cxgb4: add support for high priority filters

T6 has a separate region known as high priority filter region
that allows classifying packets going through ULD path. So,
query firmware for HPFILTER resources and enable the high
priority offload filter support when it is available.
Signed-off-by: Shahjada Abul Husain <shahjada@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/chelsio/cxgb4/cxgb4.h

@@ -56,6 +56,7 @@
 #include <asm/io.h>
 #include "t4_chip_type.h"
 #include "cxgb4_uld.h"
+#include "t4fw_api.h"
 
 #define CH_WARN(adap, fmt, ...) dev_warn(adap->pdev_dev, fmt, ## __VA_ARGS__)
 extern struct list_head adapter_list;
@@ -68,6 +69,16 @@ extern struct mutex uld_mutex;
 #define ETHTXQ_STOP_THRES \
 	(1 + DIV_ROUND_UP((3 * MAX_SKB_FRAGS) / 2 + (MAX_SKB_FRAGS & 1), 8))
 
+#define FW_PARAM_DEV(param) \
+	(FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
+	 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
+
+#define FW_PARAM_PFVF(param) \
+	(FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
+	 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param) |  \
+	 FW_PARAMS_PARAM_Y_V(0) | \
+	 FW_PARAMS_PARAM_Z_V(0))
+
 enum {
 	MAX_NPORTS	= 4,     /* max # of ports */
 	SERNUM_LEN	= 24,    /* Serial # length */

drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c

@@ -3240,6 +3240,9 @@ static int tid_info_show(struct seq_file *seq, void *v)
 		seq_printf(seq, "SFTID range: %u..%u in use: %u\n",
 			   t->sftid_base, t->sftid_base + t->nsftids - 2,
 			   t->sftids_in_use);
+	if (t->nhpftids)
+		seq_printf(seq, "HPFTID range: %u..%u\n", t->hpftid_base,
+			   t->hpftid_base + t->nhpftids - 1);
 	if (t->ntids)
 		seq_printf(seq, "HW TID usage: %u IP users, %u IPv6 users\n",
 			   t4_read_reg(adap, LE_DB_ACT_CNT_IPV4_A),

drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c

@@ -804,6 +804,26 @@ static int setup_ppod_edram(struct adapter *adap)
 	return 0;
 }
 
+static void adap_config_hpfilter(struct adapter *adapter)
+{
+	u32 param, val = 0;
+	int ret;
+
+	/* Enable HP filter region. Older fw will fail this request and
+	 * it is fine.
+	 */
+	param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
+	ret = t4_set_params(adapter, adapter->mbox, adapter->pf, 0,
+			    1, &param, &val);
+
+	/* An error means FW doesn't know about HP filter support,
+	 * it's not a problem, don't return an error.
+	 */
+	if (ret < 0)
+		dev_err(adapter->pdev_dev,
+			"HP filter region isn't supported by FW\n");
+}
+
 /**
  *	cxgb4_write_rss - write the RSS table for a given port
  *	@pi: the port
@@ -1518,6 +1538,7 @@ static int tid_init(struct tid_info *t)
 	struct adapter *adap = container_of(t, struct adapter, tids);
 	unsigned int max_ftids = t->nftids + t->nsftids;
 	unsigned int natids = t->natids;
+	unsigned int hpftid_bmap_size;
 	unsigned int eotid_bmap_size;
 	unsigned int stid_bmap_size;
 	unsigned int ftid_bmap_size;
@@ -1525,12 +1546,15 @@ static int tid_init(struct tid_info *t)
 
