keystone2: add keystone multicore navigator driver
authorVitaly Andrianov <vitalya@ti.com>
Tue, 1 Apr 2014 19:01:12 +0000 (15:01 -0400)
committerTom Rini <trini@ti.com>
Thu, 17 Apr 2014 21:24:39 +0000 (17:24 -0400)
Multicore navigator consists of Network Coprocessor (NetCP) and
Queue Manager sub system. More details on the hardware can
be obtained from the following links:-

Network Coprocessor: http://www.ti.com/lit/pdf/sprugz6
Multicore Navigator: http://www.ti.com/lit/pdf/sprugr9

Multicore navigator driver implements APIs to configure
the Queue Manager and NetCP Pkt DMA.

Signed-off-by: Vitaly Andrianov <vitalya@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Acked-by: Tom Rini <trini@ti.com>
arch/arm/cpu/armv7/keystone/Makefile
arch/arm/cpu/armv7/keystone/keystone_nav.c [new file with mode: 0644]
arch/arm/include/asm/arch-keystone/keystone_nav.h [new file with mode: 0644]

index 05b385204de0535029d5e5e5e337dca7b14807c2..b1bd0224eae39d29f11ed5e7de0d8e0577e0228f 100644 (file)
@@ -11,6 +11,7 @@ obj-y += psc.o
 obj-y  += clock.o
 obj-y  += cmd_clock.o
 obj-y  += cmd_mon.o
+obj-y  += keystone_nav.o
 obj-y  += msmc.o
 obj-$(CONFIG_SPL_BUILD)        += spl.o
 obj-y  += ddr3.o
diff --git a/arch/arm/cpu/armv7/keystone/keystone_nav.c b/arch/arm/cpu/armv7/keystone/keystone_nav.c
new file mode 100644 (file)
index 0000000..39d6f99
--- /dev/null
@@ -0,0 +1,376 @@
+/*
+ * Multicore Navigator driver for TI Keystone 2 devices.
+ *
+ * (C) Copyright 2012-2014
+ *     Texas Instruments Incorporated, <www.ti.com>
+ *
+ * SPDX-License-Identifier:     GPL-2.0+
+ */
+#include <common.h>
+#include <asm/io.h>
+#include <asm/arch/keystone_nav.h>
+
+static int soc_type =
+#ifdef CONFIG_SOC_K2HK
+       k2hk;
+#endif
+
+struct qm_config k2hk_qm_memmap = {
+       .stat_cfg       = 0x02a40000,
+       .queue          = (struct qm_reg_queue *)0x02a80000,
+       .mngr_vbusm     = 0x23a80000,
+       .i_lram         = 0x00100000,
+       .proxy          = (struct qm_reg_queue *)0x02ac0000,
+       .status_ram     = 0x02a06000,
+       .mngr_cfg       = (struct qm_cfg_reg *)0x02a02000,
+       .intd_cfg       = 0x02a0c000,
+       .desc_mem       = (struct descr_mem_setup_reg *)0x02a03000,
+       .region_num     = 64,
+       .pdsp_cmd       = 0x02a20000,
+       .pdsp_ctl       = 0x02a0f000,
+       .pdsp_iram      = 0x02a10000,
+       .qpool_num      = 4000,
+};
+
+/*
+ * We are going to use only one type of descriptors - host packet
+ * descriptors. We staticaly allocate memory for them here
+ */
+struct qm_host_desc desc_pool[HDESC_NUM] __aligned(sizeof(struct qm_host_desc));
+
+static struct qm_config *qm_cfg;
+
+inline int num_of_desc_to_reg(int num_descr)
+{
+       int j, num;
+
+       for (j = 0, num = 32; j < 15; j++, num *= 2) {
+               if (num_descr <= num)
+                       return j;
+       }
+
+       return 15;
+}
+
+static int _qm_init(struct qm_config *cfg)
+{
+       u32     j;
+
+       if (cfg == NULL)
+               return QM_ERR;
+
+       qm_cfg = cfg;
+
+       qm_cfg->mngr_cfg->link_ram_base0        = qm_cfg->i_lram;
+       qm_cfg->mngr_cfg->link_ram_size0        = HDESC_NUM * 8;
