From 8e737145e8b27f04fe1c348ac99644f6cad4cfbb Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:32:05 +0100
Subject: [PATCH 01/11] sfc_ef100: check firmware version at start-of-day

Early in EF100 development there was a different format of event
 descriptor; if the NIC is somehow running the very old firmware
 which will use that format, fail the probe.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c | 40 ++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 6a00f2a2dc2b5..75131bcf4f1a0 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -485,6 +485,36 @@ const struct efx_nic_type ef100_pf_nic_type = {
 
 };
 
+static int compare_versions(const char *a, const char *b)
+{
+	int a_major, a_minor, a_point, a_patch;
+	int b_major, b_minor, b_point, b_patch;
+	int a_matched, b_matched;
+
+	a_matched = sscanf(a, "%d.%d.%d.%d", &a_major, &a_minor, &a_point, &a_patch);
+	b_matched = sscanf(b, "%d.%d.%d.%d", &b_major, &b_minor, &b_point, &b_patch);
+
+	if (a_matched == 4 && b_matched != 4)
+		return +1;
+
+	if (a_matched != 4 && b_matched == 4)
+		return -1;
+
+	if (a_matched != 4 && b_matched != 4)
+		return 0;
+
+	if (a_major != b_major)
+		return a_major - b_major;
+
+	if (a_minor != b_minor)
+		return a_minor - b_minor;
+
+	if (a_point != b_point)
+		return a_point - b_point;
+
+	return a_patch - b_patch;
+}
+
 /*	NIC probe and remove
  */
 static int ef100_probe_main(struct efx_nic *efx)
@@ -492,6 +522,7 @@ static int ef100_probe_main(struct efx_nic *efx)
 	unsigned int bar_size = resource_size(&efx->pci_dev->resource[efx->mem_bar]);
 	struct net_device *net_dev = efx->net_dev;
 	struct ef100_nic_data *nic_data;
+	char fw_version[32];
 	int i, rc;
 
 	if (WARN_ON(bar_size == 0))
@@ -562,6 +593,15 @@ static int ef100_probe_main(struct efx_nic *efx)
 		goto fail;
 	efx->port_num = rc;
 
+	efx_mcdi_print_fwver(efx, fw_version, sizeof(fw_version));
+	netif_dbg(efx, drv, efx->net_dev, "Firmware version %s\n", fw_version);
+
+	if (compare_versions(fw_version, "1.1.0.1000") < 0) {
+		netif_info(efx, drv, efx->net_dev, "Firmware uses old event descriptors\n");
+		rc = -EINVAL;
+		goto fail;
+	}
+
 	rc = ef100_phy_probe(efx);
 	if (rc)
 		goto fail;

From 4496363bec32df36df51db7f915a11d3b38615e3 Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:32:16 +0100
Subject: [PATCH 02/11] sfc_ef100: fail the probe if NIC uses unsol_ev credits

In the future, EF100 is planned to have a credit-based scheme for
 handling unsolicited events, which drivers will need to use in order
 to function correctly.  However, current EF100 hardware does not yet
 generate unsolicited events and the credit scheme has not yet been
 implemented in firmware.  To prevent compatibility problems later if
 the current driver is used with future firmware which does implement
 it, we check for the corresponding capability flag (which that
 future firmware will set), and if found, we refuse to probe.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 75131bcf4f1a0..c2bec2bdbc1f6 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -602,6 +602,12 @@ static int ef100_probe_main(struct efx_nic *efx)
 		goto fail;
 	}
 
+	if (efx_has_cap(efx, UNSOL_EV_CREDIT_SUPPORTED)) {
+		netif_info(efx, drv, efx->net_dev, "Firmware uses unsolicited-event credits\n");
+		rc = -EINVAL;
+		goto fail;
+	}
+
 	rc = ef100_phy_probe(efx);
 	if (rc)
 		goto fail;

From adcfc3482ffff813fa2c34e5902005853f79c2aa Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:33:20 +0100
Subject: [PATCH 03/11] sfc_ef100: read Design Parameters at probe time

Several parts of the EF100 architecture are parameterised (to allow
 varying capabilities on FPGAs according to resource constraints), and
 these parameters are exposed to the driver through a TLV-encoded
 region of the BAR.
For the most part we either don't care about these values at all or
 just need to sanity-check them against the driver's assumptions, but
 there are a number of TSO limits which we record so that we will be
 able to check against them in the TX path when handling GSO skbs.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c | 216 +++++++++++++++++++++++++++
 drivers/net/ethernet/sfc/ef100_nic.h |   4 +
 2 files changed, 220 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index c2bec2bdbc1f6..9b5e4b42fe51b 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -515,6 +515,208 @@ static int compare_versions(const char *a, const char *b)
 	return a_patch - b_patch;
 }
 
+enum ef100_tlv_state_machine {
+	EF100_TLV_TYPE,
+	EF100_TLV_TYPE_CONT,
+	EF100_TLV_LENGTH,
+	EF100_TLV_VALUE
+};
+
+struct ef100_tlv_state {
+	enum ef100_tlv_state_machine state;
+	u64 value;
+	u32 value_offset;
+	u16 type;
+	u8 len;
+};
+
+static int ef100_tlv_feed(struct ef100_tlv_state *state, u8 byte)
+{
+	switch (state->state) {
+	case EF100_TLV_TYPE:
+		state->type = byte & 0x7f;
+		state->state = (byte & 0x80) ? EF100_TLV_TYPE_CONT
+					     : EF100_TLV_LENGTH;
+		/* Clear ready to read in a new entry */
+		state->value = 0;
+		state->value_offset = 0;
+		return 0;
+	case EF100_TLV_TYPE_CONT:
+		state->type |= byte << 7;
+		state->state = EF100_TLV_LENGTH;
+		return 0;
+	case EF100_TLV_LENGTH:
+		state->len = byte;
+		/* We only handle TLVs that fit in a u64 */
+		if (state->len > sizeof(state->value))
+			return -EOPNOTSUPP;
+		/* len may be zero, implying a value of zero */
+		state->state = state->len ? EF100_TLV_VALUE : EF100_TLV_TYPE;
+		return 0;
+	case EF100_TLV_VALUE:
+		state->value |= ((u64)byte) << (state->value_offset * 8);
+		state->value_offset++;
+		if (state->value_offset >= state->len)
+			state->state = EF100_TLV_TYPE;
+		return 0;
+	default: /* state machine error, can't happen */
+		WARN_ON_ONCE(1);
+		return -EIO;
+	}
+}
+
+static int ef100_process_design_param(struct efx_nic *efx,
+				      const struct ef100_tlv_state *reader)
+{
+	struct ef100_nic_data *nic_data = efx->nic_data;
+
+	switch (reader->type) {
+	case ESE_EF100_DP_GZ_PAD: /* padding, skip it */
+		return 0;
+	case ESE_EF100_DP_GZ_PARTIAL_TSTAMP_SUB_NANO_BITS:
+		/* Driver doesn't support timestamping yet, so we don't care */
+		return 0;
+	case ESE_EF100_DP_GZ_EVQ_UNSOL_CREDIT_SEQ_BITS:
+		/* Driver doesn't support unsolicited-event credits yet, so
+		 * we don't care
+		 */
+		return 0;
+	case ESE_EF100_DP_GZ_NMMU_GROUP_SIZE:
+		/* Driver doesn't manage the NMMU (so we don't care) */
+		return 0;
+	case ESE_EF100_DP_GZ_RX_L4_CSUM_PROTOCOLS:
+		/* Driver uses CHECKSUM_COMPLETE, so we don't care about
+		 * protocol checksum validation
+		 */
+		return 0;
+	case ESE_EF100_DP_GZ_TSO_MAX_HDR_LEN:
+		nic_data->tso_max_hdr_len = min_t(u64, reader->value, 0xffff);
+		return 0;
+	case ESE_EF100_DP_GZ_TSO_MAX_HDR_NUM_SEGS:
+		/* We always put HDR_NUM_SEGS=1 in our TSO descriptors */
+		if (!reader->value) {
+			netif_err(efx, probe, efx->net_dev,
+				  "TSO_MAX_HDR_NUM_SEGS < 1\n");
+			return -EOPNOTSUPP;
+		}
+		return 0;
+	case ESE_EF100_DP_GZ_RXQ_SIZE_GRANULARITY:
+	case ESE_EF100_DP_GZ_TXQ_SIZE_GRANULARITY:
+		/* Our TXQ and RXQ sizes are always power-of-two and thus divisible by
+		 * EFX_MIN_DMAQ_SIZE, so we just need to check that
+		 * EFX_MIN_DMAQ_SIZE is divisible by GRANULARITY.
+		 * This is very unlikely to fail.
+		 */
+		if (EFX_MIN_DMAQ_SIZE % reader->value) {
+			netif_err(efx, probe, efx->net_dev,
+				  "%s size granularity is %llu, can't guarantee safety\n",
+				  reader->type == ESE_EF100_DP_GZ_RXQ_SIZE_GRANULARITY ? "RXQ" : "TXQ",
+				  reader->value);
+			return -EOPNOTSUPP;
+		}
+		return 0;
+	case ESE_EF100_DP_GZ_TSO_MAX_PAYLOAD_LEN:
+		nic_data->tso_max_payload_len = min_t(u64, reader->value, GSO_MAX_SIZE);
+		efx->net_dev->gso_max_size = nic_data->tso_max_payload_len;
+		return 0;
+	case ESE_EF100_DP_GZ_TSO_MAX_PAYLOAD_NUM_SEGS:
+		nic_data->tso_max_payload_num_segs = min_t(u64, reader->value, 0xffff);
+		efx->net_dev->gso_max_segs = nic_data->tso_max_payload_num_segs;
+		return 0;
+	case ESE_EF100_DP_GZ_TSO_MAX_NUM_FRAMES:
+		nic_data->tso_max_frames = min_t(u64, reader->value, 0xffff);
+		return 0;
+	case ESE_EF100_DP_GZ_COMPAT:
+		if (reader->value) {
+			netif_err(efx, probe, efx->net_dev,
+				  "DP_COMPAT has unknown bits %#llx, driver not compatible with this hw\n",
+				  reader->value);
+			return -EOPNOTSUPP;
+		}
+		return 0;
+	case ESE_EF100_DP_GZ_MEM2MEM_MAX_LEN:
+		/* Driver doesn't use mem2mem transfers */
+		return 0;
+	case ESE_EF100_DP_GZ_EVQ_TIMER_TICK_NANOS:
+		/* Driver doesn't currently use EVQ_TIMER */
+		return 0;
+	case ESE_EF100_DP_GZ_NMMU_PAGE_SIZES:
+		/* Driver doesn't manage the NMMU (so we don't care) */
+		return 0;
+	case ESE_EF100_DP_GZ_VI_STRIDES:
+		/* We never try to set the VI stride, and we don't rely on
+		 * being able to find VIs past VI 0 until after we've learned
+		 * the current stride from MC_CMD_GET_CAPABILITIES.
+		 * So the value of this shouldn't matter.
+		 */
+		if (reader->value != ESE_EF100_DP_GZ_VI_STRIDES_DEFAULT)
+			netif_dbg(efx, probe, efx->net_dev,
+				  "NIC has other than default VI_STRIDES (mask "
+				  "%#llx), early probing might use wrong one\n",
+				  reader->value);
+		return 0;
+	case ESE_EF100_DP_GZ_RX_MAX_RUNT:
+		/* Driver doesn't look at L2_STATUS:LEN_ERR bit, so we don't
+		 * care whether it indicates runt or overlength for any given
+		 * packet, so we don't care about this parameter.
+		 */
+		return 0;
+	default:
+		/* Host interface says "Drivers should ignore design parameters
+		 * that they do not recognise."
+		 */
+		netif_dbg(efx, probe, efx->net_dev,
+			  "Ignoring unrecognised design parameter %u\n",
+			  reader->type);
+		return 0;
+	}
+}
+
+static int ef100_check_design_params(struct efx_nic *efx)
+{
+	struct ef100_tlv_state reader = {};
+	u32 total_len, offset = 0;
+	efx_dword_t reg;
+	int rc = 0, i;
+	u32 data;
+
+	efx_readd(efx, &reg, ER_GZ_PARAMS_TLV_LEN);
+	total_len = EFX_DWORD_FIELD(reg, EFX_DWORD_0);
+	netif_dbg(efx, probe, efx->net_dev, "%u bytes of design parameters\n",
+		  total_len);
+	while (offset < total_len) {
+		efx_readd(efx, &reg, ER_GZ_PARAMS_TLV + offset);
+		data = EFX_DWORD_FIELD(reg, EFX_DWORD_0);
+		for (i = 0; i < sizeof(data); i++) {
+			rc = ef100_tlv_feed(&reader, data);
+			/* Got a complete value? */
+			if (!rc && reader.state == EF100_TLV_TYPE)
+				rc = ef100_process_design_param(efx, &reader);
+			if (rc)
+				goto out;
+			data >>= 8;
+			offset++;
+		}
+	}
+	/* Check we didn't end halfway through a TLV entry, which could either
+	 * mean that the TLV stream is truncated or just that it's corrupted
+	 * and our state machine is out of sync.
+	 */
+	if (reader.state != EF100_TLV_TYPE) {
+		if (reader.state == EF100_TLV_TYPE_CONT)
+			netif_err(efx, probe, efx->net_dev,
+				  "truncated design parameter (incomplete type %u)\n",
+				  reader.type);
+		else
+			netif_err(efx, probe, efx->net_dev,
+				  "truncated design parameter %u\n",
+				  reader.type);
+		rc = -EIO;
+	}
+out:
+	return rc;
+}
+
 /*	NIC probe and remove
  */
 static int ef100_probe_main(struct efx_nic *efx)
@@ -536,6 +738,20 @@ static int ef100_probe_main(struct efx_nic *efx)
 	net_dev->features |= efx->type->offload_features;
 	net_dev->hw_features |= efx->type->offload_features;
 
