net: ethernet: ti: introduce am654 common platform time sync driver
authorGrygorii Strashko <grygorii.strashko@ti.com>
Fri, 1 May 2020 20:50:06 +0000 (23:50 +0300)
committerDavid S. Miller <davem@davemloft.net>
Mon, 4 May 2020 19:02:02 +0000 (12:02 -0700)
The CPTS module is used to facilitate host control of time sync operations.
Main features of CPTS module are:
- selection of multiple external clock sources
- control of time sync events via interrupt or polling
- 64-bit timestamp mode in ns with HW PPM and nudge adjustment.
- hardware timestamp ext. inputs (HWx_TS_PUSH)
- timestamp Generator function outputs (TS_GENFx)
Depending on integration it enables compliance with the IEEE 1588-2008
standard for a precision clock synchronization protocol, Ethernet Enhanced
Scheduled Traffic Operations (CPTS_ESTFn) and PCIe Subsystem Precision Time
Measurement (PTM).

Introduced driver provides Linux PTP hardware clock for each CPTS device
and network packets timestamping where applicable. CPTS PTP hardware clock
supports following operations:
    - Set time
    - Get time
    - Shift the clock by a given offset atomically
    - Adjust clock frequency
    - Time stamp external events
    - Periodic output signals

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/ti/Kconfig
drivers/net/ethernet/ti/Makefile
drivers/net/ethernet/ti/am65-cpts.c [new file with mode: 0644]
drivers/net/ethernet/ti/am65-cpts.h [new file with mode: 0644]

index 7b0ad77..2b7590c 100644 (file)
@@ -110,6 +110,20 @@ config TI_K3_AM65_CPSW_NUSS
          To compile this driver as a module, choose M here: the module
          will be called ti-am65-cpsw-nuss.
 
+config TI_K3_AM65_CPTS
+       tristate "TI K3 AM65x CPTS"
+       depends on ARCH_K3 && OF && PTP_1588_CLOCK
+       depends on PTP_1588_CLOCK
+       select NET_PTP_CLASSIFY
+       help
+         Say y here to support the TI K3 AM65x CPTS with 1588 features such as
+         PTP hardware clock for each CPTS device and network packets
+         timestamping where applicable.
+         Depending on integration CPTS blocks enable compliance with
+         the IEEE 1588-2008 standard for a precision clock synchronization
+         protocol, Ethernet Enhanced Scheduled Traffic Operations (CPTS_ESTFn)
+         and PCIe Subsystem Precision Time Measurement (PTM).
+
 config TI_KEYSTONE_NETCP
        tristate "TI Keystone NETCP Core Support"
        select TI_DAVINCI_MDIO
index 5379219..bf86067 100644 (file)
@@ -26,3 +26,4 @@ keystone_netcp_ethss-y := netcp_ethss.o netcp_sgmii.o netcp_xgbepcsr.o cpsw_ale.
 
 obj-$(CONFIG_TI_K3_AM65_CPSW_NUSS) += ti-am65-cpsw-nuss.o
 ti-am65-cpsw-nuss-y := am65-cpsw-nuss.o cpsw_sl.o am65-cpsw-ethtool.o cpsw_ale.o k3-cppi-desc-pool.o
+obj-$(CONFIG_TI_K3_AM65_CPTS) += am65-cpts.o
diff --git a/drivers/net/ethernet/ti/am65-cpts.c b/drivers/net/ethernet/ti/am65-cpts.c
new file mode 100644 (file)
index 0000000..370162c
--- /dev/null
@@ -0,0 +1,1041 @@
+// SPDX-License-Identifier: GPL-2.0
+/* TI K3 AM65x Common Platform Time Sync
+ *
+ * Copyright (C) 2020 Texas Instruments Incorporated - http://www.ti.com
+ *
+ */
+
+#include <linux/clk.h>
+#include <linux/clk-provider.h>
+#include <linux/err.h>
+#include <linux/if_vlan.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/netdevice.h>
+#include <linux/net_tstamp.h>
+#include <linux/of.h>
+#include <linux/of_irq.h>
+#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
+#include <linux/ptp_classify.h>
+#include <linux/ptp_clock_kernel.h>
+
+#include "am65-cpts.h"
+
+struct am65_genf_regs {
+       u32 comp_lo;    /* Comparison Low Value 0:31 */
+       u32 comp_hi;    /* Comparison High Value 32:63 */
+       u32 control;    /* control */
+       u32 length;     /* Length */
+       u32 ppm_low;    /* PPM Load Low Value 0:31 */
+       u32 ppm_hi;     /* PPM Load High Value 32:63 */
+       u32 ts_nudge;   /* Nudge value */
+} __aligned(32) __packed;
+
+#define AM65_CPTS_GENF_MAX_NUM 9
+#define AM65_CPTS_ESTF_MAX_NUM 8
+
+struct am65_cpts_regs {
+       u32 idver;              /* Identification and version */
+       u32 control;            /* Time sync control */
+       u32 rftclk_sel;         /* Reference Clock Select Register */
+       u32 ts_push;            /* Time stamp event push */
+       u32 ts_load_val_lo;     /* Time Stamp Load Low Value 0:31 */
+       u32 ts_load_en;         /* Time stamp load enable */
+       u32 ts_comp_lo;         /* Time Stamp Comparison Low Value 0:31 */
+       u32 ts_comp_length;     /* Time Stamp Comparison Length */
+       u32 intstat_raw;        /* Time sync interrupt status raw */
+       u32 intstat_masked;     /* Time sync interrupt status masked */
+       u32 int_enable;         /* Time sync interrupt enable */
+       u32 ts_comp_nudge;      /* Time Stamp Comparison Nudge Value */
+       u32 event_pop;          /* Event interrupt pop */
+       u32 event_0;            /* Event Time Stamp lo 0:31 */
+       u32 event_1;            /* Event Type Fields */
+       u32 event_2;            /* Event Type Fields domain */
+       u32 event_3;            /* Event Time Stamp hi 32:63 */
+       u32 ts_load_val_hi;     /* Time Stamp Load High Value 32:63 */
+       u32 ts_comp_hi;         /* Time Stamp Comparison High Value 32:63 */
+       u32 ts_add_val;         /* Time Stamp Add value */
