[PATCH for-next v3 0/7] RDMA/bnxt_re: Doorbell Drop Prevention

[PATCH for-next v3 0/7] RDMA/bnxt_re: Doorbell Drop Prevention

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 1940 bytes --]

The idea behind this series is to prevent Doorbell drops
on some of the Broadcom adapters that require Doorbell
moderation. This is achieved by pacing the doorbell writes
into the hardware FIFO. The rate at which individual doorbells
are written needs to be dynamically adjusted, because
it depends on the ability of the hardware to drain the
FIFO and on the number and behavior of individual
doorbell writers. When congestion is detected by the user
library, it notifies the driver and driver adjust the
pacing parameters dynamically in a shared page, which will
be used for pacing the Doorbells.

Currently this feature is targeted only for user applications.
The corresponding user lib patch is in the pull request.
https://github.com/linux-rdma/rdma-core/pull/1360

Thanks,
Selvin Xavier

v2 -> v3:
     Fix the build warning
     Reported-by: kernel test robot <lkp@intel.com>

v1 -> v2:
     Rebased the patches on top of the latest for-next branch

Chandramohan Akula (7):
  bnxt_en: Update HW interface headers
  bnxt_en: Share the bar0 address with the RoCE driver
  RDMA/bnxt_re: Initialize Doorbell pacing feature
  RDMA/bnxt_re: Enable pacing support for the user apps
  RDMA/bnxt_re: Update alloc_page uapi for pacing
  RDMA/bnxt_re: Implement doorbell pacing algorithm
  RDMA/bnxt_re: Add a new uapi for driver notification

 drivers/infiniband/hw/bnxt_re/bnxt_re.h       |  27 ++++
 drivers/infiniband/hw/bnxt_re/ib_verbs.c      |  52 +++++-
 drivers/infiniband/hw/bnxt_re/ib_verbs.h      |   2 +
 drivers/infiniband/hw/bnxt_re/main.c          | 220 ++++++++++++++++++++++++++
 drivers/infiniband/hw/bnxt_re/qplib_res.h     |  19 +++
 drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h |  54 +++++++
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c |   2 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h |   1 +
 include/uapi/rdma/bnxt_re-abi.h               |   7 +
 9 files changed, 380 insertions(+), 4 deletions(-)

-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 1/7] bnxt_en: Update HW interface headers

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 3567 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

Updating the HW structures for the doorbell pacing related
information. Newly added interface structures will be used in
the followup patches.

CC: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h | 54 +++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h
index b31de4c..a2d3a80 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h
@@ -3721,6 +3721,60 @@ struct hwrm_func_backing_store_qcaps_v2_output {
 	u8	valid;
 };
 
+/* hwrm_func_dbr_pacing_qcfg_input (size:128b/16B) */
+struct hwrm_func_dbr_pacing_qcfg_input {
+	__le16  req_type;
+	__le16  cmpl_ring;
+	__le16  seq_id;
+	__le16  target_id;
+	__le64  resp_addr;
+};
+
+/* hwrm_func_dbr_pacing_qcfg_output (size:512b/64B) */
+struct hwrm_func_dbr_pacing_qcfg_output {
+	__le16  error_code;
+	__le16  req_type;
+	__le16  seq_id;
+	__le16  resp_len;
+	u8      flags;
+#define FUNC_DBR_PACING_QCFG_RESP_FLAGS_DBR_NQ_EVENT_ENABLED     0x1UL
+	u8      unused_0[7];
+	__le32  dbr_stat_db_fifo_reg;
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK    0x3UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_SFT     0
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_PCIE_CFG  0x0UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_GRC       0x1UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_BAR0      0x2UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_BAR1      0x3UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_LAST     \
+		FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_BAR1
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_MASK          0xfffffffcUL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SFT           2
+	__le32  dbr_stat_db_fifo_reg_watermark_mask;
+	u8      dbr_stat_db_fifo_reg_watermark_shift;
+	u8      unused_1[3];
+	__le32  dbr_stat_db_fifo_reg_fifo_room_mask;
+	u8      dbr_stat_db_fifo_reg_fifo_room_shift;
+	u8      unused_2[3];
+	__le32  dbr_throttling_aeq_arm_reg;
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_MASK    0x3UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_SFT     0
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_PCIE_CFG  0x0UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_GRC       0x1UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_BAR0      0x2UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_BAR1      0x3UL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_LAST	\
+		FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SPACE_BAR1
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_MASK          0xfffffffcUL
+#define FUNC_DBR_PACING_QCFG_RESP_DBR_THROTTLING_AEQ_ARM_REG_ADDR_SFT           2
+	u8      dbr_throttling_aeq_arm_reg_val;
+	u8      unused_3[7];
+	__le32  primary_nq_id;
+	__le32  pacing_threshold;
+	u8      unused_4[7];
+	u8      valid;
+};
+
 /* hwrm_func_drv_if_change_input (size:192b/24B) */
 struct hwrm_func_drv_if_change_input {
 	__le16	req_type;
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 2/7] bnxt_en: Share the bar0 address with the RoCE driver

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 1506 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

Add a parameter in the bnxt_en_dev structure to share the bar0 address
with RoCE driver.

