[net-next PATCH 0/3] Dynamic LMTST region setup

[net-next PATCH 0/3] Dynamic LMTST region setup

[Geetha sowjanya]

This patch series allows RVU PF/VF to allocate memory for
LMTST operations instead of using memory reserved by firmware
which is mapped as device memory.
The LMTST mapping table contains the RVU PF/VF LMTST memory base
address entries. This table is used by hardware for LMTST operations.
Patch1 introduces new mailbox message to update the LMTST table with
the new allocated memory address.

Geetha sowjanya (2):
  octeontx2-af: Support configurable LMTST regions
  octeontx2-pf: Use runtime allocated LMTLINE region

Harman Kalra (1):
  octeontx2-af: cn10k: setting up lmtst map table

 .../net/ethernet/marvell/octeontx2/af/mbox.h  |  10 +
 .../net/ethernet/marvell/octeontx2/af/rvu.c   |   1 +
 .../net/ethernet/marvell/octeontx2/af/rvu.h   |   4 +
 .../ethernet/marvell/octeontx2/af/rvu_cn10k.c | 202 ++++++++++++++++++
 .../ethernet/marvell/octeontx2/af/rvu_reg.h   |  10 +
 .../marvell/octeontx2/af/rvu_struct.h         |   3 +-
 .../ethernet/marvell/octeontx2/nic/cn10k.c    |  87 +++-----
 .../ethernet/marvell/octeontx2/nic/cn10k.h    |   3 +-
 .../marvell/octeontx2/nic/otx2_common.h       |   7 +-
 .../ethernet/marvell/octeontx2/nic/otx2_pf.c  |  17 +-
 .../marvell/octeontx2/nic/otx2_txrx.h         |   1 +
 .../ethernet/marvell/octeontx2/nic/otx2_vf.c  |  12 +-
 12 files changed, 283 insertions(+), 74 deletions(-)

-- 
2.17.1

[net-next PATCH 1/3] octeontx2-af: cn10k: Setting up lmtst map table

[Geetha sowjanya]

From: Harman Kalra <hkalra@marvell.com>

Introducing a new mailbox to support updating lmt entries
and common lmt base address scheme i.e. multiple pcifuncs
can share lmt region to reduce L1 cache pressure for application.
Parameters passed to mailbox includes the primary pcifunc
value whose lmt regions will be shared by other secondary
pcifuncs. Here secondary pcifunc will be the one who is
calling the mailbox.
For example:
By default each pcifunc has its own LMT base address:
        PCIFUNC1    LMT_BASE_ADDR A
        PCIFUNC2    LMT_BASE_ADDR B
        PCIFUNC3    LMT_BASE_ADDR C
        PCIFUNC4    LMT_BASE_ADDR D
Application will choose PCIFUNC1 as base/primary pcifunc
and as and when other pcifunc(secondary pcifuncs) gets
probed, this mailbox will be called and LMTST table will
be updated as:
        PCIFUNC1    LMT_BASE_ADDR A
        PCIFUNC2    LMT_BASE_ADDR A
        PCIFUNC3    LMT_BASE_ADDR A
        PCIFUNC4    LMT_BASE_ADDR A

On FLR lmtst map table gets resetted to the default lmt
base addresses for all secondary pcifuncs.

Signed-off-by: Harman Kalra <hkalra@marvell.com>
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
---
 .../net/ethernet/marvell/octeontx2/af/mbox.h  |   7 +
 .../net/ethernet/marvell/octeontx2/af/rvu.c   |   1 +
 .../net/ethernet/marvell/octeontx2/af/rvu.h   |   4 +
 .../ethernet/marvell/octeontx2/af/rvu_cn10k.c | 140 ++++++++++++++++++
 .../ethernet/marvell/octeontx2/af/rvu_reg.h   |   5 +
 .../marvell/octeontx2/af/rvu_struct.h         |   3 +-
 6 files changed, 159 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
index 770d86262838..638db868125a 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
@@ -134,6 +134,8 @@ M(MSIX_OFFSET,		0x005, msix_offset, msg_req, msix_offset_rsp)	\
 M(VF_FLR,		0x006, vf_flr, msg_req, msg_rsp)		\
 M(PTP_OP,		0x007, ptp_op, ptp_req, ptp_rsp)		\
 M(GET_HW_CAP,		0x008, get_hw_cap, msg_req, get_hw_cap_rsp)	\
+M(LMTST_TBL_SETUP,	0x00a, lmtst_tbl_setup, lmtst_tbl_setup_req,    \
+				msg_rsp)				\
 M(SET_VF_PERM,		0x00b, set_vf_perm, set_vf_perm, msg_rsp)	\
 /* CGX mbox IDs (range 0x200 - 0x3FF) */				\
 M(CGX_START_RXTX,	0x200, cgx_start_rxtx, msg_req, msg_rsp)	\
@@ -1278,6 +1280,11 @@ struct set_vf_perm  {
 	u64	flags;
 };
 