+       u32 ts_ppm_low;         /* Time Stamp PPM Load Low Value 0:31 */
+       u32 ts_ppm_hi;          /* Time Stamp PPM Load High Value 32:63 */
+       u32 ts_nudge;           /* Time Stamp Nudge value */
+       u32 reserv[33];
+       struct am65_genf_regs genf[AM65_CPTS_GENF_MAX_NUM];
+       struct am65_genf_regs estf[AM65_CPTS_ESTF_MAX_NUM];
+};
+
+/* CONTROL_REG */
+#define AM65_CPTS_CONTROL_EN                   BIT(0)
+#define AM65_CPTS_CONTROL_INT_TEST             BIT(1)
+#define AM65_CPTS_CONTROL_TS_COMP_POLARITY     BIT(2)
+#define AM65_CPTS_CONTROL_TSTAMP_EN            BIT(3)
+#define AM65_CPTS_CONTROL_SEQUENCE_EN          BIT(4)
+#define AM65_CPTS_CONTROL_64MODE               BIT(5)
+#define AM65_CPTS_CONTROL_TS_COMP_TOG          BIT(6)
+#define AM65_CPTS_CONTROL_TS_PPM_DIR           BIT(7)
+#define AM65_CPTS_CONTROL_HW1_TS_PUSH_EN       BIT(8)
+#define AM65_CPTS_CONTROL_HW2_TS_PUSH_EN       BIT(9)
+#define AM65_CPTS_CONTROL_HW3_TS_PUSH_EN       BIT(10)
+#define AM65_CPTS_CONTROL_HW4_TS_PUSH_EN       BIT(11)
+#define AM65_CPTS_CONTROL_HW5_TS_PUSH_EN       BIT(12)
+#define AM65_CPTS_CONTROL_HW6_TS_PUSH_EN       BIT(13)
+#define AM65_CPTS_CONTROL_HW7_TS_PUSH_EN       BIT(14)
+#define AM65_CPTS_CONTROL_HW8_TS_PUSH_EN       BIT(15)
+#define AM65_CPTS_CONTROL_HW1_TS_PUSH_OFFSET   (8)
+
+#define AM65_CPTS_CONTROL_TS_SYNC_SEL_MASK     (0xF)
+#define AM65_CPTS_CONTROL_TS_SYNC_SEL_SHIFT    (28)
+
+/* RFTCLK_SEL_REG */
+#define AM65_CPTS_RFTCLK_SEL_MASK              (0x1F)
+
+/* TS_PUSH_REG */
+#define AM65_CPTS_TS_PUSH                      BIT(0)
+
+/* TS_LOAD_EN_REG */
+#define AM65_CPTS_TS_LOAD_EN                   BIT(0)
+
+/* INTSTAT_RAW_REG */
+#define AM65_CPTS_INTSTAT_RAW_TS_PEND          BIT(0)
+
+/* INTSTAT_MASKED_REG */
+#define AM65_CPTS_INTSTAT_MASKED_TS_PEND       BIT(0)
+
+/* INT_ENABLE_REG */
+#define AM65_CPTS_INT_ENABLE_TS_PEND_EN                BIT(0)
+
+/* TS_COMP_NUDGE_REG */
+#define AM65_CPTS_TS_COMP_NUDGE_MASK           (0xFF)
+
+/* EVENT_POP_REG */
+#define AM65_CPTS_EVENT_POP                    BIT(0)
+
+/* EVENT_1_REG */
+#define AM65_CPTS_EVENT_1_SEQUENCE_ID_MASK     GENMASK(15, 0)
+
+#define AM65_CPTS_EVENT_1_MESSAGE_TYPE_MASK    GENMASK(19, 16)
+#define AM65_CPTS_EVENT_1_MESSAGE_TYPE_SHIFT   (16)
+
+#define AM65_CPTS_EVENT_1_EVENT_TYPE_MASK      GENMASK(23, 20)
+#define AM65_CPTS_EVENT_1_EVENT_TYPE_SHIFT     (20)
+
+#define AM65_CPTS_EVENT_1_PORT_NUMBER_MASK     GENMASK(28, 24)
+#define AM65_CPTS_EVENT_1_PORT_NUMBER_SHIFT    (24)
+
+/* EVENT_2_REG */
+#define AM65_CPTS_EVENT_2_REG_DOMAIN_MASK      (0xFF)
+#define AM65_CPTS_EVENT_2_REG_DOMAIN_SHIFT     (0)
+
+enum {
+       AM65_CPTS_EV_PUSH,      /* Time Stamp Push Event */
+       AM65_CPTS_EV_ROLL,      /* Time Stamp Rollover Event */
+       AM65_CPTS_EV_HALF,      /* Time Stamp Half Rollover Event */
+       AM65_CPTS_EV_HW,                /* Hardware Time Stamp Push Event */
+       AM65_CPTS_EV_RX,                /* Ethernet Receive Event */
+       AM65_CPTS_EV_TX,                /* Ethernet Transmit Event */
+       AM65_CPTS_EV_TS_COMP,   /* Time Stamp Compare Event */
+       AM65_CPTS_EV_HOST,      /* Host Transmit Event */
+};
+
+struct am65_cpts_event {
+       struct list_head list;
+       unsigned long tmo;
+       u32 event1;
+       u32 event2;
+       u64 timestamp;
+};
+
+#define AM65_CPTS_FIFO_DEPTH           (16)
+#define AM65_CPTS_MAX_EVENTS           (32)
+#define AM65_CPTS_EVENT_RX_TX_TIMEOUT  (20) /* ms */
+#define AM65_CPTS_SKB_TX_WORK_TIMEOUT  1 /* jiffies */
+#define AM65_CPTS_MIN_PPM              0x400
+
+struct am65_cpts {
+       struct device *dev;
+       struct am65_cpts_regs __iomem *reg;
+       struct ptp_clock_info ptp_info;
+       struct ptp_clock *ptp_clock;
+       int phc_index;
+       struct clk_hw *clk_mux_hw;
+       struct device_node *clk_mux_np;
+       struct clk *refclk;
+       u32 refclk_freq;
+       struct list_head events;
+       struct list_head pool;
+       struct am65_cpts_event pool_data[AM65_CPTS_MAX_EVENTS];
+       spinlock_t lock; /* protects events lists*/
+       u32 ext_ts_inputs;
+       u32 genf_num;
+       u32 ts_add_val;
+       int irq;
+       struct mutex ptp_clk_lock; /* PHC access sync */
+       u64 timestamp;
+       u32 genf_enable;
+       u32 hw_ts_enable;
+       struct sk_buff_head txq;
+};
+
+struct am65_cpts_skb_cb_data {
+       unsigned long tmo;
+       u32 skb_mtype_seqid;
+};
+
+#define am65_cpts_write32(c, v, r) writel(v, &(c)->reg->r)
+#define am65_cpts_read32(c, r) readl(&(c)->reg->r)
+
+static void am65_cpts_settime(struct am65_cpts *cpts, u64 start_tstamp)
+{
+       u32 val;
+
+       val = upper_32_bits(start_tstamp);
+       am65_cpts_write32(cpts, val, ts_load_val_hi);
+       val = lower_32_bits(start_tstamp);
+       am65_cpts_write32(cpts, val, ts_load_val_lo);
+
+       am65_cpts_write32(cpts, AM65_CPTS_TS_LOAD_EN, ts_load_en);
+}
+
+static void am65_cpts_set_add_val(struct am65_cpts *cpts)
+{
+       /* select coefficient according to the rate */
+       cpts->ts_add_val = (NSEC_PER_SEC / cpts->refclk_freq - 1) & 0x7;
+
+       am65_cpts_write32(cpts, cpts->ts_add_val, ts_add_val);
+}
+
+static void am65_cpts_disable(struct am65_cpts *cpts)
+{
+       am65_cpts_write32(cpts, 0, control);
