE;
+ mbox->reg.bar_reg = ioremap(bar_reg, mbox->reg.len);
+ if (!mbox->reg.bar_reg) {
+ dev_err(&pdev->dev,
+ "CMDQ BAR region %d mapping failed\n",
+ mbox->reg.bar_id);
+ return -ENOMEM;
+ }
+
+ mbox->prod = (void __iomem *)(mbox->reg.bar_reg +
+ BNG_FW_PF_VF_COMM_PROD_OFFSET);
+ mbox->db = (void __iomem *)(mbox->reg.bar_reg + BNG_FW_COMM_TRIG_OFFSET);
+ return 0;
+}
+
+static irqreturn_t bng_re_creq_irq(int irq, void *dev_instance)
+{
+ struct bng_re_rcfw *rcfw = dev_instance;
+ struct bng_re_creq_ctx *creq;
+ struct bng_re_hwq *hwq;
+ u32 sw_cons;
+
+ creq = &rcfw->creq;
+ hwq = &creq->hwq;
+ /* Prefetch the CREQ element */
+ sw_cons = HWQ_CMP(hwq->cons, hwq);
+ bng_re_get_qe(hwq, sw_cons, NULL);
+
+ tasklet_schedule(&creq->creq_tasklet);
+ return IRQ_HANDLED;
+}
+
+int bng_re_rcfw_start_irq(struct bng_re_rcfw *rcfw, int msix_vector,
+ bool need_init)
+{
+ struct bng_re_creq_ctx *creq;
+ struct bng_re_res *res;
+ int rc;
+
+ creq = &rcfw->creq;
+ res = rcfw->res;
+
+ if (creq->irq_handler_avail)
+ return -EFAULT;
+
+ creq->msix_vec = msix_vector;
+ if (need_init)
+ tasklet_setup(&creq->creq_tasklet, bng_re_service_creq);
+ else
+ tasklet_enable(&creq->creq_tasklet);
+
+ creq->irq_name = kasprintf(GFP_KERNEL, "bng_re-creq@pci:%s",
+ pci_name(res->pdev));
+ if (!creq->irq_name)
+ return -ENOMEM;
+ rc = request_irq(creq->msix_vec, bng_re_creq_irq, 0,
+ creq->irq_name, rcfw);
+ if (rc) {
+ kfree(creq->irq_name);
+ creq->irq_name = NULL;
+ tasklet_disable(&creq->creq_tasklet);
+ return rc;
+ }
+ creq->irq_handler_avail = true;
+
+ bng_re_ring_nq_db(&creq->creq_db.dbinfo, res->cctx, true);
+ atomic_inc(&rcfw->rcfw_intr_enabled);
+
+ return 0;
+}
+
+static int bng_re_map_creq_db(struct bng_re_rcfw *rcfw, u32 reg_offt)
+{
+ struct bng_re_creq_db *creq_db;
+ resource_size_t bar_reg;
+ struct pci_dev *pdev;
+
+ pdev = rcfw->pdev;
+ creq_db = &rcfw->creq.creq_db;
+
+ creq_db->dbinfo.flags = 0;
+ creq_db->reg.bar_id = BNG_FW_COMM_CONS_PCI_BAR_REGION;
+ creq_db->reg.bar_base = pci_resource_start(pdev, creq_db->reg.bar_id);
+ if (!creq_db->reg.bar_id)
+ dev_err(&pdev->dev,
+ "CREQ BAR region %d resc start is 0!",
+ creq_db->reg.bar_id);
+
+ bar_reg = creq_db->reg.bar_base + reg_offt;
+
+ creq_db->reg.len = BNG_FW_CREQ_DB_LEN;
+ creq_db->reg.bar_reg = ioremap(bar_reg, creq_db->reg.len);
+ if (!creq_db->reg.bar_reg) {
+ dev_err(&pdev->dev,
+ "CREQ BAR region %d mapping failed",
+ creq_db->reg.bar_id);
+ return -ENOMEM;
+ }
+ creq_db->dbinfo.db = creq_db->reg.bar_reg;
+ creq_db->dbinfo.hwq = &rcfw->creq.hwq;
+ creq_db->dbinfo.xid = rcfw->creq.ring_id;
+ return 0;
+}
+
+void bng_re_rcfw_stop_irq(struct bng_re_rcfw *rcfw, bool kill)
+{
+ struct bng_re_creq_ctx *creq;
+
+ creq = &rcfw->creq;
+
+ if (!creq->irq_handler_avail)
+ return;
+
+ creq->irq_handler_avail = false;
+ /* Mask h/w interrupts */
+ bng_re_ring_nq_db(&creq->creq_db.dbinfo, rcfw->res->cctx, false);
+ /* Sync with last running IRQ-handler */
+ synchronize_irq(creq->msix_vec);
+ free_irq(creq->msix_vec, rcfw);
+ kfree(creq->irq_name);
+ creq->irq_name = NULL;
+ atomic_set(&rcfw->rcfw_intr_enabled, 0);
+ if (kill)
+ tasklet_kill(&creq->creq_tasklet);
+ tasklet_disable(&creq->creq_tasklet);
+}
+
+void bng_re_disable_rcfw_channel(struct bng_re_rcfw *rcfw)
+{
+ struct bng_re_creq_ctx *creq;
+ struct bng_re_cmdq_ctx *cmdq;
+
+ creq = &rcfw->creq;
+ cmdq = &rcfw->cmdq;
+ /* Make sure the HW channel is stopped! */
+ bng_re_rcfw_stop_irq(rcfw, true);
+
+ iounmap(cmdq->cmdq_mbox.reg.bar_reg);
+ iounmap(creq->creq_db.reg.bar_reg);
+
+ cmdq->cmdq_mbox.reg.bar_reg = NULL;
+ creq->creq_db.reg.bar_reg = NULL;
+ creq->msix_vec = 0;
+}
+
