Merge branch 'octeontx2-af-miscellaneous-changes-for-cpt'

Srujana Challa says:

====================
octeontx2-af: Miscellaneous changes for CPT

This patchset consists of miscellaneous changes for CPT.
- Adds a new mailbox to reset the requested CPT LF.
- Modify FLR sequence as per HW team suggested.
- Adds support to recover CPT engines when they gets fault.
- Updates CPT inbound inline IPsec configuration mailbox,
  as per new generation of the OcteonTX2 chips.
- Adds a new mailbox to return CPT FLT Interrupt info.
====================

Link: https://lore.kernel.org/r/20230118120354.1017961-1-schalla@marvell.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/marvell/octeontx2/af/mbox.h

@@ -195,6 +195,9 @@ M(CPT_STATS,            0xA05, cpt_sts, cpt_sts_req, cpt_sts_rsp)	\
 M(CPT_RXC_TIME_CFG,     0xA06, cpt_rxc_time_cfg, cpt_rxc_time_cfg_req,  \
 			       msg_rsp)                                 \
 M(CPT_CTX_CACHE_SYNC,   0xA07, cpt_ctx_cache_sync, msg_req, msg_rsp)    \
+M(CPT_LF_RESET,         0xA08, cpt_lf_reset, cpt_lf_rst_req, msg_rsp)	\
+M(CPT_FLT_ENG_INFO,     0xA09, cpt_flt_eng_info, cpt_flt_eng_info_req,	\
+			       cpt_flt_eng_info_rsp)			\
 /* SDP mbox IDs (range 0x1000 - 0x11FF) */				\
 M(SET_SDP_CHAN_INFO, 0x1000, set_sdp_chan_info, sdp_chan_info_msg, msg_rsp) \
 M(GET_SDP_CHAN_INFO, 0x1001, get_sdp_chan_info, msg_req, sdp_get_chan_info_msg) \
@@ -1613,6 +1616,8 @@ struct cpt_lf_alloc_req_msg {
 	u16 sso_pf_func;
 	u16 eng_grpmsk;
 	int blkaddr;
+	u8 ctx_ilen_valid : 1;
+	u8 ctx_ilen : 7;
 };
 
 #define CPT_INLINE_INBOUND      0
@@ -1696,6 +1701,28 @@ struct cpt_inst_lmtst_req {
 	u64 rsvd;
 };
 
+/* Mailbox message format to request for CPT LF reset */
+struct cpt_lf_rst_req {
+	struct mbox_msghdr hdr;
+	u32 slot;
+	u32 rsvd;
+};
+
+/* Mailbox message format to request for CPT faulted engines */
+struct cpt_flt_eng_info_req {
+	struct mbox_msghdr hdr;
+	int blkaddr;
+	bool reset;
+	u32 rsvd;
+};
+
+struct cpt_flt_eng_info_rsp {
+	struct mbox_msghdr hdr;
+	u64 flt_eng_map[CPT_10K_AF_INT_VEC_RVU];
+	u64 rcvrd_eng_map[CPT_10K_AF_INT_VEC_RVU];
+	u64 rsvd;
+};
+
 struct sdp_node_info {
 	/* Node to which this PF belons to */
 	u8 node_id;

drivers/net/ethernet/marvell/octeontx2/af/rvu.c

@@ -1164,8 +1164,16 @@ static int rvu_setup_hw_resources(struct rvu *rvu)
 		goto nix_err;
 	}
 
+	err = rvu_cpt_init(rvu);
+	if (err) {
+		dev_err(rvu->dev, "%s: Failed to initialize cpt\n", __func__);
+		goto mcs_err;
+	}
+
 	return 0;
 
