sfc: Add support for IEEE-1588 PTP
authorStuart Hodgson <smhodgson@solarflare.com>
Mon, 3 Sep 2012 10:09:36 +0000 (11:09 +0100)
committerBen Hutchings <bhutchings@solarflare.com>
Wed, 19 Sep 2012 01:54:12 +0000 (02:54 +0100)
Add PTP IEEE-1588 support and make accesible via the PHC subsystem.

This work is based on prior code by Andrew Jackson

Signed-off-by: Stuart Hodgson <smhodgson@solarflare.com>
[bwh:
 - Add byte order conversion in efx_ptp_send_times()
 - Simplify conversion of PPS event times
 - Add the built-in vs module check to CONFIG_SFC_PTP dependencies]
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
drivers/net/ethernet/sfc/Kconfig
drivers/net/ethernet/sfc/Makefile
drivers/net/ethernet/sfc/efx.c
drivers/net/ethernet/sfc/ethtool.c
drivers/net/ethernet/sfc/mcdi.c
drivers/net/ethernet/sfc/mcdi_pcol.h
drivers/net/ethernet/sfc/net_driver.h
drivers/net/ethernet/sfc/nic.h
drivers/net/ethernet/sfc/ptp.c [new file with mode: 0644]
drivers/net/ethernet/sfc/siena.c
drivers/net/ethernet/sfc/tx.c

index fb3cbc27063cccb5ae821989c516195527e32e65..25906c1d1b1590825502bc192dffa1aa8d8e494e 100644 (file)
@@ -34,3 +34,10 @@ config SFC_SRIOV
          This enables support for the SFC9000 I/O Virtualization
          features, allowing accelerated network performance in
          virtualized environments.
+config SFC_PTP
+       bool "Solarflare SFC9000-family PTP support"
+       depends on SFC && PTP_1588_CLOCK && !(SFC=y && PTP_1588_CLOCK=m)
+       default y
+       ---help---
+         This enables support for the Precision Time Protocol (PTP)
+         on SFC9000-family NICs
index ea1f8db5731811cbe7c7005931b1c4e4e8db5593..e11f2ecf69d9b5dfd0ae5600e33bb0fe1d7382fc 100644 (file)
@@ -5,5 +5,6 @@ sfc-y                   += efx.o nic.o falcon.o siena.o tx.o rx.o filter.o \
                           mcdi.o mcdi_phy.o mcdi_mon.o
 sfc-$(CONFIG_SFC_MTD)  += mtd.o
 sfc-$(CONFIG_SFC_SRIOV)        += siena_sriov.o
+sfc-$(CONFIG_SFC_PTP)  += ptp.o
 
 obj-$(CONFIG_SFC)      += sfc.o
index 8b79a6413fe4568b37693ff0c2b8a8b4296f6594..96bd980e828da5d28b0bed6329cb85e3655d598f 100644 (file)
@@ -1779,6 +1779,9 @@ static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
        struct efx_nic *efx = netdev_priv(net_dev);
        struct mii_ioctl_data *data = if_mii(ifr);
 
+       if (cmd == SIOCSHWTSTAMP)
+               return efx_ptp_ioctl(efx, ifr, cmd);
+
        /* Convert phy_id from older PRTAD/DEVAD format */
        if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
            (data->phy_id & 0xfc00) == 0x0400)
index f8e7e204981fce68937354ac3cab5bfefd24b389..9df556c01b8e26412d047603bda7c6c51b9cef65 100644 (file)
@@ -1174,6 +1174,7 @@ const struct ethtool_ops efx_ethtool_ops = {
        .get_rxfh_indir_size    = efx_ethtool_get_rxfh_indir_size,
        .get_rxfh_indir         = efx_ethtool_get_rxfh_indir,
        .set_rxfh_indir         = efx_ethtool_set_rxfh_indir,
+       .get_ts_info            = efx_ptp_get_ts_info,
        .get_module_info        = efx_ethtool_get_module_info,
        .get_module_eeprom      = efx_ethtool_get_module_eeprom,
 };
index 2707e86def9ec21c5dbc0bd76611c9188151cbc1..294df4bca4a6a448705a5fe8d62b5ea216459741 100644 (file)
@@ -578,6 +578,11 @@ void efx_mcdi_process_event(struct efx_channel *channel,
        case MCDI_EVENT_CODE_FLR:
                efx_sriov_flr(efx, MCDI_EVENT_FIELD(*event, FLR_VF));
                break;
+       case MCDI_EVENT_CODE_PTP_RX:
+       case MCDI_EVENT_CODE_PTP_FAULT:
+       case MCDI_EVENT_CODE_PTP_PPS:
+               efx_ptp_event(efx, event);
+               break;
 
        default:
                netif_err(efx, hw, efx->net_dev, "Unknown MCDI event 0x%x\n",
index 50389326bec8ce229d011819f790a9e95d9bc566..9d426d0457bdd5140709b1da8186d817d5b2bbd2 100644 (file)
 #define          MCDI_EVENT_CODE_TX_FLUSH  0xc /* enum */
 #define          MCDI_EVENT_CODE_PTP_RX  0xd /* enum */
 #define          MCDI_EVENT_CODE_PTP_FAULT  0xe /* enum */
+#define          MCDI_EVENT_CODE_PTP_PPS  0xf /* enum */
 #define       MCDI_EVENT_CMDDONE_DATA_OFST 0
 #define       MCDI_EVENT_CMDDONE_DATA_LBN 0
 #define       MCDI_EVENT_CMDDONE_DATA_WIDTH 32
index 0f0926e6896357de71c2dc9d350ca7c3a28b1461..797dbed25d94f3322ec9fcbe6f589d54992e779f 100644 (file)
@@ -56,7 +56,8 @@
 #define EFX_MAX_CHANNELS 32U
 #define EFX_MAX_RX_QUEUES EFX_MAX_CHANNELS
 #define EFX_EXTRA_CHANNEL_IOV  0
-#define EFX_MAX_EXTRA_CHANNELS 1U
+#define EFX_EXTRA_CHANNEL_PTP  1
+#define EFX_MAX_EXTRA_CHANNELS 2U
 
 /* Checksum generation is a per-queue option in hardware, so each
  * queue visible to the networking core is backed by two hardware TX
@@ -68,6 +69,9 @@
 #define EFX_TXQ_TYPES          4
 #define EFX_MAX_TX_QUEUES      (EFX_TXQ_TYPES * EFX_MAX_CHANNELS)
 
+/* Forward declare Precision Time Protocol (PTP) support structure. */
+struct efx_ptp_data;
+
 struct efx_self_tests;
 
 /**
@@ -736,6 +740,7 @@ struct vfdi_status;
  *     %local_addr_list. Protected by %local_lock.
  * @local_lock: Mutex protecting %local_addr_list and %local_page_list.
  * @peer_work: Work item to broadcast peer addresses to VMs.
+ * @ptp_data: PTP state data
  * @monitor_work: Hardware monitor workitem
  * @biu_lock: BIU (bus interface unit) lock
  * @last_irq_cpu: Last CPU to handle a possible test interrupt.  This
@@ -863,6 +868,10 @@ struct efx_nic {
        struct work_struct peer_work;
 #endif
 
+#ifdef CONFIG_SFC_PTP
+       struct efx_ptp_data *ptp_data;
+#endif
+
        /* The following fields may be written more often */
 
        struct delayed_work monitor_work ____cacheline_aligned_in_smp;
@@ -1125,5 +1134,13 @@ static inline void clear_bit_le(unsigned nr, unsigned char *addr)
 #define EFX_MAX_FRAME_LEN(mtu) \
        ((((mtu) + ETH_HLEN + VLAN_HLEN + 4/* FCS */ + 7) & ~7) + 16)
 
+static inline bool efx_xmit_with_hwtstamp(struct sk_buff *skb)
+{
+       return skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP;
+}
+static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
+{
+       skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+}
 
 #endif /* EFX_NET_DRIVER_H */
index bab5cd9f5740bb8e9e7476849f089ba513346b9f..438cef11f7270bd620ed04c42fb1c8ee8a9d3621 100644 (file)
@@ -11,6 +11,7 @@
 #ifndef EFX_NIC_H
 #define EFX_NIC_H
 
+#include <linux/net_tstamp.h>
 #include <linux/i2c-algo-bit.h>
 #include "net_driver.h"
 #include "efx.h"
@@ -250,6 +251,41 @@ extern int efx_sriov_get_vf_config(struct net_device *dev, int vf,
 extern int efx_sriov_set_vf_spoofchk(struct net_device *net_dev, int vf,
                                     bool spoofchk);
 