+struct lmtst_tbl_setup_req {
+	struct mbox_msghdr hdr;
+	u16 base_pcifunc;
+};
+
 /* CPT mailbox error codes
  * Range 901 - 1000.
  */
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu.c b/drivers/net/ethernet/marvell/octeontx2/af/rvu.c
index 0b092949d7ac..10cddf1ac7b9 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu.c
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu.c
@@ -2333,6 +2333,7 @@ static void __rvu_flr_handler(struct rvu *rvu, u16 pcifunc)
 	rvu_blklf_teardown(rvu, pcifunc, BLKADDR_SSOW);
 	rvu_blklf_teardown(rvu, pcifunc, BLKADDR_SSO);
 	rvu_blklf_teardown(rvu, pcifunc, BLKADDR_NPA);
+	rvu_reset_lmt_map_tbl(rvu, pcifunc);
 	rvu_detach_rsrcs(rvu, NULL, pcifunc);
 	mutex_unlock(&rvu->flr_lock);
 }
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu.h b/drivers/net/ethernet/marvell/octeontx2/af/rvu.h
index 9e5d9ba6f01e..3c0a7e981f72 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu.h
@@ -243,6 +243,7 @@ struct rvu_pfvf {
 	u8	nix_blkaddr; /* BLKADDR_NIX0/1 assigned to this PF */
 	u8	nix_rx_intf; /* NIX0_RX/NIX1_RX interface to NPC */
 	u8	nix_tx_intf; /* NIX0_TX/NIX1_TX interface to NPC */
+	u64     lmt_base_addr; /* Preseving the pcifunc's lmtst base addr*/
 	unsigned long flags;
 };
 
@@ -754,6 +755,9 @@ int rvu_cpt_lf_teardown(struct rvu *rvu, u16 pcifunc, int lf, int slot);
 int rvu_set_channels_base(struct rvu *rvu);
 void rvu_program_channels(struct rvu *rvu);
 
+/* CN10K RVU - LMT*/
+void rvu_reset_lmt_map_tbl(struct rvu *rvu, u16 pcifunc);
+
 #ifdef CONFIG_DEBUG_FS
 void rvu_dbg_init(struct rvu *rvu);
 void rvu_dbg_exit(struct rvu *rvu);
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
index 7d9e71c6965f..87f56e1f32e3 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
@@ -10,6 +10,146 @@
 #include "cgx.h"
 #include "rvu_reg.h"
 