+mcs_err:
+	rvu_mcs_exit(rvu);
 nix_err:
 	rvu_nix_freemem(rvu);
 npa_err:

drivers/net/ethernet/marvell/octeontx2/af/rvu.h

@@ -108,6 +108,8 @@ struct rvu_block {
 	u64  lfreset_reg;
 	unsigned char name[NAME_SIZE];
 	struct rvu *rvu;
+	u64 cpt_flt_eng_map[3];
+	u64 cpt_rcvrd_eng_map[3];
 };
 
 struct nix_mcast {
@@ -459,6 +461,7 @@ struct rvu {
 	struct rvu_pfvf		*pf;
 	struct rvu_pfvf		*hwvf;
 	struct mutex		rsrc_lock; /* Serialize resource alloc/free */
+	struct mutex		alias_lock; /* Serialize bar2 alias access */
 	int			vfs; /* Number of VFs attached to RVU */
 	int			nix_blkaddr[MAX_NIX_BLKS];
 
@@ -510,6 +513,7 @@ struct rvu {
 	struct ptp		*ptp;
 
 	int			mcs_blk_cnt;
+	int			cpt_pf_num;
 
 #ifdef CONFIG_DEBUG_FS
 	struct rvu_debugfs	rvu_dbg;
@@ -524,6 +528,8 @@ struct rvu {
 	struct list_head	mcs_intrq_head;
 	/* mcs interrupt queue lock */
 	spinlock_t		mcs_intrq_lock;
+	/* CPT interrupt lock */
+	spinlock_t		cpt_intr_lock;
 };
 
 static inline void rvu_write64(struct rvu *rvu, u64 block, u64 offset, u64 val)
@@ -546,6 +552,17 @@ static inline u64 rvupf_read64(struct rvu *rvu, u64 offset)
 	return readq(rvu->pfreg_base + offset);
 }
 
+static inline void rvu_bar2_sel_write64(struct rvu *rvu, u64 block, u64 offset, u64 val)
+{
+	/* HW requires read back of RVU_AF_BAR2_SEL register to make sure completion of
+	 * write operation.
+	 */
+	rvu_write64(rvu, block, offset, val);
+	rvu_read64(rvu, block, offset);
+	/* Barrier to ensure read completes before accessing LF registers */
+	mb();
+}
+
 /* Silicon revisions */
 static inline bool is_rvu_pre_96xx_C0(struct rvu *rvu)
 {
@@ -865,6 +882,7 @@ void rvu_cpt_unregister_interrupts(struct rvu *rvu);
 int rvu_cpt_lf_teardown(struct rvu *rvu, u16 pcifunc, int blkaddr, int lf,
 			int slot);
 int rvu_cpt_ctx_flush(struct rvu *rvu, u16 pcifunc);
+int rvu_cpt_init(struct rvu *rvu);
 
 /* CN10K RVU */
 int rvu_set_channels_base(struct rvu *rvu);

drivers/net/ethernet/marvell/octeontx2/af/rvu_cpt.c

@@ -17,7 +17,7 @@
 #define	PCI_DEVID_OTX2_CPT10K_PF 0xA0F2
 
 /* Length of initial context fetch in 128 byte words */
-#define CPT_CTX_ILEN    2ULL
+#define CPT_CTX_ILEN    1ULL
 
 #define cpt_get_eng_sts(e_min, e_max, rsp, etype)                   \
 ({                                                                  \
@@ -37,34 +37,68 @@
 	(_rsp)->free_sts_##etype = free_sts;                        \
 })
 