+struct ethtool_ts_info;
+#ifdef CONFIG_SFC_PTP
+extern void efx_ptp_probe(struct efx_nic *efx);
+extern int efx_ptp_ioctl(struct efx_nic *efx, struct ifreq *ifr, int cmd);
+extern int efx_ptp_get_ts_info(struct net_device *net_dev,
+                              struct ethtool_ts_info *ts_info);
+extern bool efx_ptp_is_ptp_tx(struct efx_nic *efx, struct sk_buff *skb);
+extern int efx_ptp_tx(struct efx_nic *efx, struct sk_buff *skb);
+extern void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev);
+#else
+static inline void efx_ptp_probe(struct efx_nic *efx) {}
+static inline int efx_ptp_ioctl(struct efx_nic *efx, struct ifreq *ifr, int cmd)
+{
+       return -EOPNOTSUPP;
+}
+static inline int efx_ptp_get_ts_info(struct net_device *net_dev,
+                                     struct ethtool_ts_info *ts_info)
+{
+       ts_info->so_timestamping = (SOF_TIMESTAMPING_SOFTWARE |
+                                   SOF_TIMESTAMPING_RX_SOFTWARE);
+       ts_info->phc_index = -1;
+
+       return 0;
+}
+static inline bool efx_ptp_is_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
+{
+       return false;
+}
+static inline int efx_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
+{
+       return NETDEV_TX_OK;
+}
+static inline void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev) {}
+#endif
+
 extern const struct efx_nic_type falcon_a1_nic_type;
 extern const struct efx_nic_type falcon_b0_nic_type;
 extern const struct efx_nic_type siena_a0_nic_type;
diff --git a/drivers/net/ethernet/sfc/ptp.c b/drivers/net/ethernet/sfc/ptp.c
new file mode 100644 (file)
index 0000000..2b07a4e
--- /dev/null
@@ -0,0 +1,1483 @@
+/****************************************************************************
+ * Driver for Solarflare Solarstorm network controllers and boards
+ * Copyright 2011 Solarflare Communications Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation, incorporated herein by reference.
+ */
+
+/* Theory of operation:
+ *
+ * PTP support is assisted by firmware running on the MC, which provides
+ * the hardware timestamping capabilities.  Both transmitted and received
+ * PTP event packets are queued onto internal queues for subsequent processing;
+ * this is because the MC operations are relatively long and would block
+ * block NAPI/interrupt operation.
+ *
+ * Receive event processing:
+ *     The event contains the packet's UUID and sequence number, together
+ *     with the hardware timestamp.  The PTP receive packet queue is searched
+ *     for this UUID/sequence number and, if found, put on a pending queue.
+ *     Packets not matching are delivered without timestamps (MCDI events will
+ *     always arrive after the actual packet).
+ *     It is important for the operation of the PTP protocol that the ordering
+ *     of packets between the event and general port is maintained.
+ *
+ * Work queue processing:
+ *     If work waiting, synchronise host/hardware time
+ *
+ *     Transmit: send packet through MC, which returns the transmission time
+ *     that is converted to an appropriate timestamp.
+ *
+ *     Receive: the packet's reception time is converted to an appropriate
+ *     timestamp.
+ */
+#include <linux/ip.h>
+#include <linux/udp.h>
+#include <linux/time.h>
+#include <linux/ktime.h>
+#include <linux/module.h>
+#include <linux/net_tstamp.h>
+#include <linux/pps_kernel.h>
+#include <linux/ptp_clock_kernel.h>
+#include "net_driver.h"
+#include "efx.h"
+#include "mcdi.h"
+#include "mcdi_pcol.h"
+#include "io.h"
+#include "regs.h"
+#include "nic.h"
+
+/* Maximum number of events expected to make up a PTP event */
+#define        MAX_EVENT_FRAGS                 3
+
+/* Maximum delay, ms, to begin synchronisation */
+#define        MAX_SYNCHRONISE_WAIT_MS         2
+
+/* How long, at most, to spend synchronising */
+#define        SYNCHRONISE_PERIOD_NS           250000
+
+/* How often to update the shared memory time */
+#define        SYNCHRONISATION_GRANULARITY_NS  200
+
+/* Minimum permitted length of a (corrected) synchronisation time */
+#define        MIN_SYNCHRONISATION_NS          120
+
+/* Maximum permitted length of a (corrected) synchronisation time */
+#define        MAX_SYNCHRONISATION_NS          1000
+
+/* How many (MC) receive events that can be queued */
+#define        MAX_RECEIVE_EVENTS              8
+
+/* Length of (modified) moving average. */
+#define        AVERAGE_LENGTH                  16
+
+/* How long an unmatched event or packet can be held */
+#define PKT_EVENT_LIFETIME_MS          10
+
+/* Offsets into PTP packet for identification.  These offsets are from the
+ * start of the IP header, not the MAC header.  Note that neither PTP V1 nor
+ * PTP V2 permit the use of IPV4 options.
+ */
+#define PTP_DPORT_OFFSET       22
+
+#define PTP_V1_VERSION_LENGTH  2
+#define PTP_V1_VERSION_OFFSET  28
+
+#define PTP_V1_UUID_LENGTH     6
+#define PTP_V1_UUID_OFFSET     50
+
+#define PTP_V1_SEQUENCE_LENGTH 2
+#define PTP_V1_SEQUENCE_OFFSET 58
+
+/* The minimum length of a PTP V1 packet for offsets, etc. to be valid:
+ * includes IP header.
+ */
+#define        PTP_V1_MIN_LENGTH       64
+
+#define PTP_V2_VERSION_LENGTH  1
+#define PTP_V2_VERSION_OFFSET  29
+
+/* Although PTP V2 UUIDs are comprised a ClockIdentity (8) and PortNumber (2),
+ * the MC only captures the last six bytes of the clock identity. These values
+ * reflect those, not the ones used in the standard.  The standard permits
+ * mapping of V1 UUIDs to V2 UUIDs with these same values.
+ */
+#define PTP_V2_MC_UUID_LENGTH  6
+#define PTP_V2_MC_UUID_OFFSET  50
+
+#define PTP_V2_SEQUENCE_LENGTH 2
+#define PTP_V2_SEQUENCE_OFFSET 58
+
+/* The minimum length of a PTP V2 packet for offsets, etc. to be valid:
+ * includes IP header.
+ */
+#define        PTP_V2_MIN_LENGTH       63
+
+#define        PTP_MIN_LENGTH          63
+
+#define PTP_ADDRESS            0xe0000181      /* 224.0.1.129 */
+#define PTP_EVENT_PORT         319
+#define PTP_GENERAL_PORT       320
+
+/* Annoyingly the format of the version numbers are different between
+ * versions 1 and 2 so it isn't possible to simply look for 1 or 2.
+ */
+#define        PTP_VERSION_V1          1
+
+#define        PTP_VERSION_V2          2
+#define        PTP_VERSION_V2_MASK     0x0f
+
+enum ptp_packet_state {
+       PTP_PACKET_STATE_UNMATCHED = 0,
+       PTP_PACKET_STATE_MATCHED,
+       PTP_PACKET_STATE_TIMED_OUT,
+       PTP_PACKET_STATE_MATCH_UNWANTED
+};
+
+/* NIC synchronised with single word of time only comprising
+ * partial seconds and full nanoseconds: 10^9 ~ 2^30 so 2 bits for seconds.
+ */
+#define        MC_NANOSECOND_BITS      30
+#define        MC_NANOSECOND_MASK      ((1 << MC_NANOSECOND_BITS) - 1)
+#define        MC_SECOND_MASK          ((1 << (32 - MC_NANOSECOND_BITS)) - 1)
+
+/* Maximum parts-per-billion adjustment that is acceptable */
+#define MAX_PPB                        1000000
+
+/* Number of bits required to hold the above */
+#define        MAX_PPB_BITS            20
+
+/* Number of extra bits allowed when calculating fractional ns.
+ * EXTRA_BITS + MC_CMD_PTP_IN_ADJUST_BITS + MAX_PPB_BITS should
+ * be less than 63.
+ */
+#define        PPB_EXTRA_BITS          2
+
+/* Precalculate scale word to avoid long long division at runtime */
+#define        PPB_SCALE_WORD  ((1LL << (PPB_EXTRA_BITS + MC_CMD_PTP_IN_ADJUST_BITS +\
+                       MAX_PPB_BITS)) / 1000000000LL)
+
+#define PTP_SYNC_ATTEMPTS      4
+
+/**
+ * struct efx_ptp_match - Matching structure, stored in sk_buff's cb area.
+ * @words: UUID and (partial) sequence number
+ * @expiry: Time after which the packet should be delivered irrespective of
+ *            event arrival.
+ * @state: The state of the packet - whether it is ready for processing or
+ *         whether that is of no interest.
+ */
+struct efx_ptp_match {
+       u32 words[DIV_ROUND_UP(PTP_V1_UUID_LENGTH, 4)];
+       unsigned long expiry;
+       enum ptp_packet_state state;
+};
+
+/**
+ * struct efx_ptp_event_rx - A PTP receive event (from MC)
+ * @seq0: First part of (PTP) UUID
+ * @seq1: Second part of (PTP) UUID and sequence number
+ * @hwtimestamp: Event timestamp
+ */
+struct efx_ptp_event_rx {
+       struct list_head link;
+       u32 seq0;
+       u32 seq1;
+       ktime_t hwtimestamp;
+       unsigned long expiry;