+/* RVU LMTST */
+#define LMT_TBL_OP_READ		0
+#define LMT_TBL_OP_WRITE	1
+#define LMT_MAP_TABLE_SIZE	(128 * 1024)
+#define LMT_MAPTBL_ENTRY_SIZE	16
+
+/* Function to perform operations (read/write) on lmtst map table */
+static int lmtst_map_table_ops(struct rvu *rvu, u32 index, u64 *val,
+			       int lmt_tbl_op)
+{
+	void __iomem *lmt_map_base;
+	u64 tbl_base;
+
+	tbl_base = rvu_read64(rvu, BLKADDR_APR, APR_AF_LMT_MAP_BASE);
+
+	lmt_map_base = ioremap_wc(tbl_base, LMT_MAP_TABLE_SIZE);
+	if (!lmt_map_base) {
+		dev_err(rvu->dev, "Failed to setup lmt map table mapping!!\n");
+		return -ENOMEM;
+	}
+
+	if (lmt_tbl_op == LMT_TBL_OP_READ) {
+		*val = readq(lmt_map_base + index);
+	} else {
+		writeq((*val), (lmt_map_base + index));
+		/* Flushing the AP interceptor cache to make APR_LMT_MAP_ENTRY_S
+		 * changes effective. Write 1 for flush and read is being used as a
+		 * barrier and sets up a data dependency. Write to 0 after a write
+		 * to 1 to complete the flush.
+		 */
+		rvu_write64(rvu, BLKADDR_APR, APR_AF_LMT_CTL, BIT_ULL(0));
+		rvu_read64(rvu, BLKADDR_APR, APR_AF_LMT_CTL);
+		rvu_write64(rvu, BLKADDR_APR, APR_AF_LMT_CTL, 0x00);
+	}
+
+	iounmap(lmt_map_base);
+	return 0;
+}
+
+static u32 rvu_get_lmtst_tbl_index(struct rvu *rvu, u16 pcifunc)
+{
+	return ((rvu_get_pf(pcifunc) * rvu->hw->total_vfs) +
+		(pcifunc & RVU_PFVF_FUNC_MASK)) * LMT_MAPTBL_ENTRY_SIZE;
+}
+
+int rvu_mbox_handler_lmtst_tbl_setup(struct rvu *rvu,
+				     struct lmtst_tbl_setup_req *req,
+				     struct msg_rsp *rsp)
+{
+	struct rvu_pfvf *pfvf = rvu_get_pfvf(rvu, req->hdr.pcifunc);
+	u32 pri_tbl_idx, sec_tbl_idx;
+	int err = 0;
+	u64 val;
+
+	/* Reconfiguring lmtst map table in lmt region shared mode i.e. make
+	 * multiple PF_FUNCs to share an LMTLINE region, so primary/base
+	 * pcifunc (which is passed as an argument to mailbox) is the one
+	 * whose lmt base address will be shared among other secondary
+	 * pcifunc (will be the one who is calling this mailbox).
+	 */
+	if (req->base_pcifunc) {
+		/* Calculating the LMT table index equivalent to primary
+		 * pcifunc.
+		 */
+		pri_tbl_idx = rvu_get_lmtst_tbl_index(rvu, req->base_pcifunc);
+
+		/* Truncating secondary pcifunc to calculate the LMT table index
+		 * equivalent to secondary pcifunc.
+		 */
+		sec_tbl_idx = rvu_get_lmtst_tbl_index(rvu, req->hdr.pcifunc);
+		/* Read the base lmt addr of the secondary pcifunc */
+		err = lmtst_map_table_ops(rvu, sec_tbl_idx, &val,
+					  LMT_TBL_OP_READ);
+		if (err) {
+			dev_err(rvu->dev,
+				"Failed to read LMT map table: index 0x%x err %d\n",
+				sec_tbl_idx, err);
+			goto error;
+		}
+
+		/* Storing the seondary's lmt base address as this needs to be
+		 * reverted in FLR. Also making sure this default value doesn't
+		 * get overwritten on multiple calls to this mailbox.
+		 */
+		if (!pfvf->lmt_base_addr)
+			pfvf->lmt_base_addr = val;
+
+		/* Read the base lmt addr of the primary pcifunc */
+		err = lmtst_map_table_ops(rvu, pri_tbl_idx, &val,
+					  LMT_TBL_OP_READ);
+		if (err) {
+			dev_err(rvu->dev,
+				"Failed to read LMT map table: index 0x%x err %d\n",
+				pri_tbl_idx, err);
+			goto error;
+		}
+
+		/* Update the base lmt addr of secondary with primary's base
+		 * lmt addr.
+		 */
+		err = lmtst_map_table_ops(rvu, sec_tbl_idx, &val,
+					  LMT_TBL_OP_WRITE);
+		if (err) {
+			dev_err(rvu->dev,
+				"Failed to update LMT map table: index 0x%x err %d\n",
+				sec_tbl_idx, err);
+			goto error;
+		}
+	}
+
+error:
+	return err;
+}
+
+/* Resetting the lmtst map table to original base addresses */
+void rvu_reset_lmt_map_tbl(struct rvu *rvu, u16 pcifunc)
+{
+	struct rvu_pfvf *pfvf = rvu_get_pfvf(rvu, pcifunc);
+	u32 tbl_idx;
+	int err;
+
+	if (is_rvu_otx2(rvu))
+		return;
+
+	if (pfvf->lmt_base_addr) {
+		/* This corresponds to lmt map table index */
+		tbl_idx = rvu_get_lmtst_tbl_index(rvu, pcifunc);
+		/* Reverting back original lmt base addr for respective
+		 * pcifunc.
+		 */
+		err = lmtst_map_table_ops(rvu, tbl_idx, &pfvf->lmt_base_addr,
+					  LMT_TBL_OP_WRITE);
+		if (err)
+			dev_err(rvu->dev,
+				"Failed to update LMT map table: index 0x%x err %d\n",
+				tbl_idx, err);
+		pfvf->lmt_base_addr = 0;
+	}
+}
+
 int rvu_set_channels_base(struct rvu *rvu)
 {
 	struct rvu_hwinfo *hw = rvu->hw;
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h b/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h
index 76837d5e19c6..61bafe956aae 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h
@@ -692,4 +692,9 @@
 #define LBK_LINK_CFG_ID_MASK		GENMASK_ULL(11, 6)
 #define LBK_LINK_CFG_BASE_MASK		GENMASK_ULL(5, 0)
 
+/* APR */
+#define	APR_AF_LMT_CFG			(0x000ull)
+#define	APR_AF_LMT_MAP_BASE		(0x008ull)
+#define	APR_AF_LMT_CTL			(0x010ull)
+
 #endif /* RVU_REG_H */
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_struct.h b/drivers/net/ethernet/marvell/octeontx2/af/rvu_struct.h
index 14aa8e37ea41..5bbe6727d11d 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_struct.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_struct.h
@@ -35,7 +35,8 @@ enum rvu_block_addr_e {
 	BLKADDR_NDC_NPA0	= 0xeULL,
 	BLKADDR_NDC_NIX1_RX	= 0x10ULL,
 	BLKADDR_NDC_NIX1_TX	= 0x11ULL,
-	BLK_COUNT		= 0x12ULL,
+	BLKADDR_APR		= 0x16ULL,
+	BLK_COUNT		= 0x17ULL,
 };
 
 /* RVU Block Type Enumeration */
-- 
2.17.1

[net-next PATCH 2/3] octeontx2-af: cn10k: Support configurable LMTST regions

[Geetha sowjanya]

This patch extends the lmtst_tbl_setup_req mbox to support run time
LMTST configuration.
RVU PF/VF and DPDK/ODP allocates a LMT region, creates a translation
entry for a device via VFIO IOCTLs.
This IOVA is shared with AF through above mbox. AF then uses
RVU_SMMU transulation Widget and gets PA for the IOVA and updates
the LMTtable entry for that device.

Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
---
 .../net/ethernet/marvell/octeontx2/af/mbox.h  |   3 +
 .../ethernet/marvell/octeontx2/af/rvu_cn10k.c | 130 +++++++++++++-----
 .../ethernet/marvell/octeontx2/af/rvu_reg.h   |   5 +
 3 files changed, 103 insertions(+), 35 deletions(-)

diff --git a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
index 638db868125a..9672cbf8a90a 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
@@ -1283,6 +1283,9 @@ struct set_vf_perm  {
 struct lmtst_tbl_setup_req {
 	struct mbox_msghdr hdr;
 	u16 base_pcifunc;
+	u8  use_local_lmt_region;
+	u64 lmt_iova;
+	u64 rsvd[4];
 };
 
 /* CPT mailbox error codes
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
index 87f56e1f32e3..8d48b64485c6 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
@@ -55,14 +55,101 @@ static u32 rvu_get_lmtst_tbl_index(struct rvu *rvu, u16 pcifunc)
 		(pcifunc & RVU_PFVF_FUNC_MASK)) * LMT_MAPTBL_ENTRY_SIZE;
 }
 
+static int rvu_get_lmtaddr(struct rvu *rvu, u16 pcifunc,
+			   u64 iova, u64 *lmt_addr)
+{
+	u64 pa, val, pf;
+	int err;
+
+	if (!iova) {
+		dev_err(rvu->dev, "%s Requested Null address for transulation\n", __func__);
+		return -EINVAL;
+	}
+
+	rvu_write64(rvu, BLKADDR_RVUM, RVU_AF_SMMU_ADDR_REQ, iova);
+	pf = rvu_get_pf(pcifunc) & 0x1F;
+	val = BIT_ULL(63) | BIT_ULL(14) | BIT_ULL(13) | pf << 8 |
+	      ((pcifunc & RVU_PFVF_FUNC_MASK) & 0xFF);
+	rvu_write64(rvu, BLKADDR_RVUM, RVU_AF_SMMU_TXN_REQ, val);
+
+	err = rvu_poll_reg(rvu, BLKADDR_RVUM, RVU_AF_SMMU_ADDR_RSP_STS, BIT_ULL(0), false);
+	if (err) {
+		dev_err(rvu->dev, "%s LMTLINE iova transulation failed\n", __func__);
+		return err;
+	}
+	val = rvu_read64(rvu, BLKADDR_RVUM, RVU_AF_SMMU_ADDR_RSP_STS);
+	if (val & ~0x1ULL) {
+		dev_err(rvu->dev, "%s LMTLINE iova transulation failed err:%llx\n", __func__, val);
+		return -EIO;
+	}
+	/* PA[51:12] = RVU_AF_SMMU_TLN_FLIT1[60:21]
+	 * PA[11:0] = IOVA[11:0]
+	 */
+	pa = rvu_read64(rvu, BLKADDR_RVUM, RVU_AF_SMMU_TLN_FLIT1) >> 21;
+	pa &= GENMASK_ULL(39, 0);
+	*lmt_addr = (pa << 12) | (iova  & 0xFFF);
+
+	return 0;
+}
+
+static int rvu_update_lmtaddr(struct rvu *rvu, u16 pcifunc, u64 lmt_addr)
+{
+	struct rvu_pfvf *pfvf = rvu_get_pfvf(rvu, pcifunc);
+	u32 tbl_idx;
+	int err = 0;
+	u64 val;
+
+	/* Read the current lmt addr of pcifunc */
+	tbl_idx = rvu_get_lmtst_tbl_index(rvu, pcifunc);
+	err = lmtst_map_table_ops(rvu, tbl_idx, &val, LMT_TBL_OP_READ);
+	if (err) {
+		dev_err(rvu->dev,
+			"Failed to read LMT map table: index 0x%x err %d\n",
+			tbl_idx, err);
+		return err;
+	}
+
+	/* Storing the seondary's lmt base address as this needs to be
+	 * reverted in FLR. Also making sure this default value doesn't
+	 * get overwritten on multiple calls to this mailbox.
+	 */
+	if (!pfvf->lmt_base_addr)
+		pfvf->lmt_base_addr = val;
+
+	/* Update the LMT table with new addr */
+	err = lmtst_map_table_ops(rvu, tbl_idx, &lmt_addr, LMT_TBL_OP_WRITE);
+	if (err) {
+		dev_err(rvu->dev,
+			"Failed to update LMT map table: index 0x%x err %d\n",
+			tbl_idx, err);
+		return err;
+	}
+	return 0;
+}
+
 int rvu_mbox_handler_lmtst_tbl_setup(struct rvu *rvu,
 				     struct lmtst_tbl_setup_req *req,
 				     struct msg_rsp *rsp)
 {
-	struct rvu_pfvf *pfvf = rvu_get_pfvf(rvu, req->hdr.pcifunc);
-	u32 pri_tbl_idx, sec_tbl_idx;
+	u64 lmt_addr, val;
+	u32 pri_tbl_idx;
 	int err = 0;
-	u64 val;
+
+	/* Check if PF_FUNC wants to use it's own local memory as LMTLINE
+	 * region, if so, convert that IOVA to physical address and
+	 * populate LMT table with that address
+	 */
+	if (req->use_local_lmt_region) {
+		err = rvu_get_lmtaddr(rvu, req->hdr.pcifunc,
+				      req->lmt_iova, &lmt_addr);
+		if (err < 0)
+			return err;
+
+		/* Update the lmt addr for this PFFUNC in the LMT table */
+		err = rvu_update_lmtaddr(rvu, req->hdr.pcifunc, lmt_addr);
+		if (err)
+			return err;
+	}
 