+	/* Populate design-parameter defaults */
+	nic_data->tso_max_hdr_len = ESE_EF100_DP_GZ_TSO_MAX_HDR_LEN_DEFAULT;
+	nic_data->tso_max_frames = ESE_EF100_DP_GZ_TSO_MAX_NUM_FRAMES_DEFAULT;
+	nic_data->tso_max_payload_num_segs = ESE_EF100_DP_GZ_TSO_MAX_PAYLOAD_NUM_SEGS_DEFAULT;
+	nic_data->tso_max_payload_len = ESE_EF100_DP_GZ_TSO_MAX_PAYLOAD_LEN_DEFAULT;
+	net_dev->gso_max_segs = ESE_EF100_DP_GZ_TSO_MAX_HDR_NUM_SEGS_DEFAULT;
+	/* Read design parameters */
+	rc = ef100_check_design_params(efx);
+	if (rc) {
+		netif_err(efx, probe, efx->net_dev,
+			  "Unsupported design parameters\n");
+		goto fail;
+	}
+
 	/* we assume later that we can copy from this buffer in dwords */
 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
 
diff --git a/drivers/net/ethernet/sfc/ef100_nic.h b/drivers/net/ethernet/sfc/ef100_nic.h
index 6367bbb2c9b33..c8816bc6ae78f 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.h
+++ b/drivers/net/ethernet/sfc/ef100_nic.h
@@ -26,6 +26,10 @@ struct ef100_nic_data {
 	u16 warm_boot_count;
 	u8 port_id[ETH_ALEN];
 	DECLARE_BITMAP(evq_phases, EFX_MAX_CHANNELS);
+	u16 tso_max_hdr_len;
+	u16 tso_max_payload_num_segs;
+	u16 tso_max_frames;
+	unsigned int tso_max_payload_len;
 };
 
 #define efx_ef100_has_cap(caps, flag) \

From d19a5372186336df8a90391c1ae2011e03310dca Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:34:00 +0100
Subject: [PATCH 04/11] sfc_ef100: TX path for EF100 NICs

Includes checksum offload and TSO, so declare those in our netdev features.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c  |   8 +
 drivers/net/ethernet/sfc/ef100_tx.c   | 367 +++++++++++++++++++++++++-
 drivers/net/ethernet/sfc/ef100_tx.h   |   4 +
 drivers/net/ethernet/sfc/net_driver.h |  21 ++
 drivers/net/ethernet/sfc/tx_common.c  |   1 +
 5 files changed, 396 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 9b5e4b42fe51b..fcc5b0de76d71 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -263,6 +263,9 @@ static int ef100_ev_process(struct efx_channel *channel, int quota)
 		case ESE_GZ_EF100_EV_MCDI:
 			efx_mcdi_process_event(channel, p_event);
 			break;
+		case ESE_GZ_EF100_EV_TX_COMPLETION:
+			ef100_ev_tx(channel, p_event);
+			break;
 		case ESE_GZ_EF100_EV_DRIVER:
 			netif_info(efx, drv, efx->net_dev,
 				   "Driver initiated event " EFX_QWORD_FMT "\n",
@@ -436,10 +439,15 @@ static unsigned int ef100_check_caps(const struct efx_nic *efx,
 
 /*	NIC level access functions
  */
+#define EF100_OFFLOAD_FEATURES	(NETIF_F_HW_CSUM |			\
+	NETIF_F_HIGHDMA | NETIF_F_SG | NETIF_F_FRAGLIST |		\
+	NETIF_F_TSO_ECN | NETIF_F_TSO_MANGLEID | NETIF_F_HW_VLAN_CTAG_TX)
+
 const struct efx_nic_type ef100_pf_nic_type = {
 	.revision = EFX_REV_EF100,
 	.is_vf = false,
 	.probe = ef100_probe_pf,
+	.offload_features = EF100_OFFLOAD_FEATURES,
 	.mcdi_max_ver = 2,
 	.mcdi_request = ef100_mcdi_request,
 	.mcdi_poll_response = ef100_mcdi_poll_response,
diff --git a/drivers/net/ethernet/sfc/ef100_tx.c b/drivers/net/ethernet/sfc/ef100_tx.c
index 15e646f8c3e0d..a09546e434085 100644
--- a/drivers/net/ethernet/sfc/ef100_tx.c
+++ b/drivers/net/ethernet/sfc/ef100_tx.c
@@ -9,14 +9,24 @@
  * by the Free Software Foundation, incorporated herein by reference.
  */
 
+#include <net/ip6_checksum.h>
+
 #include "net_driver.h"
 #include "tx_common.h"
 #include "nic_common.h"
+#include "mcdi_functions.h"
+#include "ef100_regs.h"
+#include "io.h"
 #include "ef100_tx.h"
+#include "ef100_nic.h"
 
-/* TX queue stubs */
 int ef100_tx_probe(struct efx_tx_queue *tx_queue)
 {
+	/* Allocate an extra descriptor for the QMDA status completion entry */
+	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
+				    (tx_queue->ptr_mask + 2) *
+				    sizeof(efx_oword_t),
+				    GFP_KERNEL);
 	return 0;
 }
 