+       qm_cfg->mngr_cfg->link_ram_base1        = 0;
+       qm_cfg->mngr_cfg->link_ram_size1        = 0;
+       qm_cfg->mngr_cfg->link_ram_base2        = 0;
+
+       qm_cfg->desc_mem[0].base_addr = (u32)desc_pool;
+       qm_cfg->desc_mem[0].start_idx = 0;
+       qm_cfg->desc_mem[0].desc_reg_size =
+               (((sizeof(struct qm_host_desc) >> 4) - 1) << 16) |
+               num_of_desc_to_reg(HDESC_NUM);
+
+       memset(desc_pool, 0, sizeof(desc_pool));
+       for (j = 0; j < HDESC_NUM; j++)
+               qm_push(&desc_pool[j], qm_cfg->qpool_num);
+
+       return QM_OK;
+}
+
+int qm_init(void)
+{
+       switch (soc_type) {
+       case k2hk:
+               return _qm_init(&k2hk_qm_memmap);
+       }
+
+       return QM_ERR;
+}
+
+void qm_close(void)
+{
+       u32     j;
+
+       if (qm_cfg == NULL)
+               return;
+
+       queue_close(qm_cfg->qpool_num);
+
+       qm_cfg->mngr_cfg->link_ram_base0        = 0;
+       qm_cfg->mngr_cfg->link_ram_size0        = 0;
+       qm_cfg->mngr_cfg->link_ram_base1        = 0;
+       qm_cfg->mngr_cfg->link_ram_size1        = 0;
+       qm_cfg->mngr_cfg->link_ram_base2        = 0;
+
+       for (j = 0; j < qm_cfg->region_num; j++) {
+               qm_cfg->desc_mem[j].base_addr = 0;
+               qm_cfg->desc_mem[j].start_idx = 0;
+               qm_cfg->desc_mem[j].desc_reg_size = 0;
+       }
+
+       qm_cfg = NULL;
+}
+
+void qm_push(struct qm_host_desc *hd, u32 qnum)
+{
+       u32 regd;
+
+       if (!qm_cfg)
+               return;
+
+       cpu_to_bus((u32 *)hd, sizeof(struct qm_host_desc)/4);
+       regd = (u32)hd | ((sizeof(struct qm_host_desc) >> 4) - 1);
+       writel(regd, &qm_cfg->queue[qnum].ptr_size_thresh);
+}
+
+void qm_buff_push(struct qm_host_desc *hd, u32 qnum,
+                   void *buff_ptr, u32 buff_len)
+{
+       hd->orig_buff_len = buff_len;
+       hd->buff_len = buff_len;
+       hd->orig_buff_ptr = (u32)buff_ptr;
+       hd->buff_ptr = (u32)buff_ptr;
+       qm_push(hd, qnum);
+}
+
+struct qm_host_desc *qm_pop(u32 qnum)
+{
+       u32 uhd;
+
+       if (!qm_cfg)
+               return NULL;
+
+       uhd = readl(&qm_cfg->queue[qnum].ptr_size_thresh) & ~0xf;
+       if (uhd)
+               cpu_to_bus((u32 *)uhd, sizeof(struct qm_host_desc)/4);
+
+       return (struct qm_host_desc *)uhd;
+}
+
+struct qm_host_desc *qm_pop_from_free_pool(void)
+{
+       if (!qm_cfg)
+               return NULL;
+
+       return qm_pop(qm_cfg->qpool_num);
+}
+
+void queue_close(u32 qnum)
+{
+       struct qm_host_desc *hd;
+
+       while ((hd = qm_pop(qnum)))
+               ;
+}
+
+/*
+ * DMA API
+ */
+
+struct pktdma_cfg k2hk_netcp_pktdma = {
+       .global         = (struct global_ctl_regs *)0x02004000,
+       .tx_ch          = (struct tx_chan_regs *)0x02004400,
+       .tx_ch_num      = 9,
+       .rx_ch          = (struct rx_chan_regs *)0x02004800,
+       .rx_ch_num      = 26,
+       .tx_sched       = (u32 *)0x02004c00,
+       .rx_flows       = (struct rx_flow_regs *)0x02005000,
+       .rx_flow_num    = 32,
+       .rx_free_q      = 4001,