CC: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c | 2 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
index 852eb44..6ba2b93 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
@@ -345,7 +345,7 @@ static void bnxt_set_edev_info(struct bnxt_en_dev *edev, struct bnxt *bp)
 	edev->hw_ring_stats_size = bp->hw_ring_stats_size;
 	edev->pf_port_id = bp->pf.port_id;
 	edev->en_state = bp->state;
-
+	edev->bar0 = bp->bar0;
 	edev->ulp_tbl->msix_requested = bnxt_get_ulp_msix_num(bp);
 }
 
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h
index 80cbc4b..6ff77f0 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h
@@ -81,6 +81,7 @@ struct bnxt_en_dev {
 							 * mode only. Will be
 							 * updated in resume.
 							 */
+	void __iomem                    *bar0;
 };
 
 static inline bool bnxt_ulp_registered(struct bnxt_en_dev *edev)
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 3/7] RDMA/bnxt_re: Initialize Doorbell pacing feature

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 7923 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

Checks for pacing feature capability and get the doorbell pacing
configuration using FW commands. Allocate a page and initialize
the pacing parameters for the applications. Cleanup the page and
de-initialize the pacing during device removal.

Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/bnxt_re.h   | 22 +++++++
 drivers/infiniband/hw/bnxt_re/main.c      | 96 +++++++++++++++++++++++++++++++
 drivers/infiniband/hw/bnxt_re/qplib_res.h | 19 ++++++
 3 files changed, 137 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index ea81b24..1543f80 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -112,6 +112,27 @@ struct bnxt_re_gsi_context {
 #define BNXT_RE_NQ_IDX			1
 #define BNXT_RE_GEN_P5_MAX_VF		64
 
+struct bnxt_re_pacing {
+	u64 dbr_db_fifo_reg_off;
+	void *dbr_page;
+	u64 dbr_bar_addr;
+	u32 pacing_algo_th;
+	u32 do_pacing_save;
+	u32 dbq_pacing_time; /* ms */
+	u32 dbr_def_do_pacing;
+	bool dbr_pacing;
+};
+
+#define BNXT_RE_DBR_PACING_TIME 5 /* ms */
+#define BNXT_RE_PACING_ALGO_THRESHOLD 250 /* Entries in DB FIFO */
+#define BNXT_RE_PACING_ALARM_TH_MULTIPLE 2 /* Multiple of pacing algo threshold */
+/* Default do_pacing value when there is no congestion */
+#define BNXT_RE_DBR_DO_PACING_NO_CONGESTION 0x7F /* 1 in 512 probability */
+#define BNXT_RE_DB_FIFO_ROOM_MASK 0x1FFF8000
+#define BNXT_RE_MAX_FIFO_DEPTH 0x2c00
+#define BNXT_RE_DB_FIFO_ROOM_SHIFT 15
+#define BNXT_RE_GRC_FIFO_REG_BASE 0x2000
+
 struct bnxt_re_dev {
 	struct ib_device		ibdev;
 	struct list_head		list;
@@ -171,6 +192,7 @@ struct bnxt_re_dev {
 	atomic_t nq_alloc_cnt;
 	u32 is_virtfn;
 	u32 num_vfs;
+	struct bnxt_re_pacing pacing;
 };
 
 #define to_bnxt_re_dev(ptr, member)	\
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index b42166f..13cd84d 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -432,9 +432,92 @@ int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev)
 		return rc;
 	cctx->modes.db_push = le32_to_cpu(resp.flags) & FUNC_QCAPS_RESP_FLAGS_WCB_PUSH_MODE;
 
+	cctx->modes.dbr_pacing =
+		le32_to_cpu(resp.flags_ext2) & FUNC_QCAPS_RESP_FLAGS_EXT2_DBR_PACING_EXT_SUPPORTED ?
+		true : false;
 	return 0;
 }
 