@@ -27,10 +37,286 @@ void ef100_tx_init(struct efx_tx_queue *tx_queue)
 		netdev_get_tx_queue(tx_queue->efx->net_dev,
 				    tx_queue->channel->channel -
 				    tx_queue->efx->tx_channel_offset);
+
+	if (efx_mcdi_tx_init(tx_queue, false))
+		netdev_WARN(tx_queue->efx->net_dev,
+			    "failed to initialise TXQ %d\n", tx_queue->queue);
+}
+
+static bool ef100_tx_can_tso(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
+{
+	struct efx_nic *efx = tx_queue->efx;
+	struct ef100_nic_data *nic_data;
+	struct efx_tx_buffer *buffer;
+	struct tcphdr *tcphdr;
+	struct iphdr *iphdr;
+	size_t header_len;
+	u32 mss;
+
+	nic_data = efx->nic_data;
+
+	if (!skb_is_gso_tcp(skb))
+		return false;
+	if (!(efx->net_dev->features & NETIF_F_TSO))
+		return false;
+
+	mss = skb_shinfo(skb)->gso_size;
+	if (unlikely(mss < 4)) {
+		WARN_ONCE(1, "MSS of %u is too small for TSO\n", mss);
+		return false;
+	}
+
+	header_len = efx_tx_tso_header_length(skb);
+	if (header_len > nic_data->tso_max_hdr_len)
+		return false;
+
+	if (skb_shinfo(skb)->gso_segs > nic_data->tso_max_payload_num_segs) {
+		/* net_dev->gso_max_segs should've caught this */
+		WARN_ON_ONCE(1);
+		return false;
+	}
+
+	if (skb->data_len / mss > nic_data->tso_max_frames)
+		return false;
+
+	/* net_dev->gso_max_size should've caught this */
+	if (WARN_ON_ONCE(skb->data_len > nic_data->tso_max_payload_len))
+		return false;
+
+	/* Reserve an empty buffer for the TSO V3 descriptor.
+	 * Convey the length of the header since we already know it.
+	 */
+	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
+	buffer->flags = EFX_TX_BUF_TSO_V3 | EFX_TX_BUF_CONT;
+	buffer->len = header_len;
+	buffer->unmap_len = 0;
+	buffer->skb = skb;
+	++tx_queue->insert_count;
+
+	/* Adjust the TCP checksum to exclude the total length, since we set
+	 * ED_INNER_IP_LEN in the descriptor.
+	 */
+	tcphdr = tcp_hdr(skb);
+	if (skb_is_gso_v6(skb)) {
+		tcphdr->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
+						 &ipv6_hdr(skb)->daddr,
+						 0, IPPROTO_TCP, 0);
+	} else {
+		iphdr = ip_hdr(skb);
+		tcphdr->check = ~csum_tcpudp_magic(iphdr->saddr, iphdr->daddr,
+						   0, IPPROTO_TCP, 0);
+	}
+	return true;
+}
+
+static efx_oword_t *ef100_tx_desc(struct efx_tx_queue *tx_queue, unsigned int index)
+{
+	if (likely(tx_queue->txd.buf.addr))
+		return ((efx_oword_t *)tx_queue->txd.buf.addr) + index;
+	else
+		return NULL;
+}
+
+void ef100_notify_tx_desc(struct efx_tx_queue *tx_queue)
+{
+	unsigned int write_ptr;
+	efx_dword_t reg;
+
+	if (unlikely(tx_queue->notify_count == tx_queue->write_count))
+		return;
+
+	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
+	/* The write pointer goes into the high word */
+	EFX_POPULATE_DWORD_1(reg, ERF_GZ_TX_RING_PIDX, write_ptr);
+	efx_writed_page(tx_queue->efx, &reg,
+			ER_GZ_TX_RING_DOORBELL, tx_queue->queue);
+	tx_queue->notify_count = tx_queue->write_count;
+	tx_queue->xmit_more_available = false;
+}
+
+static void ef100_tx_push_buffers(struct efx_tx_queue *tx_queue)
+{
+	ef100_notify_tx_desc(tx_queue);
+	++tx_queue->pushes;
+}
+
+static void ef100_set_tx_csum_partial(const struct sk_buff *skb,
+				      struct efx_tx_buffer *buffer, efx_oword_t *txd)
+{
+	efx_oword_t csum;
+	int csum_start;
+
+	if (!skb || skb->ip_summed != CHECKSUM_PARTIAL)
+		return;
+
+	/* skb->csum_start has the offset from head, but we need the offset
+	 * from data.
+	 */
+	csum_start = skb_checksum_start_offset(skb);
+	EFX_POPULATE_OWORD_3(csum,
+			     ESF_GZ_TX_SEND_CSO_PARTIAL_EN, 1,
+			     ESF_GZ_TX_SEND_CSO_PARTIAL_START_W,
+			     csum_start >> 1,
+			     ESF_GZ_TX_SEND_CSO_PARTIAL_CSUM_W,
+			     skb->csum_offset >> 1);
+	EFX_OR_OWORD(*txd, *txd, csum);
+}
+
+static void ef100_set_tx_hw_vlan(const struct sk_buff *skb, efx_oword_t *txd)
+{
+	u16 vlan_tci = skb_vlan_tag_get(skb);
+	efx_oword_t vlan;
+
+	EFX_POPULATE_OWORD_2(vlan,
+			     ESF_GZ_TX_SEND_VLAN_INSERT_EN, 1,
+			     ESF_GZ_TX_SEND_VLAN_INSERT_TCI, vlan_tci);
+	EFX_OR_OWORD(*txd, *txd, vlan);
+}
+
+static void ef100_make_send_desc(struct efx_nic *efx,
+				 const struct sk_buff *skb,
+				 struct efx_tx_buffer *buffer, efx_oword_t *txd,
+				 unsigned int segment_count)
+{
+	/* TX send descriptor */
+	EFX_POPULATE_OWORD_3(*txd,
+			     ESF_GZ_TX_SEND_NUM_SEGS, segment_count,
+			     ESF_GZ_TX_SEND_LEN, buffer->len,
+			     ESF_GZ_TX_SEND_ADDR, buffer->dma_addr);
+
+	if (likely(efx->net_dev->features & NETIF_F_HW_CSUM))
+		ef100_set_tx_csum_partial(skb, buffer, txd);
+	if (efx->net_dev->features & NETIF_F_HW_VLAN_CTAG_TX &&
+	    skb && skb_vlan_tag_present(skb))
+		ef100_set_tx_hw_vlan(skb, txd);
+}
+
+static void ef100_make_tso_desc(struct efx_nic *efx,
+				const struct sk_buff *skb,
+				struct efx_tx_buffer *buffer, efx_oword_t *txd,
+				unsigned int segment_count)
+{
+	u32 mangleid = (efx->net_dev->features & NETIF_F_TSO_MANGLEID) ||
+		skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID ?
+		ESE_GZ_TX_DESC_IP4_ID_NO_OP :
+		ESE_GZ_TX_DESC_IP4_ID_INC_MOD16;
+	u16 vlan_enable =  efx->net_dev->features & NETIF_F_HW_VLAN_CTAG_TX ?
+		skb_vlan_tag_present(skb) : 0;
+	unsigned int len, ip_offset, tcp_offset, payload_segs;
+	u16 vlan_tci = skb_vlan_tag_get(skb);
+	u32 mss = skb_shinfo(skb)->gso_size;
+
+	len = skb->len - buffer->len;
+	/* We use 1 for the TSO descriptor and 1 for the header */
+	payload_segs = segment_count - 2;
+	ip_offset =  skb_network_offset(skb);
+	tcp_offset = skb_transport_offset(skb);
+
+	EFX_POPULATE_OWORD_13(*txd,
+			      ESF_GZ_TX_DESC_TYPE, ESE_GZ_TX_DESC_TYPE_TSO,
+			      ESF_GZ_TX_TSO_MSS, mss,
+			      ESF_GZ_TX_TSO_HDR_NUM_SEGS, 1,
+			      ESF_GZ_TX_TSO_PAYLOAD_NUM_SEGS, payload_segs,
+			      ESF_GZ_TX_TSO_HDR_LEN_W, buffer->len >> 1,
+			      ESF_GZ_TX_TSO_PAYLOAD_LEN, len,
+			      ESF_GZ_TX_TSO_CSO_INNER_L4, 1,
+			      ESF_GZ_TX_TSO_INNER_L3_OFF_W, ip_offset >> 1,
+			      ESF_GZ_TX_TSO_INNER_L4_OFF_W, tcp_offset >> 1,
+			      ESF_GZ_TX_TSO_ED_INNER_IP4_ID, mangleid,
+			      ESF_GZ_TX_TSO_ED_INNER_IP_LEN, 1,
+			      ESF_GZ_TX_TSO_VLAN_INSERT_EN, vlan_enable,
+			      ESF_GZ_TX_TSO_VLAN_INSERT_TCI, vlan_tci
+		);
+}
+
+static void ef100_tx_make_descriptors(struct efx_tx_queue *tx_queue,
+				      const struct sk_buff *skb,
+				      unsigned int segment_count)
+{
+	unsigned int old_write_count = tx_queue->write_count;
+	unsigned int new_write_count = old_write_count;
+	struct efx_tx_buffer *buffer;
+	unsigned int next_desc_type;
+	unsigned int write_ptr;
+	efx_oword_t *txd;
+	unsigned int nr_descs = tx_queue->insert_count - old_write_count;
+
+	if (unlikely(nr_descs == 0))
+		return;
+
+	if (segment_count)
+		next_desc_type = ESE_GZ_TX_DESC_TYPE_TSO;
+	else
+		next_desc_type = ESE_GZ_TX_DESC_TYPE_SEND;
+
+	/* if it's a raw write (such as XDP) then always SEND single frames */
+	if (!skb)
+		nr_descs = 1;
+
+	do {
+		write_ptr = new_write_count & tx_queue->ptr_mask;
+		buffer = &tx_queue->buffer[write_ptr];
+		txd = ef100_tx_desc(tx_queue, write_ptr);
+		++new_write_count;
+
+		/* Create TX descriptor ring entry */
+		tx_queue->packet_write_count = new_write_count;
+
+		switch (next_desc_type) {
+		case ESE_GZ_TX_DESC_TYPE_SEND:
+			ef100_make_send_desc(tx_queue->efx, skb,
+					     buffer, txd, nr_descs);
+			break;
+		case ESE_GZ_TX_DESC_TYPE_TSO:
+			/* TX TSO descriptor */
+			WARN_ON_ONCE(!(buffer->flags & EFX_TX_BUF_TSO_V3));
+			ef100_make_tso_desc(tx_queue->efx, skb,
+					    buffer, txd, nr_descs);
+			break;
+		default:
+			/* TX segment descriptor */
+			EFX_POPULATE_OWORD_3(*txd,
+					     ESF_GZ_TX_DESC_TYPE, ESE_GZ_TX_DESC_TYPE_SEG,
+					     ESF_GZ_TX_SEG_LEN, buffer->len,
+					     ESF_GZ_TX_SEG_ADDR, buffer->dma_addr);
+		}
+		/* if it's a raw write (such as XDP) then always SEND */
+		next_desc_type = skb ? ESE_GZ_TX_DESC_TYPE_SEG :
+				       ESE_GZ_TX_DESC_TYPE_SEND;
+
+	} while (new_write_count != tx_queue->insert_count);
+
+	wmb(); /* Ensure descriptors are written before they are fetched */
+
+	tx_queue->write_count = new_write_count;
+
+	/* The write_count above must be updated before reading
+	 * channel->holdoff_doorbell to avoid a race with the
+	 * completion path, so ensure these operations are not
+	 * re-ordered.  This also flushes the update of write_count
+	 * back into the cache.
+	 */
+	smp_mb();
 }
 
 void ef100_tx_write(struct efx_tx_queue *tx_queue)
 {
+	ef100_tx_make_descriptors(tx_queue, NULL, 0);
+	ef100_tx_push_buffers(tx_queue);
+}
+
+void ef100_ev_tx(struct efx_channel *channel, const efx_qword_t *p_event)
+{
+	unsigned int tx_done =
+		EFX_QWORD_FIELD(*p_event, ESF_GZ_EV_TXCMPL_NUM_DESC);
+	unsigned int qlabel =
+		EFX_QWORD_FIELD(*p_event, ESF_GZ_EV_TXCMPL_Q_LABEL);
+	struct efx_tx_queue *tx_queue =
+		efx_channel_get_tx_queue(channel, qlabel);
+	unsigned int tx_index = (tx_queue->read_count + tx_done - 1) &
+				tx_queue->ptr_mask;
+
+	efx_xmit_done(tx_queue, tx_index);
 }
 