-static irqreturn_t rvu_cpt_af_flt_intr_handler(int irq, void *ptr)
+static irqreturn_t cpt_af_flt_intr_handler(int vec, void *ptr)
 {
 	struct rvu_block *block = ptr;
 	struct rvu *rvu = block->rvu;
 	int blkaddr = block->addr;
-	u64 reg0, reg1, reg2;
-
-	reg0 = rvu_read64(rvu, blkaddr, CPT_AF_FLTX_INT(0));
-	reg1 = rvu_read64(rvu, blkaddr, CPT_AF_FLTX_INT(1));
-	if (!is_rvu_otx2(rvu)) {
-		reg2 = rvu_read64(rvu, blkaddr, CPT_AF_FLTX_INT(2));
-		dev_err_ratelimited(rvu->dev,
-				    "Received CPTAF FLT irq : 0x%llx, 0x%llx, 0x%llx",
-				     reg0, reg1, reg2);
-	} else {
-		dev_err_ratelimited(rvu->dev,
-				    "Received CPTAF FLT irq : 0x%llx, 0x%llx",
-				     reg0, reg1);
+	u64 reg, val;
+	int i, eng;
+	u8 grp;
+
+	reg = rvu_read64(rvu, blkaddr, CPT_AF_FLTX_INT(vec));
+	dev_err_ratelimited(rvu->dev, "Received CPTAF FLT%d irq : 0x%llx", vec, reg);
+
+	i = -1;
+	while ((i = find_next_bit((unsigned long *)&reg, 64, i + 1)) < 64) {
+		switch (vec) {
+		case 0:
+			eng = i;
+			break;
+		case 1:
+			eng = i + 64;
+			break;
+		case 2:
+			eng = i + 128;
+			break;
+		}
+		grp = rvu_read64(rvu, blkaddr, CPT_AF_EXEX_CTL2(eng)) & 0xFF;
+		/* Disable and enable the engine which triggers fault */
+		rvu_write64(rvu, blkaddr, CPT_AF_EXEX_CTL2(eng), 0x0);
+		val = rvu_read64(rvu, blkaddr, CPT_AF_EXEX_CTL(eng));
+		rvu_write64(rvu, blkaddr, CPT_AF_EXEX_CTL(eng), val & ~1ULL);
+
+		rvu_write64(rvu, blkaddr, CPT_AF_EXEX_CTL2(eng), grp);
+		rvu_write64(rvu, blkaddr, CPT_AF_EXEX_CTL(eng), val | 1ULL);
+
+		spin_lock(&rvu->cpt_intr_lock);
+		block->cpt_flt_eng_map[vec] |= BIT_ULL(i);
+		val = rvu_read64(rvu, blkaddr, CPT_AF_EXEX_STS(eng));
+		val = val & 0x3;
+		if (val == 0x1 || val == 0x2)
+			block->cpt_rcvrd_eng_map[vec] |= BIT_ULL(i);
+		spin_unlock(&rvu->cpt_intr_lock);
 	}
-
-	rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT(0), reg0);
-	rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT(1), reg1);
-	if (!is_rvu_otx2(rvu))
-		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT(2), reg2);
+	rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT(vec), reg);
 
 	return IRQ_HANDLED;
 }
 
+static irqreturn_t rvu_cpt_af_flt0_intr_handler(int irq, void *ptr)
+{
+	return cpt_af_flt_intr_handler(CPT_AF_INT_VEC_FLT0, ptr);
+}
+
+static irqreturn_t rvu_cpt_af_flt1_intr_handler(int irq, void *ptr)
+{
+	return cpt_af_flt_intr_handler(CPT_AF_INT_VEC_FLT1, ptr);
+}
+
+static irqreturn_t rvu_cpt_af_flt2_intr_handler(int irq, void *ptr)
+{
+	return cpt_af_flt_intr_handler(CPT_10K_AF_INT_VEC_FLT2, ptr);
+}
+
 static irqreturn_t rvu_cpt_af_rvu_intr_handler(int irq, void *ptr)
 {
 	struct rvu_block *block = ptr;
@@ -119,8 +153,10 @@ static void cpt_10k_unregister_interrupts(struct rvu_block *block, int off)
 	int i;
 
 	/* Disable all CPT AF interrupts */
-	for (i = 0; i < CPT_10K_AF_INT_VEC_RVU; i++)
-		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1C(i), 0x1);
+	rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1C(0), ~0ULL);
+	rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1C(1), ~0ULL);
+	rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1C(2), 0xFFFF);
+
 	rvu_write64(rvu, blkaddr, CPT_AF_RVU_INT_ENA_W1C, 0x1);
 	rvu_write64(rvu, blkaddr, CPT_AF_RAS_INT_ENA_W1C, 0x1);
 
@@ -151,7 +187,7 @@ static void cpt_unregister_interrupts(struct rvu *rvu, int blkaddr)
 