 	/* Reconfiguring lmtst map table in lmt region shared mode i.e. make
 	 * multiple PF_FUNCs to share an LMTLINE region, so primary/base
@@ -76,27 +163,6 @@ int rvu_mbox_handler_lmtst_tbl_setup(struct rvu *rvu,
 		 */
 		pri_tbl_idx = rvu_get_lmtst_tbl_index(rvu, req->base_pcifunc);
 
-		/* Truncating secondary pcifunc to calculate the LMT table index
-		 * equivalent to secondary pcifunc.
-		 */
-		sec_tbl_idx = rvu_get_lmtst_tbl_index(rvu, req->hdr.pcifunc);
-		/* Read the base lmt addr of the secondary pcifunc */
-		err = lmtst_map_table_ops(rvu, sec_tbl_idx, &val,
-					  LMT_TBL_OP_READ);
-		if (err) {
-			dev_err(rvu->dev,
-				"Failed to read LMT map table: index 0x%x err %d\n",
-				sec_tbl_idx, err);
-			goto error;
-		}
-
-		/* Storing the seondary's lmt base address as this needs to be
-		 * reverted in FLR. Also making sure this default value doesn't
-		 * get overwritten on multiple calls to this mailbox.
-		 */
-		if (!pfvf->lmt_base_addr)
-			pfvf->lmt_base_addr = val;
-
 		/* Read the base lmt addr of the primary pcifunc */
 		err = lmtst_map_table_ops(rvu, pri_tbl_idx, &val,
 					  LMT_TBL_OP_READ);
@@ -104,24 +170,18 @@ int rvu_mbox_handler_lmtst_tbl_setup(struct rvu *rvu,
 			dev_err(rvu->dev,
 				"Failed to read LMT map table: index 0x%x err %d\n",
 				pri_tbl_idx, err);
-			goto error;
+			return err;
 		}
 
 		/* Update the base lmt addr of secondary with primary's base
 		 * lmt addr.
 		 */
-		err = lmtst_map_table_ops(rvu, sec_tbl_idx, &val,
-					  LMT_TBL_OP_WRITE);
-		if (err) {
-			dev_err(rvu->dev,
-				"Failed to update LMT map table: index 0x%x err %d\n",
-				sec_tbl_idx, err);
-			goto error;
-		}
+		err = rvu_update_lmtaddr(rvu, req->hdr.pcifunc, val);
+		if (err)
+			return err;
 	}
 
-error:
-	return err;
+	return 0;
 }
 
 /* Resetting the lmtst map table to original base addresses */
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h b/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h
index 61bafe956aae..8b01ef6e2c99 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h
@@ -49,6 +49,11 @@
 #define RVU_AF_PFX_VF_BAR4_ADDR             (0x5400 | (a) << 4)
 #define RVU_AF_PFX_VF_BAR4_CFG              (0x5600 | (a) << 4)
 #define RVU_AF_PFX_LMTLINE_ADDR             (0x5800 | (a) << 4)
+#define RVU_AF_SMMU_ADDR_REQ		    (0x6000)
+#define RVU_AF_SMMU_TXN_REQ		    (0x6008)
+#define RVU_AF_SMMU_ADDR_RSP_STS	    (0x6010)
+#define RVU_AF_SMMU_ADDR_TLN		    (0x6018)
+#define RVU_AF_SMMU_TLN_FLIT1		    (0x6030)
 
 /* Admin function's privileged PF/VF registers */
 #define RVU_PRIV_CONST                      (0x8000000)
-- 
2.17.1

[net-next PATCH 3/3] octeontx2-pf: cn10k: Use runtime allocated LMTLINE region

[Geetha sowjanya]

The current driver uses static LMTST region allocated by firmware.
This memory gets populated as PF/VF BAR2. RVU PF/VF driver ioremap
the memory as device memory for NIX/NPA operation. Since the memory
is mapped as device memory we see performance degration. To address
this issue this patch implements runtime memory allocation.
RVU PF/VF allocates memory during device probe and share the base
address with RVU AF. RVU AF then configure the LMT MAP table
accordingly.

Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
---
 .../ethernet/marvell/octeontx2/nic/cn10k.c    | 87 ++++++++-----------
 .../ethernet/marvell/octeontx2/nic/cn10k.h    |  3 +-
 .../marvell/octeontx2/nic/otx2_common.h       |  7 +-
 .../ethernet/marvell/octeontx2/nic/otx2_pf.c  | 17 ++--
 .../marvell/octeontx2/nic/otx2_txrx.h         |  1 +
 .../ethernet/marvell/octeontx2/nic/otx2_vf.c  | 12 ++-
 6 files changed, 54 insertions(+), 73 deletions(-)

diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c
index 1b08896b46d2..184de9466286 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c
@@ -22,69 +22,52 @@ static struct dev_hw_ops cn10k_hw_ops = {
 	.refill_pool_ptrs = cn10k_refill_pool_ptrs,
 };
 
-int cn10k_pf_lmtst_init(struct otx2_nic *pf)
+int cn10k_lmtst_init(struct otx2_nic *pfvf)
 {
-	int size, num_lines;
-	u64 base;
 
-	if (!test_bit(CN10K_LMTST, &pf->hw.cap_flag)) {
-		pf->hw_ops = &otx2_hw_ops;
+	struct lmtst_tbl_setup_req *req;
+	int qcount, err;
+
+	if (!test_bit(CN10K_LMTST, &pfvf->hw.cap_flag)) {
+		pfvf->hw_ops = &otx2_hw_ops;
 		return 0;
 	}
 
-	pf->hw_ops = &cn10k_hw_ops;
-	base = pci_resource_start(pf->pdev, PCI_MBOX_BAR_NUM) +
-		       (MBOX_SIZE * (pf->total_vfs + 1));
-
-	size = pci_resource_len(pf->pdev, PCI_MBOX_BAR_NUM) -
-	       (MBOX_SIZE * (pf->total_vfs + 1));
-
-	pf->hw.lmt_base = ioremap(base, size);
+	pfvf->hw_ops = &cn10k_hw_ops;
+	qcount = pfvf->hw.max_queues;
+	/* LMTST lines allocation
+	 * qcount = num_online_cpus();
+	 * NPA = TX + RX + XDP.
+	 * NIX = TX * 32 (For Burst SQE flush).
+	 */
+	pfvf->tot_lmt_lines = (qcount * 3) + (qcount * 32);
+	pfvf->npa_lmt_lines = qcount * 3;
+	pfvf->nix_lmt_size =  LMT_BURST_SIZE * LMT_LINE_SIZE;
 
-	if (!pf->hw.lmt_base) {
-		dev_err(pf->dev, "Unable to map PF LMTST region\n");
+	mutex_lock(&pfvf->mbox.lock);
+	req = otx2_mbox_alloc_msg_lmtst_tbl_setup(&pfvf->mbox);
+	if (!req) {
+		mutex_unlock(&pfvf->mbox.lock);
 		return -ENOMEM;
 	}
 
-	/* FIXME: Get the num of LMTST lines from LMT table */
-	pf->tot_lmt_lines = size / LMT_LINE_SIZE;
-	num_lines = (pf->tot_lmt_lines - NIX_LMTID_BASE) /
-			    pf->hw.tx_queues;
-	/* Number of LMT lines per SQ queues */
-	pf->nix_lmt_lines = num_lines > 32 ? 32 : num_lines;
-
-	pf->nix_lmt_size = pf->nix_lmt_lines * LMT_LINE_SIZE;
-	return 0;
-}
+	req->use_local_lmt_region = true;
 
-int cn10k_vf_lmtst_init(struct otx2_nic *vf)
-{
-	int size, num_lines;
-
-	if (!test_bit(CN10K_LMTST, &vf->hw.cap_flag)) {
-		vf->hw_ops = &otx2_hw_ops;
-		return 0;
+	err = qmem_alloc(pfvf->dev, &pfvf->dync_lmt, pfvf->tot_lmt_lines,
+			 LMT_LINE_SIZE);
+	if (err) {
+		mutex_unlock(&pfvf->mbox.lock);
+		return err;
 	}
+	pfvf->hw.lmt_base = (u64 *)pfvf->dync_lmt->base;
+	req->lmt_iova = (u64)pfvf->dync_lmt->iova;
 
-	vf->hw_ops = &cn10k_hw_ops;
-	size = pci_resource_len(vf->pdev, PCI_MBOX_BAR_NUM);
-	vf->hw.lmt_base = ioremap_wc(pci_resource_start(vf->pdev,
-							PCI_MBOX_BAR_NUM),
-				     size);
-	if (!vf->hw.lmt_base) {
-		dev_err(vf->dev, "Unable to map VF LMTST region\n");
-		return -ENOMEM;
-	}
+	err = otx2_sync_mbox_msg(&pfvf->mbox);
+	mutex_unlock(&pfvf->mbox.lock);
 