 /* Add a socket buffer to a TX queue
@@ -42,10 +328,81 @@ void ef100_tx_write(struct efx_tx_queue *tx_queue)
  */
 int ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 {
-	/* Stub.  No TX path yet. */
+	unsigned int old_insert_count = tx_queue->insert_count;
 	struct efx_nic *efx = tx_queue->efx;
+	bool xmit_more = netdev_xmit_more();
+	unsigned int fill_level;
+	unsigned int segments;
+	int rc;
+
+	if (!tx_queue->buffer || !tx_queue->ptr_mask) {
+		netif_stop_queue(efx->net_dev);
+		dev_kfree_skb_any(skb);
+		return -ENODEV;
+	}
+
+	segments = skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 0;
+	if (segments == 1)
+		segments = 0;	/* Don't use TSO/GSO for a single segment. */
+	if (segments && !ef100_tx_can_tso(tx_queue, skb)) {
+		rc = efx_tx_tso_fallback(tx_queue, skb);
+		tx_queue->tso_fallbacks++;
+		if (rc)
+			goto err;
+		else
+			return 0;
+	}
+
+	/* Map for DMA and create descriptors */
+	rc = efx_tx_map_data(tx_queue, skb, segments);
+	if (rc)
+		goto err;
+	ef100_tx_make_descriptors(tx_queue, skb, segments);
+
+	fill_level = efx_channel_tx_fill_level(tx_queue->channel);
+	if (fill_level > efx->txq_stop_thresh) {
+		netif_tx_stop_queue(tx_queue->core_txq);
+		/* Re-read after a memory barrier in case we've raced with
+		 * the completion path. Otherwise there's a danger we'll never
+		 * restart the queue if all completions have just happened.
+		 */
+		smp_mb();
+		fill_level = efx_channel_tx_fill_level(tx_queue->channel);
+		if (fill_level < efx->txq_stop_thresh)
+			netif_tx_start_queue(tx_queue->core_txq);
+	}
+
+	if (__netdev_tx_sent_queue(tx_queue->core_txq, skb->len, xmit_more))
+		tx_queue->xmit_more_available = false; /* push doorbell */
+	else if (tx_queue->write_count - tx_queue->notify_count > 255)
+		/* Ensure we never push more than 256 packets at once */
+		tx_queue->xmit_more_available = false; /* push */
+	else
+		tx_queue->xmit_more_available = true; /* don't push yet */
+
+	if (!tx_queue->xmit_more_available)
+		ef100_tx_push_buffers(tx_queue);
+
+	if (segments) {
+		tx_queue->tso_bursts++;
+		tx_queue->tso_packets += segments;
+		tx_queue->tx_packets  += segments;
+	} else {
+		tx_queue->tx_packets++;
+	}
+	return 0;
+
+err:
+	efx_enqueue_unwind(tx_queue, old_insert_count);
+	if (!IS_ERR_OR_NULL(skb))
+		dev_kfree_skb_any(skb);
 
-	netif_stop_queue(efx->net_dev);
-	dev_kfree_skb_any(skb);
-	return -ENODEV;
+	/* If we're not expecting another transmit and we had something to push
+	 * on this queue then we need to push here to get the previous packets
+	 * out.  We only enter this branch from before the 'Update BQL' section
+	 * above, so xmit_more_available still refers to the old state.
+	 */
+	if (tx_queue->xmit_more_available && !xmit_more)
+		ef100_tx_push_buffers(tx_queue);
+	return rc;
 }
diff --git a/drivers/net/ethernet/sfc/ef100_tx.h b/drivers/net/ethernet/sfc/ef100_tx.h
index 9a472f7aff432..fa23e430bdd72 100644
--- a/drivers/net/ethernet/sfc/ef100_tx.h
+++ b/drivers/net/ethernet/sfc/ef100_tx.h
@@ -17,6 +17,10 @@
 int ef100_tx_probe(struct efx_tx_queue *tx_queue);
 void ef100_tx_init(struct efx_tx_queue *tx_queue);
 void ef100_tx_write(struct efx_tx_queue *tx_queue);
+void ef100_notify_tx_desc(struct efx_tx_queue *tx_queue);
+unsigned int ef100_tx_max_skb_descs(struct efx_nic *efx);
+
+void ef100_ev_tx(struct efx_channel *channel, const efx_qword_t *p_event);
 
 netdev_tx_t ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb);
 #endif
diff --git a/drivers/net/ethernet/sfc/net_driver.h b/drivers/net/ethernet/sfc/net_driver.h
index 9791fac0b6496..7bb7ecb480ae1 100644
--- a/drivers/net/ethernet/sfc/net_driver.h
+++ b/drivers/net/ethernet/sfc/net_driver.h
@@ -173,6 +173,7 @@ struct efx_tx_buffer {
 #define EFX_TX_BUF_MAP_SINGLE	8	/* buffer was mapped with dma_map_single() */
 #define EFX_TX_BUF_OPTION	0x10	/* empty buffer for option descriptor */
 #define EFX_TX_BUF_XDP		0x20	/* buffer was sent with XDP */
+#define EFX_TX_BUF_TSO_V3	0x40	/* empty buffer for a TSO_V3 descriptor */
 
 /**
  * struct efx_tx_queue - An Efx TX queue
@@ -245,6 +246,7 @@ struct efx_tx_buffer {
  * @pio_packets: Number of times the TX PIO feature has been used
  * @xmit_more_available: Are any packets waiting to be pushed to the NIC
  * @cb_packets: Number of times the TX copybreak feature has been used
+ * @notify_count: Count of notified descriptors to the NIC
  * @empty_read_count: If the completion path has seen the queue as empty
  *	and the transmission path has not yet checked this, the value of
  *	@read_count bitwise-added to %EFX_EMPTY_COUNT_VALID; otherwise 0.
@@ -292,6 +294,7 @@ struct efx_tx_queue {
 	unsigned int pio_packets;
 	bool xmit_more_available;
 	unsigned int cb_packets;
+	unsigned int notify_count;
 	/* Statistics to supplement MAC stats */
 	unsigned long tx_packets;
 
@@ -1669,6 +1672,24 @@ static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
 	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
 }
 
+/* Get the max fill level of the TX queues on this channel */
+static inline unsigned int
+efx_channel_tx_fill_level(struct efx_channel *channel)
+{
+	struct efx_tx_queue *tx_queue;
+	unsigned int fill_level = 0;
+
+	/* This function is currently only used by EF100, which maybe
+	 * could do something simpler and just compute the fill level
+	 * of the single TXQ that's really in use.
+	 */
+	efx_for_each_channel_tx_queue(tx_queue, channel)
+		fill_level = max(fill_level,
+				 tx_queue->insert_count - tx_queue->read_count);
+
+	return fill_level;
+}
+
 /* Get all supported features.
  * If a feature is not fixed, it is present in hw_features.
  * If a feature is fixed, it does not present in hw_features, but
diff --git a/drivers/net/ethernet/sfc/tx_common.c b/drivers/net/ethernet/sfc/tx_common.c
index 11b64c609550a..793e234819a8c 100644
--- a/drivers/net/ethernet/sfc/tx_common.c
+++ b/drivers/net/ethernet/sfc/tx_common.c
@@ -71,6 +71,7 @@ void efx_init_tx_queue(struct efx_tx_queue *tx_queue)
 		  "initialising TX queue %d\n", tx_queue->queue);
 
 	tx_queue->insert_count = 0;
+	tx_queue->notify_count = 0;
 	tx_queue->write_count = 0;
 	tx_queue->packet_write_count = 0;
 	tx_queue->old_write_count = 0;

From a9dc3d5612ce6b48bd1d230d0e3c21478a9538b3 Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:34:47 +0100
Subject: [PATCH 05/11] sfc_ef100: RX filter table management and related
 gubbins

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_netdev.c | 10 ++++
 drivers/net/ethernet/sfc/ef100_nic.c    | 67 +++++++++++++++++++++++++
 2 files changed, 77 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_netdev.c b/drivers/net/ethernet/sfc/ef100_netdev.c
index ec9ca81fed85a..362a915c836af 100644
--- a/drivers/net/ethernet/sfc/ef100_netdev.c
+++ b/drivers/net/ethernet/sfc/ef100_netdev.c
@@ -89,6 +89,7 @@ static int ef100_net_stop(struct net_device *net_dev)
 	efx_disable_interrupts(efx);
 	efx_clear_interrupt_affinity(efx);
 	efx_nic_fini_interrupt(efx);
+	efx_remove_filters(efx);
 	efx_fini_napi(efx);
 	efx_remove_channels(efx);
 	efx_mcdi_free_vis(efx);
@@ -138,6 +139,10 @@ static int ef100_net_open(struct net_device *net_dev)
 
 	efx_init_napi(efx);
 
+	rc = efx_probe_filters(efx);
+	if (rc)
+		goto fail;
+
 	rc = efx_nic_init_interrupt(efx);
 	if (rc)
 		goto fail;
@@ -207,8 +212,13 @@ static const struct net_device_ops ef100_netdev_ops = {
 	.ndo_open               = ef100_net_open,
 	.ndo_stop               = ef100_net_stop,
 	.ndo_start_xmit         = ef100_hard_start_xmit,
+	.ndo_validate_addr      = eth_validate_addr,
+	.ndo_set_rx_mode        = efx_set_rx_mode, /* Lookout */
 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
 	.ndo_get_phys_port_name = efx_get_phys_port_name,
+#ifdef CONFIG_RFS_ACCEL
+	.ndo_rx_flow_steer      = efx_filter_rfs,
+#endif
 };
 
 /*	Netdev registration
diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index fcc5b0de76d71..728e2ffd1d77d 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -347,6 +347,37 @@ static int ef100_phy_probe(struct efx_nic *efx)
 	return 0;
 }
 