 	/* Disable all CPT AF interrupts */
 	for (i = 0; i < CPT_AF_INT_VEC_RVU; i++)
-		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1C(i), 0x1);
+		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1C(i), ~0ULL);
 	rvu_write64(rvu, blkaddr, CPT_AF_RVU_INT_ENA_W1C, 0x1);
 	rvu_write64(rvu, blkaddr, CPT_AF_RAS_INT_ENA_W1C, 0x1);
 
@@ -172,16 +208,31 @@ static int cpt_10k_register_interrupts(struct rvu_block *block, int off)
 {
 	struct rvu *rvu = block->rvu;
 	int blkaddr = block->addr;
+	irq_handler_t flt_fn;
 	int i, ret;
 
 	for (i = CPT_10K_AF_INT_VEC_FLT0; i < CPT_10K_AF_INT_VEC_RVU; i++) {
 		sprintf(&rvu->irq_name[(off + i) * NAME_SIZE], "CPTAF FLT%d", i);
+
+		switch (i) {
+		case CPT_10K_AF_INT_VEC_FLT0:
+			flt_fn = rvu_cpt_af_flt0_intr_handler;
+			break;
+		case CPT_10K_AF_INT_VEC_FLT1:
+			flt_fn = rvu_cpt_af_flt1_intr_handler;
+			break;
+		case CPT_10K_AF_INT_VEC_FLT2:
+			flt_fn = rvu_cpt_af_flt2_intr_handler;
+			break;
+		}
 		ret = rvu_cpt_do_register_interrupt(block, off + i,
-						    rvu_cpt_af_flt_intr_handler,
-						    &rvu->irq_name[(off + i) * NAME_SIZE]);
+						    flt_fn, &rvu->irq_name[(off + i) * NAME_SIZE]);
 		if (ret)
 			goto err;
-		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1S(i), 0x1);
+		if (i == CPT_10K_AF_INT_VEC_FLT2)
+			rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1S(i), 0xFFFF);
+		else
+			rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1S(i), ~0ULL);
 	}
 
 	ret = rvu_cpt_do_register_interrupt(block, off + CPT_10K_AF_INT_VEC_RVU,
@@ -208,8 +259,8 @@ static int cpt_register_interrupts(struct rvu *rvu, int blkaddr)
 {
 	struct rvu_hwinfo *hw = rvu->hw;
 	struct rvu_block *block;
+	irq_handler_t flt_fn;
 	int i, offs, ret = 0;
-	char irq_name[16];
 
 	if (!is_block_implemented(rvu->hw, blkaddr))
 		return 0;
@@ -226,13 +277,20 @@ static int cpt_register_interrupts(struct rvu *rvu, int blkaddr)
 		return cpt_10k_register_interrupts(block, offs);
 
 	for (i = CPT_AF_INT_VEC_FLT0; i < CPT_AF_INT_VEC_RVU; i++) {
-		snprintf(irq_name, sizeof(irq_name), "CPTAF FLT%d", i);
+		sprintf(&rvu->irq_name[(offs + i) * NAME_SIZE], "CPTAF FLT%d", i);
+		switch (i) {
+		case CPT_AF_INT_VEC_FLT0:
+			flt_fn = rvu_cpt_af_flt0_intr_handler;
+			break;
+		case CPT_AF_INT_VEC_FLT1:
+			flt_fn = rvu_cpt_af_flt1_intr_handler;
+			break;
+		}
 		ret = rvu_cpt_do_register_interrupt(block, offs + i,
-						    rvu_cpt_af_flt_intr_handler,
-						    irq_name);
+						    flt_fn, &rvu->irq_name[(offs + i) * NAME_SIZE]);
 		if (ret)
 			goto err;
-		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1S(i), 0x1);
+		rvu_write64(rvu, blkaddr, CPT_AF_FLTX_INT_ENA_W1S(i), ~0ULL);
 	}
 
 	ret = rvu_cpt_do_register_interrupt(block, offs + CPT_AF_INT_VEC_RVU,
@@ -290,7 +348,7 @@ static int get_cpt_pf_num(struct rvu *rvu)
 