-	vf->tot_lmt_lines = size / LMT_LINE_SIZE;
-	/* LMTST lines per SQ */
-	num_lines = (vf->tot_lmt_lines - NIX_LMTID_BASE) /
-			    vf->hw.tx_queues;
-	vf->nix_lmt_lines = num_lines > 32 ? 32 : num_lines;
-	vf->nix_lmt_size = vf->nix_lmt_lines * LMT_LINE_SIZE;
 	return 0;
 }
-EXPORT_SYMBOL(cn10k_vf_lmtst_init);
+EXPORT_SYMBOL(cn10k_lmtst_init);
 
 int cn10k_sq_aq_init(void *dev, u16 qidx, u16 sqb_aura)
 {
@@ -93,9 +76,11 @@ int cn10k_sq_aq_init(void *dev, u16 qidx, u16 sqb_aura)
 	struct otx2_snd_queue *sq;
 
 	sq = &pfvf->qset.sq[qidx];
-	sq->lmt_addr = (__force u64 *)((u64)pfvf->hw.nix_lmt_base +
+	sq->lmt_addr = (u64 *)((u64)pfvf->hw.nix_lmt_base +
 			       (qidx * pfvf->nix_lmt_size));
 
+	sq->lmt_id = pfvf->npa_lmt_lines + (qidx * LMT_BURST_SIZE);
+
 	/* Get memory to put this msg */
 	aq = otx2_mbox_alloc_msg_nix_cn10k_aq_enq(&pfvf->mbox);
 	if (!aq)
@@ -158,15 +143,13 @@ void cn10k_refill_pool_ptrs(void *dev, struct otx2_cq_queue *cq)
 
 void cn10k_sqe_flush(void *dev, struct otx2_snd_queue *sq, int size, int qidx)
 {
-	struct otx2_nic *pfvf = dev;
-	int lmt_id = NIX_LMTID_BASE + (qidx * pfvf->nix_lmt_lines);
 	u64 val = 0, tar_addr = 0;
 
 	/* FIXME: val[0:10] LMT_ID.
 	 * [12:15] no of LMTST - 1 in the burst.
 	 * [19:63] data size of each LMTST in the burst except first.
 	 */
-	val = (lmt_id & 0x7FF);
+	val = (sq->lmt_id & 0x7FF);
 	/* Target address for LMTST flush tells HW how many 128bit
 	 * words are present.
 	 * tar_addr[6:4] size of first LMTST - 1 in units of 128b.
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.h b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.h
index 71292a4cf1f3..1a1ae334477d 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.h
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.h
@@ -12,8 +12,7 @@
 void cn10k_refill_pool_ptrs(void *dev, struct otx2_cq_queue *cq);
 void cn10k_sqe_flush(void *dev, struct otx2_snd_queue *sq, int size, int qidx);
 int cn10k_sq_aq_init(void *dev, u16 qidx, u16 sqb_aura);
-int cn10k_pf_lmtst_init(struct otx2_nic *pf);
-int cn10k_vf_lmtst_init(struct otx2_nic *vf);
+int cn10k_lmtst_init(struct otx2_nic *pfvf);
 int cn10k_free_all_ipolicers(struct otx2_nic *pfvf);
 int cn10k_alloc_matchall_ipolicer(struct otx2_nic *pfvf);
 int cn10k_free_matchall_ipolicer(struct otx2_nic *pfvf);
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h
index 234b330f3183..20a9c69f020f 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h
@@ -218,8 +218,8 @@ struct otx2_hw {
 	unsigned long		cap_flag;
 
 #define LMT_LINE_SIZE		128
-#define NIX_LMTID_BASE		72 /* RX + TX + XDP */
-	void __iomem		*lmt_base;
+#define LMT_BURST_SIZE		32 /* 32 LMTST lines for burst SQE flush */
+	u64			*lmt_base;
 	u64			*npa_lmt_base;
 	u64			*nix_lmt_base;
 };
@@ -363,8 +363,9 @@ struct otx2_nic {
 	/* Block address of NIX either BLKADDR_NIX0 or BLKADDR_NIX1 */
 	int			nix_blkaddr;
 	/* LMTST Lines info */
+	struct qmem		*dync_lmt;
 	u16			tot_lmt_lines;
-	u16			nix_lmt_lines;
+	u16			npa_lmt_lines;
 	u32			nix_lmt_size;
 
 	struct otx2_ptp		*ptp;
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c
index 59912f73417b..088c28df849d 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c
@@ -1533,10 +1533,10 @@ int otx2_open(struct net_device *netdev)
 