+static int ef100_filter_table_probe(struct efx_nic *efx)
+{
+	return efx_mcdi_filter_table_probe(efx, true);
+}
+
+static int ef100_filter_table_up(struct efx_nic *efx)
+{
+	int rc;
+
+	rc = efx_mcdi_filter_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
+	if (rc) {
+		efx_mcdi_filter_table_down(efx);
+		return rc;
+	}
+
+	rc = efx_mcdi_filter_add_vlan(efx, 0);
+	if (rc) {
+		efx_mcdi_filter_del_vlan(efx, EFX_FILTER_VID_UNSPEC);
+		efx_mcdi_filter_table_down(efx);
+	}
+
+	return rc;
+}
+
+static void ef100_filter_table_down(struct efx_nic *efx)
+{
+	efx_mcdi_filter_del_vlan(efx, 0);
+	efx_mcdi_filter_del_vlan(efx, EFX_FILTER_VID_UNSPEC);
+	efx_mcdi_filter_table_down(efx);
+}
+
 /*	Other
  */
 static int ef100_reconfigure_mac(struct efx_nic *efx, bool mtu_only)
@@ -393,12 +424,24 @@ static int ef100_reset(struct efx_nic *efx, enum reset_type reset_type)
 		__clear_bit(reset_type, &efx->reset_pending);
 		rc = dev_open(efx->net_dev, NULL);
 	} else if (reset_type == RESET_TYPE_ALL) {
+		/* A RESET_TYPE_ALL will cause filters to be removed, so we remove filters
+		 * and reprobe after reset to avoid removing filters twice
+		 */
+		down_read(&efx->filter_sem);
+		ef100_filter_table_down(efx);
+		up_read(&efx->filter_sem);
 		rc = efx_mcdi_reset(efx, reset_type);
 		if (rc)
 			return rc;
 
 		netif_device_attach(efx->net_dev);
 
+		down_read(&efx->filter_sem);
+		rc = ef100_filter_table_up(efx);
+		up_read(&efx->filter_sem);
+		if (rc)
+			return rc;
+
 		rc = dev_open(efx->net_dev, NULL);
 	} else {
 		rc = 1;	/* Leave the device closed */
@@ -480,6 +523,20 @@ const struct efx_nic_type ef100_pf_nic_type = {
 	.rx_remove = efx_mcdi_rx_remove,
 	.rx_write = ef100_rx_write,
 	.rx_packet = __ef100_rx_packet,
+	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
+	.filter_table_probe = ef100_filter_table_up,
+	.filter_table_restore = efx_mcdi_filter_table_restore,
+	.filter_table_remove = ef100_filter_table_down,
+	.filter_insert = efx_mcdi_filter_insert,
+	.filter_remove_safe = efx_mcdi_filter_remove_safe,
+	.filter_get_safe = efx_mcdi_filter_get_safe,
+	.filter_clear_rx = efx_mcdi_filter_clear_rx,
+	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
+	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
+	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
+#ifdef CONFIG_RFS_ACCEL
+	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
+#endif
 
 	.get_phys_port_id = efx_ef100_get_phys_port_id,
 
@@ -840,6 +897,12 @@ static int ef100_probe_main(struct efx_nic *efx)
 	if (rc)
 		goto fail;
 
+	down_write(&efx->filter_sem);
+	rc = ef100_filter_table_probe(efx);
+	up_write(&efx->filter_sem);
+	if (rc)
+		goto fail;
+
 	rc = ef100_register_netdev(efx);
 	if (rc)
 		goto fail;
@@ -877,6 +940,10 @@ void ef100_remove(struct efx_nic *efx)
 	struct ef100_nic_data *nic_data = efx->nic_data;
 
 	ef100_unregister_netdev(efx);
+
+	down_write(&efx->filter_sem);
+	efx_mcdi_filter_table_remove(efx);
+	up_write(&efx->filter_sem);
 	efx_fini_channels(efx);
 	kfree(efx->phy_data);
 	efx->phy_data = NULL;

From 8e57daf70671e482209b5d231a181f04845cf73e Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:36:28 +0100
Subject: [PATCH 06/11] sfc_ef100: RX path for EF100

Includes RSS spreading.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c |  25 ++++-
 drivers/net/ethernet/sfc/ef100_rx.c  | 150 +++++++++++++++++++++++++--
 drivers/net/ethernet/sfc/ef100_rx.h  |   1 +
 3 files changed, 167 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 728e2ffd1d77d..bb753856d88fe 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -260,6 +260,10 @@ static int ef100_ev_process(struct efx_channel *channel, int quota)
 		ev_type = EFX_QWORD_FIELD(*p_event, ESF_GZ_E_TYPE);
 
 		switch (ev_type) {
+		case ESE_GZ_EF100_EV_RX_PKTS:
+			efx_ef100_ev_rx(channel, p_event);
+			++spent;
+			break;
 		case ESE_GZ_EF100_EV_MCDI:
 			efx_mcdi_process_event(channel, p_event);
 			break;
@@ -482,9 +486,10 @@ static unsigned int ef100_check_caps(const struct efx_nic *efx,
 
 /*	NIC level access functions
  */
-#define EF100_OFFLOAD_FEATURES	(NETIF_F_HW_CSUM |			\
+#define EF100_OFFLOAD_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_RXCSUM |	\
 	NETIF_F_HIGHDMA | NETIF_F_SG | NETIF_F_FRAGLIST |		\
-	NETIF_F_TSO_ECN | NETIF_F_TSO_MANGLEID | NETIF_F_HW_VLAN_CTAG_TX)
+	NETIF_F_RXHASH | NETIF_F_RXFCS | NETIF_F_TSO_ECN | NETIF_F_RXALL | \
+	NETIF_F_TSO_MANGLEID | NETIF_F_HW_VLAN_CTAG_TX)
 
 const struct efx_nic_type ef100_pf_nic_type = {
 	.revision = EFX_REV_EF100,
@@ -540,6 +545,16 @@ const struct efx_nic_type ef100_pf_nic_type = {
 
 	.get_phys_port_id = efx_ef100_get_phys_port_id,
 
+	.rx_prefix_size = ESE_GZ_RX_PKT_PREFIX_LEN,
+	.rx_hash_offset = ESF_GZ_RX_PREFIX_RSS_HASH_LBN / 8,
+	.rx_ts_offset = ESF_GZ_RX_PREFIX_PARTIAL_TSTAMP_LBN / 8,
+	.rx_hash_key_size = 40,
+	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
+	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
+	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
+	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
+	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
+
 	.reconfigure_mac = ef100_reconfigure_mac,
 
 	/* Per-type bar/size configuration not used on ef100. Location of
@@ -903,6 +918,12 @@ static int ef100_probe_main(struct efx_nic *efx)
 	if (rc)
 		goto fail;
 
+	netdev_rss_key_fill(efx->rss_context.rx_hash_key,
+			    sizeof(efx->rss_context.rx_hash_key));
+
+	/* Don't fail init if RSS setup doesn't work. */
+	efx_mcdi_push_default_indir_table(efx, efx->n_rx_channels);
+
 	rc = ef100_register_netdev(efx);
 	if (rc)
 		goto fail;
diff --git a/drivers/net/ethernet/sfc/ef100_rx.c b/drivers/net/ethernet/sfc/ef100_rx.c
index 4223a38f46d3f..13ba1a4f66fca 100644
--- a/drivers/net/ethernet/sfc/ef100_rx.c
+++ b/drivers/net/ethernet/sfc/ef100_rx.c
@@ -12,20 +12,156 @@
 #include "ef100_rx.h"
 #include "rx_common.h"
 #include "efx.h"
+#include "nic_common.h"
+#include "mcdi_functions.h"
+#include "ef100_regs.h"
+#include "ef100_nic.h"
+#include "io.h"
 
-/* RX stubs */
+/* Get the value of a field in the RX prefix */
+#define PREFIX_OFFSET_W(_f)	(ESF_GZ_RX_PREFIX_ ## _f ## _LBN / 32)
+#define PREFIX_OFFSET_B(_f)	(ESF_GZ_RX_PREFIX_ ## _f ## _LBN % 32)
+#define PREFIX_WIDTH_MASK(_f)	((1UL << ESF_GZ_RX_PREFIX_ ## _f ## _WIDTH) - 1)
+#define PREFIX_WORD(_p, _f)	le32_to_cpu((__force __le32)(_p)[PREFIX_OFFSET_W(_f)])
+#define PREFIX_FIELD(_p, _f)	((PREFIX_WORD(_p, _f) >> PREFIX_OFFSET_B(_f)) & \
+				 PREFIX_WIDTH_MASK(_f))
 
-void ef100_rx_write(struct efx_rx_queue *rx_queue)
+#define ESF_GZ_RX_PREFIX_NT_OR_INNER_L3_CLASS_LBN	\
+		(ESF_GZ_RX_PREFIX_CLASS_LBN + ESF_GZ_RX_PREFIX_HCLASS_NT_OR_INNER_L3_CLASS_LBN)
+#define ESF_GZ_RX_PREFIX_NT_OR_INNER_L3_CLASS_WIDTH	\
+		ESF_GZ_RX_PREFIX_HCLASS_NT_OR_INNER_L3_CLASS_WIDTH
+
+static bool check_fcs(struct efx_channel *channel, u32 *prefix)
 {
+	u16 rxclass;
+	u8 l2status;
+
+	rxclass = le16_to_cpu((__force __le16)PREFIX_FIELD(prefix, CLASS));
+	l2status = PREFIX_FIELD(&rxclass, HCLASS_L2_STATUS);
+
+	if (likely(l2status == ESE_GZ_RH_HCLASS_L2_STATUS_OK))
+		/* Everything is ok */
+		return 0;
+
+	if (l2status == ESE_GZ_RH_HCLASS_L2_STATUS_FCS_ERR)
+		channel->n_rx_eth_crc_err++;
+	return 1;
 }
 
 void __ef100_rx_packet(struct efx_channel *channel)
 {
-	/* Stub.  No RX path yet.  Discard the buffer. */
-	struct efx_rx_buffer *rx_buf = efx_rx_buffer(&channel->rx_queue,
-						     channel->rx_pkt_index);
-	struct efx_rx_queue *rx_queue = efx_channel_get_rx_queue(channel);
+	struct efx_rx_buffer *rx_buf = efx_rx_buffer(&channel->rx_queue, channel->rx_pkt_index);
+	struct efx_nic *efx = channel->efx;
+	u8 *eh = efx_rx_buf_va(rx_buf);
+	__wsum csum = 0;
+	u32 *prefix;
+
+	prefix = (u32 *)(eh - ESE_GZ_RX_PKT_PREFIX_LEN);
+
+	if (check_fcs(channel, prefix) &&
+	    unlikely(!(efx->net_dev->features & NETIF_F_RXALL)))
+		goto out;
+
+	rx_buf->len = le16_to_cpu((__force __le16)PREFIX_FIELD(prefix, LENGTH));
+	if (rx_buf->len <= sizeof(struct ethhdr)) {
+		if (net_ratelimit())
+			netif_err(channel->efx, rx_err, channel->efx->net_dev,
+				  "RX packet too small (%d)\n", rx_buf->len);
+		++channel->n_rx_frm_trunc;
+		goto out;
+	}
+
+	if (likely(efx->net_dev->features & NETIF_F_RXCSUM)) {
+		if (PREFIX_FIELD(prefix, NT_OR_INNER_L3_CLASS) == 1) {
+			++channel->n_rx_ip_hdr_chksum_err;
+		} else {
+			u16 sum = be16_to_cpu((__force __be16)PREFIX_FIELD(prefix, CSUM_FRAME));
 