 static bool is_cpt_pf(struct rvu *rvu, u16 pcifunc)
 {
-	int cpt_pf_num = get_cpt_pf_num(rvu);
+	int cpt_pf_num = rvu->cpt_pf_num;
 
 	if (rvu_get_pf(pcifunc) != cpt_pf_num)
 		return false;
@@ -302,7 +360,7 @@ static bool is_cpt_pf(struct rvu *rvu, u16 pcifunc)
 
 static bool is_cpt_vf(struct rvu *rvu, u16 pcifunc)
 {
-	int cpt_pf_num = get_cpt_pf_num(rvu);
+	int cpt_pf_num = rvu->cpt_pf_num;
 
 	if (rvu_get_pf(pcifunc) != cpt_pf_num)
 		return false;
@@ -371,8 +429,12 @@ int rvu_mbox_handler_cpt_lf_alloc(struct rvu *rvu,
 
 		/* Set CPT LF group and priority */
 		val = (u64)req->eng_grpmsk << 48 | 1;
-		if (!is_rvu_otx2(rvu))
-			val |= (CPT_CTX_ILEN << 17);
+		if (!is_rvu_otx2(rvu)) {
+			if (req->ctx_ilen_valid)
+				val |= (req->ctx_ilen << 17);
+			else
+				val |= (CPT_CTX_ILEN << 17);
+		}
 
 		rvu_write64(rvu, blkaddr, CPT_AF_LFX_CTL(cptlf), val);
 
@@ -762,10 +824,21 @@ int rvu_mbox_handler_cpt_sts(struct rvu *rvu, struct cpt_sts_req *req,
 #define RXC_ZOMBIE_COUNT  GENMASK_ULL(60, 48)
 
 static void cpt_rxc_time_cfg(struct rvu *rvu, struct cpt_rxc_time_cfg_req *req,
-			     int blkaddr)
+			     int blkaddr, struct cpt_rxc_time_cfg_req *save)
 {
 	u64 dfrg_reg;
 
+	if (save) {
+		/* Save older config */
+		dfrg_reg = rvu_read64(rvu, blkaddr, CPT_AF_RXC_DFRG);
+		save->zombie_thres = FIELD_GET(RXC_ZOMBIE_THRES, dfrg_reg);
+		save->zombie_limit = FIELD_GET(RXC_ZOMBIE_LIMIT, dfrg_reg);
+		save->active_thres = FIELD_GET(RXC_ACTIVE_THRES, dfrg_reg);
+		save->active_limit = FIELD_GET(RXC_ACTIVE_LIMIT, dfrg_reg);
+
+		save->step = rvu_read64(rvu, blkaddr, CPT_AF_RXC_TIME_CFG);
+	}
+
 	dfrg_reg = FIELD_PREP(RXC_ZOMBIE_THRES, req->zombie_thres);
 	dfrg_reg |= FIELD_PREP(RXC_ZOMBIE_LIMIT, req->zombie_limit);
 	dfrg_reg |= FIELD_PREP(RXC_ACTIVE_THRES, req->active_thres);
@@ -790,7 +863,7 @@ int rvu_mbox_handler_cpt_rxc_time_cfg(struct rvu *rvu,
 	    !is_cpt_vf(rvu, req->hdr.pcifunc))
 		return CPT_AF_ERR_ACCESS_DENIED;
 
-	cpt_rxc_time_cfg(rvu, req, blkaddr);
+	cpt_rxc_time_cfg(rvu, req, blkaddr, NULL);
 
 	return 0;
 }
@@ -801,9 +874,67 @@ int rvu_mbox_handler_cpt_ctx_cache_sync(struct rvu *rvu, struct msg_req *req,
 	return rvu_cpt_ctx_flush(rvu, req->hdr.pcifunc);
 }
 