 	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag)) {
 		/* Reserve LMT lines for NPA AURA batch free */
-		pf->hw.npa_lmt_base = (__force u64 *)pf->hw.lmt_base;
+		pf->hw.npa_lmt_base = pf->hw.lmt_base;
 		/* Reserve LMT lines for NIX TX */
-		pf->hw.nix_lmt_base = (__force u64 *)((u64)pf->hw.npa_lmt_base +
-				      (NIX_LMTID_BASE * LMT_LINE_SIZE));
+		pf->hw.nix_lmt_base = (u64 *)((u64)pf->hw.npa_lmt_base +
+				      (pf->npa_lmt_lines * LMT_LINE_SIZE));
 	}
 
 	err = otx2_init_hw_resources(pf);
@@ -2526,7 +2526,7 @@ static int otx2_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (err)
 		goto err_detach_rsrc;
 
-	err = cn10k_pf_lmtst_init(pf);
+	err = cn10k_lmtst_init(pf);
 	if (err)
 		goto err_detach_rsrc;
 
@@ -2630,8 +2630,8 @@ static int otx2_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 err_ptp_destroy:
 	otx2_ptp_destroy(pf);
 err_detach_rsrc:
-	if (hw->lmt_base)
-		iounmap(hw->lmt_base);
+	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
+		qmem_free(pf->dev, pf->dync_lmt);
 	otx2_detach_resources(&pf->mbox);
 err_disable_mbox_intr:
 	otx2_disable_mbox_intr(pf);
@@ -2772,9 +2772,8 @@ static void otx2_remove(struct pci_dev *pdev)
 	otx2_mcam_flow_del(pf);
 	otx2_shutdown_tc(pf);
 	otx2_detach_resources(&pf->mbox);
-	if (pf->hw.lmt_base)
-		iounmap(pf->hw.lmt_base);
-
+	if (test_bit(CN10K_LMTST, &pf->hw.cap_flag))
+		qmem_free(pf->dev, pf->dync_lmt);
 	otx2_disable_mbox_intr(pf);
 	otx2_pfaf_mbox_destroy(pf);
 	pci_free_irq_vectors(pf->pdev);
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h
index 52486c1f0973..2f144e2cf436 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h
@@ -83,6 +83,7 @@ struct otx2_snd_queue {
 	u16			num_sqbs;
 	u16			sqe_thresh;
 	u8			sqe_per_sqb;
+	u32			lmt_id;
 	u64			 io_addr;
 	u64			*aura_fc_addr;
 	u64			*lmt_addr;
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_vf.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_vf.c
index 13a908f75ba0..a8bee5aefec1 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_vf.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_vf.c
@@ -609,7 +609,7 @@ static int otx2vf_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (err)
 		goto err_detach_rsrc;
 
-	err = cn10k_vf_lmtst_init(vf);
+	err = cn10k_lmtst_init(vf);
 	if (err)
 		goto err_detach_rsrc;
 
@@ -667,8 +667,8 @@ static int otx2vf_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 err_unreg_netdev:
 	unregister_netdev(netdev);
 err_detach_rsrc:
-	if (hw->lmt_base)
-		iounmap(hw->lmt_base);
+	if (test_bit(CN10K_LMTST, &vf->hw.cap_flag))
+		qmem_free(vf->dev, vf->dync_lmt);
 	otx2_detach_resources(&vf->mbox);
 err_disable_mbox_intr:
 	otx2vf_disable_mbox_intr(vf);
@@ -700,10 +700,8 @@ static void otx2vf_remove(struct pci_dev *pdev)
 		destroy_workqueue(vf->otx2_wq);
 	otx2vf_disable_mbox_intr(vf);
 	otx2_detach_resources(&vf->mbox);
-
-	if (vf->hw.lmt_base)
-		iounmap(vf->hw.lmt_base);
-
+	if (test_bit(CN10K_LMTST, &vf->hw.cap_flag))
+		qmem_free(vf->dev, vf->dync_lmt);
 	otx2vf_vfaf_mbox_destroy(vf);
 	pci_free_irq_vectors(vf->pdev);
 	pci_set_drvdata(pdev, NULL);
-- 
2.17.1

Re: [net-next PATCH 0/3] Dynamic LMTST region setup

[patchwork-bot+netdevbpf]

Hello:

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

On Tue, 29 Jun 2021 22:30:03 +0530 you wrote:
> This patch series allows RVU PF/VF to allocate memory for
> LMTST operations instead of using memory reserved by firmware
> which is mapped as device memory.
> The LMTST mapping table contains the RVU PF/VF LMTST memory base
> address entries. This table is used by hardware for LMTST operations.
> Patch1 introduces new mailbox message to update the LMTST table with
> the new allocated memory address.
> 
> [...]

Here is the summary with links:
  - [net-next,1/3] octeontx2-af: cn10k: Setting up lmtst map table
    https://git.kernel.org/netdev/net-next/c/873a1e3d207a
  - [net-next,2/3] octeontx2-af: cn10k: Support configurable LMTST regions
    https://git.kernel.org/netdev/net-next/c/893ae97214c3
  - [net-next,3/3] octeontx2-pf: cn10k: Use runtime allocated LMTLINE region
    https://git.kernel.org/netdev/net-next/c/5c0512072f65

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