-	efx_free_rx_buffers(rx_queue, rx_buf, 1);
+			csum = (__force __wsum) sum;
+		}
+	}
+
+	if (channel->type->receive_skb) {
+		struct efx_rx_queue *rx_queue =
+			efx_channel_get_rx_queue(channel);
+
+		/* no support for special channels yet, so just discard */
+		WARN_ON_ONCE(1);
+		efx_free_rx_buffers(rx_queue, rx_buf, 1);
+		goto out;
+	}
+
+	efx_rx_packet_gro(channel, rx_buf, channel->rx_pkt_n_frags, eh, csum);
+
+out:
 	channel->rx_pkt_n_frags = 0;
 }
+
+static void ef100_rx_packet(struct efx_rx_queue *rx_queue, unsigned int index)
+{
+	struct efx_rx_buffer *rx_buf = efx_rx_buffer(rx_queue, index);
+	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
+	struct efx_nic *efx = rx_queue->efx;
+
+	++rx_queue->rx_packets;
+
+	netif_vdbg(efx, rx_status, efx->net_dev,
+		   "RX queue %d received id %x\n",
+		   efx_rx_queue_index(rx_queue), index);
+
+	efx_sync_rx_buffer(efx, rx_buf, efx->rx_dma_len);
+
+	prefetch(efx_rx_buf_va(rx_buf));
+
+	rx_buf->page_offset += efx->rx_prefix_size;
+
+	efx_recycle_rx_pages(channel, rx_buf, 1);
+
+	efx_rx_flush_packet(channel);
+	channel->rx_pkt_n_frags = 1;
+	channel->rx_pkt_index = index;
+}
+
+void efx_ef100_ev_rx(struct efx_channel *channel, const efx_qword_t *p_event)
+{
+	struct efx_rx_queue *rx_queue = efx_channel_get_rx_queue(channel);
+	unsigned int n_packets =
+		EFX_QWORD_FIELD(*p_event, ESF_GZ_EV_RXPKTS_NUM_PKT);
+	int i;
+
+	WARN_ON_ONCE(!n_packets);
+	if (n_packets > 1)
+		++channel->n_rx_merge_events;
+
+	channel->irq_mod_score += 2 * n_packets;
+
+	for (i = 0; i < n_packets; ++i) {
+		ef100_rx_packet(rx_queue,
+				rx_queue->removed_count & rx_queue->ptr_mask);
+		++rx_queue->removed_count;
+	}
+}
+
+void ef100_rx_write(struct efx_rx_queue *rx_queue)
+{
+	struct efx_rx_buffer *rx_buf;
+	unsigned int idx;
+	efx_qword_t *rxd;
+	efx_dword_t rxdb;
+
+	while (rx_queue->notified_count != rx_queue->added_count) {
+		idx = rx_queue->notified_count & rx_queue->ptr_mask;
+		rx_buf = efx_rx_buffer(rx_queue, idx);
+		rxd = efx_rx_desc(rx_queue, idx);
+
+		EFX_POPULATE_QWORD_1(*rxd, ESF_GZ_RX_BUF_ADDR, rx_buf->dma_addr);
+
+		++rx_queue->notified_count;
+	}
+
+	wmb();
+	EFX_POPULATE_DWORD_1(rxdb, ERF_GZ_RX_RING_PIDX,
+			     rx_queue->added_count & rx_queue->ptr_mask);
+	efx_writed_page(rx_queue->efx, &rxdb,
+			ER_GZ_RX_RING_DOORBELL, efx_rx_queue_index(rx_queue));
+}
diff --git a/drivers/net/ethernet/sfc/ef100_rx.h b/drivers/net/ethernet/sfc/ef100_rx.h
index b5dadf741aa09..f2f266863966a 100644
--- a/drivers/net/ethernet/sfc/ef100_rx.h
+++ b/drivers/net/ethernet/sfc/ef100_rx.h
@@ -14,6 +14,7 @@
 
 #include "net_driver.h"
 
+void efx_ef100_ev_rx(struct efx_channel *channel, const efx_qword_t *p_event);
 void ef100_rx_write(struct efx_rx_queue *rx_queue);
 void __ef100_rx_packet(struct efx_channel *channel);
 

From b780feac367e5b77028b8088829b7958944bd71a Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:36:44 +0100
Subject: [PATCH 07/11] sfc_ef100: plumb in fini_dmaq

Bring down the TX and RX queues at ifdown, so that we can then fini the
 EVQs (otherwise the MC would return EBUSY because they're still in use).

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index bb753856d88fe..1953e16b2b96c 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -528,6 +528,7 @@ const struct efx_nic_type ef100_pf_nic_type = {
 	.rx_remove = efx_mcdi_rx_remove,
 	.rx_write = ef100_rx_write,
 	.rx_packet = __ef100_rx_packet,
+	.fini_dmaq = efx_fini_dmaq,
 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
 	.filter_table_probe = ef100_filter_table_up,
 	.filter_table_restore = efx_mcdi_filter_table_restore,

From b593b6f1b4921700c00394d35e098259e3d04913 Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:37:20 +0100
Subject: [PATCH 08/11] sfc_ef100: statistics gathering

MAC stats work much the same as on EF10, with a periodic DMA to a region
 specified via an MCDI.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_netdev.c |   6 +
 drivers/net/ethernet/sfc/ef100_nic.c    | 170 ++++++++++++++++++++++++
 drivers/net/ethernet/sfc/ef100_nic.h    |  41 ++++++
 3 files changed, 217 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_netdev.c b/drivers/net/ethernet/sfc/ef100_netdev.c
index 362a915c836af..63c311ba28b93 100644
--- a/drivers/net/ethernet/sfc/ef100_netdev.c
+++ b/drivers/net/ethernet/sfc/ef100_netdev.c
@@ -86,6 +86,7 @@ static int ef100_net_stop(struct net_device *net_dev)
 
 	netif_stop_queue(net_dev);
 	efx_stop_all(efx);
+	efx_mcdi_mac_fini_stats(efx);
 	efx_disable_interrupts(efx);
 	efx_clear_interrupt_affinity(efx);
 	efx_nic_fini_interrupt(efx);
@@ -157,6 +158,10 @@ static int ef100_net_open(struct net_device *net_dev)
 	 */
 	(void) efx_mcdi_poll_reboot(efx);
 
+	rc = efx_mcdi_mac_init_stats(efx);
+	if (rc)
+		goto fail;
+
 	efx_start_all(efx);
 
 	/* Link state detection is normally event-driven; we have
@@ -212,6 +217,7 @@ static const struct net_device_ops ef100_netdev_ops = {
 	.ndo_open               = ef100_net_open,
 	.ndo_stop               = ef100_net_stop,
 	.ndo_start_xmit         = ef100_hard_start_xmit,
+	.ndo_get_stats64        = efx_net_stats,
 	.ndo_validate_addr      = eth_validate_addr,
 	.ndo_set_rx_mode        = efx_set_rx_mode, /* Lookout */
 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 1953e16b2b96c..c81da0b2f0c1b 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -453,6 +453,171 @@ static int ef100_reset(struct efx_nic *efx, enum reset_type reset_type)
 	return rc;
 }
 
