octeontx2-af: configure default CPT credits for CN10KA B0

The maximum CPT credits that RXC can use are now configurable on CN10KA B0
through a hardware CSR. This patch sets the default value to optimize peak
performance, aligning it with other chip versions.
Signed-off-by: Srujana Challa <schalla@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

@@ -691,6 +691,26 @@ static inline bool is_cnf10ka_a0(struct rvu *rvu)
 	return false;
 }
 
+static inline bool is_cn10ka_a0(struct rvu *rvu)
+{
+	struct pci_dev *pdev = rvu->pdev;
+
+	if (pdev->subsystem_device == PCI_SUBSYS_DEVID_CN10K_A &&
+	    (pdev->revision & 0x0F) == 0x0)
+		return true;
+	return false;
+}
+
+static inline bool is_cn10ka_a1(struct rvu *rvu)
+{
+	struct pci_dev *pdev = rvu->pdev;
+
+	if (pdev->subsystem_device == PCI_SUBSYS_DEVID_CN10K_A &&
+	    (pdev->revision & 0x0F) == 0x1)
+		return true;
+	return false;
+}
+
 static inline bool is_cn10kb(struct rvu *rvu)
 {
 	struct pci_dev *pdev = rvu->pdev;

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

@@ -22,6 +22,9 @@
 /* Interrupt vector count of CPT RVU and RAS interrupts */
 #define CPT_10K_AF_RVU_RAS_INT_VEC_CNT  2
 
+/* Default CPT_AF_RXC_CFG1:max_rxc_icb_cnt */
+#define CPT_DFLT_MAX_RXC_ICB_CNT  0xC0ULL
+
 #define cpt_get_eng_sts(e_min, e_max, rsp, etype)                   \
 ({                                                                  \
 	u64 free_sts = 0, busy_sts = 0;                             \
@@ -737,6 +740,7 @@ static bool validate_and_update_reg_offset(struct rvu *rvu,
 		case CPT_AF_BLK_RST:
 		case CPT_AF_CONSTANTS1:
 		case CPT_AF_CTX_FLUSH_TIMER:
+		case CPT_AF_RXC_CFG1:
 			return true;
 		}
 
@@ -1285,9 +1289,12 @@ int rvu_cpt_ctx_flush(struct rvu *rvu, u16 pcifunc)
 	return 0;
 }
 
+#define MAX_RXC_ICB_CNT  GENMASK_ULL(40, 32)
+
 int rvu_cpt_init(struct rvu *rvu)
 {
 	struct rvu_hwinfo *hw = rvu->hw;
+	u64 reg_val;
 
 	/* Retrieve CPT PF number */
 	rvu->cpt_pf_num = get_cpt_pf_num(rvu);
@@ -1295,6 +1302,17 @@ int rvu_cpt_init(struct rvu *rvu)
 	    !is_cn10kb(rvu))
 		hw->cap.cpt_rxc = true;
 
+	if (hw->cap.cpt_rxc && !is_cn10ka_a0(rvu) && !is_cn10ka_a1(rvu)) {
+		/* Set CPT_AF_RXC_CFG1:max_rxc_icb_cnt to 0xc0 to not effect
+		 * inline inbound peak performance
+		 */
+		reg_val = rvu_read64(rvu, BLKADDR_CPT0, CPT_AF_RXC_CFG1);
+		reg_val &= ~MAX_RXC_ICB_CNT;
+		reg_val |= FIELD_PREP(MAX_RXC_ICB_CNT,
+				      CPT_DFLT_MAX_RXC_ICB_CNT);
+		rvu_write64(rvu, BLKADDR_CPT0, CPT_AF_RXC_CFG1, reg_val);
+	}
+
 	spin_lock_init(&rvu->cpt_intr_lock);
 
 	return 0;

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

@@ -545,6 +545,7 @@
 #define CPT_AF_CTX_PSH_PC               (0x49450ull)
 #define CPT_AF_CTX_PSH_LATENCY_PC       (0x49458ull)
 #define CPT_AF_CTX_CAM_DATA(a)          (0x49800ull | (u64)(a) << 3)
+#define CPT_AF_RXC_CFG1                 (0x50000ull)
 #define CPT_AF_RXC_TIME                 (0x50010ull)
 #define CPT_AF_RXC_TIME_CFG             (0x50018ull)
 #define CPT_AF_RXC_DFRG                 (0x50020ull)