+       .rx_rcv_q       = 4002,
+       .tx_snd_q       = 648,
+};
+
+struct pktdma_cfg *netcp;
+
+static int netcp_rx_disable(void)
+{
+       u32 j, v, k;
+
+       for (j = 0; j < netcp->rx_ch_num; j++) {
+               v = readl(&netcp->rx_ch[j].cfg_a);
+               if (!(v & CPDMA_CHAN_A_ENABLE))
+                       continue;
+
+               writel(v | CPDMA_CHAN_A_TDOWN, &netcp->rx_ch[j].cfg_a);
+               for (k = 0; k < TDOWN_TIMEOUT_COUNT; k++) {
+                       udelay(100);
+                       v = readl(&netcp->rx_ch[j].cfg_a);
+                       if (!(v & CPDMA_CHAN_A_ENABLE))
+                               continue;
+               }
+               /* TODO: teardown error on if TDOWN_TIMEOUT_COUNT is reached */
+       }
+
+       /* Clear all of the flow registers */
+       for (j = 0; j < netcp->rx_flow_num; j++) {
+               writel(0, &netcp->rx_flows[j].control);
+               writel(0, &netcp->rx_flows[j].tags);
+               writel(0, &netcp->rx_flows[j].tag_sel);
+               writel(0, &netcp->rx_flows[j].fdq_sel[0]);
+               writel(0, &netcp->rx_flows[j].fdq_sel[1]);
+               writel(0, &netcp->rx_flows[j].thresh[0]);
+               writel(0, &netcp->rx_flows[j].thresh[1]);
+               writel(0, &netcp->rx_flows[j].thresh[2]);
+       }
+
+       return QM_OK;
+}
+
+static int netcp_tx_disable(void)
+{
+       u32 j, v, k;
+
+       for (j = 0; j < netcp->tx_ch_num; j++) {
+               v = readl(&netcp->tx_ch[j].cfg_a);
+               if (!(v & CPDMA_CHAN_A_ENABLE))
+                       continue;
+
+               writel(v | CPDMA_CHAN_A_TDOWN, &netcp->tx_ch[j].cfg_a);
+               for (k = 0; k < TDOWN_TIMEOUT_COUNT; k++) {
+                       udelay(100);
+                       v = readl(&netcp->tx_ch[j].cfg_a);
+                       if (!(v & CPDMA_CHAN_A_ENABLE))
+                               continue;
+               }
+               /* TODO: teardown error on if TDOWN_TIMEOUT_COUNT is reached */
+       }
+
+       return QM_OK;
+}
+
+static int _netcp_init(struct pktdma_cfg *netcp_cfg,
+                      struct rx_buff_desc *rx_buffers)
+{
+       u32 j, v;
+       struct qm_host_desc *hd;
+       u8 *rx_ptr;
+
+       if (netcp_cfg == NULL || rx_buffers == NULL ||
+           rx_buffers->buff_ptr == NULL || qm_cfg == NULL)
+               return QM_ERR;
+
+       netcp = netcp_cfg;
+       netcp->rx_flow = rx_buffers->rx_flow;
+
+       /* init rx queue */
+       rx_ptr = rx_buffers->buff_ptr;
+
+       for (j = 0; j < rx_buffers->num_buffs; j++) {
+               hd = qm_pop(qm_cfg->qpool_num);
+               if (hd == NULL)
+                       return QM_ERR;
+
+               qm_buff_push(hd, netcp->rx_free_q,
+                            rx_ptr, rx_buffers->buff_len);
+
+               rx_ptr += rx_buffers->buff_len;
+       }
+
+       netcp_rx_disable();
+
+       /* configure rx channels */
+       v = CPDMA_REG_VAL_MAKE_RX_FLOW_A(1, 1, 0, 0, 0, 0, 0, netcp->rx_rcv_q);
+       writel(v, &netcp->rx_flows[netcp->rx_flow].control);
+       writel(0, &netcp->rx_flows[netcp->rx_flow].tags);