 	stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
 	ftid_bmap_size = BITS_TO_LONGS(t->nftids);
+	hpftid_bmap_size = BITS_TO_LONGS(t->nhpftids);
 	eotid_bmap_size = BITS_TO_LONGS(t->neotids);
 	size = t->ntids * sizeof(*t->tid_tab) +
 	       natids * sizeof(*t->atid_tab) +
 	       t->nstids * sizeof(*t->stid_tab) +
 	       t->nsftids * sizeof(*t->stid_tab) +
 	       stid_bmap_size * sizeof(long) +
+	       t->nhpftids * sizeof(*t->hpftid_tab) +
+	       hpftid_bmap_size * sizeof(long) +
 	       max_ftids * sizeof(*t->ftid_tab) +
 	       ftid_bmap_size * sizeof(long) +
 	       t->neotids * sizeof(*t->eotid_tab) +
@@ -1543,7 +1567,9 @@ static int tid_init(struct tid_info *t)
 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
 	t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
 	t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
-	t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
+	t->hpftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
+	t->hpftid_bmap = (unsigned long *)&t->hpftid_tab[t->nhpftids];
+	t->ftid_tab = (struct filter_entry *)&t->hpftid_bmap[hpftid_bmap_size];
 	t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
 	t->eotid_tab = (struct eotid_entry *)&t->ftid_bmap[ftid_bmap_size];
 	t->eotid_bmap = (unsigned long *)&t->eotid_tab[t->neotids];
@@ -1578,6 +1604,8 @@ static int tid_init(struct tid_info *t)
 			bitmap_zero(t->eotid_bmap, t->neotids);
 	}
 
+	if (t->nhpftids)
+		bitmap_zero(t->hpftid_bmap, t->nhpftids);
 	bitmap_zero(t->ftid_bmap, t->nftids);
 	return 0;
 }
@@ -4351,6 +4379,7 @@ static int adap_init0_config(struct adapter *adapter, int reset)
 			"HMA configuration failed with error %d\n", ret);
 
 	if (is_t6(adapter->params.chip)) {
+		adap_config_hpfilter(adapter);
 		ret = setup_ppod_edram(adapter);
 		if (!ret)
 			dev_info(adapter->pdev_dev, "Successfully enabled "
@@ -4660,16 +4689,6 @@ static int adap_init0(struct adapter *adap, int vpd_skip)
 	/*
 	 * Grab some of our basic fundamental operating parameters.
 	 */
-#define FW_PARAM_DEV(param) \
-	(FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
-	FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
-
-#define FW_PARAM_PFVF(param) \
-	FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
-	FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)|  \
-	FW_PARAMS_PARAM_Y_V(0) | \
-	FW_PARAMS_PARAM_Z_V(0)
-
 	params[0] = FW_PARAM_PFVF(EQ_START);
 	params[1] = FW_PARAM_PFVF(L2T_START);
 	params[2] = FW_PARAM_PFVF(L2T_END);
@@ -4687,6 +4706,16 @@ static int adap_init0(struct adapter *adap, int vpd_skip)
 	adap->sge.ingr_start = val[5];
 
 	if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
+		params[0] = FW_PARAM_PFVF(HPFILTER_START);
+		params[1] = FW_PARAM_PFVF(HPFILTER_END);
+		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
+				      params, val);
+		if (ret < 0)
+			goto bye;
+
+		adap->tids.hpftid_base = val[0];
+		adap->tids.nhpftids = val[1] - val[0] + 1;
+
 		/* Read the raw mps entries. In T6, the last 2 tcam entries
 		 * are reserved for raw mac addresses (rawf = 2, one per port).
 		 */
@@ -5050,8 +5079,6 @@ static int adap_init0(struct adapter *adap, int vpd_skip)
 		}
 		adap->params.crypto = ntohs(caps_cmd.cryptocaps);
 	}
-#undef FW_PARAM_PFVF
-#undef FW_PARAM_DEV
 
 	/* The MTU/MSS Table is initialized by now, so load their values.  If
 	 * we're initializing the adapter, then we'll make any modifications

drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c

@@ -672,10 +672,13 @@ int cxgb4_tc_flower_replace(struct net_device *dev,
 		 * 0 to driver. However, the hardware TCAM index
 		 * starts from 0. Hence, the -1 here.
 		 */
-		if (cls->common.prio <= adap->tids.nftids)
+		if (cls->common.prio <= adap->tids.nftids) {
 			fidx = cls->common.prio - 1;
-		else
+			if (fidx < adap->tids.nhpftids)
+				fs->prio = 1;
+		} else {
 			fidx = cxgb4_get_free_ftid(dev, inet_family);
+		}
 
 		/* Only insert FLOWER rule if its priority doesn't
 		 * conflict with existing rules in the LETCAM.

drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_matchall.c

@@ -156,6 +156,8 @@ static int cxgb4_matchall_alloc_filter(struct net_device *dev,
 	fs = &tc_port_matchall->ingress.fs;
 	memset(fs, 0, sizeof(*fs));
 