+static void ef100_common_stat_mask(unsigned long *mask)
+{
+	__set_bit(EF100_STAT_port_rx_packets, mask);
+	__set_bit(EF100_STAT_port_tx_packets, mask);
+	__set_bit(EF100_STAT_port_rx_bytes, mask);
+	__set_bit(EF100_STAT_port_tx_bytes, mask);
+	__set_bit(EF100_STAT_port_rx_multicast, mask);
+	__set_bit(EF100_STAT_port_rx_bad, mask);
+	__set_bit(EF100_STAT_port_rx_align_error, mask);
+	__set_bit(EF100_STAT_port_rx_overflow, mask);
+}
+
+static void ef100_ethtool_stat_mask(unsigned long *mask)
+{
+	__set_bit(EF100_STAT_port_tx_pause, mask);
+	__set_bit(EF100_STAT_port_tx_unicast, mask);
+	__set_bit(EF100_STAT_port_tx_multicast, mask);
+	__set_bit(EF100_STAT_port_tx_broadcast, mask);
+	__set_bit(EF100_STAT_port_tx_lt64, mask);
+	__set_bit(EF100_STAT_port_tx_64, mask);
+	__set_bit(EF100_STAT_port_tx_65_to_127, mask);
+	__set_bit(EF100_STAT_port_tx_128_to_255, mask);
+	__set_bit(EF100_STAT_port_tx_256_to_511, mask);
+	__set_bit(EF100_STAT_port_tx_512_to_1023, mask);
+	__set_bit(EF100_STAT_port_tx_1024_to_15xx, mask);
+	__set_bit(EF100_STAT_port_tx_15xx_to_jumbo, mask);
+	__set_bit(EF100_STAT_port_rx_good, mask);
+	__set_bit(EF100_STAT_port_rx_pause, mask);
+	__set_bit(EF100_STAT_port_rx_unicast, mask);
+	__set_bit(EF100_STAT_port_rx_broadcast, mask);
+	__set_bit(EF100_STAT_port_rx_lt64, mask);
+	__set_bit(EF100_STAT_port_rx_64, mask);
+	__set_bit(EF100_STAT_port_rx_65_to_127, mask);
+	__set_bit(EF100_STAT_port_rx_128_to_255, mask);
+	__set_bit(EF100_STAT_port_rx_256_to_511, mask);
+	__set_bit(EF100_STAT_port_rx_512_to_1023, mask);
+	__set_bit(EF100_STAT_port_rx_1024_to_15xx, mask);
+	__set_bit(EF100_STAT_port_rx_15xx_to_jumbo, mask);
+	__set_bit(EF100_STAT_port_rx_gtjumbo, mask);
+	__set_bit(EF100_STAT_port_rx_bad_gtjumbo, mask);
+	__set_bit(EF100_STAT_port_rx_length_error, mask);
+	__set_bit(EF100_STAT_port_rx_nodesc_drops, mask);
+	__set_bit(GENERIC_STAT_rx_nodesc_trunc, mask);
+	__set_bit(GENERIC_STAT_rx_noskb_drops, mask);
+}
+
+#define EF100_DMA_STAT(ext_name, mcdi_name)			\
+	[EF100_STAT_ ## ext_name] =				\
+	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
+
+static const struct efx_hw_stat_desc ef100_stat_desc[EF100_STAT_COUNT] = {
+	EF100_DMA_STAT(port_tx_bytes, TX_BYTES),
+	EF100_DMA_STAT(port_tx_packets, TX_PKTS),
+	EF100_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
+	EF100_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
+	EF100_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
+	EF100_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
+	EF100_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
+	EF100_DMA_STAT(port_tx_64, TX_64_PKTS),
+	EF100_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
+	EF100_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
+	EF100_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
+	EF100_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
+	EF100_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
+	EF100_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
+	EF100_DMA_STAT(port_rx_bytes, RX_BYTES),
+	EF100_DMA_STAT(port_rx_packets, RX_PKTS),
+	EF100_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
+	EF100_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
+	EF100_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
+	EF100_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
+	EF100_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
+	EF100_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
+	EF100_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
+	EF100_DMA_STAT(port_rx_64, RX_64_PKTS),
+	EF100_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
+	EF100_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
+	EF100_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
+	EF100_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
+	EF100_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
+	EF100_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
+	EF100_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
+	EF100_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
+	EF100_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
+	EF100_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
+	EF100_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
+	EF100_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
+	EFX_GENERIC_SW_STAT(rx_nodesc_trunc),
+	EFX_GENERIC_SW_STAT(rx_noskb_drops),
+};
+
+static size_t ef100_describe_stats(struct efx_nic *efx, u8 *names)
+{
+	DECLARE_BITMAP(mask, EF100_STAT_COUNT) = {};
+
+	ef100_ethtool_stat_mask(mask);
+	return efx_nic_describe_stats(ef100_stat_desc, EF100_STAT_COUNT,
+				      mask, names);
+}
+
+static size_t ef100_update_stats_common(struct efx_nic *efx, u64 *full_stats,
+					struct rtnl_link_stats64 *core_stats)
+{
+	struct ef100_nic_data *nic_data = efx->nic_data;
+	DECLARE_BITMAP(mask, EF100_STAT_COUNT) = {};
+	size_t stats_count = 0, index;
+	u64 *stats = nic_data->stats;
+
+	ef100_ethtool_stat_mask(mask);
+
+	if (full_stats) {
+		for_each_set_bit(index, mask, EF100_STAT_COUNT) {
+			if (ef100_stat_desc[index].name) {
+				*full_stats++ = stats[index];
+				++stats_count;
+			}
+		}
+	}
+
+	if (!core_stats)
+		return stats_count;
+
+	core_stats->rx_packets = stats[EF100_STAT_port_rx_packets];
+	core_stats->tx_packets = stats[EF100_STAT_port_tx_packets];
+	core_stats->rx_bytes = stats[EF100_STAT_port_rx_bytes];
+	core_stats->tx_bytes = stats[EF100_STAT_port_tx_bytes];
+	core_stats->rx_dropped = stats[EF100_STAT_port_rx_nodesc_drops] +
+				 stats[GENERIC_STAT_rx_nodesc_trunc] +
+				 stats[GENERIC_STAT_rx_noskb_drops];
+	core_stats->multicast = stats[EF100_STAT_port_rx_multicast];
+	core_stats->rx_length_errors =
+			stats[EF100_STAT_port_rx_gtjumbo] +
+			stats[EF100_STAT_port_rx_length_error];
+	core_stats->rx_crc_errors = stats[EF100_STAT_port_rx_bad];
+	core_stats->rx_frame_errors =
+			stats[EF100_STAT_port_rx_align_error];
+	core_stats->rx_fifo_errors = stats[EF100_STAT_port_rx_overflow];
+	core_stats->rx_errors = (core_stats->rx_length_errors +
+				 core_stats->rx_crc_errors +
+				 core_stats->rx_frame_errors);
+
+	return stats_count;
+}
+
+static size_t ef100_update_stats(struct efx_nic *efx,
+				 u64 *full_stats,
+				 struct rtnl_link_stats64 *core_stats)
+{
+	__le64 *mc_stats = kmalloc(array_size(efx->num_mac_stats, sizeof(__le64)), GFP_ATOMIC);
+	struct ef100_nic_data *nic_data = efx->nic_data;
+	DECLARE_BITMAP(mask, EF100_STAT_COUNT) = {};
+	u64 *stats = nic_data->stats;
+
+	ef100_common_stat_mask(mask);
+	ef100_ethtool_stat_mask(mask);
+
+	efx_nic_copy_stats(efx, mc_stats);
+	efx_nic_update_stats(ef100_stat_desc, EF100_STAT_COUNT, mask,
+			     stats, mc_stats, false);
+
+	kfree(mc_stats);
+
+	return ef100_update_stats_common(efx, full_stats, core_stats);
+}
+
 static int efx_ef100_get_phys_port_id(struct efx_nic *efx,
 				      struct netdev_phys_item_id *ppid)
 {
@@ -557,6 +722,11 @@ const struct efx_nic_type ef100_pf_nic_type = {
 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
 
 	.reconfigure_mac = ef100_reconfigure_mac,
+	.describe_stats = ef100_describe_stats,
+	.start_stats = efx_mcdi_mac_start_stats,
+	.update_stats = ef100_update_stats,
+	.pull_stats = efx_mcdi_mac_pull_stats,
+	.stop_stats = efx_mcdi_mac_stop_stats,
 
 	/* Per-type bar/size configuration not used on ef100. Location of
 	 * registers is defined by extended capabilities.
diff --git a/drivers/net/ethernet/sfc/ef100_nic.h b/drivers/net/ethernet/sfc/ef100_nic.h
index c8816bc6ae78f..7c2d374900743 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.h
+++ b/drivers/net/ethernet/sfc/ef100_nic.h
@@ -17,6 +17,46 @@ extern const struct efx_nic_type ef100_pf_nic_type;
 int ef100_probe_pf(struct efx_nic *efx);
 void ef100_remove(struct efx_nic *efx);
 
+enum {
+	EF100_STAT_port_tx_bytes = GENERIC_STAT_COUNT,
+	EF100_STAT_port_tx_packets,
+	EF100_STAT_port_tx_pause,
+	EF100_STAT_port_tx_unicast,
+	EF100_STAT_port_tx_multicast,
+	EF100_STAT_port_tx_broadcast,
+	EF100_STAT_port_tx_lt64,
+	EF100_STAT_port_tx_64,
+	EF100_STAT_port_tx_65_to_127,
+	EF100_STAT_port_tx_128_to_255,
+	EF100_STAT_port_tx_256_to_511,
+	EF100_STAT_port_tx_512_to_1023,
+	EF100_STAT_port_tx_1024_to_15xx,
+	EF100_STAT_port_tx_15xx_to_jumbo,
+	EF100_STAT_port_rx_bytes,
+	EF100_STAT_port_rx_packets,
+	EF100_STAT_port_rx_good,
+	EF100_STAT_port_rx_bad,
+	EF100_STAT_port_rx_pause,
+	EF100_STAT_port_rx_unicast,
+	EF100_STAT_port_rx_multicast,
+	EF100_STAT_port_rx_broadcast,
+	EF100_STAT_port_rx_lt64,
+	EF100_STAT_port_rx_64,
+	EF100_STAT_port_rx_65_to_127,
+	EF100_STAT_port_rx_128_to_255,
+	EF100_STAT_port_rx_256_to_511,
+	EF100_STAT_port_rx_512_to_1023,
+	EF100_STAT_port_rx_1024_to_15xx,
+	EF100_STAT_port_rx_15xx_to_jumbo,
+	EF100_STAT_port_rx_gtjumbo,
+	EF100_STAT_port_rx_bad_gtjumbo,
+	EF100_STAT_port_rx_align_error,
+	EF100_STAT_port_rx_length_error,
+	EF100_STAT_port_rx_overflow,
+	EF100_STAT_port_rx_nodesc_drops,
+	EF100_STAT_COUNT
+};
+
 struct ef100_nic_data {
 	struct efx_nic *efx;
 	struct efx_buffer mcdi_buf;
@@ -26,6 +66,7 @@ struct ef100_nic_data {
 	u16 warm_boot_count;
 	u8 port_id[ETH_ALEN];
 	DECLARE_BITMAP(evq_phases, EFX_MAX_CHANNELS);
+	u64 stats[EF100_STAT_COUNT];
 	u16 tso_max_hdr_len;
 	u16 tso_max_payload_num_segs;
 	u16 tso_max_frames;

From 43c3df0d56474dd27e1f2a1ca70e9060d341e695 Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:37:50 +0100
Subject: [PATCH 09/11] sfc_ef100: functions for selftests

Self-tests for event and interrupt reception and NVRAM.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c | 47 ++++++++++++++++++++++++++++
 1 file changed, 47 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index c81da0b2f0c1b..90fc44052abf1 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -632,6 +632,50 @@ static int efx_ef100_get_phys_port_id(struct efx_nic *efx,
 	return 0;
 }
 