+static void bng_re_start_rcfw(struct bng_re_rcfw *rcfw)
+{
+ struct bng_re_cmdq_ctx *cmdq;
+ struct bng_re_creq_ctx *creq;
+ struct bng_re_cmdq_mbox *mbox;
+ struct cmdq_init init = {0};
+
+ cmdq = &rcfw->cmdq;
+ creq = &rcfw->creq;
+ mbox = &cmdq->cmdq_mbox;
+
+ init.cmdq_pbl = cpu_to_le64(cmdq->hwq.pbl[BNG_PBL_LVL_0].pg_map_arr[0]);
+ init.cmdq_size_cmdq_lvl =
+ cpu_to_le16(((rcfw->cmdq_depth <<
+ CMDQ_INIT_CMDQ_SIZE_SFT) &
+ CMDQ_INIT_CMDQ_SIZE_MASK) |
+ ((cmdq->hwq.level <<
+ CMDQ_INIT_CMDQ_LVL_SFT) &
+ CMDQ_INIT_CMDQ_LVL_MASK));
+ init.creq_ring_id = cpu_to_le16(creq->ring_id);
+ /* Write to the mailbox register */
+ __iowrite32_copy(mbox->reg.bar_reg, &init, sizeof(init) / 4);
+}
+
+int bng_re_enable_fw_channel(struct bng_re_rcfw *rcfw,
+ int msix_vector,
+ int cp_bar_reg_off)
+{
+ struct bng_re_cmdq_ctx *cmdq;
+ int rc;
+
+ cmdq = &rcfw->cmdq;
+
+ /* Assign defaults */
+ cmdq->seq_num = 0;
+ set_bit(FIRMWARE_FIRST_FLAG, &cmdq->flags);
+ init_waitqueue_head(&cmdq->waitq);
+
+ rc = bng_re_map_cmdq_mbox(rcfw);
+ if (rc)
+ return rc;
+
+ rc = bng_re_map_creq_db(rcfw, cp_bar_reg_off);
+ if (rc)
+ return rc;
+
+ rc = bng_re_rcfw_start_irq(rcfw, msix_vector, true);
+ if (rc) {
+ dev_err(&rcfw->pdev->dev,
+ "Failed to request IRQ for CREQ rc = 0x%x\n", rc);
+ bng_re_disable_rcfw_channel(rcfw);
+ return rc;
+ }
+
+ bng_re_start_rcfw(rcfw);
+ return 0;
+}
+
+int bng_re_deinit_rcfw(struct bng_re_rcfw *rcfw)
+{
+ struct creq_deinitialize_fw_resp resp = {};
+ struct cmdq_deinitialize_fw req = {};
+ struct bng_re_cmdqmsg msg = {};
+ int rc;
+
+ bng_re_rcfw_cmd_prep((struct cmdq_base *)&req,
+ CMDQ_BASE_OPCODE_DEINITIALIZE_FW,
+ sizeof(req));
+ bng_re_fill_cmdqmsg(&msg, &req, &resp, NULL,
+ sizeof(req), sizeof(resp), 0);
+ rc = bng_re_rcfw_send_message(rcfw, &msg);
+ if (rc)
+ return rc;
+
+ clear_bit(FIRMWARE_INITIALIZED_FLAG, &rcfw->cmdq.flags);
+ return 0;
+}
+static inline bool _is_hw_retx_supported(u16 dev_cap_flags)
+{
+ return dev_cap_flags &
+ (CREQ_QUERY_FUNC_RESP_SB_HW_REQUESTER_RETX_ENABLED |
+ CREQ_QUERY_FUNC_RESP_SB_HW_RESPONDER_RETX_ENABLED);
+}
+
+#define BNG_RE_HW_RETX(a) _is_hw_retx_supported((a))
+static inline bool _is_optimize_modify_qp_supported(u16 dev_cap_ext_flags2)
+{
+ return dev_cap_ext_flags2 &
+ CREQ_QUERY_FUNC_RESP_SB_OPTIMIZE_MODIFY_QP_SUPPORTED;
+}
+
+int bng_re_init_rcfw(struct bng_re_rcfw *rcfw,
+ struct bng_re_stats *stats_ctx)
+{
+ struct creq_initialize_fw_resp resp = {};
+ struct cmdq_initialize_fw req = {};
+ struct bng_re_cmdqmsg msg = {};
+ int rc;
+ u16 flags = 0;
+
+ bng_re_rcfw_cmd_prep((struct cmdq_base *)&req,
+ CMDQ_BASE_OPCODE_INITIALIZE_FW,
+ sizeof(req));
+ /* Supply (log-base-2-of-host-page-size - base-page-shift)
+ * to bono to adjust the doorbell page sizes.
+ */
+ req.log2_dbr_pg_size = cpu_to_le16(PAGE_SHIFT -
+ BNG_FW_DBR_BASE_PAGE_SHIFT);
+ if (BNG_RE_HW_RETX(rcfw->res->dattr->dev_cap_flags))
+ flags |= CMDQ_INITIALIZE_FW_FLAGS_HW_REQUESTER_RETX_SUPPORTED;
+ if (_is_optimize_modify_qp_supported(rcfw->res->dattr->dev_cap_flags2))
+ flags |= CMDQ_INITIALIZE_FW_FLAGS_OPTIMIZE_MODIFY_QP_SUPPORTED;
+ req.flags |= cpu_to_le16(flags);
+ req.stat_ctx_id = cpu_to_le32(stats_ctx->fw_id);
+ bng_re_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
+ rc = bng_re_rcfw_send_message(rcfw, &msg);
+ if (rc)
+ return rc;
+ set_bit(FIRMWARE_INITIALIZED_FLAG, &rcfw->cmdq.flags);
+ return 0;
+}
diff --git a/drivers/infiniband/hw/bng_re/bng_fw.h b/drivers/infiniband/hw/bng_re/bng_fw.h
new file mode 100644
index 000000000000..c89c926ec2fc
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_fw.h
@@ -0,0 +1,211 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+// Copyright (c) 2025 Broadcom.