+static int bnxt_re_hwrm_dbr_pacing_qcfg(struct bnxt_re_dev *rdev)
+{
+	struct hwrm_func_dbr_pacing_qcfg_output resp = {};
+	struct hwrm_func_dbr_pacing_qcfg_input req = {};
+	struct bnxt_en_dev *en_dev = rdev->en_dev;
+	struct bnxt_qplib_chip_ctx *cctx;
+	struct bnxt_fw_msg fw_msg = {};
+	int rc;
+
+	cctx = rdev->chip_ctx;
+	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_DBR_PACING_QCFG);
+	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
+			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+	rc = bnxt_send_msg(en_dev, &fw_msg);
+	if (rc)
+		return rc;
+
+	if ((le32_to_cpu(resp.dbr_stat_db_fifo_reg) &
+	    FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK) ==
+		FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_GRC)
+		cctx->dbr_stat_db_fifo =
+			le32_to_cpu(resp.dbr_stat_db_fifo_reg) &
+			~FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK;
+	return 0;
+}
+
+/* Update the pacing tunable parameters to the default values */
+static void bnxt_re_set_default_pacing_data(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
+
+	pacing_data->do_pacing = rdev->pacing.dbr_def_do_pacing;
+	pacing_data->pacing_th = rdev->pacing.pacing_algo_th;
+	pacing_data->alarm_th =
+		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
+}
+
+static int bnxt_re_initialize_dbr_pacing(struct bnxt_re_dev *rdev)
+{
+	if (bnxt_re_hwrm_dbr_pacing_qcfg(rdev))
+		return -EIO;
+
+	/* Allocate a page for app use */
+	rdev->pacing.dbr_page = (void *)__get_free_page(GFP_KERNEL);
+	if (!rdev->pacing.dbr_page)
+		return -ENOMEM;
+
+	memset((u8 *)rdev->pacing.dbr_page, 0, PAGE_SIZE);
+	rdev->qplib_res.pacing_data = (struct bnxt_qplib_db_pacing_data *)rdev->pacing.dbr_page;
+
+	/* MAP HW window 2 for reading db fifo depth */
+	writel(rdev->chip_ctx->dbr_stat_db_fifo & BNXT_GRC_BASE_MASK,
+	       rdev->en_dev->bar0 + BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
+	rdev->pacing.dbr_db_fifo_reg_off =
+		(rdev->chip_ctx->dbr_stat_db_fifo & BNXT_GRC_OFFSET_MASK) +
+		 BNXT_RE_GRC_FIFO_REG_BASE;
+	rdev->pacing.dbr_bar_addr =
+		pci_resource_start(rdev->qplib_res.pdev, 0) + rdev->pacing.dbr_db_fifo_reg_off;
+
+	rdev->pacing.pacing_algo_th = BNXT_RE_PACING_ALGO_THRESHOLD;
+	rdev->pacing.dbq_pacing_time = BNXT_RE_DBR_PACING_TIME;
+	rdev->pacing.dbr_def_do_pacing = BNXT_RE_DBR_DO_PACING_NO_CONGESTION;
+	rdev->pacing.do_pacing_save = rdev->pacing.dbr_def_do_pacing;
+	rdev->qplib_res.pacing_data->fifo_max_depth = BNXT_RE_MAX_FIFO_DEPTH;
+	rdev->qplib_res.pacing_data->fifo_room_mask = BNXT_RE_DB_FIFO_ROOM_MASK;
+	rdev->qplib_res.pacing_data->fifo_room_shift = BNXT_RE_DB_FIFO_ROOM_SHIFT;
+	rdev->qplib_res.pacing_data->grc_reg_offset = rdev->pacing.dbr_db_fifo_reg_off;
+	bnxt_re_set_default_pacing_data(rdev);
+	return 0;
+}
+
+static void bnxt_re_deinitialize_dbr_pacing(struct bnxt_re_dev *rdev)
+{
+	if (rdev->pacing.dbr_page)
+		free_page((u64)rdev->pacing.dbr_page);
+
+	rdev->pacing.dbr_page = NULL;
+	rdev->pacing.dbr_pacing = false;
+}
+
 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
 				 u16 fw_ring_id, int type)
 {
@@ -1217,6 +1300,9 @@ static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev)
 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags))
 		rdev->num_msix = 0;
 
+	if (rdev->pacing.dbr_pacing)
+		bnxt_re_deinitialize_dbr_pacing(rdev);
+
 	bnxt_re_destroy_chip_ctx(rdev);
 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags))
 		bnxt_unregister_dev(rdev->en_dev);
@@ -1309,6 +1395,16 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 wqe_mode)
 		goto free_ring;
 	}
 