+};
+
+/**
+ * struct efx_ptp_timeset - Synchronisation between host and MC
+ * @host_start: Host time immediately before hardware timestamp taken
+ * @seconds: Hardware timestamp, seconds
+ * @nanoseconds: Hardware timestamp, nanoseconds
+ * @host_end: Host time immediately after hardware timestamp taken
+ * @waitns: Number of nanoseconds between hardware timestamp being read and
+ *          host end time being seen
+ * @window: Difference of host_end and host_start
+ * @valid: Whether this timeset is valid
+ */
+struct efx_ptp_timeset {
+       u32 host_start;
+       u32 seconds;
+       u32 nanoseconds;
+       u32 host_end;
+       u32 waitns;
+       u32 window;     /* Derived: end - start, allowing for wrap */
+};
+
+/**
+ * struct efx_ptp_data - Precision Time Protocol (PTP) state
+ * @channel: The PTP channel
+ * @rxq: Receive queue (awaiting timestamps)
+ * @txq: Transmit queue
+ * @evt_list: List of MC receive events awaiting packets
+ * @evt_free_list: List of free events
+ * @evt_lock: Lock for manipulating evt_list and evt_free_list
+ * @rx_evts: Instantiated events (on evt_list and evt_free_list)
+ * @workwq: Work queue for processing pending PTP operations
+ * @work: Work task
+ * @reset_required: A serious error has occurred and the PTP task needs to be
+ *                  reset (disable, enable).
+ * @rxfilter_event: Receive filter when operating
+ * @rxfilter_general: Receive filter when operating
+ * @config: Current timestamp configuration
+ * @enabled: PTP operation enabled
+ * @mode: Mode in which PTP operating (PTP version)
+ * @evt_frags: Partly assembled PTP events
+ * @evt_frag_idx: Current fragment number
+ * @evt_code: Last event code
+ * @start: Address at which MC indicates ready for synchronisation
+ * @host_time_pps: Host time at last PPS
+ * @last_sync_ns: Last number of nanoseconds between readings when synchronising
+ * @base_sync_ns: Number of nanoseconds for last synchronisation.
+ * @base_sync_valid: Whether base_sync_time is valid.
+ * @current_adjfreq: Current ppb adjustment.
+ * @phc_clock: Pointer to registered phc device
+ * @phc_clock_info: Registration structure for phc device
+ * @pps_work: pps work task for handling pps events
+ * @pps_workwq: pps work queue
+ * @nic_ts_enabled: Flag indicating if NIC generated TS events are handled
+ * @txbuf: Buffer for use when transmitting (PTP) packets to MC (avoids
+ *         allocations in main data path).
+ * @debug_ptp_dir: PTP debugfs directory
+ * @missed_rx_sync: Number of packets received without syncrhonisation.
+ * @good_syncs: Number of successful synchronisations.
+ * @no_time_syncs: Number of synchronisations with no good times.
+ * @bad_sync_durations: Number of synchronisations with bad durations.
+ * @bad_syncs: Number of failed synchronisations.
+ * @last_sync_time: Number of nanoseconds for last synchronisation.
+ * @sync_timeouts: Number of synchronisation timeouts
+ * @fast_syncs: Number of synchronisations requiring short delay
+ * @min_sync_delta: Minimum time between event and synchronisation
+ * @max_sync_delta: Maximum time between event and synchronisation
+ * @average_sync_delta: Average time between event and synchronisation.
+ *                      Modified moving average.
+ * @last_sync_delta: Last time between event and synchronisation
+ * @mc_stats: Context value for MC statistics
+ * @timeset: Last set of synchronisation statistics.
+ */
+struct efx_ptp_data {
+       struct efx_channel *channel;
+       struct sk_buff_head rxq;
+       struct sk_buff_head txq;
+       struct list_head evt_list;
+       struct list_head evt_free_list;
+       spinlock_t evt_lock;
+       struct efx_ptp_event_rx rx_evts[MAX_RECEIVE_EVENTS];
+       struct workqueue_struct *workwq;
+       struct work_struct work;
+       bool reset_required;
+       u32 rxfilter_event;
+       u32 rxfilter_general;
+       bool rxfilter_installed;
+       struct hwtstamp_config config;
+       bool enabled;
+       unsigned int mode;
+       efx_qword_t evt_frags[MAX_EVENT_FRAGS];
+       int evt_frag_idx;
+       int evt_code;
+       struct efx_buffer start;
+       struct pps_event_time host_time_pps;
+       unsigned last_sync_ns;
+       unsigned base_sync_ns;
+       bool base_sync_valid;
+       s64 current_adjfreq;
+       struct ptp_clock *phc_clock;
+       struct ptp_clock_info phc_clock_info;
+       struct work_struct pps_work;
+       struct workqueue_struct *pps_workwq;
+       bool nic_ts_enabled;
+       u8 txbuf[ALIGN(MC_CMD_PTP_IN_TRANSMIT_LEN(
+                              MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM), 4)];
+       struct efx_ptp_timeset
+       timeset[MC_CMD_PTP_OUT_SYNCHRONIZE_TIMESET_MAXNUM];
+};
+
+static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta);
+static int efx_phc_adjtime(struct ptp_clock_info *ptp, s64 delta);
+static int efx_phc_gettime(struct ptp_clock_info *ptp, struct timespec *ts);
+static int efx_phc_settime(struct ptp_clock_info *ptp,
+                          const struct timespec *e_ts);
+static int efx_phc_enable(struct ptp_clock_info *ptp,
+                         struct ptp_clock_request *request, int on);
+
+/* Enable MCDI PTP support. */
+static int efx_ptp_enable(struct efx_nic *efx)
+{
+       u8 inbuf[MC_CMD_PTP_IN_ENABLE_LEN];
+
+       MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_ENABLE);
+       MCDI_SET_DWORD(inbuf, PTP_IN_ENABLE_QUEUE,
+                      efx->ptp_data->channel->channel);
+       MCDI_SET_DWORD(inbuf, PTP_IN_ENABLE_MODE, efx->ptp_data->mode);
+
+       return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
+                           NULL, 0, NULL);
+}
+
+/* Disable MCDI PTP support.
+ *
+ * Note that this function should never rely on the presence of ptp_data -
+ * may be called before that exists.
+ */
+static int efx_ptp_disable(struct efx_nic *efx)
+{
+       u8 inbuf[MC_CMD_PTP_IN_DISABLE_LEN];
+
+       MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_DISABLE);
+       return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
+                           NULL, 0, NULL);
+}
+
+static void efx_ptp_deliver_rx_queue(struct sk_buff_head *q)
+{
+       struct sk_buff *skb;
+
+       while ((skb = skb_dequeue(q))) {
+               local_bh_disable();
+               netif_receive_skb(skb);
+               local_bh_enable();
+       }
+}
+
+static void efx_ptp_handle_no_channel(struct efx_nic *efx)
+{
+       netif_err(efx, drv, efx->net_dev,
+                 "ERROR: PTP requires MSI-X and 1 additional interrupt"
+                 "vector. PTP disabled\n");
+}
+
+/* Repeatedly send the host time to the MC which will capture the hardware
+ * time.
+ */
+static void efx_ptp_send_times(struct efx_nic *efx,
+                              struct pps_event_time *last_time)
+{
+       struct pps_event_time now;
+       struct timespec limit;
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       struct timespec start;
+       int *mc_running = ptp->start.addr;
+
+       pps_get_ts(&now);
+       start = now.ts_real;
+       limit = now.ts_real;
+       timespec_add_ns(&limit, SYNCHRONISE_PERIOD_NS);
+
+       /* Write host time for specified period or until MC is done */
+       while ((timespec_compare(&now.ts_real, &limit) < 0) &&
+              ACCESS_ONCE(*mc_running)) {
+               struct timespec update_time;
+               unsigned int host_time;
+
+               /* Don't update continuously to avoid saturating the PCIe bus */
+               update_time = now.ts_real;
+               timespec_add_ns(&update_time, SYNCHRONISATION_GRANULARITY_NS);
+               do {
+                       pps_get_ts(&now);
+               } while ((timespec_compare(&now.ts_real, &update_time) < 0) &&
+                        ACCESS_ONCE(*mc_running));
+
+               /* Synchronise NIC with single word of time only */
+               host_time = (now.ts_real.tv_sec << MC_NANOSECOND_BITS |
+                            now.ts_real.tv_nsec);
+               /* Update host time in NIC memory */
+               _efx_writed(efx, cpu_to_le32(host_time),
+                           FR_CZ_MC_TREG_SMEM + MC_SMEM_P0_PTP_TIME_OFST);
+       }
+       *last_time = now;
+}
+
+/* Read a timeset from the MC's results and partial process. */