+
+#ifndef __BNG_FW_H__
+#define __BNG_FW_H__
+
+#include "bng_tlv.h"
+
+/* FW DB related */
+#define BNG_FW_CMDQ_TRIG_VAL 1
+#define BNG_FW_COMM_PCI_BAR_REGION 0
+#define BNG_FW_COMM_CONS_PCI_BAR_REGION 2
+#define BNG_FW_DBR_BASE_PAGE_SHIFT 12
+#define BNG_FW_COMM_SIZE 0x104
+#define BNG_FW_COMM_BASE_OFFSET 0x600
+#define BNG_FW_COMM_TRIG_OFFSET 0x100
+#define BNG_FW_PF_VF_COMM_PROD_OFFSET 0xc
+#define BNG_FW_CREQ_DB_LEN 8
+
+/* CREQ */
+#define BNG_FW_CREQE_MAX_CNT (64 * 1024)
+#define BNG_FW_CREQE_UNITS 16
+#define BNG_FW_CREQ_ENTRY_POLL_BUDGET 0x100
+#define BNG_FW_CREQ_CMP_VALID(hdr, pass) \
+ (!!((hdr)->v & CREQ_BASE_V) == \
+ !((pass) & BNG_RE_FLAG_EPOCH_CONS_MASK))
+#define BNG_FW_CREQ_ENTRY_POLL_BUDGET 0x100
+
+/* CMDQ */
+struct bng_fw_cmdqe {
+ u8 data[16];
+};
+
+#define BNG_FW_CMDQE_MAX_CNT 8192
+#define BNG_FW_CMDQE_UNITS sizeof(struct bng_fw_cmdqe)
+#define BNG_FW_CMDQE_BYTES(depth) ((depth) * BNG_FW_CMDQE_UNITS)
+
+#define BNG_FW_MAX_COOKIE_VALUE (BNG_FW_CMDQE_MAX_CNT - 1)
+#define BNG_FW_CMD_IS_BLOCKING 0x8000
+
+/* Crsq buf is 1024-Byte */
+struct bng_re_crsbe {
+ u8 data[1024];
+};
+
+
+static inline u32 bng_fw_cmdqe_npages(u32 depth)
+{
+ u32 npages;
+
+ npages = BNG_FW_CMDQE_BYTES(depth) / PAGE_SIZE;
+ if (BNG_FW_CMDQE_BYTES(depth) % PAGE_SIZE)
+ npages++;
+ return npages;
+}
+
+static inline u32 bng_fw_cmdqe_page_size(u32 depth)
+{
+ return (bng_fw_cmdqe_npages(depth) * PAGE_SIZE);
+}
+struct bng_re_cmdq_mbox {
+ struct bng_re_reg_desc reg;
+ void __iomem *prod;
+ void __iomem *db;
+};
+
+/* HWQ */
+struct bng_re_cmdq_ctx {
+ struct bng_re_hwq hwq;
+ struct bng_re_cmdq_mbox cmdq_mbox;
+ unsigned long flags;
+#define FIRMWARE_INITIALIZED_FLAG (0)
+#define FIRMWARE_STALL_DETECTED (3)
+#define FIRMWARE_FIRST_FLAG (31)
+ wait_queue_head_t waitq;
+ u32 seq_num;
+};
+
+struct bng_re_creq_db {
+ struct bng_re_reg_desc reg;
+ struct bng_re_db_info dbinfo;
+};
+
+struct bng_re_creq_stat {
+ u64 creq_qp_event_processed;
+ u64 creq_func_event_processed;
+};
+
+struct bng_re_creq_ctx {
+ struct bng_re_hwq hwq;
+ struct bng_re_creq_db creq_db;
+ struct bng_re_creq_stat stats;
+ struct tasklet_struct creq_tasklet;
+ u16 ring_id;
+ int msix_vec;
+ bool irq_handler_avail;
+ char *irq_name;
+};
+
+struct bng_re_crsqe {
+ struct creq_qp_event *resp;
+ u32 req_size;
+ /* Free slots at the time of submission */
+ u32 free_slots;
+ u8 opcode;
+ bool is_waiter_alive;
+ bool is_in_used;
+};
+
+struct bng_re_rcfw_sbuf {
+ void *sb;
+ dma_addr_t dma_addr;
+ u32 size;
+};
+
+/* RoCE FW Communication Channels */
+struct bng_re_rcfw {
+ struct pci_dev *pdev;
+ struct bng_re_res *res;
+ struct bng_re_cmdq_ctx cmdq;
+ struct bng_re_creq_ctx creq;
+ struct bng_re_crsqe *crsqe_tbl;
+ /* To synchronize the qp-handle hash table */
+ spinlock_t tbl_lock;
+ u32 cmdq_depth;
+ /* cached from chip cctx for quick reference in slow path */
+ u16 max_timeout;
+ atomic_t rcfw_intr_enabled;
+};
+
+struct bng_re_cmdqmsg {
+ struct cmdq_base *req;
+ struct creq_base *resp;
+ void *sb;
+ u32 req_sz;
+ u32 res_sz;
+ u8 block;
+};
+
+static inline void bng_re_rcfw_cmd_prep(struct cmdq_base *req,
+ u8 opcode, u8 cmd_size)
+{
+ req->opcode = opcode;
+ req->cmd_size = cmd_size;
+}
+
+static inline void bng_re_fill_cmdqmsg(struct bng_re_cmdqmsg *msg,
+ void *req, void *resp, void *sb,
+ u32 req_sz, u32 res_sz, u8 block)
+{
+ msg->req = req;
+ msg->resp = resp;
+ msg->sb = sb;
+ msg->req_sz = req_sz;
+ msg->res_sz = res_sz;
+ msg->block = block;
+}
+
+/* Get the number of command units required for the req. The
+ * function returns correct value only if called before
+ * setting using bng_re_set_cmd_slots
+ */
+static inline u32 bng_re_get_cmd_slots(struct cmdq_base *req)
+{
+ u32 cmd_units = 0;
+
+ if (HAS_TLV_HEADER(req)) {
+ struct roce_tlv *tlv_req = (struct roce_tlv *)req;
+
+ cmd_units = tlv_req->total_size;
+ } else {
+ cmd_units = (req->cmd_size + BNG_FW_CMDQE_UNITS - 1) /
+ BNG_FW_CMDQE_UNITS;
+ }
+
+ return cmd_units;
+}
+
+static inline u32 bng_re_set_cmd_slots(struct cmdq_base *req)
+{
+ u32 cmd_byte = 0;
+
+ if (HAS_TLV_HEADER(req)) {
+ struct roce_tlv *tlv_req = (struct roce_tlv *)req;
+
+ cmd_byte = tlv_req->total_size * BNG_FW_CMDQE_UNITS;
+ } else {
+ cmd_byte = req->cmd_size;
+ req->cmd_size = (req->cmd_size + BNG_FW_CMDQE_UNITS - 1) /
+ BNG_FW_CMDQE_UNITS;
+ }
+
+ return cmd_byte;
+}
+
+void bng_re_free_rcfw_channel(struct bng_re_rcfw *rcfw);
+int bng_re_alloc_fw_channel(struct bng_re_res *res,
+ struct bng_re_rcfw *rcfw);
+int bng_re_enable_fw_channel(struct bng_re_rcfw *rcfw,
+ int msix_vector,
+ int cp_bar_reg_off);
+void bng_re_disable_rcfw_channel(struct bng_re_rcfw *rcfw);
+int bng_re_rcfw_start_irq(struct bng_re_rcfw *rcfw, int msix_vector,
+ bool need_init);
+void bng_re_rcfw_stop_irq(struct bng_re_rcfw *rcfw, bool kill);
+int bng_re_rcfw_send_message(struct bng_re_rcfw *rcfw,
+ struct bng_re_cmdqmsg *msg);
+int bng_re_init_rcfw(struct bng_re_rcfw *rcfw,
+ struct bng_re_stats *stats_ctx);
+int bng_re_deinit_rcfw(struct bng_re_rcfw *rcfw);
+#endif
diff --git a/drivers/infiniband/hw/bng_re/bng_re.h b/drivers/infiniband/hw/bng_re/bng_re.h
new file mode 100644
index 000000000000..dae4862621a7
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_re.h
@@ -0,0 +1,85 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+// Copyright (c) 2025 Broadcom.