+       am65_cpts_write32(cpts, 0, int_enable);
+}
+
+static int am65_cpts_event_get_port(struct am65_cpts_event *event)
+{
+       return (event->event1 & AM65_CPTS_EVENT_1_PORT_NUMBER_MASK) >>
+               AM65_CPTS_EVENT_1_PORT_NUMBER_SHIFT;
+}
+
+static int am65_cpts_event_get_type(struct am65_cpts_event *event)
+{
+       return (event->event1 & AM65_CPTS_EVENT_1_EVENT_TYPE_MASK) >>
+               AM65_CPTS_EVENT_1_EVENT_TYPE_SHIFT;
+}
+
+static int am65_cpts_cpts_purge_events(struct am65_cpts *cpts)
+{
+       struct list_head *this, *next;
+       struct am65_cpts_event *event;
+       int removed = 0;
+
+       list_for_each_safe(this, next, &cpts->events) {
+               event = list_entry(this, struct am65_cpts_event, list);
+               if (time_after(jiffies, event->tmo)) {
+                       list_del_init(&event->list);
+                       list_add(&event->list, &cpts->pool);
+                       ++removed;
+               }
+       }
+
+       if (removed)
+               dev_dbg(cpts->dev, "event pool cleaned up %d\n", removed);
+       return removed ? 0 : -1;
+}
+
+static bool am65_cpts_fifo_pop_event(struct am65_cpts *cpts,
+                                    struct am65_cpts_event *event)
+{
+       u32 r = am65_cpts_read32(cpts, intstat_raw);
+
+       if (r & AM65_CPTS_INTSTAT_RAW_TS_PEND) {
+               event->timestamp = am65_cpts_read32(cpts, event_0);
+               event->event1 = am65_cpts_read32(cpts, event_1);
+               event->event2 = am65_cpts_read32(cpts, event_2);
+               event->timestamp |= (u64)am65_cpts_read32(cpts, event_3) << 32;
+               am65_cpts_write32(cpts, AM65_CPTS_EVENT_POP, event_pop);
+               return false;
+       }
+       return true;
+}
+
+static int am65_cpts_fifo_read(struct am65_cpts *cpts)
+{
+       struct ptp_clock_event pevent;
+       struct am65_cpts_event *event;
+       bool schedule = false;
+       int i, type, ret = 0;
+       unsigned long flags;
+
+       spin_lock_irqsave(&cpts->lock, flags);
+       for (i = 0; i < AM65_CPTS_FIFO_DEPTH; i++) {
+               event = list_first_entry_or_null(&cpts->pool,
+                                                struct am65_cpts_event, list);
+
+               if (!event) {
+                       if (am65_cpts_cpts_purge_events(cpts)) {
+                               dev_err(cpts->dev, "cpts: event pool empty\n");
+                               ret = -1;
+                               goto out;
+                       }
+                       continue;
+               }
+
+               if (am65_cpts_fifo_pop_event(cpts, event))
+                       break;
+
+               type = am65_cpts_event_get_type(event);
+               switch (type) {
+               case AM65_CPTS_EV_PUSH:
+                       cpts->timestamp = event->timestamp;
+                       dev_dbg(cpts->dev, "AM65_CPTS_EV_PUSH t:%llu\n",
+                               cpts->timestamp);
+                       break;
+               case AM65_CPTS_EV_RX:
+               case AM65_CPTS_EV_TX:
+                       event->tmo = jiffies +
+                               msecs_to_jiffies(AM65_CPTS_EVENT_RX_TX_TIMEOUT);
+
+                       list_del_init(&event->list);
+                       list_add_tail(&event->list, &cpts->events);
+
+                       dev_dbg(cpts->dev,
+                               "AM65_CPTS_EV_TX e1:%08x e2:%08x t:%lld\n",
+                               event->event1, event->event2,
+                               event->timestamp);
+                       schedule = true;
+                       break;
+               case AM65_CPTS_EV_HW:
+                       pevent.index = am65_cpts_event_get_port(event) - 1;
+                       pevent.timestamp = event->timestamp;
+                       pevent.type = PTP_CLOCK_EXTTS;
+                       dev_dbg(cpts->dev, "AM65_CPTS_EV_HW p:%d t:%llu\n",
+                               pevent.index, event->timestamp);
+
+                       ptp_clock_event(cpts->ptp_clock, &pevent);
+                       break;
+               case AM65_CPTS_EV_HOST:
+                       break;
+               case AM65_CPTS_EV_ROLL:
+               case AM65_CPTS_EV_HALF:
+               case AM65_CPTS_EV_TS_COMP:
+                       dev_dbg(cpts->dev,
+                               "AM65_CPTS_EVT: %d e1:%08x e2:%08x t:%lld\n",
+                               type,
+                               event->event1, event->event2,
+                               event->timestamp);
+                       break;
+               default:
+                       dev_err(cpts->dev, "cpts: unknown event type\n");
+                       ret = -1;
+                       goto out;
+               }
+       }
+
+out:
+       spin_unlock_irqrestore(&cpts->lock, flags);
+
+       if (schedule)
+               ptp_schedule_worker(cpts->ptp_clock, 0);
+
+       return ret;
+}
+
+static u64 am65_cpts_gettime(struct am65_cpts *cpts,
+                            struct ptp_system_timestamp *sts)
+{
+       unsigned long flags;
+       u64 val = 0;
+
+       /* temporarily disable cpts interrupt to avoid intentional
+        * doubled read. Interrupt can be in-flight - it's Ok.