+int rvu_mbox_handler_cpt_lf_reset(struct rvu *rvu, struct cpt_lf_rst_req *req,
+				  struct msg_rsp *rsp)
+{
+	u16 pcifunc = req->hdr.pcifunc;
+	struct rvu_block *block;
+	int cptlf, blkaddr, ret;
+	u16 actual_slot;
+	u64 ctl, ctl2;
+
+	blkaddr = rvu_get_blkaddr_from_slot(rvu, BLKTYPE_CPT, pcifunc,
+					    req->slot, &actual_slot);
+	if (blkaddr < 0)
+		return CPT_AF_ERR_LF_INVALID;
+
+	block = &rvu->hw->block[blkaddr];
+
+	cptlf = rvu_get_lf(rvu, block, pcifunc, actual_slot);
+	if (cptlf < 0)
+		return CPT_AF_ERR_LF_INVALID;
+	ctl = rvu_read64(rvu, blkaddr, CPT_AF_LFX_CTL(cptlf));
+	ctl2 = rvu_read64(rvu, blkaddr, CPT_AF_LFX_CTL2(cptlf));
+
+	ret = rvu_lf_reset(rvu, block, cptlf);
+	if (ret)
+		dev_err(rvu->dev, "Failed to reset blkaddr %d LF%d\n",
+			block->addr, cptlf);
+
+	rvu_write64(rvu, blkaddr, CPT_AF_LFX_CTL(cptlf), ctl);
+	rvu_write64(rvu, blkaddr, CPT_AF_LFX_CTL2(cptlf), ctl2);
+
+	return 0;
+}
+
+int rvu_mbox_handler_cpt_flt_eng_info(struct rvu *rvu, struct cpt_flt_eng_info_req *req,
+				      struct cpt_flt_eng_info_rsp *rsp)
+{
+	struct rvu_block *block;
+	unsigned long flags;
+	int blkaddr, vec;
+
+	blkaddr = validate_and_get_cpt_blkaddr(req->blkaddr);
+	if (blkaddr < 0)
+		return blkaddr;
+
+	block = &rvu->hw->block[blkaddr];
+	for (vec = 0; vec < CPT_10K_AF_INT_VEC_RVU; vec++) {
+		spin_lock_irqsave(&rvu->cpt_intr_lock, flags);
+		rsp->flt_eng_map[vec] = block->cpt_flt_eng_map[vec];
+		rsp->rcvrd_eng_map[vec] = block->cpt_rcvrd_eng_map[vec];
+		if (req->reset) {
+			block->cpt_flt_eng_map[vec] = 0x0;
+			block->cpt_rcvrd_eng_map[vec] = 0x0;
+		}
+		spin_unlock_irqrestore(&rvu->cpt_intr_lock, flags);
+	}
+	return 0;
+}
+
 static void cpt_rxc_teardown(struct rvu *rvu, int blkaddr)
 {
-	struct cpt_rxc_time_cfg_req req;
+	struct cpt_rxc_time_cfg_req req, prev;
 	int timeout = 2000;
 	u64 reg;
 
@@ -819,7 +950,7 @@ static void cpt_rxc_teardown(struct rvu *rvu, int blkaddr)
 	req.active_thres = 1;
 	req.active_limit = 1;
 
-	cpt_rxc_time_cfg(rvu, &req, blkaddr);
+	cpt_rxc_time_cfg(rvu, &req, blkaddr, &prev);
 
 	do {
 		reg = rvu_read64(rvu, blkaddr, CPT_AF_RXC_ACTIVE_STS);
@@ -845,70 +976,68 @@ static void cpt_rxc_teardown(struct rvu *rvu, int blkaddr)
 
 	if (timeout == 0)
 		dev_warn(rvu->dev, "Poll for RXC zombie count hits hard loop counter\n");
+
+	/* Restore config */
+	cpt_rxc_time_cfg(rvu, &prev, blkaddr, NULL);
 }
 