+
+#ifndef __BNG_RE_H__
+#define __BNG_RE_H__
+
+#include "bng_res.h"
+
+#define BNG_RE_ADEV_NAME "bng_en"
+
+#define BNG_RE_DESC "Broadcom 800G RoCE Driver"
+
+#define rdev_to_dev(rdev) ((rdev) ? (&(rdev)->ibdev.dev) : NULL)
+
+#define BNG_RE_MIN_MSIX 2
+#define BNG_RE_MAX_MSIX BNGE_MAX_ROCE_MSIX
+
+#define BNG_RE_CREQ_NQ_IDX 0
+
+#define BNGE_INVALID_STATS_CTX_ID -1
+/* NQ specific structures */
+struct bng_re_nq_db {
+ struct bng_re_reg_desc reg;
+ struct bng_re_db_info dbinfo;
+};
+
+struct bng_re_nq {
+ struct pci_dev *pdev;
+ struct bng_re_res *res;
+ char *name;
+ struct bng_re_hwq hwq;
+ struct bng_re_nq_db nq_db;
+ u16 ring_id;
+ int msix_vec;
+ cpumask_t mask;
+ struct tasklet_struct nq_tasklet;
+ bool requested;
+ int budget;
+ u32 load;
+
+ struct workqueue_struct *cqn_wq;
+};
+
+struct bng_re_nq_record {
+ struct bnge_msix_info msix_entries[BNG_RE_MAX_MSIX];
+ struct bng_re_nq nq[BNG_RE_MAX_MSIX];
+ int num_msix;
+ /* serialize NQ access */
+ struct mutex load_lock;
+};
+
+struct bng_re_en_dev_info {
+ struct bng_re_dev *rdev;
+ struct bnge_auxr_dev *auxr_dev;
+};
+
+struct bng_re_ring_attr {
+ dma_addr_t *dma_arr;
+ int pages;
+ int type;
+ u32 depth;
+ u32 lrid; /* Logical ring id */
+ u8 mode;
+};
+
+struct bng_re_dev {
+ struct ib_device ibdev;
+ unsigned long flags;
+#define BNG_RE_FLAG_NETDEV_REGISTERED 0
+#define BNG_RE_FLAG_RCFW_CHANNEL_EN 1
+ struct net_device *netdev;
+ struct auxiliary_device *adev;
+ struct bnge_auxr_dev *aux_dev;
+ struct bng_re_chip_ctx *chip_ctx;
+ int fn_id;
+ struct bng_re_res bng_res;
+ struct bng_re_rcfw rcfw;
+ struct bng_re_nq_record *nqr;
+ /* Device Resources */
+ struct bng_re_dev_attr *dev_attr;
+ struct dentry *dbg_root;
+ struct bng_re_stats stats_ctx;
+};
+
+#endif
diff --git a/drivers/infiniband/hw/bng_re/bng_res.c b/drivers/infiniband/hw/bng_re/bng_res.c
new file mode 100644
index 000000000000..c50823758b53
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_res.c
@@ -0,0 +1,279 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2025 Broadcom.
+
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+#include <rdma/ib_umem.h>
+
+#include <linux/bnxt/hsi.h>
+#include "bng_res.h"
+#include "roce_hsi.h"
+
+/* Stats */
+void bng_re_free_stats_ctx_mem(struct pci_dev *pdev,
+ struct bng_re_stats *stats)
+{
+ if (stats->dma) {
+ dma_free_coherent(&pdev->dev, stats->size,
+ stats->dma, stats->dma_map);
+ }
+ memset(stats, 0, sizeof(*stats));
+ stats->fw_id = -1;
+}
+
+int bng_re_alloc_stats_ctx_mem(struct pci_dev *pdev,
+ struct bng_re_chip_ctx *cctx,
+ struct bng_re_stats *stats)
+{
+ memset(stats, 0, sizeof(*stats));
+ stats->fw_id = -1;
+ stats->size = cctx->hw_stats_size;
+ stats->dma = dma_alloc_coherent(&pdev->dev, stats->size,
+ &stats->dma_map, GFP_KERNEL);
+ if (!stats->dma)
+ return -ENOMEM;
+
+ return 0;
+}
+
+static void bng_free_pbl(struct bng_re_res *res, struct bng_re_pbl *pbl)
+{
+ struct pci_dev *pdev = res->pdev;
+ int i;
+
+ for (i = 0; i < pbl->pg_count; i++) {
+ if (pbl->pg_arr[i])
+ dma_free_coherent(&pdev->dev, pbl->pg_size,
+ (void *)((unsigned long)
+ pbl->pg_arr[i] &
+ PAGE_MASK),
+ pbl->pg_map_arr[i]);
+ else
+ dev_warn(&pdev->dev,
+ "PBL free pg_arr[%d] empty?!\n", i);
+ pbl->pg_arr[i] = NULL;
+ }
+
+ vfree(pbl->pg_arr);
+ pbl->pg_arr = NULL;
+ vfree(pbl->pg_map_arr);
+ pbl->pg_map_arr = NULL;
+ pbl->pg_count = 0;
+ pbl->pg_size = 0;
+}
+
+static int bng_alloc_pbl(struct bng_re_res *res,
+ struct bng_re_pbl *pbl,
+ struct bng_re_sg_info *sginfo)
+{
+ struct pci_dev *pdev = res->pdev;
+ u32 pages;
+ int i;
+
+ if (sginfo->nopte)
+ return 0;
+ pages = sginfo->npages;
+
+ /* page ptr arrays */
+ pbl->pg_arr = vmalloc_array(pages, sizeof(void *));
+ if (!pbl->pg_arr)
+ return -ENOMEM;
+
+ pbl->pg_map_arr = vmalloc_array(pages, sizeof(dma_addr_t));
+ if (!pbl->pg_map_arr) {
+ vfree(pbl->pg_arr);
+ pbl->pg_arr = NULL;
+ return -ENOMEM;
+ }
+ pbl->pg_count = 0;
+ pbl->pg_size = sginfo->pgsize;
+
+ for (i = 0; i < pages; i++) {
+ pbl->pg_arr[i] = dma_alloc_coherent(&pdev->dev,
+ pbl->pg_size,
+ &pbl->pg_map_arr[i],
+ GFP_KERNEL);
+ if (!pbl->pg_arr[i])
+ goto fail;
+ pbl->pg_count++;
+ }
+
+ return 0;
+fail:
+ bng_free_pbl(res, pbl);
+ return -ENOMEM;
+}
+
+void bng_re_free_hwq(struct bng_re_res *res,
+ struct bng_re_hwq *hwq)
+{
+ int i;
+
+ if (!hwq->max_elements)
+ return;
+ if (hwq->level >= BNG_PBL_LVL_MAX)
+ return;
+
+ for (i = 0; i < hwq->level + 1; i++)
+ bng_free_pbl(res, &hwq->pbl[i]);
+
+ hwq->level = BNG_PBL_LVL_MAX;
+ hwq->max_elements = 0;
+ hwq->element_size = 0;
+ hwq->prod = 0;
+ hwq->cons = 0;
+}
+
+/* All HWQs are power of 2 in size */