+	if (bnxt_qplib_dbr_pacing_en(rdev->chip_ctx)) {
+		rc = bnxt_re_initialize_dbr_pacing(rdev);
+		if (!rc) {
+			rdev->pacing.dbr_pacing = true;
+		} else {
+			ibdev_err(&rdev->ibdev,
+				  "DBR pacing disabled with error : %d\n", rc);
+			rdev->pacing.dbr_pacing = false;
+		}
+	}
 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
 				     rdev->is_virtfn);
 	if (rc)
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h
index d850a55..57161d3 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h
@@ -48,6 +48,7 @@ extern const struct bnxt_qplib_gid bnxt_qplib_gid_zero;
 struct bnxt_qplib_drv_modes {
 	u8	wqe_mode;
 	bool db_push;
+	bool dbr_pacing;
 };
 
 struct bnxt_qplib_chip_ctx {
@@ -58,6 +59,17 @@ struct bnxt_qplib_chip_ctx {
 	u16	hwrm_cmd_max_timeout;
 	struct bnxt_qplib_drv_modes modes;
 	u64	hwrm_intf_ver;
+	u32     dbr_stat_db_fifo;
+};
+
+struct bnxt_qplib_db_pacing_data {
+	u32 do_pacing;
+	u32 pacing_th;
+	u32 alarm_th;
+	u32 fifo_max_depth;
+	u32 fifo_room_mask;
+	u32 fifo_room_shift;
+	u32 grc_reg_offset;
 };
 
 #define BNXT_QPLIB_DBR_PF_DB_OFFSET     0x10000
@@ -271,6 +283,7 @@ struct bnxt_qplib_res {
 	struct mutex                    dpi_tbl_lock;
 	bool				prio;
 	bool                            is_vf;
+	struct bnxt_qplib_db_pacing_data *pacing_data;
 };
 
 static inline bool bnxt_qplib_is_chip_gen_p5(struct bnxt_qplib_chip_ctx *cctx)
@@ -467,4 +480,10 @@ static inline bool _is_ext_stats_supported(u16 dev_cap_flags)
 	return dev_cap_flags &
 		CREQ_QUERY_FUNC_RESP_SB_EXT_STATS;
 }
+
+static inline u8 bnxt_qplib_dbr_pacing_en(struct bnxt_qplib_chip_ctx *cctx)
+{
+	return cctx->modes.dbr_pacing;
+}
+
 #endif /* __BNXT_QPLIB_RES_H__ */
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 4/7] RDMA/bnxt_re: Enable pacing support for the user apps

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 1374 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

Report the pacing capability to the user applications.

Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c | 2 ++
 include/uapi/rdma/bnxt_re-abi.h          | 1 +
 2 files changed, 3 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index abef0b8..86b71c7 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -4075,6 +4075,8 @@ int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata)
 		goto cfail;
 	}
 	uctx->shpage_mmap = &entry->rdma_entry;
+	if (rdev->pacing.dbr_pacing)
+		resp.comp_mask |= BNXT_RE_UCNTX_CMASK_DBR_PACING_ENABLED;
 
 	rc = ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)));
 	if (rc) {
diff --git a/include/uapi/rdma/bnxt_re-abi.h b/include/uapi/rdma/bnxt_re-abi.h
index 8a2a1d4..060bf1d 100644
--- a/include/uapi/rdma/bnxt_re-abi.h
+++ b/include/uapi/rdma/bnxt_re-abi.h
@@ -53,6 +53,7 @@ enum {
 	BNXT_RE_UCNTX_CMASK_HAVE_CCTX = 0x1ULL,
 	BNXT_RE_UCNTX_CMASK_HAVE_MODE = 0x02ULL,
 	BNXT_RE_UCNTX_CMASK_WC_DPI_ENABLED = 0x04ULL,
+	BNXT_RE_UCNTX_CMASK_DBR_PACING_ENABLED = 0x08ULL,
 };
 
 enum bnxt_re_wqe_mode {
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 5/7] RDMA/bnxt_re: Update alloc_page uapi for pacing

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 3946 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

Update the alloc_page uapi functionality for handling the
mapping of doorbell pacing shared page and bar address.

Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c | 35 +++++++++++++++++++++++++++++---
 drivers/infiniband/hw/bnxt_re/ib_verbs.h |  2 ++
 include/uapi/rdma/bnxt_re-abi.h          |  2 ++
 3 files changed, 36 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 86b71c7..0703163 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -565,6 +565,8 @@ bnxt_re_mmap_entry_insert(struct bnxt_re_ucontext *uctx, u64 mem_offset,
 		break;
 	case BNXT_RE_MMAP_UC_DB:
 	case BNXT_RE_MMAP_WC_DB:
+	case BNXT_RE_MMAP_DBR_BAR:
+	case BNXT_RE_MMAP_DBR_PAGE:
 		ret = rdma_user_mmap_entry_insert(&uctx->ib_uctx,
 						  &entry->rdma_entry, PAGE_SIZE);
 		break;
@@ -4149,6 +4151,19 @@ int bnxt_re_mmap(struct ib_ucontext *ib_uctx, struct vm_area_struct *vma)
 	case BNXT_RE_MMAP_SH_PAGE:
 		ret = vm_insert_page(vma, vma->vm_start, virt_to_page(uctx->shpg));
 		break;
+	case BNXT_RE_MMAP_DBR_BAR:
+		pfn = bnxt_entry->mem_offset >> PAGE_SHIFT;
+		ret = rdma_user_mmap_io(ib_uctx, vma, pfn, PAGE_SIZE,
+					pgprot_noncached(vma->vm_page_prot),
+					rdma_entry);
+		break;
+	case BNXT_RE_MMAP_DBR_PAGE:
+		/* Driver doesn't expect write access for user space */
+		if (vma->vm_flags & VM_WRITE)
+			return -EFAULT;
+		ret = vm_insert_page(vma, vma->vm_start,
+				     virt_to_page((void *)bnxt_entry->mem_offset));
+		break;
 	default:
 		ret = -EINVAL;
 		break;
@@ -4180,7 +4195,7 @@ static int UVERBS_HANDLER(BNXT_RE_METHOD_ALLOC_PAGE)(struct uverbs_attr_bundle *
 	u64 mmap_offset;
 	u32 length;
 	u32 dpi;
-	u64 dbr;
+	u64 addr;
 	int err;
 
 	uctx = container_of(ib_uverbs_get_ucontext(attrs), struct bnxt_re_ucontext, ib_uctx);
@@ -4202,19 +4217,30 @@ static int UVERBS_HANDLER(BNXT_RE_METHOD_ALLOC_PAGE)(struct uverbs_attr_bundle *
 				return -ENOMEM;
 			length = PAGE_SIZE;
 			dpi = uctx->wcdpi.dpi;
-			dbr = (u64)uctx->wcdpi.umdbr;
+			addr = (u64)uctx->wcdpi.umdbr;
 			mmap_flag = BNXT_RE_MMAP_WC_DB;
 		} else {
 			return -EINVAL;
 		}
 
 		break;
+	case BNXT_RE_ALLOC_DBR_BAR_PAGE:
+		length = PAGE_SIZE;
+		addr = (u64)rdev->pacing.dbr_bar_addr;
+		mmap_flag = BNXT_RE_MMAP_DBR_BAR;
+		break;
+
+	case BNXT_RE_ALLOC_DBR_PAGE:
+		length = PAGE_SIZE;
+		addr = (u64)rdev->pacing.dbr_page;
+		mmap_flag = BNXT_RE_MMAP_DBR_PAGE;
+		break;
 
 	default:
 		return -EOPNOTSUPP;
 	}
 
-	entry = bnxt_re_mmap_entry_insert(uctx, dbr, mmap_flag, &mmap_offset);
+	entry = bnxt_re_mmap_entry_insert(uctx, addr, mmap_flag, &mmap_offset);
 	if (!entry)
 		return -ENOMEM;
 
@@ -4254,6 +4280,9 @@ static int alloc_page_obj_cleanup(struct ib_uobject *uobject,
 			uctx->wcdpi.dbr = NULL;
 		}
 		break;
+	case BNXT_RE_MMAP_DBR_BAR:
+	case BNXT_RE_MMAP_DBR_PAGE:
+		break;
 	default:
 		goto exit;
 	}
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
index 32d9e9d..f392a09 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
@@ -146,6 +146,8 @@ enum bnxt_re_mmap_flag {
 	BNXT_RE_MMAP_SH_PAGE,
 	BNXT_RE_MMAP_UC_DB,
 	BNXT_RE_MMAP_WC_DB,
+	BNXT_RE_MMAP_DBR_PAGE,
+	BNXT_RE_MMAP_DBR_BAR,
 };
 
 struct bnxt_re_user_mmap_entry {
diff --git a/include/uapi/rdma/bnxt_re-abi.h b/include/uapi/rdma/bnxt_re-abi.h
index 060bf1d..78a324f 100644
--- a/include/uapi/rdma/bnxt_re-abi.h
+++ b/include/uapi/rdma/bnxt_re-abi.h
@@ -136,6 +136,8 @@ enum bnxt_re_objects {
 
 enum bnxt_re_alloc_page_type {
 	BNXT_RE_ALLOC_WC_PAGE = 0,
+	BNXT_RE_ALLOC_DBR_BAR_PAGE,
+	BNXT_RE_ALLOC_DBR_PAGE,
 };
 
 enum bnxt_re_var_alloc_page_attrs {
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 6/7] RDMA/bnxt_re: Implement doorbell pacing algorithm

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 7186 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

User applications alert the driver when the Doorbell FIFO
reaches the alarm threshold. The driver updates the pacing
parameters in the shared page to do the maximum pacing
by the application till the DB FIFO congestion reduces to
pacing threshold. Driver keeps checking the DB FIFO depth
at the pacing interval and gradually adjusts the pacing level.
Once the pacing level reaches default values (no congestion in
the FIFO) pacing gets completed.

Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/bnxt_re.h |   5 ++
 drivers/infiniband/hw/bnxt_re/main.c    | 124 ++++++++++++++++++++++++++++++++
 2 files changed, 129 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index 1543f80..2175103 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -121,8 +121,10 @@ struct bnxt_re_pacing {
 	u32 dbq_pacing_time; /* ms */
 	u32 dbr_def_do_pacing;
 	bool dbr_pacing;
+	struct mutex dbq_lock; /* synchronize db pacing algo */
 };
 
+#define BNXT_RE_MAX_DBR_DO_PACING 0xFFFF
 #define BNXT_RE_DBR_PACING_TIME 5 /* ms */
 #define BNXT_RE_PACING_ALGO_THRESHOLD 250 /* Entries in DB FIFO */
 #define BNXT_RE_PACING_ALARM_TH_MULTIPLE 2 /* Multiple of pacing algo threshold */
@@ -193,6 +195,8 @@ struct bnxt_re_dev {
 	u32 is_virtfn;
 	u32 num_vfs;
 	struct bnxt_re_pacing pacing;
+	struct work_struct dbq_fifo_check_work;
+	struct delayed_work dbq_pacing_work;
 };
 
 #define to_bnxt_re_dev(ptr, member)	\
@@ -203,6 +207,7 @@ struct bnxt_re_dev {
 #define BNXT_RE_ROCEV2_IPV6_PACKET	3
 
 #define BNXT_RE_CHECK_RC(x) ((x) && ((x) != -ETIMEDOUT))
+void bnxt_re_pacing_alert(struct bnxt_re_dev *rdev);
 
 static inline struct device *rdev_to_dev(struct bnxt_re_dev *rdev)
 {
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 13cd84d..6469811 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -475,6 +475,125 @@ static void bnxt_re_set_default_pacing_data(struct bnxt_re_dev *rdev)
 		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
 }
 