+        */
+       am65_cpts_write32(cpts, 0, int_enable);
+
+       /* use spin_lock_irqsave() here as it has to run very fast */
+       spin_lock_irqsave(&cpts->lock, flags);
+       ptp_read_system_prets(sts);
+       am65_cpts_write32(cpts, AM65_CPTS_TS_PUSH, ts_push);
+       am65_cpts_read32(cpts, ts_push);
+       ptp_read_system_postts(sts);
+       spin_unlock_irqrestore(&cpts->lock, flags);
+
+       am65_cpts_fifo_read(cpts);
+
+       am65_cpts_write32(cpts, AM65_CPTS_INT_ENABLE_TS_PEND_EN, int_enable);
+
+       val = cpts->timestamp;
+
+       return val;
+}
+
+static irqreturn_t am65_cpts_interrupt(int irq, void *dev_id)
+{
+       struct am65_cpts *cpts = dev_id;
+
+       if (am65_cpts_fifo_read(cpts))
+               dev_dbg(cpts->dev, "cpts: unable to obtain a time stamp\n");
+
+       return IRQ_HANDLED;
+}
+
+/* PTP clock operations */
+static int am65_cpts_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb)
+{
+       struct am65_cpts *cpts = container_of(ptp, struct am65_cpts, ptp_info);
+       int neg_adj = 0;
+       u64 adj_period;
+       u32 val;
+
+       if (ppb < 0) {
+               neg_adj = 1;
+               ppb = -ppb;
+       }
+
+       /* base freq = 1GHz = 1 000 000 000
+        * ppb_norm = ppb * base_freq / clock_freq;
+        * ppm_norm = ppb_norm / 1000
+        * adj_period = 1 000 000 / ppm_norm
+        * adj_period = 1 000 000 000 / ppb_norm
+        * adj_period = 1 000 000 000 / (ppb * base_freq / clock_freq)
+        * adj_period = (1 000 000 000 * clock_freq) / (ppb * base_freq)
+        * adj_period = clock_freq / ppb
+        */
+       adj_period = div_u64(cpts->refclk_freq, ppb);
+
+       mutex_lock(&cpts->ptp_clk_lock);
+
+       val = am65_cpts_read32(cpts, control);
+       if (neg_adj)
+               val |= AM65_CPTS_CONTROL_TS_PPM_DIR;
+       else
+               val &= ~AM65_CPTS_CONTROL_TS_PPM_DIR;
+       am65_cpts_write32(cpts, val, control);
+
+       val = upper_32_bits(adj_period) & 0x3FF;
+       am65_cpts_write32(cpts, val, ts_ppm_hi);
+       val = lower_32_bits(adj_period);
+       am65_cpts_write32(cpts, val, ts_ppm_low);
+
+       mutex_unlock(&cpts->ptp_clk_lock);
+
+       return 0;
+}
+
+static int am65_cpts_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
+{
+       struct am65_cpts *cpts = container_of(ptp, struct am65_cpts, ptp_info);
+       s64 ns;
+
+       mutex_lock(&cpts->ptp_clk_lock);
+       ns = am65_cpts_gettime(cpts, NULL);
+       ns += delta;
+       am65_cpts_settime(cpts, ns);
+       mutex_unlock(&cpts->ptp_clk_lock);
+
+       return 0;
+}
+
+static int am65_cpts_ptp_gettimex(struct ptp_clock_info *ptp,
+                                 struct timespec64 *ts,
+                                 struct ptp_system_timestamp *sts)
+{
+       struct am65_cpts *cpts = container_of(ptp, struct am65_cpts, ptp_info);
+       u64 ns;
+
+       mutex_lock(&cpts->ptp_clk_lock);
+       ns = am65_cpts_gettime(cpts, sts);
+       mutex_unlock(&cpts->ptp_clk_lock);
+       *ts = ns_to_timespec64(ns);
+
+       return 0;
+}
+
+static int am65_cpts_ptp_settime(struct ptp_clock_info *ptp,
+                                const struct timespec64 *ts)
+{
+       struct am65_cpts *cpts = container_of(ptp, struct am65_cpts, ptp_info);
+       u64 ns;
+
+       ns = timespec64_to_ns(ts);
+       mutex_lock(&cpts->ptp_clk_lock);
+       am65_cpts_settime(cpts, ns);
+       mutex_unlock(&cpts->ptp_clk_lock);
+
+       return 0;
+}
+
+static void am65_cpts_extts_enable_hw(struct am65_cpts *cpts, u32 index, int on)
+{
+       u32 v;
+
+       v = am65_cpts_read32(cpts, control);
+       if (on) {
+               v |= BIT(AM65_CPTS_CONTROL_HW1_TS_PUSH_OFFSET + index);
+               cpts->hw_ts_enable |= BIT(index);
+       } else {
+               v &= ~BIT(AM65_CPTS_CONTROL_HW1_TS_PUSH_OFFSET + index);
+               cpts->hw_ts_enable &= ~BIT(index);
+       }
+       am65_cpts_write32(cpts, v, control);
+}
+
+static int am65_cpts_extts_enable(struct am65_cpts *cpts, u32 index, int on)
+{
+       if (!!(cpts->hw_ts_enable & BIT(index)) == !!on)
+               return 0;
+
+       mutex_lock(&cpts->ptp_clk_lock);
+       am65_cpts_extts_enable_hw(cpts, index, on);
+       mutex_unlock(&cpts->ptp_clk_lock);
+
+       dev_dbg(cpts->dev, "%s: ExtTS:%u %s\n",
+               __func__, index, on ? "enabled" : "disabled");
+
+       return 0;
+}
+
+static void am65_cpts_perout_enable_hw(struct am65_cpts *cpts,
+                                      struct ptp_perout_request *req, int on)
+{
+       u64 ns_period, ns_start, cycles;
+       struct timespec64 ts;