+int bng_re_alloc_init_hwq(struct bng_re_hwq *hwq,
+ struct bng_re_hwq_attr *hwq_attr)
+{
+ u32 npages, pg_size;
+ struct bng_re_sg_info sginfo = {};
+ u32 depth, stride, npbl, npde;
+ dma_addr_t *src_phys_ptr, **dst_virt_ptr;
+ struct bng_re_res *res;
+ struct pci_dev *pdev;
+ int i, rc, lvl;
+
+ res = hwq_attr->res;
+ pdev = res->pdev;
+ pg_size = hwq_attr->sginfo->pgsize;
+ hwq->level = BNG_PBL_LVL_MAX;
+
+ depth = roundup_pow_of_two(hwq_attr->depth);
+ stride = roundup_pow_of_two(hwq_attr->stride);
+
+ npages = (depth * stride) / pg_size;
+ if ((depth * stride) % pg_size)
+ npages++;
+ if (!npages)
+ return -EINVAL;
+ hwq_attr->sginfo->npages = npages;
+
+ if (npages == MAX_PBL_LVL_0_PGS && !hwq_attr->sginfo->nopte) {
+ /* This request is Level 0, map PTE */
+ rc = bng_alloc_pbl(res, &hwq->pbl[BNG_PBL_LVL_0], hwq_attr->sginfo);
+ if (rc)
+ goto fail;
+ hwq->level = BNG_PBL_LVL_0;
+ goto done;
+ }
+
+ if (npages >= MAX_PBL_LVL_0_PGS) {
+ if (npages > MAX_PBL_LVL_1_PGS) {
+ u32 flag = PTU_PTE_VALID;
+ /* 2 levels of indirection */
+ npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
+ if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
+ npbl++;
+ npde = npbl >> MAX_PDL_LVL_SHIFT;
+ if (npbl % BIT(MAX_PDL_LVL_SHIFT))
+ npde++;
+ /* Alloc PDE pages */
+ sginfo.pgsize = npde * pg_size;
+ sginfo.npages = 1;
+ rc = bng_alloc_pbl(res, &hwq->pbl[BNG_PBL_LVL_0], &sginfo);
+ if (rc)
+ goto fail;
+
+ /* Alloc PBL pages */
+ sginfo.npages = npbl;
+ sginfo.pgsize = PAGE_SIZE;
+ rc = bng_alloc_pbl(res, &hwq->pbl[BNG_PBL_LVL_1], &sginfo);
+ if (rc)
+ goto fail;
+ /* Fill PDL with PBL page pointers */
+ dst_virt_ptr =
+ (dma_addr_t **)hwq->pbl[BNG_PBL_LVL_0].pg_arr;
+ src_phys_ptr = hwq->pbl[BNG_PBL_LVL_1].pg_map_arr;
+ for (i = 0; i < hwq->pbl[BNG_PBL_LVL_1].pg_count; i++)
+ dst_virt_ptr[0][i] = src_phys_ptr[i] | flag;
+
+ /* Alloc or init PTEs */
+ rc = bng_alloc_pbl(res, &hwq->pbl[BNG_PBL_LVL_2],
+ hwq_attr->sginfo);
+ if (rc)
+ goto fail;
+ hwq->level = BNG_PBL_LVL_2;
+ if (hwq_attr->sginfo->nopte)
+ goto done;
+ /* Fill PBLs with PTE pointers */
+ dst_virt_ptr =
+ (dma_addr_t **)hwq->pbl[BNG_PBL_LVL_1].pg_arr;
+ src_phys_ptr = hwq->pbl[BNG_PBL_LVL_2].pg_map_arr;
+ for (i = 0; i < hwq->pbl[BNG_PBL_LVL_2].pg_count; i++) {
+ dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
+ src_phys_ptr[i] | PTU_PTE_VALID;
+ }
+ if (hwq_attr->type == BNG_HWQ_TYPE_QUEUE) {
+ /* Find the last pg of the size */
+ i = hwq->pbl[BNG_PBL_LVL_2].pg_count;
+ dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
+ PTU_PTE_LAST;
+ if (i > 1)
+ dst_virt_ptr[PTR_PG(i - 2)]
+ [PTR_IDX(i - 2)] |=
+ PTU_PTE_NEXT_TO_LAST;
+ }
+ } else { /* pages < 512 npbl = 1, npde = 0 */
+ u32 flag = PTU_PTE_VALID;
+
+ /* 1 level of indirection */
+ npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
+ if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
+ npbl++;
+ sginfo.npages = npbl;
+ sginfo.pgsize = PAGE_SIZE;
+ /* Alloc PBL page */
+ rc = bng_alloc_pbl(res, &hwq->pbl[BNG_PBL_LVL_0], &sginfo);
+ if (rc)
+ goto fail;
+ /* Alloc or init PTEs */
+ rc = bng_alloc_pbl(res, &hwq->pbl[BNG_PBL_LVL_1],
+ hwq_attr->sginfo);
+ if (rc)
+ goto fail;
+ hwq->level = BNG_PBL_LVL_1;
+ if (hwq_attr->sginfo->nopte)
+ goto done;
+ /* Fill PBL with PTE pointers */
+ dst_virt_ptr =
+ (dma_addr_t **)hwq->pbl[BNG_PBL_LVL_0].pg_arr;
+ src_phys_ptr = hwq->pbl[BNG_PBL_LVL_1].pg_map_arr;
+ for (i = 0; i < hwq->pbl[BNG_PBL_LVL_1].pg_count; i++)
+ dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
+ src_phys_ptr[i] | flag;
+ if (hwq_attr->type == BNG_HWQ_TYPE_QUEUE) {
+ /* Find the last pg of the size */
+ i = hwq->pbl[BNG_PBL_LVL_1].pg_count;
+ dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
+ PTU_PTE_LAST;
+ if (i > 1)
+ dst_virt_ptr[PTR_PG(i - 2)]
+ [PTR_IDX(i - 2)] |=
+ PTU_PTE_NEXT_TO_LAST;
+ }
+ }
+ }
+done:
+ hwq->prod = 0;
+ hwq->cons = 0;
+ hwq->pdev = pdev;
+ hwq->depth = hwq_attr->depth;
+ hwq->max_elements = hwq->depth;
+ hwq->element_size = stride;
+ hwq->qe_ppg = pg_size / stride;
+ /* For direct access to the elements */
+ lvl = hwq->level;
+ if (hwq_attr->sginfo->nopte && hwq->level)
+ lvl = hwq->level - 1;
+ hwq->pbl_ptr = hwq->pbl[lvl].pg_arr;
+ hwq->pbl_dma_ptr = hwq->pbl[lvl].pg_map_arr;
+ spin_lock_init(&hwq->lock);
+
+ return 0;
+fail:
+ bng_re_free_hwq(res, hwq);
+ return -ENOMEM;
+}
diff --git a/drivers/infiniband/hw/bng_re/bng_res.h b/drivers/infiniband/hw/bng_re/bng_res.h
new file mode 100644
index 000000000000..9997f86d6a0e
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_res.h
@@ -0,0 +1,215 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+// Copyright (c) 2025 Broadcom.