+	if (fidx < adap->tids.nhpftids)
+		fs->prio = 1;
 	fs->tc_prio = cls->common.prio;
 	fs->tc_cookie = cls->cookie;
 	fs->hitcnts = 1;

drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c

@@ -176,7 +176,7 @@ int cxgb4_config_knode(struct net_device *dev, struct tc_cls_u32_offload *cls)
 	/* Only insert U32 rule if its priority doesn't conflict with
 	 * existing rules in the LETCAM.
 	 */
-	if (filter_id >= adapter->tids.nftids ||
+	if (filter_id >= adapter->tids.nftids + adapter->tids.nhpftids ||
 	    !cxgb4_filter_prio_in_range(dev, filter_id, cls->common.prio)) {
 		NL_SET_ERR_MSG_MOD(extack,
 				   "No free LETCAM index available");
@@ -199,6 +199,8 @@ int cxgb4_config_knode(struct net_device *dev, struct tc_cls_u32_offload *cls)
 
 	memset(&fs, 0, sizeof(fs));
 
+	if (filter_id < adapter->tids.nhpftids)
+		fs.prio = 1;
 	fs.tc_prio = cls->common.prio;
 	fs.tc_cookie = cls->knode.handle;
 
@@ -355,6 +357,7 @@ int cxgb4_delete_knode(struct net_device *dev, struct tc_cls_u32_offload *cls)
 	unsigned int filter_id, max_tids, i, j;
 	struct cxgb4_link *link = NULL;
 	struct cxgb4_tc_u32_table *t;
+	struct filter_entry *f;
 	u32 handle, uhtid;
 	int ret;
 
@@ -363,8 +366,15 @@ int cxgb4_delete_knode(struct net_device *dev, struct tc_cls_u32_offload *cls)
 
 	/* Fetch the location to delete the filter. */
 	filter_id = TC_U32_NODE(cls->knode.handle) - 1;
-	if (filter_id >= adapter->tids.nftids ||
-	    cls->knode.handle != adapter->tids.ftid_tab[filter_id].fs.tc_cookie)
+	if (filter_id >= adapter->tids.nftids + adapter->tids.nhpftids)
+		return -ERANGE;
+
+	if (filter_id < adapter->tids.nhpftids)
+		f = &adapter->tids.hpftid_tab[filter_id];
+	else
+		f = &adapter->tids.ftid_tab[filter_id - adapter->tids.nhpftids];
+
+	if (cls->knode.handle != f->fs.tc_cookie)
 		return -ERANGE;
 
 	t = adapter->tc_u32;
@@ -445,7 +455,7 @@ void cxgb4_cleanup_tc_u32(struct adapter *adap)
 
 struct cxgb4_tc_u32_table *cxgb4_init_tc_u32(struct adapter *adap)
 {
-	unsigned int max_tids = adap->tids.nftids;
+	unsigned int max_tids = adap->tids.nftids + adap->tids.nhpftids;
 	struct cxgb4_tc_u32_table *t;
 	unsigned int i;
 

drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h

@@ -111,6 +111,11 @@ struct tid_info {
 	unsigned int natids;
 	unsigned int atid_base;
 
+	struct filter_entry *hpftid_tab;
+	unsigned long *hpftid_bmap;
+	unsigned int nhpftids;
+	unsigned int hpftid_base;
+
 	struct filter_entry *ftid_tab;
 	unsigned long *ftid_bmap;
 	unsigned int nftids;

drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h

@@ -1321,6 +1321,7 @@ enum fw_params_param_dev {
 	FW_PARAMS_PARAM_DEV_RDMA_WRITE_WITH_IMM = 0x21,
 	FW_PARAMS_PARAM_DEV_PPOD_EDRAM  = 0x23,
 	FW_PARAMS_PARAM_DEV_RI_WRITE_CMPL_WR    = 0x24,
+	FW_PARAMS_PARAM_DEV_HPFILTER_REGION_SUPPORT = 0x26,
 	FW_PARAMS_PARAM_DEV_OPAQUE_VIID_SMT_EXTN = 0x27,
 	FW_PARAMS_PARAM_DEV_HASHFILTER_WITH_OFLD = 0x28,
 	FW_PARAMS_PARAM_DEV_DBQ_TIMER	= 0x29,