+static void __wait_for_fifo_occupancy_below_th(struct bnxt_re_dev *rdev)
+{
+	u32 read_val, fifo_occup;
+
+	/* loop shouldn't run infintely as the occupancy usually goes
+	 * below pacing algo threshold as soon as pacing kicks in.
+	 */
+	while (1) {
+		read_val = readl(rdev->en_dev->bar0 + rdev->pacing.dbr_db_fifo_reg_off);
+		fifo_occup = BNXT_RE_MAX_FIFO_DEPTH -
+			((read_val & BNXT_RE_DB_FIFO_ROOM_MASK) >>
+			 BNXT_RE_DB_FIFO_ROOM_SHIFT);
+		/* Fifo occupancy cannot be greater the MAX FIFO depth */
+		if (fifo_occup > BNXT_RE_MAX_FIFO_DEPTH)
+			break;
+
+		if (fifo_occup < rdev->qplib_res.pacing_data->pacing_th)
+			break;
+	}
+}
+
+static void bnxt_re_db_fifo_check(struct work_struct *work)
+{
+	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
+			dbq_fifo_check_work);
+	struct bnxt_qplib_db_pacing_data *pacing_data;
+	u32 pacing_save;
+
+	if (!mutex_trylock(&rdev->pacing.dbq_lock))
+		return;
+	pacing_data = rdev->qplib_res.pacing_data;
+	pacing_save = rdev->pacing.do_pacing_save;
+	__wait_for_fifo_occupancy_below_th(rdev);
+	cancel_delayed_work_sync(&rdev->dbq_pacing_work);
+	if (pacing_save > rdev->pacing.dbr_def_do_pacing) {
+		/* Double the do_pacing value during the congestion */
+		pacing_save = pacing_save << 1;
+	} else {
+		/*
+		 * when a new congestion is detected increase the do_pacing
+		 * by 8 times. And also increase the pacing_th by 4 times. The
+		 * reason to increase pacing_th is to give more space for the
+		 * queue to oscillate down without getting empty, but also more
+		 * room for the queue to increase without causing another alarm.
+		 */
+		pacing_save = pacing_save << 3;
+		pacing_data->pacing_th = rdev->pacing.pacing_algo_th * 4;
+	}
+
+	if (pacing_save > BNXT_RE_MAX_DBR_DO_PACING)
+		pacing_save = BNXT_RE_MAX_DBR_DO_PACING;
+
+	pacing_data->do_pacing = pacing_save;
+	rdev->pacing.do_pacing_save = pacing_data->do_pacing;
+	pacing_data->alarm_th =
+		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
+	schedule_delayed_work(&rdev->dbq_pacing_work,
+			      msecs_to_jiffies(rdev->pacing.dbq_pacing_time));
+	mutex_unlock(&rdev->pacing.dbq_lock);
+}
+
+static void bnxt_re_pacing_timer_exp(struct work_struct *work)
+{
+	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
+			dbq_pacing_work.work);
+	struct bnxt_qplib_db_pacing_data *pacing_data;
+	u32 read_val, fifo_occup;
+
+	if (!mutex_trylock(&rdev->pacing.dbq_lock))
+		return;
+
+	pacing_data = rdev->qplib_res.pacing_data;
+	read_val = readl(rdev->en_dev->bar0 + rdev->pacing.dbr_db_fifo_reg_off);
+	fifo_occup = BNXT_RE_MAX_FIFO_DEPTH -
+		((read_val & BNXT_RE_DB_FIFO_ROOM_MASK) >>
+		 BNXT_RE_DB_FIFO_ROOM_SHIFT);
+
+	if (fifo_occup > pacing_data->pacing_th)
+		goto restart_timer;
+
+	/*
+	 * Instead of immediately going back to the default do_pacing
+	 * reduce it by 1/8 times and restart the timer.
+	 */
+	pacing_data->do_pacing = pacing_data->do_pacing - (pacing_data->do_pacing >> 3);
+	pacing_data->do_pacing = max_t(u32, rdev->pacing.dbr_def_do_pacing, pacing_data->do_pacing);
+	if (pacing_data->do_pacing <= rdev->pacing.dbr_def_do_pacing) {
+		bnxt_re_set_default_pacing_data(rdev);
+		goto dbq_unlock;
+	}
+
+restart_timer:
+	schedule_delayed_work(&rdev->dbq_pacing_work,
+			      msecs_to_jiffies(rdev->pacing.dbq_pacing_time));
+dbq_unlock:
+	rdev->pacing.do_pacing_save = pacing_data->do_pacing;
+	mutex_unlock(&rdev->pacing.dbq_lock);
+}
+
+void bnxt_re_pacing_alert(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_db_pacing_data *pacing_data;
+
+	if (!rdev->pacing.dbr_pacing)
+		return;
+	mutex_lock(&rdev->pacing.dbq_lock);
+	pacing_data = rdev->qplib_res.pacing_data;
+
+	/*
+	 * Increase the alarm_th to max so that other user lib instances do not
+	 * keep alerting the driver.
+	 */
+	pacing_data->alarm_th = BNXT_RE_MAX_FIFO_DEPTH;
+	pacing_data->do_pacing = BNXT_RE_MAX_DBR_DO_PACING;
+	cancel_work_sync(&rdev->dbq_fifo_check_work);
+	schedule_work(&rdev->dbq_fifo_check_work);
+	mutex_unlock(&rdev->pacing.dbq_lock);
+}
+
 static int bnxt_re_initialize_dbr_pacing(struct bnxt_re_dev *rdev)
 {
 	if (bnxt_re_hwrm_dbr_pacing_qcfg(rdev))
@@ -506,11 +625,16 @@ static int bnxt_re_initialize_dbr_pacing(struct bnxt_re_dev *rdev)
 	rdev->qplib_res.pacing_data->fifo_room_shift = BNXT_RE_DB_FIFO_ROOM_SHIFT;
 	rdev->qplib_res.pacing_data->grc_reg_offset = rdev->pacing.dbr_db_fifo_reg_off;
 	bnxt_re_set_default_pacing_data(rdev);
+	/* Initialize worker for DBR Pacing */
+	INIT_WORK(&rdev->dbq_fifo_check_work, bnxt_re_db_fifo_check);
+	INIT_DELAYED_WORK(&rdev->dbq_pacing_work, bnxt_re_pacing_timer_exp);
 	return 0;
 }
 
 static void bnxt_re_deinitialize_dbr_pacing(struct bnxt_re_dev *rdev)
 {
+	cancel_work_sync(&rdev->dbq_fifo_check_work);
+	cancel_delayed_work_sync(&rdev->dbq_pacing_work);
 	if (rdev->pacing.dbr_page)
 		free_page((u64)rdev->pacing.dbr_page);
 
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

[PATCH for-next v3 7/7] RDMA/bnxt_re: Add a new uapi for driver notification

[Selvin Xavier]

[-- Attachment #1: Type: text/plain, Size: 2376 bytes --]

From: Chandramohan Akula <chandramohan.akula@broadcom.com>

Add driver notify uapi for application notifying
the driver about the doorbell FIFO congestion.