+static void efx_ptp_read_timeset(u8 *data, struct efx_ptp_timeset *timeset)
+{
+       unsigned start_ns, end_ns;
+
+       timeset->host_start = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_HOSTSTART);
+       timeset->seconds = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_SECONDS);
+       timeset->nanoseconds = MCDI_DWORD(data,
+                                        PTP_OUT_SYNCHRONIZE_NANOSECONDS);
+       timeset->host_end = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_HOSTEND),
+       timeset->waitns = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_WAITNS);
+
+       /* Ignore seconds */
+       start_ns = timeset->host_start & MC_NANOSECOND_MASK;
+       end_ns = timeset->host_end & MC_NANOSECOND_MASK;
+       /* Allow for rollover */
+       if (end_ns < start_ns)
+               end_ns += NSEC_PER_SEC;
+       /* Determine duration of operation */
+       timeset->window = end_ns - start_ns;
+}
+
+/* Process times received from MC.
+ *
+ * Extract times from returned results, and establish the minimum value
+ * seen.  The minimum value represents the "best" possible time and events
+ * too much greater than this are rejected - the machine is, perhaps, too
+ * busy. A number of readings are taken so that, hopefully, at least one good
+ * synchronisation will be seen in the results.
+ */
+static int efx_ptp_process_times(struct efx_nic *efx, u8 *synch_buf,
+                                size_t response_length,
+                                const struct pps_event_time *last_time)
+{
+       unsigned number_readings = (response_length /
+                              MC_CMD_PTP_OUT_SYNCHRONIZE_TIMESET_LEN);
+       unsigned i;
+       unsigned min;
+       unsigned min_set = 0;
+       unsigned total;
+       unsigned ngood = 0;
+       unsigned last_good = 0;
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       bool min_valid = false;
+       u32 last_sec;
+       u32 start_sec;
+       struct timespec delta;
+
+       if (number_readings == 0)
+               return -EAGAIN;
+
+       /* Find minimum value in this set of results, discarding clearly
+        * erroneous results.
+        */
+       for (i = 0; i < number_readings; i++) {
+               efx_ptp_read_timeset(synch_buf, &ptp->timeset[i]);
+               synch_buf += MC_CMD_PTP_OUT_SYNCHRONIZE_TIMESET_LEN;
+               if (ptp->timeset[i].window > SYNCHRONISATION_GRANULARITY_NS) {
+                       if (min_valid) {
+                               if (ptp->timeset[i].window < min_set)
+                                       min_set = ptp->timeset[i].window;
+                       } else {
+                               min_valid = true;
+                               min_set = ptp->timeset[i].window;
+                       }
+               }
+       }
+
+       if (min_valid) {
+               if (ptp->base_sync_valid && (min_set > ptp->base_sync_ns))
+                       min = ptp->base_sync_ns;
+               else
+                       min = min_set;
+       } else {
+               min = SYNCHRONISATION_GRANULARITY_NS;
+       }
+
+       /* Discard excessively long synchronise durations.  The MC times
+        * when it finishes reading the host time so the corrected window
+        * time should be fairly constant for a given platform.
+        */
+       total = 0;
+       for (i = 0; i < number_readings; i++)
+               if (ptp->timeset[i].window > ptp->timeset[i].waitns) {
+                       unsigned win;
+
+                       win = ptp->timeset[i].window - ptp->timeset[i].waitns;
+                       if (win >= MIN_SYNCHRONISATION_NS &&
+                           win < MAX_SYNCHRONISATION_NS) {
+                               total += ptp->timeset[i].window;
+                               ngood++;
+                               last_good = i;
+                       }
+               }
+
+       if (ngood == 0) {
+               netif_warn(efx, drv, efx->net_dev,
+                          "PTP no suitable synchronisations %dns %dns\n",
+                          ptp->base_sync_ns, min_set);
+               return -EAGAIN;
+       }
+
+       /* Average minimum this synchronisation */
+       ptp->last_sync_ns = DIV_ROUND_UP(total, ngood);
+       if (!ptp->base_sync_valid || (ptp->last_sync_ns < ptp->base_sync_ns)) {
+               ptp->base_sync_valid = true;
+               ptp->base_sync_ns = ptp->last_sync_ns;
+       }
+
+       /* Calculate delay from actual PPS to last_time */
+       delta.tv_nsec =
+               ptp->timeset[last_good].nanoseconds +
+               last_time->ts_real.tv_nsec -
+               (ptp->timeset[last_good].host_start & MC_NANOSECOND_MASK);
+
+       /* It is possible that the seconds rolled over between taking
+        * the start reading and the last value written by the host.  The
+        * timescales are such that a gap of more than one second is never
+        * expected.
+        */
+       start_sec = ptp->timeset[last_good].host_start >> MC_NANOSECOND_BITS;
+       last_sec = last_time->ts_real.tv_sec & MC_SECOND_MASK;
+       if (start_sec != last_sec) {
+               if (((start_sec + 1) & MC_SECOND_MASK) != last_sec) {
+                       netif_warn(efx, hw, efx->net_dev,
+                                  "PTP bad synchronisation seconds\n");
+                       return -EAGAIN;
+               } else {
+                       delta.tv_sec = 1;
+               }
+       } else {
+               delta.tv_sec = 0;
+       }
+
+       ptp->host_time_pps = *last_time;
+       pps_sub_ts(&ptp->host_time_pps, delta);
+
+       return 0;
+}
+
+/* Synchronize times between the host and the MC */
+static int efx_ptp_synchronize(struct efx_nic *efx, unsigned int num_readings)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       u8 synch_buf[MC_CMD_PTP_OUT_SYNCHRONIZE_LENMAX];
+       size_t response_length;
+       int rc;
+       unsigned long timeout;
+       struct pps_event_time last_time = {};
+       unsigned int loops = 0;
+       int *start = ptp->start.addr;
+
+       MCDI_SET_DWORD(synch_buf, PTP_IN_OP, MC_CMD_PTP_OP_SYNCHRONIZE);
+       MCDI_SET_DWORD(synch_buf, PTP_IN_SYNCHRONIZE_NUMTIMESETS,
+                      num_readings);
+       MCDI_SET_DWORD(synch_buf, PTP_IN_SYNCHRONIZE_START_ADDR_LO,
+                      (u32)ptp->start.dma_addr);
+       MCDI_SET_DWORD(synch_buf, PTP_IN_SYNCHRONIZE_START_ADDR_HI,
+                      (u32)((u64)ptp->start.dma_addr >> 32));
+
+       /* Clear flag that signals MC ready */
+       ACCESS_ONCE(*start) = 0;
+       efx_mcdi_rpc_start(efx, MC_CMD_PTP, synch_buf,
+                          MC_CMD_PTP_IN_SYNCHRONIZE_LEN);
+
+       /* Wait for start from MCDI (or timeout) */
+       timeout = jiffies + msecs_to_jiffies(MAX_SYNCHRONISE_WAIT_MS);
+       while (!ACCESS_ONCE(*start) && (time_before(jiffies, timeout))) {
+               udelay(20);     /* Usually start MCDI execution quickly */
+               loops++;
+       }
+
+       if (ACCESS_ONCE(*start))
+               efx_ptp_send_times(efx, &last_time);
+
+       /* Collect results */
+       rc = efx_mcdi_rpc_finish(efx, MC_CMD_PTP,
+                                MC_CMD_PTP_IN_SYNCHRONIZE_LEN,
+                                synch_buf, sizeof(synch_buf),
+                                &response_length);
+       if (rc == 0)
+               rc = efx_ptp_process_times(efx, synch_buf, response_length,
+                                          &last_time);
+
+       return rc;
+}
+
+/* Transmit a PTP packet, via the MCDI interface, to the wire. */
+static int efx_ptp_xmit_skb(struct efx_nic *efx, struct sk_buff *skb)
+{
+       u8 *txbuf = efx->ptp_data->txbuf;
+       struct skb_shared_hwtstamps timestamps;
+       int rc = -EIO;
+       /* MCDI driver requires word aligned lengths */
+       size_t len = ALIGN(MC_CMD_PTP_IN_TRANSMIT_LEN(skb->len), 4);
+       u8 txtime[MC_CMD_PTP_OUT_TRANSMIT_LEN];
+
+       MCDI_SET_DWORD(txbuf, PTP_IN_OP, MC_CMD_PTP_OP_TRANSMIT);
+       MCDI_SET_DWORD(txbuf, PTP_IN_TRANSMIT_LENGTH, skb->len);
+       if (skb_shinfo(skb)->nr_frags != 0) {
+               rc = skb_linearize(skb);
+               if (rc != 0)
+                       goto fail;