+       u32 val;
+
+       if (on) {
+               ts.tv_sec = req->period.sec;
+               ts.tv_nsec = req->period.nsec;
+               ns_period = timespec64_to_ns(&ts);
+
+               cycles = (ns_period * cpts->refclk_freq) / NSEC_PER_SEC;
+
+               ts.tv_sec = req->start.sec;
+               ts.tv_nsec = req->start.nsec;
+               ns_start = timespec64_to_ns(&ts);
+
+               val = upper_32_bits(ns_start);
+               am65_cpts_write32(cpts, val, genf[req->index].comp_hi);
+               val = lower_32_bits(ns_start);
+               am65_cpts_write32(cpts, val, genf[req->index].comp_lo);
+               val = lower_32_bits(cycles);
+               am65_cpts_write32(cpts, val, genf[req->index].length);
+
+               cpts->genf_enable |= BIT(req->index);
+       } else {
+               am65_cpts_write32(cpts, 0, genf[req->index].length);
+
+               cpts->genf_enable &= ~BIT(req->index);
+       }
+}
+
+static int am65_cpts_perout_enable(struct am65_cpts *cpts,
+                                  struct ptp_perout_request *req, int on)
+{
+       if (!!(cpts->genf_enable & BIT(req->index)) == !!on)
+               return 0;
+
+       mutex_lock(&cpts->ptp_clk_lock);
+       am65_cpts_perout_enable_hw(cpts, req, on);
+       mutex_unlock(&cpts->ptp_clk_lock);
+
+       dev_dbg(cpts->dev, "%s: GenF:%u %s\n",
+               __func__, req->index, on ? "enabled" : "disabled");
+
+       return 0;
+}
+
+static int am65_cpts_ptp_enable(struct ptp_clock_info *ptp,
+                               struct ptp_clock_request *rq, int on)
+{
+       struct am65_cpts *cpts = container_of(ptp, struct am65_cpts, ptp_info);
+
+       switch (rq->type) {
+       case PTP_CLK_REQ_EXTTS:
+               return am65_cpts_extts_enable(cpts, rq->extts.index, on);
+       case PTP_CLK_REQ_PEROUT:
+               return am65_cpts_perout_enable(cpts, &rq->perout, on);
+       default:
+               break;
+       }
+
+       return -EOPNOTSUPP;
+}
+
+static long am65_cpts_ts_work(struct ptp_clock_info *ptp);
+
+static struct ptp_clock_info am65_ptp_info = {
+       .owner          = THIS_MODULE,
+       .name           = "CTPS timer",
+       .adjfreq        = am65_cpts_ptp_adjfreq,
+       .adjtime        = am65_cpts_ptp_adjtime,
+       .gettimex64     = am65_cpts_ptp_gettimex,
+       .settime64      = am65_cpts_ptp_settime,
+       .enable         = am65_cpts_ptp_enable,
+       .do_aux_work    = am65_cpts_ts_work,
+};
+
+static bool am65_cpts_match_tx_ts(struct am65_cpts *cpts,
+                                 struct am65_cpts_event *event)
+{
+       struct sk_buff_head txq_list;
+       struct sk_buff *skb, *tmp;
+       unsigned long flags;
+       bool found = false;
+       u32 mtype_seqid;
+
+       mtype_seqid = event->event1 &
+                     (AM65_CPTS_EVENT_1_MESSAGE_TYPE_MASK |
+                      AM65_CPTS_EVENT_1_EVENT_TYPE_MASK |
+                      AM65_CPTS_EVENT_1_SEQUENCE_ID_MASK);
+
+       __skb_queue_head_init(&txq_list);
+
+       spin_lock_irqsave(&cpts->txq.lock, flags);
+       skb_queue_splice_init(&cpts->txq, &txq_list);
+       spin_unlock_irqrestore(&cpts->txq.lock, flags);
+
+       /* no need to grab txq.lock as access is always done under cpts->lock */
+       skb_queue_walk_safe(&txq_list, skb, tmp) {
+               struct skb_shared_hwtstamps ssh;
+               struct am65_cpts_skb_cb_data *skb_cb =
+                                       (struct am65_cpts_skb_cb_data *)skb->cb;
+
+               if (mtype_seqid == skb_cb->skb_mtype_seqid) {
+                       u64 ns = event->timestamp;
+
+                       memset(&ssh, 0, sizeof(ssh));
+                       ssh.hwtstamp = ns_to_ktime(ns);
+                       skb_tstamp_tx(skb, &ssh);
+                       found = true;
+                       __skb_unlink(skb, &txq_list);
+                       dev_consume_skb_any(skb);
+                       dev_dbg(cpts->dev,
+                               "match tx timestamp mtype_seqid %08x\n",
+                               mtype_seqid);
+                       break;
+               }
+
+               if (time_after(jiffies, skb_cb->tmo)) {
+                       /* timeout any expired skbs over 100 ms */
+                       dev_dbg(cpts->dev,
+                               "expiring tx timestamp mtype_seqid %08x\n",
+                               mtype_seqid);
+                       __skb_unlink(skb, &txq_list);
+                       dev_consume_skb_any(skb);
+               }
+       }
+
+       spin_lock_irqsave(&cpts->txq.lock, flags);