+static int efx_ef100_irq_test_generate(struct efx_nic *efx)
+{
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
+
+	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
+
+	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
+	return efx_mcdi_rpc_quiet(efx, MC_CMD_TRIGGER_INTERRUPT,
+				  inbuf, sizeof(inbuf), NULL, 0, NULL);
+}
+
+#define EFX_EF100_TEST 1
+
+static void efx_ef100_ev_test_generate(struct efx_channel *channel)
+{
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
+	struct efx_nic *efx = channel->efx;
+	efx_qword_t event;
+	int rc;
+
+	EFX_POPULATE_QWORD_2(event,
+			     ESF_GZ_E_TYPE, ESE_GZ_EF100_EV_DRIVER,
+			     ESF_GZ_DRIVER_DATA, EFX_EF100_TEST);
+
+	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
+
+	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
+	 * already swapped the data to little-endian order.
+	 */
+	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
+	       sizeof(efx_qword_t));
+
+	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
+			  NULL, 0, NULL);
+	if (rc && (rc != -ENETDOWN))
+		goto fail;
+
+	return;
+
+fail:
+	WARN_ON(true);
+	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+}
+
 static unsigned int ef100_check_caps(const struct efx_nic *efx,
 				     u8 flag, u32 offset)
 {
@@ -668,6 +712,7 @@ const struct efx_nic_type ef100_pf_nic_type = {
 	.mcdi_poll_reboot = ef100_mcdi_poll_reboot,
 	.mcdi_reboot_detected = ef100_mcdi_reboot_detected,
 	.irq_enable_master = efx_port_dummy_op_void,
+	.irq_test_generate = efx_ef100_irq_test_generate,
 	.irq_disable_non_ev = efx_port_dummy_op_void,
 	.push_irq_moderation = efx_channel_dummy_op_void,
 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
@@ -684,6 +729,7 @@ const struct efx_nic_type ef100_pf_nic_type = {
 	.irq_handle_msi = ef100_msi_interrupt,
 	.ev_process = ef100_ev_process,
 	.ev_read_ack = ef100_ev_read_ack,
+	.ev_test_generate = efx_ef100_ev_test_generate,
 	.tx_probe = ef100_tx_probe,
 	.tx_init = ef100_tx_init,
 	.tx_write = ef100_tx_write,
@@ -722,6 +768,7 @@ const struct efx_nic_type ef100_pf_nic_type = {
 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
 
 	.reconfigure_mac = ef100_reconfigure_mac,
+	.test_nvram = efx_new_mcdi_nvram_test_all,
 	.describe_stats = ef100_describe_stats,
 	.start_stats = efx_mcdi_mac_start_stats,
 	.update_stats = ef100_update_stats,

From ef2c57b956143c300adef4b89438e1da3db4cf32 Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:38:49 +0100
Subject: [PATCH 10/11] sfc_ef100: read pf_index at probe time

We'll need it later, for VF representors.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100_nic.c | 4 ++++
 drivers/net/ethernet/sfc/ef100_nic.h | 1 +
 2 files changed, 5 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 90fc44052abf1..10748efbf98e7 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -1096,6 +1096,10 @@ static int ef100_probe_main(struct efx_nic *efx)
 	if (rc)
 		goto fail;
 
+	rc = efx_get_pf_index(efx, &nic_data->pf_index);
+	if (rc)
+		goto fail;
+
 	rc = efx_ef100_init_datapath_caps(efx);
 	if (rc < 0)
 		goto fail;
diff --git a/drivers/net/ethernet/sfc/ef100_nic.h b/drivers/net/ethernet/sfc/ef100_nic.h
index 7c2d374900743..4a64c94384932 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.h
+++ b/drivers/net/ethernet/sfc/ef100_nic.h
@@ -63,6 +63,7 @@ struct ef100_nic_data {
 	u32 datapath_caps;
 	u32 datapath_caps2;
 	u32 datapath_caps3;
+	unsigned int pf_index;
 	u16 warm_boot_count;
 	u8 port_id[ETH_ALEN];
 	DECLARE_BITMAP(evq_phases, EFX_MAX_CHANNELS);

From d61592a112928a11cd04081adf51ba4f752859ed Mon Sep 17 00:00:00 2001
From: Edward Cree <ecree@solarflare.com>
Date: Mon, 3 Aug 2020 21:40:01 +0100
Subject: [PATCH 11/11] sfc_ef100: add nic-type for VFs, and bind to them

We don't yet have a .sriov_configure() to create them, though.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 drivers/net/ethernet/sfc/ef100.c     |  2 +
 drivers/net/ethernet/sfc/ef100_nic.c | 77 ++++++++++++++++++++++++++++
 drivers/net/ethernet/sfc/ef100_nic.h |  2 +
 3 files changed, 81 insertions(+)

diff --git a/drivers/net/ethernet/sfc/ef100.c b/drivers/net/ethernet/sfc/ef100.c
index de611c0f94e71..9729983f48409 100644
--- a/drivers/net/ethernet/sfc/ef100.c
+++ b/drivers/net/ethernet/sfc/ef100.c
@@ -527,6 +527,8 @@ static int ef100_pci_probe(struct pci_dev *pci_dev,
 static const struct pci_device_id ef100_pci_table[] = {
 	{PCI_DEVICE(PCI_VENDOR_ID_XILINX, 0x0100),  /* Riverhead PF */
 		.driver_data = (unsigned long) &ef100_pf_nic_type },
+	{PCI_DEVICE(PCI_VENDOR_ID_XILINX, 0x1100),  /* Riverhead VF */
+		.driver_data = (unsigned long) &ef100_vf_nic_type },
 	{0}                     /* end of list */
 };
 
diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c
index 10748efbf98e7..8a2126fec078b 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.c
+++ b/drivers/net/ethernet/sfc/ef100_nic.c
@@ -783,6 +783,78 @@ const struct efx_nic_type ef100_pf_nic_type = {
 
 };
 
+const struct efx_nic_type ef100_vf_nic_type = {
+	.revision = EFX_REV_EF100,
+	.is_vf = true,
+	.probe = ef100_probe_vf,
+	.offload_features = EF100_OFFLOAD_FEATURES,
+	.mcdi_max_ver = 2,
+	.mcdi_request = ef100_mcdi_request,
+	.mcdi_poll_response = ef100_mcdi_poll_response,
+	.mcdi_read_response = ef100_mcdi_read_response,
+	.mcdi_poll_reboot = ef100_mcdi_poll_reboot,
+	.mcdi_reboot_detected = ef100_mcdi_reboot_detected,
+	.irq_enable_master = efx_port_dummy_op_void,
+	.irq_test_generate = efx_ef100_irq_test_generate,
+	.irq_disable_non_ev = efx_port_dummy_op_void,
+	.push_irq_moderation = efx_channel_dummy_op_void,
+	.min_interrupt_mode = EFX_INT_MODE_MSIX,
+	.map_reset_reason = ef100_map_reset_reason,
+	.map_reset_flags = ef100_map_reset_flags,
+	.reset = ef100_reset,
+	.check_caps = ef100_check_caps,
+	.ev_probe = ef100_ev_probe,
+	.ev_init = ef100_ev_init,
+	.ev_fini = efx_mcdi_ev_fini,
+	.ev_remove = efx_mcdi_ev_remove,
+	.irq_handle_msi = ef100_msi_interrupt,
+	.ev_process = ef100_ev_process,
+	.ev_read_ack = ef100_ev_read_ack,
+	.ev_test_generate = efx_ef100_ev_test_generate,
+	.tx_probe = ef100_tx_probe,
+	.tx_init = ef100_tx_init,
+	.tx_write = ef100_tx_write,
+	.tx_enqueue = ef100_enqueue_skb,
+	.rx_probe = efx_mcdi_rx_probe,
+	.rx_init = efx_mcdi_rx_init,
+	.rx_remove = efx_mcdi_rx_remove,
+	.rx_write = ef100_rx_write,
+	.rx_packet = __ef100_rx_packet,
+	.fini_dmaq = efx_fini_dmaq,
+	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
+	.filter_table_probe = ef100_filter_table_up,
+	.filter_table_restore = efx_mcdi_filter_table_restore,
+	.filter_table_remove = ef100_filter_table_down,
+	.filter_insert = efx_mcdi_filter_insert,
+	.filter_remove_safe = efx_mcdi_filter_remove_safe,
+	.filter_get_safe = efx_mcdi_filter_get_safe,
+	.filter_clear_rx = efx_mcdi_filter_clear_rx,
+	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
+	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
+	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
+	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
+
+	.rx_prefix_size = ESE_GZ_RX_PKT_PREFIX_LEN,
+	.rx_hash_offset = ESF_GZ_RX_PREFIX_RSS_HASH_LBN / 8,
+	.rx_ts_offset = ESF_GZ_RX_PREFIX_PARTIAL_TSTAMP_LBN / 8,
+	.rx_hash_key_size = 40,
+	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
+	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
+	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
+
+	.reconfigure_mac = ef100_reconfigure_mac,
+	.test_nvram = efx_new_mcdi_nvram_test_all,
+	.describe_stats = ef100_describe_stats,
+	.start_stats = efx_mcdi_mac_start_stats,
+	.update_stats = ef100_update_stats,
+	.pull_stats = efx_mcdi_mac_pull_stats,
+	.stop_stats = efx_mcdi_mac_stop_stats,
+
+	.mem_bar = NULL,
+	.mem_map_size = NULL,
+
+};
+
 static int compare_versions(const char *a, const char *b)
 {
 	int a_major, a_minor, a_point, a_patch;
@@ -1178,6 +1250,11 @@ int ef100_probe_pf(struct efx_nic *efx)
 	return rc;
 }
 
+int ef100_probe_vf(struct efx_nic *efx)
+{
+	return ef100_probe_main(efx);
+}
+
 void ef100_remove(struct efx_nic *efx)
 {
 	struct ef100_nic_data *nic_data = efx->nic_data;
diff --git a/drivers/net/ethernet/sfc/ef100_nic.h b/drivers/net/ethernet/sfc/ef100_nic.h
index 4a64c94384932..e799688d52647 100644
--- a/drivers/net/ethernet/sfc/ef100_nic.h
+++ b/drivers/net/ethernet/sfc/ef100_nic.h
@@ -13,8 +13,10 @@
 #include "nic_common.h"
 
 extern const struct efx_nic_type ef100_pf_nic_type;
+extern const struct efx_nic_type ef100_vf_nic_type;
 
 int ef100_probe_pf(struct efx_nic *efx);
+int ef100_probe_vf(struct efx_nic *efx);
 void ef100_remove(struct efx_nic *efx);
 
 enum {