+       }
+
+       if (skb->ip_summed == CHECKSUM_PARTIAL) {
+               rc = skb_checksum_help(skb);
+               if (rc != 0)
+                       goto fail;
+       }
+       skb_copy_from_linear_data(skb,
+                                 &txbuf[MC_CMD_PTP_IN_TRANSMIT_PACKET_OFST],
+                                 len);
+       rc = efx_mcdi_rpc(efx, MC_CMD_PTP, txbuf, len, txtime,
+                         sizeof(txtime), &len);
+       if (rc != 0)
+               goto fail;
+
+       memset(&timestamps, 0, sizeof(timestamps));
+       timestamps.hwtstamp = ktime_set(
+               MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_SECONDS),
+               MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_NANOSECONDS));
+
+       skb_tstamp_tx(skb, &timestamps);
+
+       rc = 0;
+
+fail:
+       dev_kfree_skb(skb);
+
+       return rc;
+}
+
+static void efx_ptp_drop_time_expired_events(struct efx_nic *efx)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       struct list_head *cursor;
+       struct list_head *next;
+
+       /* Drop time-expired events */
+       spin_lock_bh(&ptp->evt_lock);
+       if (!list_empty(&ptp->evt_list)) {
+               list_for_each_safe(cursor, next, &ptp->evt_list) {
+                       struct efx_ptp_event_rx *evt;
+
+                       evt = list_entry(cursor, struct efx_ptp_event_rx,
+                                        link);
+                       if (time_after(jiffies, evt->expiry)) {
+                               list_del(&evt->link);
+                               list_add(&evt->link, &ptp->evt_free_list);
+                               netif_warn(efx, hw, efx->net_dev,
+                                          "PTP rx event dropped\n");
+                       }
+               }
+       }
+       spin_unlock_bh(&ptp->evt_lock);
+}
+
+static enum ptp_packet_state efx_ptp_match_rx(struct efx_nic *efx,
+                                             struct sk_buff *skb)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       bool evts_waiting;
+       struct list_head *cursor;
+       struct list_head *next;
+       struct efx_ptp_match *match;
+       enum ptp_packet_state rc = PTP_PACKET_STATE_UNMATCHED;
+
+       spin_lock_bh(&ptp->evt_lock);
+       evts_waiting = !list_empty(&ptp->evt_list);
+       spin_unlock_bh(&ptp->evt_lock);
+
+       if (!evts_waiting)
+               return PTP_PACKET_STATE_UNMATCHED;
+
+       match = (struct efx_ptp_match *)skb->cb;
+       /* Look for a matching timestamp in the event queue */
+       spin_lock_bh(&ptp->evt_lock);
+       list_for_each_safe(cursor, next, &ptp->evt_list) {
+               struct efx_ptp_event_rx *evt;
+
+               evt = list_entry(cursor, struct efx_ptp_event_rx, link);
+               if ((evt->seq0 == match->words[0]) &&
+                   (evt->seq1 == match->words[1])) {
+                       struct skb_shared_hwtstamps *timestamps;
+
+                       /* Match - add in hardware timestamp */
+                       timestamps = skb_hwtstamps(skb);
+                       timestamps->hwtstamp = evt->hwtimestamp;
+
+                       match->state = PTP_PACKET_STATE_MATCHED;
+                       rc = PTP_PACKET_STATE_MATCHED;
+                       list_del(&evt->link);
+                       list_add(&evt->link, &ptp->evt_free_list);
+                       break;
+               }
+       }
+       spin_unlock_bh(&ptp->evt_lock);
+
+       return rc;
+}
+
+/* Process any queued receive events and corresponding packets
+ *
+ * q is returned with all the packets that are ready for delivery.
+ * true is returned if at least one of those packets requires
+ * synchronisation.
+ */
+static bool efx_ptp_process_events(struct efx_nic *efx, struct sk_buff_head *q)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       bool rc = false;
+       struct sk_buff *skb;
+
+       while ((skb = skb_dequeue(&ptp->rxq))) {
+               struct efx_ptp_match *match;
+
+               match = (struct efx_ptp_match *)skb->cb;
+               if (match->state == PTP_PACKET_STATE_MATCH_UNWANTED) {
+                       __skb_queue_tail(q, skb);
+               } else if (efx_ptp_match_rx(efx, skb) ==
+                          PTP_PACKET_STATE_MATCHED) {
+                       rc = true;
+                       __skb_queue_tail(q, skb);
+               } else if (time_after(jiffies, match->expiry)) {
+                       match->state = PTP_PACKET_STATE_TIMED_OUT;
+                       netif_warn(efx, rx_err, efx->net_dev,
+                                  "PTP packet - no timestamp seen\n");
+                       __skb_queue_tail(q, skb);
+               } else {
+                       /* Replace unprocessed entry and stop */
+                       skb_queue_head(&ptp->rxq, skb);
+                       break;
+               }
+       }
+
+       return rc;
+}
+
+/* Complete processing of a received packet */
+static inline void efx_ptp_process_rx(struct efx_nic *efx, struct sk_buff *skb)
+{
+       local_bh_disable();
+       netif_receive_skb(skb);
+       local_bh_enable();
+}
+
+static int efx_ptp_start(struct efx_nic *efx)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       struct efx_filter_spec rxfilter;
+       int rc;
+
+       ptp->reset_required = false;
+
+       /* Must filter on both event and general ports to ensure
+        * that there is no packet re-ordering.
+        */
+       efx_filter_init_rx(&rxfilter, EFX_FILTER_PRI_REQUIRED, 0,
+                          efx_rx_queue_index(
+                                  efx_channel_get_rx_queue(ptp->channel)));
+       rc = efx_filter_set_ipv4_local(&rxfilter, IPPROTO_UDP,
+                                      htonl(PTP_ADDRESS),
+                                      htons(PTP_EVENT_PORT));
+       if (rc != 0)
+               return rc;
+
+       rc = efx_filter_insert_filter(efx, &rxfilter, true);
+       if (rc < 0)
+               return rc;
+       ptp->rxfilter_event = rc;
+
+       efx_filter_init_rx(&rxfilter, EFX_FILTER_PRI_REQUIRED, 0,
+                          efx_rx_queue_index(
+                                  efx_channel_get_rx_queue(ptp->channel)));
+       rc = efx_filter_set_ipv4_local(&rxfilter, IPPROTO_UDP,
+                                      htonl(PTP_ADDRESS),
+                                      htons(PTP_GENERAL_PORT));
+       if (rc != 0)
+               goto fail;
+
+       rc = efx_filter_insert_filter(efx, &rxfilter, true);
+       if (rc < 0)
+               goto fail;
+       ptp->rxfilter_general = rc;
+
+       rc = efx_ptp_enable(efx);
+       if (rc != 0)
+               goto fail2;
+
+       ptp->evt_frag_idx = 0;
+       ptp->current_adjfreq = 0;
+       ptp->rxfilter_installed = true;
+
+       return 0;
+
+fail2:
+       efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
+                                 ptp->rxfilter_general);
+fail:
+       efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
+                                 ptp->rxfilter_event);
+
+       return rc;
+}
+
+static int efx_ptp_stop(struct efx_nic *efx)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       int rc = efx_ptp_disable(efx);
+       struct list_head *cursor;
+       struct list_head *next;
+
+       if (ptp->rxfilter_installed) {
+               efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
+                                         ptp->rxfilter_general);
+               efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
+                                         ptp->rxfilter_event);
+               ptp->rxfilter_installed = false;
+       }
+
+       /* Make sure RX packets are really delivered */
+       efx_ptp_deliver_rx_queue(&efx->ptp_data->rxq);
+       skb_queue_purge(&efx->ptp_data->txq);
+
+       /* Drop any pending receive events */
+       spin_lock_bh(&efx->ptp_data->evt_lock);
+       list_for_each_safe(cursor, next, &efx->ptp_data->evt_list) {
+               list_del(cursor);
+               list_add(cursor, &efx->ptp_data->evt_free_list);
+       }
+       spin_unlock_bh(&efx->ptp_data->evt_lock);
+
+       return rc;
+}
+
+static void efx_ptp_pps_worker(struct work_struct *work)
+{