+       skb_queue_splice(&txq_list, &cpts->txq);
+       spin_unlock_irqrestore(&cpts->txq.lock, flags);
+
+       return found;
+}
+
+static void am65_cpts_find_ts(struct am65_cpts *cpts)
+{
+       struct am65_cpts_event *event;
+       struct list_head *this, *next;
+       LIST_HEAD(events_free);
+       unsigned long flags;
+       LIST_HEAD(events);
+
+       spin_lock_irqsave(&cpts->lock, flags);
+       list_splice_init(&cpts->events, &events);
+       spin_unlock_irqrestore(&cpts->lock, flags);
+
+       list_for_each_safe(this, next, &events) {
+               event = list_entry(this, struct am65_cpts_event, list);
+               if (am65_cpts_match_tx_ts(cpts, event) ||
+                   time_after(jiffies, event->tmo)) {
+                       list_del_init(&event->list);
+                       list_add(&event->list, &events_free);
+               }
+       }
+
+       spin_lock_irqsave(&cpts->lock, flags);
+       list_splice_tail(&events, &cpts->events);
+       list_splice_tail(&events_free, &cpts->pool);
+       spin_unlock_irqrestore(&cpts->lock, flags);
+}
+
+static long am65_cpts_ts_work(struct ptp_clock_info *ptp)
+{
+       struct am65_cpts *cpts = container_of(ptp, struct am65_cpts, ptp_info);
+       unsigned long flags;
+       long delay = -1;
+
+       am65_cpts_find_ts(cpts);
+
+       spin_lock_irqsave(&cpts->txq.lock, flags);
+       if (!skb_queue_empty(&cpts->txq))
+               delay = AM65_CPTS_SKB_TX_WORK_TIMEOUT;
+       spin_unlock_irqrestore(&cpts->txq.lock, flags);
+
+       return delay;
+}
+
+/**
+ * am65_cpts_rx_enable - enable rx timestamping
+ * @cpts: cpts handle
+ * @skb: packet
+ *
+ * This functions enables rx packets timestamping. The CPTS can timestamp all
+ * rx packets.
+ */
+void am65_cpts_rx_enable(struct am65_cpts *cpts, bool en)
+{
+       u32 val;
+
+       mutex_lock(&cpts->ptp_clk_lock);
+       val = am65_cpts_read32(cpts, control);
+       if (en)
+               val |= AM65_CPTS_CONTROL_TSTAMP_EN;
+       else
+               val &= ~AM65_CPTS_CONTROL_TSTAMP_EN;
+       am65_cpts_write32(cpts, val, control);
+       mutex_unlock(&cpts->ptp_clk_lock);
+}
+EXPORT_SYMBOL_GPL(am65_cpts_rx_enable);
+
+static int am65_skb_get_mtype_seqid(struct sk_buff *skb, u32 *mtype_seqid)
+{
+       unsigned int ptp_class = ptp_classify_raw(skb);
+       u8 *msgtype, *data = skb->data;
+       unsigned int offset = 0;
+       __be16 *seqid;
+
+       if (ptp_class == PTP_CLASS_NONE)
+               return 0;
+
+       if (ptp_class & PTP_CLASS_VLAN)
+               offset += VLAN_HLEN;
+
+       switch (ptp_class & PTP_CLASS_PMASK) {
+       case PTP_CLASS_IPV4:
+               offset += ETH_HLEN + IPV4_HLEN(data + offset) + UDP_HLEN;
+               break;
+       case PTP_CLASS_IPV6:
+               offset += ETH_HLEN + IP6_HLEN + UDP_HLEN;
+               break;
+       case PTP_CLASS_L2:
+               offset += ETH_HLEN;
+               break;
+       default:
+               return 0;
+       }
+
+       if (skb->len + ETH_HLEN < offset + OFF_PTP_SEQUENCE_ID + sizeof(*seqid))
+               return 0;
+
+       if (unlikely(ptp_class & PTP_CLASS_V1))
+               msgtype = data + offset + OFF_PTP_CONTROL;
+       else
+               msgtype = data + offset;
+
+       seqid = (__be16 *)(data + offset + OFF_PTP_SEQUENCE_ID);
+       *mtype_seqid = (*msgtype << AM65_CPTS_EVENT_1_MESSAGE_TYPE_SHIFT) &
+                       AM65_CPTS_EVENT_1_MESSAGE_TYPE_MASK;
+       *mtype_seqid |= (ntohs(*seqid) & AM65_CPTS_EVENT_1_SEQUENCE_ID_MASK);
+
+       return 1;
+}
+
+/**
+ * am65_cpts_tx_timestamp - save tx packet for timestamping
+ * @cpts: cpts handle
+ * @skb: packet
+ *
+ * This functions saves tx packet for timestamping if packet can be timestamped.
+ * The future processing is done in from PTP auxiliary worker.
+ */
+void am65_cpts_tx_timestamp(struct am65_cpts *cpts, struct sk_buff *skb)
+{
+       struct am65_cpts_skb_cb_data *skb_cb = (void *)skb->cb;
+
+       if (!(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
+               return;
+
+       /* add frame to queue for processing later.
+        * The periodic FIFO check will handle this.
+        */
+       skb_get(skb);
+       /* get the timestamp for timeouts */
+       skb_cb->tmo = jiffies + msecs_to_jiffies(100);
+       skb_queue_tail(&cpts->txq, skb);
+       ptp_schedule_worker(cpts->ptp_clock, 0);
+}
+EXPORT_SYMBOL_GPL(am65_cpts_tx_timestamp);
+
+/**
+ * am65_cpts_prep_tx_timestamp - check and prepare tx packet for timestamping
+ * @cpts: cpts handle
+ * @skb: packet
+ *
+ * This functions should be called from .xmit().