+
+#ifndef __BNG_RES_H__
+#define __BNG_RES_H__
+
+#include "roce_hsi.h"
+
+#define BNG_ROCE_FW_MAX_TIMEOUT 60
+
+#define PTR_CNT_PER_PG (PAGE_SIZE / sizeof(void *))
+#define PTR_MAX_IDX_PER_PG (PTR_CNT_PER_PG - 1)
+#define PTR_PG(x) (((x) & ~PTR_MAX_IDX_PER_PG) / PTR_CNT_PER_PG)
+#define PTR_IDX(x) ((x) & PTR_MAX_IDX_PER_PG)
+
+#define HWQ_CMP(idx, hwq) ((idx) & ((hwq)->max_elements - 1))
+#define HWQ_FREE_SLOTS(hwq) (hwq->max_elements - \
+ ((HWQ_CMP(hwq->prod, hwq)\
+ - HWQ_CMP(hwq->cons, hwq))\
+ & (hwq->max_elements - 1)))
+
+#define MAX_PBL_LVL_0_PGS 1
+#define MAX_PBL_LVL_1_PGS 512
+#define MAX_PBL_LVL_1_PGS_SHIFT 9
+#define MAX_PBL_LVL_1_PGS_FOR_LVL_2 256
+#define MAX_PBL_LVL_2_PGS (256 * 512)
+#define MAX_PDL_LVL_SHIFT 9
+
+#define BNG_RE_DBR_VALID (0x1UL << 26)
+#define BNG_RE_DBR_EPOCH_SHIFT 24
+#define BNG_RE_DBR_TOGGLE_SHIFT 25
+
+#define BNG_MAX_TQM_ALLOC_REQ 48
+
+struct bng_re_reg_desc {
+ u8 bar_id;
+ resource_size_t bar_base;
+ unsigned long offset;
+ void __iomem *bar_reg;
+ size_t len;
+};
+
+struct bng_re_db_info {
+ void __iomem *db;
+ void __iomem *priv_db;
+ struct bng_re_hwq *hwq;
+ u32 xid;
+ u32 max_slot;
+ u32 flags;
+ u8 toggle;
+};
+
+enum bng_re_db_info_flags_mask {
+ BNG_RE_FLAG_EPOCH_CONS_SHIFT = 0x0UL,
+ BNG_RE_FLAG_EPOCH_PROD_SHIFT = 0x1UL,
+ BNG_RE_FLAG_EPOCH_CONS_MASK = 0x1UL,
+ BNG_RE_FLAG_EPOCH_PROD_MASK = 0x2UL,
+};
+
+enum bng_re_db_epoch_flag_shift {
+ BNG_RE_DB_EPOCH_CONS_SHIFT = BNG_RE_DBR_EPOCH_SHIFT,
+ BNG_RE_DB_EPOCH_PROD_SHIFT = (BNG_RE_DBR_EPOCH_SHIFT - 1),
+};
+
+struct bng_re_chip_ctx {
+ u16 chip_num;
+ u16 hw_stats_size;
+ u64 hwrm_intf_ver;
+ u16 hwrm_cmd_max_timeout;
+};
+
+struct bng_re_pbl {
+ u32 pg_count;
+ u32 pg_size;
+ void **pg_arr;
+ dma_addr_t *pg_map_arr;
+};
+
+enum bng_re_pbl_lvl {
+ BNG_PBL_LVL_0,
+ BNG_PBL_LVL_1,
+ BNG_PBL_LVL_2,
+ BNG_PBL_LVL_MAX
+};
+
+enum bng_re_hwq_type {
+ BNG_HWQ_TYPE_CTX,
+ BNG_HWQ_TYPE_QUEUE
+};
+
+struct bng_re_sg_info {
+ u32 npages;
+ u32 pgshft;
+ u32 pgsize;
+ bool nopte;
+};
+
+struct bng_re_hwq_attr {
+ struct bng_re_res *res;
+ struct bng_re_sg_info *sginfo;
+ enum bng_re_hwq_type type;
+ u32 depth;
+ u32 stride;
+ u32 aux_stride;
+ u32 aux_depth;
+};
+
+struct bng_re_hwq {
+ struct pci_dev *pdev;
+ /* lock to protect hwq */
+ spinlock_t lock;
+ struct bng_re_pbl pbl[BNG_PBL_LVL_MAX + 1];
+ /* Valid values: 0, 1, 2 */
+ enum bng_re_pbl_lvl level;
+ /* PBL entries */
+ void **pbl_ptr;
+ /* PBL dma_addr */
+ dma_addr_t *pbl_dma_ptr;
+ u32 max_elements;
+ u32 depth;
+ u16 element_size;
+ u32 prod;
+ u32 cons;
+ /* queue entry per page */
+ u16 qe_ppg;
+};
+
+struct bng_re_stats {
+ dma_addr_t dma_map;
+ void *dma;
+ u32 size;
+ u32 fw_id;
+};
+
+struct bng_re_res {
+ struct pci_dev *pdev;
+ struct bng_re_chip_ctx *cctx;
+ struct bng_re_dev_attr *dattr;
+};
+
+static inline void *bng_re_get_qe(struct bng_re_hwq *hwq,
+ u32 indx, u64 *pg)
+{
+ u32 pg_num, pg_idx;
+
+ pg_num = (indx / hwq->qe_ppg);
+ pg_idx = (indx % hwq->qe_ppg);
+ if (pg)
+ *pg = (u64)&hwq->pbl_ptr[pg_num];
+ return (void *)(hwq->pbl_ptr[pg_num] + hwq->element_size * pg_idx);
+}
+
+#define BNG_RE_INIT_DBHDR(xid, type, indx, toggle) \
+ (((u64)(((xid) & DBC_DBC_XID_MASK) | DBC_DBC_PATH_ROCE | \
+ (type) | BNG_RE_DBR_VALID) << 32) | (indx) | \
+ (((u32)(toggle)) << (BNG_RE_DBR_TOGGLE_SHIFT)))
+
+static inline void bng_re_ring_db(struct bng_re_db_info *info,
+ u32 type)
+{
+ u64 key = 0;
+ u32 indx;
+ u8 toggle = 0;
+
+ if (type == DBC_DBC_TYPE_CQ_ARMALL ||
+ type == DBC_DBC_TYPE_CQ_ARMSE)
+ toggle = info->toggle;
+
+ indx = (info->hwq->cons & DBC_DBC_INDEX_MASK) |
+ ((info->flags & BNG_RE_FLAG_EPOCH_CONS_MASK) <<
+ BNG_RE_DB_EPOCH_CONS_SHIFT);
+
+ key = BNG_RE_INIT_DBHDR(info->xid, type, indx, toggle);
+ writeq(key, info->db);
+}
+
+static inline void bng_re_ring_nq_db(struct bng_re_db_info *info,
+ struct bng_re_chip_ctx *cctx,
+ bool arm)
+{
+ u32 type;
+
+ type = arm ? DBC_DBC_TYPE_NQ_ARM : DBC_DBC_TYPE_NQ;
+ bng_re_ring_db(info, type);
+}
+
+static inline void bng_re_hwq_incr_cons(u32 max_elements, u32 *cons, u32 cnt,
+ u32 *dbinfo_flags)
+{
+ /* move cons and update toggle/epoch if wrap around */
+ *cons += cnt;
+ if (*cons >= max_elements) {
+ *cons %= max_elements;
+ *dbinfo_flags ^= 1UL << BNG_RE_FLAG_EPOCH_CONS_SHIFT;
+ }
+}
+
+static inline bool _is_max_srq_ext_supported(u16 dev_cap_ext_flags_2)
+{
+ return !!(dev_cap_ext_flags_2 & CREQ_QUERY_FUNC_RESP_SB_MAX_SRQ_EXTENDED);
+}
+
+void bng_re_free_hwq(struct bng_re_res *res,
+ struct bng_re_hwq *hwq);
+
+int bng_re_alloc_init_hwq(struct bng_re_hwq *hwq,
+ struct bng_re_hwq_attr *hwq_attr);
+
+void bng_re_free_stats_ctx_mem(struct pci_dev *pdev,
+ struct bng_re_stats *stats);
+
+int bng_re_alloc_stats_ctx_mem(struct pci_dev *pdev,
+ struct bng_re_chip_ctx *cctx,
+ struct bng_re_stats *stats);
+#endif
diff --git a/drivers/infiniband/hw/bng_re/bng_sp.c b/drivers/infiniband/hw/bng_re/bng_sp.c
new file mode 100644
index 000000000000..83099e05328d
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_sp.c
@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2025 Broadcom.