+       struct efx_ptp_data *ptp =
+               container_of(work, struct efx_ptp_data, pps_work);
+       struct efx_nic *efx = ptp->channel->efx;
+       struct ptp_clock_event ptp_evt;
+
+       if (efx_ptp_synchronize(efx, PTP_SYNC_ATTEMPTS))
+               return;
+
+       ptp_evt.type = PTP_CLOCK_PPSUSR;
+       ptp_evt.pps_times = ptp->host_time_pps;
+       ptp_clock_event(ptp->phc_clock, &ptp_evt);
+}
+
+/* Process any pending transmissions and timestamp any received packets.
+ */
+static void efx_ptp_worker(struct work_struct *work)
+{
+       struct efx_ptp_data *ptp_data =
+               container_of(work, struct efx_ptp_data, work);
+       struct efx_nic *efx = ptp_data->channel->efx;
+       struct sk_buff *skb;
+       struct sk_buff_head tempq;
+
+       if (ptp_data->reset_required) {
+               efx_ptp_stop(efx);
+               efx_ptp_start(efx);
+               return;
+       }
+
+       efx_ptp_drop_time_expired_events(efx);
+
+       __skb_queue_head_init(&tempq);
+       if (efx_ptp_process_events(efx, &tempq) ||
+           !skb_queue_empty(&ptp_data->txq)) {
+
+               while ((skb = skb_dequeue(&ptp_data->txq)))
+                       efx_ptp_xmit_skb(efx, skb);
+       }
+
+       while ((skb = __skb_dequeue(&tempq)))
+               efx_ptp_process_rx(efx, skb);
+}
+
+/* Initialise PTP channel and state.
+ *
+ * Setting core_index to zero causes the queue to be initialised and doesn't
+ * overlap with 'rxq0' because ptp.c doesn't use skb_record_rx_queue.
+ */
+static int efx_ptp_probe_channel(struct efx_channel *channel)
+{
+       struct efx_nic *efx = channel->efx;
+       struct efx_ptp_data *ptp;
+       int rc = 0;
+       unsigned int pos;
+
+       channel->irq_moderation = 0;
+       channel->rx_queue.core_index = 0;
+
+       ptp = kzalloc(sizeof(struct efx_ptp_data), GFP_KERNEL);
+       efx->ptp_data = ptp;
+       if (!efx->ptp_data)
+               return -ENOMEM;
+
+       rc = efx_nic_alloc_buffer(efx, &ptp->start, sizeof(int));
+       if (rc != 0)
+               goto fail1;
+
+       ptp->channel = channel;
+       skb_queue_head_init(&ptp->rxq);
+       skb_queue_head_init(&ptp->txq);
+       ptp->workwq = create_singlethread_workqueue("sfc_ptp");
+       if (!ptp->workwq) {
+               rc = -ENOMEM;
+               goto fail2;
+       }
+
+       INIT_WORK(&ptp->work, efx_ptp_worker);
+       ptp->config.flags = 0;
+       ptp->config.tx_type = HWTSTAMP_TX_OFF;
+       ptp->config.rx_filter = HWTSTAMP_FILTER_NONE;
+       INIT_LIST_HEAD(&ptp->evt_list);
+       INIT_LIST_HEAD(&ptp->evt_free_list);
+       spin_lock_init(&ptp->evt_lock);
+       for (pos = 0; pos < MAX_RECEIVE_EVENTS; pos++)
+               list_add(&ptp->rx_evts[pos].link, &ptp->evt_free_list);
+
+       ptp->phc_clock_info.owner = THIS_MODULE;
+       snprintf(ptp->phc_clock_info.name,
+                sizeof(ptp->phc_clock_info.name),
+                "%pm", efx->net_dev->perm_addr);
+       ptp->phc_clock_info.max_adj = MAX_PPB;
+       ptp->phc_clock_info.n_alarm = 0;
+       ptp->phc_clock_info.n_ext_ts = 0;
+       ptp->phc_clock_info.n_per_out = 0;
+       ptp->phc_clock_info.pps = 1;
+       ptp->phc_clock_info.adjfreq = efx_phc_adjfreq;
+       ptp->phc_clock_info.adjtime = efx_phc_adjtime;
+       ptp->phc_clock_info.gettime = efx_phc_gettime;
+       ptp->phc_clock_info.settime = efx_phc_settime;
+       ptp->phc_clock_info.enable = efx_phc_enable;
+
+       ptp->phc_clock = ptp_clock_register(&ptp->phc_clock_info);
+       if (!ptp->phc_clock)
+               goto fail3;
+
+       INIT_WORK(&ptp->pps_work, efx_ptp_pps_worker);
+       ptp->pps_workwq = create_singlethread_workqueue("sfc_pps");
+       if (!ptp->pps_workwq) {
+               rc = -ENOMEM;
+               goto fail4;
+       }
+       ptp->nic_ts_enabled = false;
+
+       return 0;
+fail4:
+       ptp_clock_unregister(efx->ptp_data->phc_clock);
+
+fail3:
+       destroy_workqueue(efx->ptp_data->workwq);
+
+fail2:
+       efx_nic_free_buffer(efx, &ptp->start);
+
+fail1:
+       kfree(efx->ptp_data);
+       efx->ptp_data = NULL;
+
+       return rc;
+}
+
+static void efx_ptp_remove_channel(struct efx_channel *channel)
+{
+       struct efx_nic *efx = channel->efx;
+
+       if (!efx->ptp_data)
+               return;
+
+       (void)efx_ptp_disable(channel->efx);
+
+       cancel_work_sync(&efx->ptp_data->work);
+       cancel_work_sync(&efx->ptp_data->pps_work);
+
+       skb_queue_purge(&efx->ptp_data->rxq);
+       skb_queue_purge(&efx->ptp_data->txq);
+
+       ptp_clock_unregister(efx->ptp_data->phc_clock);
+
+       destroy_workqueue(efx->ptp_data->workwq);
+       destroy_workqueue(efx->ptp_data->pps_workwq);
+
+       efx_nic_free_buffer(efx, &efx->ptp_data->start);
+       kfree(efx->ptp_data);
+}
+
+static void efx_ptp_get_channel_name(struct efx_channel *channel,
+                                    char *buf, size_t len)
+{
+       snprintf(buf, len, "%s-ptp", channel->efx->name);
+}
+
+/* Determine whether this packet should be processed by the PTP module
+ * or transmitted conventionally.
+ */
+bool efx_ptp_is_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
+{
+       return efx->ptp_data &&
+               efx->ptp_data->enabled &&
+               skb->len >= PTP_MIN_LENGTH &&
+               skb->len <= MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM &&
+               likely(skb->protocol == htons(ETH_P_IP)) &&
+               ip_hdr(skb)->protocol == IPPROTO_UDP &&
+               udp_hdr(skb)->dest == htons(PTP_EVENT_PORT);
+}
+
+/* Receive a PTP packet.  Packets are queued until the arrival of
+ * the receive timestamp from the MC - this will probably occur after the
+ * packet arrival because of the processing in the MC.
+ */
+static void efx_ptp_rx(struct efx_channel *channel, struct sk_buff *skb)
+{
+       struct efx_nic *efx = channel->efx;
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       struct efx_ptp_match *match = (struct efx_ptp_match *)skb->cb;
+       u8 *data;
+       unsigned int version;
+
+       match->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
+
+       /* Correct version? */
+       if (ptp->mode == MC_CMD_PTP_MODE_V1) {
+               if (skb->len < PTP_V1_MIN_LENGTH) {
+                       netif_receive_skb(skb);
+                       return;
+               }
+               version = ntohs(*(__be16 *)&skb->data[PTP_V1_VERSION_OFFSET]);
+               if (version != PTP_VERSION_V1) {
+                       netif_receive_skb(skb);
+                       return;
+               }
+       } else {
+               if (skb->len < PTP_V2_MIN_LENGTH) {
+                       netif_receive_skb(skb);
+                       return;
+               }
+               version = skb->data[PTP_V2_VERSION_OFFSET];
+
+               BUG_ON(ptp->mode != MC_CMD_PTP_MODE_V2);
+               BUILD_BUG_ON(PTP_V1_UUID_OFFSET != PTP_V2_MC_UUID_OFFSET);
+               BUILD_BUG_ON(PTP_V1_UUID_LENGTH != PTP_V2_MC_UUID_LENGTH);
+               BUILD_BUG_ON(PTP_V1_SEQUENCE_OFFSET != PTP_V2_SEQUENCE_OFFSET);
+               BUILD_BUG_ON(PTP_V1_SEQUENCE_LENGTH != PTP_V2_SEQUENCE_LENGTH);
+
+               if ((version & PTP_VERSION_V2_MASK) != PTP_VERSION_V2) {
+                       netif_receive_skb(skb);
+                       return;
+               }
+       }
+
+       /* Does this packet require timestamping? */
+       if (ntohs(*(__be16 *)&skb->data[PTP_DPORT_OFFSET]) == PTP_EVENT_PORT) {
+               struct skb_shared_hwtstamps *timestamps;
+
+               match->state = PTP_PACKET_STATE_UNMATCHED;
+
+               /* Clear all timestamps held: filled in later */
+               timestamps = skb_hwtstamps(skb);
+               memset(timestamps, 0, sizeof(*timestamps));
+
+               /* Extract UUID/Sequence information */
+               data = skb->data + PTP_V1_UUID_OFFSET;
+               match->words[0] = (data[0]         |
+                                  (data[1] << 8)  |
+                                  (data[2] << 16) |
+                                  (data[3] << 24));
+               match->words[1] = (data[4]         |
+                                  (data[5] << 8)  |
+                                  (skb->data[PTP_V1_SEQUENCE_OFFSET +