+ * It checks if packet can be timestamped, fills internal cpts data
+ * in skb-cb and marks packet as SKBTX_IN_PROGRESS.
+ */
+void am65_cpts_prep_tx_timestamp(struct am65_cpts *cpts, struct sk_buff *skb)
+{
+       struct am65_cpts_skb_cb_data *skb_cb = (void *)skb->cb;
+       int ret;
+
+       if (!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
+               return;
+
+       ret = am65_skb_get_mtype_seqid(skb, &skb_cb->skb_mtype_seqid);
+       if (!ret)
+               return;
+       skb_cb->skb_mtype_seqid |= (AM65_CPTS_EV_TX <<
+                                  AM65_CPTS_EVENT_1_EVENT_TYPE_SHIFT);
+
+       skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+}
+EXPORT_SYMBOL_GPL(am65_cpts_prep_tx_timestamp);
+
+int am65_cpts_phc_index(struct am65_cpts *cpts)
+{
+       return cpts->phc_index;
+}
+EXPORT_SYMBOL_GPL(am65_cpts_phc_index);
+
+static void cpts_free_clk_mux(void *data)
+{
+       struct am65_cpts *cpts = data;
+
+       of_clk_del_provider(cpts->clk_mux_np);
+       clk_hw_unregister_mux(cpts->clk_mux_hw);
+       of_node_put(cpts->clk_mux_np);
+}
+
+static int cpts_of_mux_clk_setup(struct am65_cpts *cpts,
+                                struct device_node *node)
+{
+       unsigned int num_parents;
+       const char **parent_names;
+       char *clk_mux_name;
+       void __iomem *reg;
+       int ret = -EINVAL;
+
+       cpts->clk_mux_np = of_get_child_by_name(node, "refclk-mux");
+       if (!cpts->clk_mux_np)
+               return 0;
+
+       num_parents = of_clk_get_parent_count(cpts->clk_mux_np);
+       if (num_parents < 1) {
+               dev_err(cpts->dev, "mux-clock %pOF must have parents\n",
+                       cpts->clk_mux_np);
+               goto mux_fail;
+       }
+
+       parent_names = devm_kcalloc(cpts->dev, sizeof(char *), num_parents,
+                                   GFP_KERNEL);
+       if (!parent_names) {
+               ret = -ENOMEM;
+               goto mux_fail;
+       }
+
+       of_clk_parent_fill(cpts->clk_mux_np, parent_names, num_parents);
+
+       clk_mux_name = devm_kasprintf(cpts->dev, GFP_KERNEL, "%s.%pOFn",
+                                     dev_name(cpts->dev), cpts->clk_mux_np);
+       if (!clk_mux_name) {
+               ret = -ENOMEM;
+               goto mux_fail;
+       }
+
+       reg = &cpts->reg->rftclk_sel;
+       /* dev must be NULL to avoid recursive incrementing
+        * of module refcnt
+        */
+       cpts->clk_mux_hw = clk_hw_register_mux(NULL, clk_mux_name,
+                                              parent_names, num_parents,
+                                              0, reg, 0, 5, 0, NULL);
+       if (IS_ERR(cpts->clk_mux_hw)) {
+               ret = PTR_ERR(cpts->clk_mux_hw);
+               goto mux_fail;
+       }
+
+       ret = of_clk_add_hw_provider(cpts->clk_mux_np, of_clk_hw_simple_get,
+                                    cpts->clk_mux_hw);
+       if (ret)
+               goto clk_hw_register;
+
+       ret = devm_add_action_or_reset(cpts->dev, cpts_free_clk_mux, cpts);
+       if (ret)
+               dev_err(cpts->dev, "failed to add clkmux reset action %d", ret);
+
+       return ret;
+
+clk_hw_register:
+       clk_hw_unregister_mux(cpts->clk_mux_hw);
+mux_fail:
+       of_node_put(cpts->clk_mux_np);
+       return ret;
+}
+
+static int am65_cpts_of_parse(struct am65_cpts *cpts, struct device_node *node)
+{
+       u32 prop[2];
+
+       if (!of_property_read_u32(node, "ti,cpts-ext-ts-inputs", &prop[0]))
+               cpts->ext_ts_inputs = prop[0];
+
+       if (!of_property_read_u32(node, "ti,cpts-periodic-outputs", &prop[0]))
+               cpts->genf_num = prop[0];
+
+       return cpts_of_mux_clk_setup(cpts, node);
+}
+
+static void am65_cpts_release(void *data)
+{
+       struct am65_cpts *cpts = data;
+
+       ptp_clock_unregister(cpts->ptp_clock);
+       am65_cpts_disable(cpts);
+       clk_disable_unprepare(cpts->refclk);
+}
+
+struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
+                                  struct device_node *node)
+{
+       struct am65_cpts *cpts;
+       int ret, i;
+
+       cpts = devm_kzalloc(dev, sizeof(*cpts), GFP_KERNEL);
+       if (!cpts)
+               return ERR_PTR(-ENOMEM);
+
+       cpts->dev = dev;
+       cpts->reg = (struct am65_cpts_regs __iomem *)regs;
+
+       cpts->irq = of_irq_get_byname(node, "cpts");
+       if (cpts->irq <= 0) {
+               ret = cpts->irq ?: -ENXIO;
+               if (ret != -EPROBE_DEFER)
+                       dev_err(dev, "Failed to get IRQ number (err = %d)\n",
+                               ret);
+               return ERR_PTR(ret);
+       }
+
+       ret = am65_cpts_of_parse(cpts, node);
+       if (ret)
+               return ERR_PTR(ret);
+
+       mutex_init(&cpts->ptp_clk_lock);
+       INIT_LIST_HEAD(&cpts->events);
+       INIT_LIST_HEAD(&cpts->pool);
+       spin_lock_init(&cpts->lock);
+       skb_queue_head_init(&cpts->txq);
+
+       for (i = 0; i < AM65_CPTS_MAX_EVENTS; i++)
+               list_add(&cpts->pool_data[i].list, &cpts->pool);
+
+       cpts->refclk = devm_get_clk_from_child(dev, node, "cpts");
+       if (IS_ERR(cpts->refclk)) {
+               ret = PTR_ERR(cpts->refclk);