-#define INPROG_INFLIGHT(reg)    ((reg) & 0x1FF)
-#define INPROG_GRB_PARTIAL(reg) ((reg) & BIT_ULL(31))
-#define INPROG_GRB(reg)         (((reg) >> 32) & 0xFF)
-#define INPROG_GWB(reg)         (((reg) >> 40) & 0xFF)
+#define INFLIGHT   GENMASK_ULL(8, 0)
+#define GRB_CNT    GENMASK_ULL(39, 32)
+#define GWB_CNT    GENMASK_ULL(47, 40)
+#define XQ_XOR     GENMASK_ULL(63, 63)
+#define DQPTR      GENMASK_ULL(19, 0)
+#define NQPTR      GENMASK_ULL(51, 32)
 
 static void cpt_lf_disable_iqueue(struct rvu *rvu, int blkaddr, int slot)
 {
-	int i = 0, hard_lp_ctr = 100000;
-	u64 inprog, grp_ptr;
-	u16 nq_ptr, dq_ptr;
+	int timeout = 1000000;
+	u64 inprog, inst_ptr;
+	u64 qsize, pending;
+	int i = 0;
 
 	/* Disable instructions enqueuing */
 	rvu_write64(rvu, blkaddr, CPT_AF_BAR2_ALIASX(slot, CPT_LF_CTL), 0x0);
 
-	/* Disable executions in the LF's queue */
 	inprog = rvu_read64(rvu, blkaddr,
 			    CPT_AF_BAR2_ALIASX(slot, CPT_LF_INPROG));
-	inprog &= ~BIT_ULL(16);
+	inprog |= BIT_ULL(16);
 	rvu_write64(rvu, blkaddr,
 		    CPT_AF_BAR2_ALIASX(slot, CPT_LF_INPROG), inprog);
 
-	/* Wait for CPT queue to become execution-quiescent */
+	qsize = rvu_read64(rvu, blkaddr,
+			   CPT_AF_BAR2_ALIASX(slot, CPT_LF_Q_SIZE)) & 0x7FFF;
 	do {
-		inprog = rvu_read64(rvu, blkaddr,
-				    CPT_AF_BAR2_ALIASX(slot, CPT_LF_INPROG));
-		if (INPROG_GRB_PARTIAL(inprog)) {
-			i = 0;
-			hard_lp_ctr--;
-		} else {
-			i++;
-		}
-
-		grp_ptr = rvu_read64(rvu, blkaddr,
-				     CPT_AF_BAR2_ALIASX(slot,
-							CPT_LF_Q_GRP_PTR));
-		nq_ptr = (grp_ptr >> 32) & 0x7FFF;
-		dq_ptr = grp_ptr & 0x7FFF;
-
-	} while (hard_lp_ctr && (i < 10) && (nq_ptr != dq_ptr));
+		inst_ptr = rvu_read64(rvu, blkaddr,
+				      CPT_AF_BAR2_ALIASX(slot, CPT_LF_Q_INST_PTR));
+		pending = (FIELD_GET(XQ_XOR, inst_ptr) * qsize * 40) +
+			  FIELD_GET(NQPTR, inst_ptr) -
+			  FIELD_GET(DQPTR, inst_ptr);
+		udelay(1);
+		timeout--;
+	} while ((pending != 0) && (timeout != 0));
 
-	if (hard_lp_ctr == 0)
-		dev_warn(rvu->dev, "CPT FLR hits hard loop counter\n");
+	if (timeout == 0)
+		dev_warn(rvu->dev, "TIMEOUT: CPT poll on pending instructions\n");
 
-	i = 0;
-	hard_lp_ctr = 100000;
+	timeout = 1000000;
+	/* Wait for CPT queue to become execution-quiescent */
 	do {
 		inprog = rvu_read64(rvu, blkaddr,
 				    CPT_AF_BAR2_ALIASX(slot, CPT_LF_INPROG));
 
-		if ((INPROG_INFLIGHT(inprog) == 0) &&
-		    (INPROG_GWB(inprog) < 40) &&
-		    ((INPROG_GRB(inprog) == 0) ||
-		     (INPROG_GRB((inprog)) == 40))) {
+		if ((FIELD_GET(INFLIGHT, inprog) == 0) &&
+		    (FIELD_GET(GRB_CNT, inprog) == 0)) {
 			i++;
 		} else {
 			i = 0;
-			hard_lp_ctr--;
+			timeout--;
 		}
-	} while (hard_lp_ctr && (i < 10));
+	} while ((timeout != 0) && (i < 10));
 