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+
+#include "bng_res.h"
+#include "bng_fw.h"
+#include "bng_sp.h"
+#include "bng_tlv.h"
+
+static bool bng_re_is_atomic_cap(struct bng_re_rcfw *rcfw)
+{
+ u16 pcie_ctl2 = 0;
+
+ pcie_capability_read_word(rcfw->pdev, PCI_EXP_DEVCTL2, &pcie_ctl2);
+ return (pcie_ctl2 & PCI_EXP_DEVCTL2_ATOMIC_REQ);
+}
+
+static void bng_re_query_version(struct bng_re_rcfw *rcfw,
+ char *fw_ver)
+{
+ struct creq_query_version_resp resp = {};
+ struct bng_re_cmdqmsg msg = {};
+ struct cmdq_query_version req = {};
+ int rc;
+
+ bng_re_rcfw_cmd_prep((struct cmdq_base *)&req,
+ CMDQ_BASE_OPCODE_QUERY_VERSION,
+ sizeof(req));
+
+ bng_re_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
+ rc = bng_re_rcfw_send_message(rcfw, &msg);
+ if (rc)
+ return;
+ fw_ver[0] = resp.fw_maj;
+ fw_ver[1] = resp.fw_minor;
+ fw_ver[2] = resp.fw_bld;
+ fw_ver[3] = resp.fw_rsvd;
+}
+
+int bng_re_get_dev_attr(struct bng_re_rcfw *rcfw)
+{
+ struct bng_re_dev_attr *attr = rcfw->res->dattr;
+ struct creq_query_func_resp resp = {};
+ struct bng_re_cmdqmsg msg = {};
+ struct creq_query_func_resp_sb *sb;
+ struct bng_re_rcfw_sbuf sbuf;
+ struct cmdq_query_func req = {};
+ u8 *tqm_alloc;
+ int i, rc;
+ u32 temp;
+
+ bng_re_rcfw_cmd_prep((struct cmdq_base *)&req,
+ CMDQ_BASE_OPCODE_QUERY_FUNC,
+ sizeof(req));
+
+ sbuf.size = ALIGN(sizeof(*sb), BNG_FW_CMDQE_UNITS);
+ sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
+ &sbuf.dma_addr, GFP_KERNEL);
+ if (!sbuf.sb)
+ return -ENOMEM;
+ sb = sbuf.sb;
+ req.resp_size = sbuf.size / BNG_FW_CMDQE_UNITS;
+ bng_re_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
+ sizeof(resp), 0);
+ rc = bng_re_rcfw_send_message(rcfw, &msg);
+ if (rc)
+ goto bail;
+ /* Extract the context from the side buffer */
+ attr->max_qp = le32_to_cpu(sb->max_qp);
+ /* max_qp value reported by FW doesn't include the QP1 */
+ attr->max_qp += 1;
+ attr->max_qp_rd_atom =
+ sb->max_qp_rd_atom > BNG_RE_MAX_OUT_RD_ATOM ?
+ BNG_RE_MAX_OUT_RD_ATOM : sb->max_qp_rd_atom;
+ attr->max_qp_init_rd_atom =
+ sb->max_qp_init_rd_atom > BNG_RE_MAX_OUT_RD_ATOM ?
+ BNG_RE_MAX_OUT_RD_ATOM : sb->max_qp_init_rd_atom;
+ attr->max_qp_wqes = le16_to_cpu(sb->max_qp_wr) - 1;
+
+ /* Adjust for max_qp_wqes for variable wqe */
+ attr->max_qp_wqes = min_t(u32, attr->max_qp_wqes, BNG_VAR_MAX_WQE - 1);
+
+ attr->max_qp_sges = min_t(u32, sb->max_sge_var_wqe, BNG_VAR_MAX_SGE);
+ attr->max_cq = le32_to_cpu(sb->max_cq);
+ attr->max_cq_wqes = le32_to_cpu(sb->max_cqe);
+ attr->max_cq_sges = attr->max_qp_sges;
+ attr->max_mr = le32_to_cpu(sb->max_mr);
+ attr->max_mw = le32_to_cpu(sb->max_mw);
+
+ attr->max_mr_size = le64_to_cpu(sb->max_mr_size);
+ attr->max_pd = 64 * 1024;
+ attr->max_raw_ethy_qp = le32_to_cpu(sb->max_raw_eth_qp);
+ attr->max_ah = le32_to_cpu(sb->max_ah);
+
+ attr->max_srq = le16_to_cpu(sb->max_srq);
+ attr->max_srq_wqes = le32_to_cpu(sb->max_srq_wr) - 1;
+ attr->max_srq_sges = sb->max_srq_sge;
+ attr->max_pkey = 1;
+ attr->max_inline_data = le32_to_cpu(sb->max_inline_data);
+ /*
+ * Read the max gid supported by HW.
+ * For each entry in HW GID in HW table, we consume 2
+ * GID entries in the kernel GID table. So max_gid reported
+ * to stack can be up to twice the value reported by the HW, up to 256 gids.