+                                             PTP_V1_SEQUENCE_LENGTH - 1] <<
+                                   16));
+       } else {
+               match->state = PTP_PACKET_STATE_MATCH_UNWANTED;
+       }
+
+       skb_queue_tail(&ptp->rxq, skb);
+       queue_work(ptp->workwq, &ptp->work);
+}
+
+/* Transmit a PTP packet.  This has to be transmitted by the MC
+ * itself, through an MCDI call.  MCDI calls aren't permitted
+ * in the transmit path so defer the actual transmission to a suitable worker.
+ */
+int efx_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+
+       skb_queue_tail(&ptp->txq, skb);
+
+       if ((udp_hdr(skb)->dest == htons(PTP_EVENT_PORT)) &&
+           (skb->len <= MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM))
+               efx_xmit_hwtstamp_pending(skb);
+       queue_work(ptp->workwq, &ptp->work);
+
+       return NETDEV_TX_OK;
+}
+
+static int efx_ptp_change_mode(struct efx_nic *efx, bool enable_wanted,
+                              unsigned int new_mode)
+{
+       if ((enable_wanted != efx->ptp_data->enabled) ||
+           (enable_wanted && (efx->ptp_data->mode != new_mode))) {
+               int rc;
+
+               if (enable_wanted) {
+                       /* Change of mode requires disable */
+                       if (efx->ptp_data->enabled &&
+                           (efx->ptp_data->mode != new_mode)) {
+                               efx->ptp_data->enabled = false;
+                               rc = efx_ptp_stop(efx);
+                               if (rc != 0)
+                                       return rc;
+                       }
+
+                       /* Set new operating mode and establish
+                        * baseline synchronisation, which must
+                        * succeed.
+                        */
+                       efx->ptp_data->mode = new_mode;
+                       rc = efx_ptp_start(efx);
+                       if (rc == 0) {
+                               rc = efx_ptp_synchronize(efx,
+                                                        PTP_SYNC_ATTEMPTS * 2);
+                               if (rc != 0)
+                                       efx_ptp_stop(efx);
+                       }
+               } else {
+                       rc = efx_ptp_stop(efx);
+               }
+
+               if (rc != 0)
+                       return rc;
+
+               efx->ptp_data->enabled = enable_wanted;
+       }
+
+       return 0;
+}
+
+static int efx_ptp_ts_init(struct efx_nic *efx, struct hwtstamp_config *init)
+{
+       bool enable_wanted = false;
+       unsigned int new_mode;
+       int rc;
+
+       if (init->flags)
+               return -EINVAL;
+
+       if ((init->tx_type != HWTSTAMP_TX_OFF) &&
+           (init->tx_type != HWTSTAMP_TX_ON))
+               return -ERANGE;
+
+       new_mode = efx->ptp_data->mode;
+       /* Determine whether any PTP HW operations are required */
+       switch (init->rx_filter) {
+       case HWTSTAMP_FILTER_NONE:
+               break;
+       case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
+       case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
+       case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
+               init->rx_filter = HWTSTAMP_FILTER_PTP_V1_L4_EVENT;
+               new_mode = MC_CMD_PTP_MODE_V1;
+               enable_wanted = true;
+               break;
+       case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
+       case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
+       case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
+       /* Although these three are accepted only IPV4 packets will be
+        * timestamped
+        */
+               init->rx_filter = HWTSTAMP_FILTER_PTP_V2_L4_EVENT;
+               new_mode = MC_CMD_PTP_MODE_V2;
+               enable_wanted = true;
+               break;
+       case HWTSTAMP_FILTER_PTP_V2_EVENT:
+       case HWTSTAMP_FILTER_PTP_V2_SYNC:
+       case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
+       case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
+       case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
+       case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
+               /* Non-IP + IPv6 timestamping not supported */
+               return -ERANGE;
+               break;
+       default:
+               return -ERANGE;
+       }
+
+       if (init->tx_type != HWTSTAMP_TX_OFF)
+               enable_wanted = true;
+
+       rc = efx_ptp_change_mode(efx, enable_wanted, new_mode);
+       if (rc != 0)
+               return rc;
+
+       efx->ptp_data->config = *init;
+
+       return 0;
+}
+
+int
+efx_ptp_get_ts_info(struct net_device *net_dev, struct ethtool_ts_info *ts_info)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+       struct efx_ptp_data *ptp = efx->ptp_data;
+
+       if (!ptp)
+               return -EOPNOTSUPP;
+
+       ts_info->so_timestamping = (SOF_TIMESTAMPING_TX_HARDWARE |
+                                   SOF_TIMESTAMPING_RX_HARDWARE |
+                                   SOF_TIMESTAMPING_RAW_HARDWARE);
+       ts_info->phc_index = ptp_clock_index(ptp->phc_clock);
+       ts_info->tx_types = 1 << HWTSTAMP_TX_OFF | 1 << HWTSTAMP_TX_ON;
+       ts_info->rx_filters = (1 << HWTSTAMP_FILTER_NONE |
+                              1 << HWTSTAMP_FILTER_PTP_V1_L4_EVENT |
+                              1 << HWTSTAMP_FILTER_PTP_V1_L4_SYNC |
+                              1 << HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ |
+                              1 << HWTSTAMP_FILTER_PTP_V2_L4_EVENT |
+                              1 << HWTSTAMP_FILTER_PTP_V2_L4_SYNC |
+                              1 << HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ);
+       return 0;
+}
+
+int efx_ptp_ioctl(struct efx_nic *efx, struct ifreq *ifr, int cmd)
+{
+       struct hwtstamp_config config;
+       int rc;
+
+       /* Not a PTP enabled port */
+       if (!efx->ptp_data)
+               return -EOPNOTSUPP;
+
+       if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
+               return -EFAULT;
+
+       rc = efx_ptp_ts_init(efx, &config);
+       if (rc != 0)
+               return rc;
+
+       return copy_to_user(ifr->ifr_data, &config, sizeof(config))
+               ? -EFAULT : 0;
+}
+
+static void ptp_event_failure(struct efx_nic *efx, int expected_frag_len)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+
+       netif_err(efx, hw, efx->net_dev,
+               "PTP unexpected event length: got %d expected %d\n",
+               ptp->evt_frag_idx, expected_frag_len);
+       ptp->reset_required = true;
+       queue_work(ptp->workwq, &ptp->work);
+}
+
+/* Process a completed receive event.  Put it on the event queue and
+ * start worker thread.  This is required because event and their
+ * correspoding packets may come in either order.
+ */
+static void ptp_event_rx(struct efx_nic *efx, struct efx_ptp_data *ptp)
+{
+       struct efx_ptp_event_rx *evt = NULL;
+
+       if (ptp->evt_frag_idx != 3) {
+               ptp_event_failure(efx, 3);
+               return;
+       }
+
+       spin_lock_bh(&ptp->evt_lock);
+       if (!list_empty(&ptp->evt_free_list)) {
+               evt = list_first_entry(&ptp->evt_free_list,
+                                      struct efx_ptp_event_rx, link);
+               list_del(&evt->link);
+
+               evt->seq0 = EFX_QWORD_FIELD(ptp->evt_frags[2], MCDI_EVENT_DATA);
+               evt->seq1 = (EFX_QWORD_FIELD(ptp->evt_frags[2],
+                                            MCDI_EVENT_SRC)        |
+                            (EFX_QWORD_FIELD(ptp->evt_frags[1],
+                                             MCDI_EVENT_SRC) << 8) |
+                            (EFX_QWORD_FIELD(ptp->evt_frags[0],
+                                             MCDI_EVENT_SRC) << 16));
+               evt->hwtimestamp = ktime_set(
+                       EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA),
+                       EFX_QWORD_FIELD(ptp->evt_frags[1], MCDI_EVENT_DATA));
+               evt->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
+               list_add_tail(&evt->link, &ptp->evt_list);
+