Signed-off-by: Chandramohan Akula <chandramohan.akula@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c | 15 +++++++++++++++
 include/uapi/rdma/bnxt_re-abi.h          |  4 ++++
 2 files changed, 19 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 0703163..ec4d163 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -4183,6 +4183,15 @@ void bnxt_re_mmap_free(struct rdma_user_mmap_entry *rdma_entry)
 	kfree(bnxt_entry);
 }
 
+static int UVERBS_HANDLER(BNXT_RE_METHOD_NOTIFY_DRV)(struct uverbs_attr_bundle *attrs)
+{
+	struct bnxt_re_ucontext *uctx;
+
+	uctx = container_of(ib_uverbs_get_ucontext(attrs), struct bnxt_re_ucontext, ib_uctx);
+	bnxt_re_pacing_alert(uctx->rdev);
+	return 0;
+}
+
 static int UVERBS_HANDLER(BNXT_RE_METHOD_ALLOC_PAGE)(struct uverbs_attr_bundle *attrs)
 {
 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs, BNXT_RE_ALLOC_PAGE_HANDLE);
@@ -4320,7 +4329,13 @@ DECLARE_UVERBS_NAMED_OBJECT(BNXT_RE_OBJECT_ALLOC_PAGE,
 			    &UVERBS_METHOD(BNXT_RE_METHOD_ALLOC_PAGE),
 			    &UVERBS_METHOD(BNXT_RE_METHOD_DESTROY_PAGE));
 
+DECLARE_UVERBS_NAMED_METHOD(BNXT_RE_METHOD_NOTIFY_DRV);
+
+DECLARE_UVERBS_GLOBAL_METHODS(BNXT_RE_OBJECT_NOTIFY_DRV,
+			      &UVERBS_METHOD(BNXT_RE_METHOD_NOTIFY_DRV));
+
 const struct uapi_definition bnxt_re_uapi_defs[] = {
 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(BNXT_RE_OBJECT_ALLOC_PAGE),
+	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(BNXT_RE_OBJECT_NOTIFY_DRV),
 	{}
 };
diff --git a/include/uapi/rdma/bnxt_re-abi.h b/include/uapi/rdma/bnxt_re-abi.h
index 78a324f..6e7c67a 100644
--- a/include/uapi/rdma/bnxt_re-abi.h
+++ b/include/uapi/rdma/bnxt_re-abi.h
@@ -132,6 +132,7 @@ enum bnxt_re_shpg_offt {
 
 enum bnxt_re_objects {
 	BNXT_RE_OBJECT_ALLOC_PAGE = (1U << UVERBS_ID_NS_SHIFT),
+	BNXT_RE_OBJECT_NOTIFY_DRV,
 };
 
 enum bnxt_re_alloc_page_type {
@@ -157,4 +158,7 @@ enum bnxt_re_alloc_page_methods {
 	BNXT_RE_METHOD_DESTROY_PAGE,
 };
 
+enum bnxt_re_notify_drv_methods {
+	BNXT_RE_METHOD_NOTIFY_DRV = (1U << UVERBS_ID_NS_SHIFT),
+};
 #endif /* __BNXT_RE_UVERBS_ABI_H__*/
-- 
2.5.5


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]

Re: [PATCH for-next v3 0/7] RDMA/bnxt_re: Doorbell Drop Prevention

[Jason Gunthorpe]

On Tue, Jul 18, 2023 at 10:02:50PM -0700, Selvin Xavier wrote:
> The idea behind this series is to prevent Doorbell drops
> on some of the Broadcom adapters that require Doorbell
> moderation. This is achieved by pacing the doorbell writes
> into the hardware FIFO. The rate at which individual doorbells
> are written needs to be dynamically adjusted, because
> it depends on the ability of the hardware to drain the
> FIFO and on the number and behavior of individual
> doorbell writers. When congestion is detected by the user
> library, it notifies the driver and driver adjust the
> pacing parameters dynamically in a shared page, which will
> be used for pacing the Doorbells.
> 
> Currently this feature is targeted only for user applications.
> The corresponding user lib patch is in the pull request.
> https://github.com/linux-rdma/rdma-core/pull/1360
> 
> Thanks,
> Selvin Xavier
> 
> v2 -> v3:
>      Fix the build warning
>      Reported-by: kernel test robot <lkp@intel.com>
> 
> v1 -> v2:
>      Rebased the patches on top of the latest for-next branch
> 
> Chandramohan Akula (7):
>   bnxt_en: Update HW interface headers
>   bnxt_en: Share the bar0 address with the RoCE driver
>   RDMA/bnxt_re: Initialize Doorbell pacing feature
>   RDMA/bnxt_re: Enable pacing support for the user apps
>   RDMA/bnxt_re: Update alloc_page uapi for pacing
>   RDMA/bnxt_re: Implement doorbell pacing algorithm
>   RDMA/bnxt_re: Add a new uapi for driver notification

Applied to for-next, thanks

Jason