-	if (hard_lp_ctr == 0)
-		dev_warn(rvu->dev, "CPT FLR hits hard loop counter\n");
+	if (timeout == 0)
+		dev_warn(rvu->dev, "TIMEOUT: CPT poll on inflight count\n");
+	/* Wait for 2 us to flush all queue writes to memory */
+	udelay(2);
 }
 
 int rvu_cpt_lf_teardown(struct rvu *rvu, u16 pcifunc, int blkaddr, int lf, int slot)
@@ -918,18 +1047,15 @@ int rvu_cpt_lf_teardown(struct rvu *rvu, u16 pcifunc, int blkaddr, int lf, int s
 	if (is_cpt_pf(rvu, pcifunc) || is_cpt_vf(rvu, pcifunc))
 		cpt_rxc_teardown(rvu, blkaddr);
 
+	mutex_lock(&rvu->alias_lock);
 	/* Enable BAR2 ALIAS for this pcifunc. */
 	reg = BIT_ULL(16) | pcifunc;
-	rvu_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, reg);
+	rvu_bar2_sel_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, reg);
 
 	cpt_lf_disable_iqueue(rvu, blkaddr, slot);
 
-	/* Set group drop to help clear out hardware */
-	reg = rvu_read64(rvu, blkaddr, CPT_AF_BAR2_ALIASX(slot, CPT_LF_INPROG));
-	reg |= BIT_ULL(17);
-	rvu_write64(rvu, blkaddr, CPT_AF_BAR2_ALIASX(slot, CPT_LF_INPROG), reg);
-
-	rvu_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, 0);
+	rvu_bar2_sel_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, 0);
+	mutex_unlock(&rvu->alias_lock);
 
 	return 0;
 }
@@ -940,7 +1066,7 @@ int rvu_cpt_lf_teardown(struct rvu *rvu, u16 pcifunc, int blkaddr, int lf, int s
 static int cpt_inline_inb_lf_cmd_send(struct rvu *rvu, int blkaddr,
 				      int nix_blkaddr)
 {
-	int cpt_pf_num = get_cpt_pf_num(rvu);
+	int cpt_pf_num = rvu->cpt_pf_num;
 	struct cpt_inst_lmtst_req *req;
 	dma_addr_t res_daddr;
 	int timeout = 3000;
@@ -1064,7 +1190,7 @@ int rvu_cpt_ctx_flush(struct rvu *rvu, u16 pcifunc)
 
 	/* Enable BAR2 ALIAS for this pcifunc. */
 	reg = BIT_ULL(16) | pcifunc;
-	rvu_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, reg);
+	rvu_bar2_sel_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, reg);
 
 	for (i = 0; i < max_ctx_entries; i++) {
 		cam_data = rvu_read64(rvu, blkaddr, CPT_AF_CTX_CAM_DATA(i));
@@ -1077,10 +1203,19 @@ int rvu_cpt_ctx_flush(struct rvu *rvu, u16 pcifunc)
 				    reg);
 		}
 	}
-	rvu_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, 0);
+	rvu_bar2_sel_write64(rvu, blkaddr, CPT_AF_BAR2_SEL, 0);
 
 unlock:
 	mutex_unlock(&rvu->rsrc_lock);
 
 	return 0;
 }
+
+int rvu_cpt_init(struct rvu *rvu)
+{
+	/* Retrieve CPT PF number */
+	rvu->cpt_pf_num = get_cpt_pf_num(rvu);
+	spin_lock_init(&rvu->cpt_intr_lock);
+
+	return 0;
+}

drivers/net/ethernet/marvell/octeontx2/af/rvu_reg.h

@@ -545,6 +545,8 @@
 
 #define CPT_LF_CTL                      0x10
 #define CPT_LF_INPROG                   0x40
+#define CPT_LF_Q_SIZE                   0x100
+#define CPT_LF_Q_INST_PTR               0x110
 #define CPT_LF_Q_GRP_PTR                0x120
 #define CPT_LF_CTX_FLUSH                0x510