+               if (ret != -EPROBE_DEFER)
+                       dev_err(dev, "Failed to get refclk %d\n", ret);
+               return ERR_PTR(ret);
+       }
+
+       ret = clk_prepare_enable(cpts->refclk);
+       if (ret) {
+               dev_err(dev, "Failed to enable refclk %d\n", ret);
+               return ERR_PTR(ret);
+       }
+
+       cpts->refclk_freq = clk_get_rate(cpts->refclk);
+
+       am65_ptp_info.max_adj = cpts->refclk_freq / AM65_CPTS_MIN_PPM;
+       cpts->ptp_info = am65_ptp_info;
+
+       if (cpts->ext_ts_inputs)
+               cpts->ptp_info.n_ext_ts = cpts->ext_ts_inputs;
+       if (cpts->genf_num)
+               cpts->ptp_info.n_per_out = cpts->genf_num;
+
+       am65_cpts_set_add_val(cpts);
+
+       am65_cpts_write32(cpts, AM65_CPTS_CONTROL_EN | AM65_CPTS_CONTROL_64MODE,
+                         control);
+       am65_cpts_write32(cpts, AM65_CPTS_INT_ENABLE_TS_PEND_EN, int_enable);
+
+       /* set time to the current system time */
+       am65_cpts_settime(cpts, ktime_to_ns(ktime_get_real()));
+
+       cpts->ptp_clock = ptp_clock_register(&cpts->ptp_info, cpts->dev);
+       if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
+               dev_err(dev, "Failed to register ptp clk %ld\n",
+                       PTR_ERR(cpts->ptp_clock));
+               if (!cpts->ptp_clock)
+                       ret = -ENODEV;
+               goto refclk_disable;
+       }
+       cpts->phc_index = ptp_clock_index(cpts->ptp_clock);
+
+       ret = devm_add_action_or_reset(dev, am65_cpts_release, cpts);
+       if (ret) {
+               dev_err(dev, "failed to add ptpclk reset action %d", ret);
+               return ERR_PTR(ret);
+       }
+
+       ret = devm_request_threaded_irq(dev, cpts->irq, NULL,
+                                       am65_cpts_interrupt,
+                                       IRQF_ONESHOT, dev_name(dev), cpts);
+       if (ret < 0) {
+               dev_err(cpts->dev, "error attaching irq %d\n", ret);
+               return ERR_PTR(ret);
+       }
+
+       dev_info(dev, "CPTS ver 0x%08x, freq:%u, add_val:%u\n",
+                am65_cpts_read32(cpts, idver),
+                cpts->refclk_freq, cpts->ts_add_val);
+
+       return cpts;
+
+refclk_disable:
+       clk_disable_unprepare(cpts->refclk);
+       return ERR_PTR(ret);
+}
+EXPORT_SYMBOL_GPL(am65_cpts_create);
+
+static int am65_cpts_probe(struct platform_device *pdev)
+{
+       struct device_node *node = pdev->dev.of_node;
+       struct device *dev = &pdev->dev;
+       struct am65_cpts *cpts;
+       struct resource *res;
+       void __iomem *base;
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cpts");
+       base = devm_ioremap_resource(dev, res);
+       if (IS_ERR(base))
+               return PTR_ERR(base);
+
+       cpts = am65_cpts_create(dev, base, node);
+       if (IS_ERR(cpts))
+               return PTR_ERR(cpts);
+
+       return 0;
+}
+
+static const struct of_device_id am65_cpts_of_match[] = {
+       { .compatible = "ti,am65-cpts", },
+       { .compatible = "ti,j721e-cpts", },
+       {},
+};
+MODULE_DEVICE_TABLE(of, am65_cpts_of_match);
+
+static struct platform_driver am65_cpts_driver = {
+       .probe          = am65_cpts_probe,
+       .driver         = {
+               .name   = "am65-cpts",
+               .of_match_table = am65_cpts_of_match,
+       },
+};
+module_platform_driver(am65_cpts_driver);
+
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Grygorii Strashko <grygorii.strashko@ti.com>");
+MODULE_DESCRIPTION("TI K3 AM65 CPTS driver");
diff --git a/drivers/net/ethernet/ti/am65-cpts.h b/drivers/net/ethernet/ti/am65-cpts.h
new file mode 100644 (file)
index 0000000..0b55dc1
--- /dev/null
@@ -0,0 +1,50 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/* TI K3 AM65 CPTS driver interface
+ *
+ * Copyright (C) 2020 Texas Instruments Incorporated - http://www.ti.com
+ */
+
+#ifndef K3_CPTS_H_
+#define K3_CPTS_H_
+
+#include <linux/device.h>
+#include <linux/of.h>
+
+struct am65_cpts;
+
+#if IS_ENABLED(CONFIG_TI_K3_AM65_CPTS)
+struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
+                                  struct device_node *node);
+int am65_cpts_phc_index(struct am65_cpts *cpts);
+void am65_cpts_tx_timestamp(struct am65_cpts *cpts, struct sk_buff *skb);
+void am65_cpts_prep_tx_timestamp(struct am65_cpts *cpts, struct sk_buff *skb);
+void am65_cpts_rx_enable(struct am65_cpts *cpts, bool en);
+#else
+static inline struct am65_cpts *am65_cpts_create(struct device *dev,
+                                                void __iomem *regs,
+                                                struct device_node *node)
+{
+       return ERR_PTR(-EOPNOTSUPP);
+}
+
+static inline int am65_cpts_phc_index(struct am65_cpts *cpts)
+{
+       return -1;
+}
+
+static inline void am65_cpts_tx_timestamp(struct am65_cpts *cpts,
+                                         struct sk_buff *skb)
+{
+}
+
+static inline void am65_cpts_prep_tx_timestamp(struct am65_cpts *cpts,
+                                              struct sk_buff *skb)
+{
+}
+
+static inline void am65_cpts_rx_enable(struct am65_cpts *cpts, bool en)
+{
+}
+#endif
+
+#endif /* K3_CPTS_H_ */