+               queue_work(ptp->workwq, &ptp->work);
+       } else {
+               netif_err(efx, rx_err, efx->net_dev, "No free PTP event");
+       }
+       spin_unlock_bh(&ptp->evt_lock);
+}
+
+static void ptp_event_fault(struct efx_nic *efx, struct efx_ptp_data *ptp)
+{
+       int code = EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA);
+       if (ptp->evt_frag_idx != 1) {
+               ptp_event_failure(efx, 1);
+               return;
+       }
+
+       netif_err(efx, hw, efx->net_dev, "PTP error %d\n", code);
+}
+
+static void ptp_event_pps(struct efx_nic *efx, struct efx_ptp_data *ptp)
+{
+       if (ptp->nic_ts_enabled)
+               queue_work(ptp->pps_workwq, &ptp->pps_work);
+}
+
+void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev)
+{
+       struct efx_ptp_data *ptp = efx->ptp_data;
+       int code = EFX_QWORD_FIELD(*ev, MCDI_EVENT_CODE);
+
+       if (!ptp->enabled)
+               return;
+
+       if (ptp->evt_frag_idx == 0) {
+               ptp->evt_code = code;
+       } else if (ptp->evt_code != code) {
+               netif_err(efx, hw, efx->net_dev,
+                         "PTP out of sequence event %d\n", code);
+               ptp->evt_frag_idx = 0;
+       }
+
+       ptp->evt_frags[ptp->evt_frag_idx++] = *ev;
+       if (!MCDI_EVENT_FIELD(*ev, CONT)) {
+               /* Process resulting event */
+               switch (code) {
+               case MCDI_EVENT_CODE_PTP_RX:
+                       ptp_event_rx(efx, ptp);
+                       break;
+               case MCDI_EVENT_CODE_PTP_FAULT:
+                       ptp_event_fault(efx, ptp);
+                       break;
+               case MCDI_EVENT_CODE_PTP_PPS:
+                       ptp_event_pps(efx, ptp);
+                       break;
+               default:
+                       netif_err(efx, hw, efx->net_dev,
+                                 "PTP unknown event %d\n", code);
+                       break;
+               }
+               ptp->evt_frag_idx = 0;
+       } else if (MAX_EVENT_FRAGS == ptp->evt_frag_idx) {
+               netif_err(efx, hw, efx->net_dev,
+                         "PTP too many event fragments\n");
+               ptp->evt_frag_idx = 0;
+       }
+}
+
+static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
+{
+       struct efx_ptp_data *ptp_data = container_of(ptp,
+                                                    struct efx_ptp_data,
+                                                    phc_clock_info);
+       struct efx_nic *efx = ptp_data->channel->efx;
+       u8 inadj[MC_CMD_PTP_IN_ADJUST_LEN];
+       s64 adjustment_ns;
+       int rc;
+
+       if (delta > MAX_PPB)
+               delta = MAX_PPB;
+       else if (delta < -MAX_PPB)
+               delta = -MAX_PPB;
+
+       /* Convert ppb to fixed point ns. */
+       adjustment_ns = (((s64)delta * PPB_SCALE_WORD) >>
+                        (PPB_EXTRA_BITS + MAX_PPB_BITS));
+
+       MCDI_SET_DWORD(inadj, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
+       MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_FREQ_LO, (u32)adjustment_ns);
+       MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_FREQ_HI,
+                      (u32)(adjustment_ns >> 32));
+       MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_SECONDS, 0);
+       MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_NANOSECONDS, 0);
+       rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inadj, sizeof(inadj),
+                         NULL, 0, NULL);
+       if (rc != 0)
+               return rc;
+
+       ptp_data->current_adjfreq = delta;
+       return 0;
+}
+
+static int efx_phc_adjtime(struct ptp_clock_info *ptp, s64 delta)
+{
+       struct efx_ptp_data *ptp_data = container_of(ptp,
+                                                    struct efx_ptp_data,
+                                                    phc_clock_info);
+       struct efx_nic *efx = ptp_data->channel->efx;
+       struct timespec delta_ts = ns_to_timespec(delta);
+       u8 inbuf[MC_CMD_PTP_IN_ADJUST_LEN];
+
+       MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
+       MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_FREQ_LO, 0);
+       MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_FREQ_HI, 0);
+       MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_SECONDS, (u32)delta_ts.tv_sec);
+       MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_NANOSECONDS, (u32)delta_ts.tv_nsec);
+       return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
+                           NULL, 0, NULL);
+}
+
+static int efx_phc_gettime(struct ptp_clock_info *ptp, struct timespec *ts)
+{
+       struct efx_ptp_data *ptp_data = container_of(ptp,
+                                                    struct efx_ptp_data,
+                                                    phc_clock_info);
+       struct efx_nic *efx = ptp_data->channel->efx;
+       u8 inbuf[MC_CMD_PTP_IN_READ_NIC_TIME_LEN];
+       u8 outbuf[MC_CMD_PTP_OUT_READ_NIC_TIME_LEN];
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_READ_NIC_TIME);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
+                         outbuf, sizeof(outbuf), NULL);
+       if (rc != 0)
+               return rc;
+
+       ts->tv_sec = MCDI_DWORD(outbuf, PTP_OUT_READ_NIC_TIME_SECONDS);
+       ts->tv_nsec = MCDI_DWORD(outbuf, PTP_OUT_READ_NIC_TIME_NANOSECONDS);
+       return 0;
+}
+
+static int efx_phc_settime(struct ptp_clock_info *ptp,
+                          const struct timespec *e_ts)
+{
+       /* Get the current NIC time, efx_phc_gettime.
+        * Subtract from the desired time to get the offset
+        * call efx_phc_adjtime with the offset
+        */
+       int rc;
+       struct timespec time_now;
+       struct timespec delta;
+
+       rc = efx_phc_gettime(ptp, &time_now);
+       if (rc != 0)
+               return rc;
+
+       delta = timespec_sub(*e_ts, time_now);
+
+       efx_phc_adjtime(ptp, timespec_to_ns(&delta));
+       if (rc != 0)
+               return rc;
+
+       return 0;
+}
+
+static int efx_phc_enable(struct ptp_clock_info *ptp,
+                         struct ptp_clock_request *request,
+                         int enable)
+{
+       struct efx_ptp_data *ptp_data = container_of(ptp,
+                                                    struct efx_ptp_data,
+                                                    phc_clock_info);
+       if (request->type != PTP_CLK_REQ_PPS)
+               return -EOPNOTSUPP;
+
+       ptp_data->nic_ts_enabled = !!enable;
+       return 0;
+}
+
+static const struct efx_channel_type efx_ptp_channel_type = {
+       .handle_no_channel      = efx_ptp_handle_no_channel,
+       .pre_probe              = efx_ptp_probe_channel,
+       .post_remove            = efx_ptp_remove_channel,
+       .get_name               = efx_ptp_get_channel_name,
+       /* no copy operation; there is no need to reallocate this channel */
+       .receive_skb            = efx_ptp_rx,
+       .keep_eventq            = false,
+};
+
+void efx_ptp_probe(struct efx_nic *efx)
+{
+       /* Check whether PTP is implemented on this NIC.  The DISABLE
+        * operation will succeed if and only if it is implemented.
+        */
+       if (efx_ptp_disable(efx) == 0)
+               efx->extra_channel_type[EFX_EXTRA_CHANNEL_PTP] =
+                       &efx_ptp_channel_type;
+}
index 6bafd216e55e6eb668d12b5e636f14975ae7f110..84b41bf08a38a22ad8e9f0954ec32631628b4ad7 100644 (file)
@@ -335,6 +335,7 @@ static int siena_probe_nic(struct efx_nic *efx)
                goto fail5;
 
        efx_sriov_probe(efx);
+       efx_ptp_probe(efx);
 
        return 0;
 
index ebca75ed78dc2d9c4afc83360e17e044522eea4e..5e090e54298e667a732c53ece789bb03e40aef5e 100644 (file)
@@ -339,6 +339,12 @@ netdev_tx_t efx_hard_start_xmit(struct sk_buff *skb,
 
        EFX_WARN_ON_PARANOID(!netif_device_present(net_dev));
 
+       /* PTP "event" packet */
+       if (unlikely(efx_xmit_with_hwtstamp(skb)) &&
+           unlikely(efx_ptp_is_ptp_tx(efx, skb))) {
+               return efx_ptp_tx(efx, skb);
+       }
+
        index = skb_get_queue_mapping(skb);
        type = skb->ip_summed == CHECKSUM_PARTIAL ? EFX_TXQ_TYPE_OFFLOAD : 0;
        if (index >= efx->n_tx_channels) {