+ */
+ attr->max_sgid = le32_to_cpu(sb->max_gid);
+ attr->max_sgid = min_t(u32, BNG_RE_NUM_GIDS_SUPPORTED, 2 * attr->max_sgid);
+ attr->dev_cap_flags = le16_to_cpu(sb->dev_cap_flags);
+ attr->dev_cap_flags2 = le16_to_cpu(sb->dev_cap_ext_flags_2);
+
+ if (_is_max_srq_ext_supported(attr->dev_cap_flags2))
+ attr->max_srq += le16_to_cpu(sb->max_srq_ext);
+
+ bng_re_query_version(rcfw, attr->fw_ver);
+ for (i = 0; i < BNG_MAX_TQM_ALLOC_REQ / 4; i++) {
+ temp = le32_to_cpu(sb->tqm_alloc_reqs[i]);
+ tqm_alloc = (u8 *)&temp;
+ attr->tqm_alloc_reqs[i * 4] = *tqm_alloc;
+ attr->tqm_alloc_reqs[i * 4 + 1] = *(++tqm_alloc);
+ attr->tqm_alloc_reqs[i * 4 + 2] = *(++tqm_alloc);
+ attr->tqm_alloc_reqs[i * 4 + 3] = *(++tqm_alloc);
+ }
+
+ attr->max_dpi = le32_to_cpu(sb->max_dpi);
+ attr->is_atomic = bng_re_is_atomic_cap(rcfw);
+bail:
+ dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
+ sbuf.sb, sbuf.dma_addr);
+ return rc;
+}
diff --git a/drivers/infiniband/hw/bng_re/bng_sp.h b/drivers/infiniband/hw/bng_re/bng_sp.h
new file mode 100644
index 000000000000..e15190515ed1
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_sp.h
@@ -0,0 +1,47 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+// Copyright (c) 2025 Broadcom.
+
+#ifndef __BNG_SP_H__
+#define __BNG_SP_H__
+
+#include "bng_fw.h"
+
+#define BNG_VAR_MAX_WQE 4352
+#define BNG_VAR_MAX_SGE 13
+
+struct bng_re_dev_attr {
+#define FW_VER_ARR_LEN 4
+ u8 fw_ver[FW_VER_ARR_LEN];
+#define BNG_RE_NUM_GIDS_SUPPORTED 256
+ u16 max_sgid;
+ u16 max_mrw;
+ u32 max_qp;
+#define BNG_RE_MAX_OUT_RD_ATOM 126
+ u32 max_qp_rd_atom;
+ u32 max_qp_init_rd_atom;
+ u32 max_qp_wqes;
+ u32 max_qp_sges;
+ u32 max_cq;
+ u32 max_cq_wqes;
+ u32 max_cq_sges;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_pd;
+ u32 max_mw;
+ u32 max_raw_ethy_qp;
+ u32 max_ah;
+ u32 max_srq;
+ u32 max_srq_wqes;
+ u32 max_srq_sges;
+ u32 max_pkey;
+ u32 max_inline_data;
+ u32 l2_db_size;
+ u8 tqm_alloc_reqs[BNG_MAX_TQM_ALLOC_REQ];
+ bool is_atomic;
+ u16 dev_cap_flags;
+ u16 dev_cap_flags2;
+ u32 max_dpi;
+};
+
+int bng_re_get_dev_attr(struct bng_re_rcfw *rcfw);
+#endif
diff --git a/drivers/infiniband/hw/bng_re/bng_tlv.h b/drivers/infiniband/hw/bng_re/bng_tlv.h
new file mode 100644
index 000000000000..278f4922962d
--- /dev/null
+++ b/drivers/infiniband/hw/bng_re/bng_tlv.h
@@ -0,0 +1,128 @@
+/* SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause */
+
+#ifndef __BNG_TLV_H__
+#define __BNG_TLV_H__
+
+#include "roce_hsi.h"
+
+struct roce_tlv {
+ struct tlv tlv;
+ u8 total_size; // in units of 16 byte chunks
+ u8 unused[7]; // for 16 byte alignment
+};
+
+/*
+ * TLV size in units of 16 byte chunks
+ */
+#define TLV_SIZE ((sizeof(struct roce_tlv) + 15) / 16)
+/*
+ * TLV length in bytes
+ */
+#define TLV_BYTES (TLV_SIZE * 16)
+
+#define HAS_TLV_HEADER(msg) (le16_to_cpu(((struct tlv *)(msg))->cmd_discr) == CMD_DISCR_TLV_ENCAP)
+#define GET_TLV_DATA(tlv) ((void *)&((uint8_t *)(tlv))[TLV_BYTES])
+
+static inline u8 __get_cmdq_base_opcode(struct cmdq_base *req, u32 size)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ return ((struct cmdq_base *)GET_TLV_DATA(req))->opcode;
+ else
+ return req->opcode;
+}
+
+static inline void __set_cmdq_base_opcode(struct cmdq_base *req,
+ u32 size, u8 val)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ ((struct cmdq_base *)GET_TLV_DATA(req))->opcode = val;
+ else
+ req->opcode = val;
+}
+
+static inline __le16 __get_cmdq_base_cookie(struct cmdq_base *req, u32 size)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ return ((struct cmdq_base *)GET_TLV_DATA(req))->cookie;
+ else
+ return req->cookie;
+}
+
+static inline void __set_cmdq_base_cookie(struct cmdq_base *req,
+ u32 size, __le16 val)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ ((struct cmdq_base *)GET_TLV_DATA(req))->cookie = val;
+ else
+ req->cookie = val;
+}
+
+static inline __le64 __get_cmdq_base_resp_addr(struct cmdq_base *req, u32 size)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ return ((struct cmdq_base *)GET_TLV_DATA(req))->resp_addr;
+ else
+ return req->resp_addr;
+}
+
+static inline void __set_cmdq_base_resp_addr(struct cmdq_base *req,
+ u32 size, __le64 val)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ ((struct cmdq_base *)GET_TLV_DATA(req))->resp_addr = val;
+ else
+ req->resp_addr = val;
+}
+
+static inline u8 __get_cmdq_base_resp_size(struct cmdq_base *req, u32 size)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ return ((struct cmdq_base *)GET_TLV_DATA(req))->resp_size;
+ else
+ return req->resp_size;
+}
+
+static inline void __set_cmdq_base_resp_size(struct cmdq_base *req,
+ u32 size, u8 val)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ ((struct cmdq_base *)GET_TLV_DATA(req))->resp_size = val;
+ else
+ req->resp_size = val;
+}
+
+static inline u8 __get_cmdq_base_cmd_size(struct cmdq_base *req, u32 size)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ return ((struct roce_tlv *)(req))->total_size;
+ else
+ return req->cmd_size;
+}
+
+static inline void __set_cmdq_base_cmd_size(struct cmdq_base *req,
+ u32 size, u8 val)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ ((struct cmdq_base *)GET_TLV_DATA(req))->cmd_size = val;
+ else
+ req->cmd_size = val;
+}
+
+static inline __le16 __get_cmdq_base_flags(struct cmdq_base *req, u32 size)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ return ((struct cmdq_base *)GET_TLV_DATA(req))->flags;
+ else
+ return req->flags;
+}
+
+static inline void __set_cmdq_base_flags(struct cmdq_base *req,
+ u32 size, __le16 val)
+{
+ if (HAS_TLV_HEADER(req) && size > TLV_BYTES)
+ ((struct cmdq_base *)GET_TLV_DATA(req))->flags = val;
+ else
+ req->flags = val;
+}
+
+#endif /* __BNG_TLV_H__ */
generated by cgit v1.2.3 (git 2.46.0) at 2025-12-05 12:46:17 +0000