+       writel(0, &netcp->rx_flows[netcp->rx_flow].tag_sel);
+
+       v = CPDMA_REG_VAL_MAKE_RX_FLOW_D(0, netcp->rx_free_q, 0,
+                                        netcp->rx_free_q);
+
+       writel(v, &netcp->rx_flows[netcp->rx_flow].fdq_sel[0]);
+       writel(v, &netcp->rx_flows[netcp->rx_flow].fdq_sel[1]);
+       writel(0, &netcp->rx_flows[netcp->rx_flow].thresh[0]);
+       writel(0, &netcp->rx_flows[netcp->rx_flow].thresh[1]);
+       writel(0, &netcp->rx_flows[netcp->rx_flow].thresh[2]);
+
+       for (j = 0; j < netcp->rx_ch_num; j++)
+               writel(CPDMA_CHAN_A_ENABLE, &netcp->rx_ch[j].cfg_a);
+
+       /* configure tx channels */
+       /* Disable loopback in the tx direction */
+       writel(0, &netcp->global->emulation_control);
+
+/* TODO: make it dependend on a soc type variable */
+#ifdef CONFIG_SOC_K2HK
+       /* Set QM base address, only for K2x devices */
+       writel(0x23a80000, &netcp->global->qm_base_addr[0]);
+#endif
+
+       /* Enable all channels. The current state isn't important */
+       for (j = 0; j < netcp->tx_ch_num; j++)  {
+               writel(0, &netcp->tx_ch[j].cfg_b);
+               writel(CPDMA_CHAN_A_ENABLE, &netcp->tx_ch[j].cfg_a);
+       }
+
+       return QM_OK;
+}
+
+int netcp_init(struct rx_buff_desc *rx_buffers)
+{
+       switch (soc_type) {
+       case k2hk:
+               _netcp_init(&k2hk_netcp_pktdma, rx_buffers);
+               return QM_OK;
+       }
+       return QM_ERR;
+}
+
+int netcp_close(void)
+{
+       if (!netcp)
+               return QM_ERR;
+
+       netcp_tx_disable();
+       netcp_rx_disable();
+
+       queue_close(netcp->rx_free_q);
+       queue_close(netcp->rx_rcv_q);
+       queue_close(netcp->tx_snd_q);
+
+       return QM_OK;
+}
+
+int netcp_send(u32 *pkt, int num_bytes, u32 swinfo2)
+{
+       struct qm_host_desc *hd;
+
+       hd = qm_pop(qm_cfg->qpool_num);
+       if (hd == NULL)
+               return QM_ERR;
+
+       hd->desc_info   = num_bytes;
+       hd->swinfo[2]   = swinfo2;
+       hd->packet_info = qm_cfg->qpool_num;
+
+       qm_buff_push(hd, netcp->tx_snd_q, pkt, num_bytes);
+
+       return QM_OK;
+}
+
+void *netcp_recv(u32 **pkt, int *num_bytes)
+{
+       struct qm_host_desc *hd;
+
+       hd = qm_pop(netcp->rx_rcv_q);
+       if (!hd)
+               return NULL;
+
+       *pkt = (u32 *)hd->buff_ptr;
+       *num_bytes = hd->desc_info & 0x3fffff;
+
+       return hd;
+}
+
+void netcp_release_rxhd(void *hd)
+{
+       struct qm_host_desc *_hd = (struct qm_host_desc *)hd;
+
+       _hd->buff_len = _hd->orig_buff_len;
+       _hd->buff_ptr = _hd->orig_buff_ptr;
+
+       qm_push(_hd, netcp->rx_free_q);
+}
diff --git a/arch/arm/include/asm/arch-keystone/keystone_nav.h b/arch/arm/include/asm/arch-keystone/keystone_nav.h
new file mode 100644 (file)
index 0000000..ab81eaf
--- /dev/null
@@ -0,0 +1,193 @@
+/*
+ * Multicore Navigator definitions
+ *
+ * (C) Copyright 2012-2014
+ *     Texas Instruments Incorporated, <www.ti.com>
+ *
+ * SPDX-License-Identifier:     GPL-2.0+
+ */
+
+#ifndef _KEYSTONE_NAV_H_
+#define _KEYSTONE_NAV_H_
+
+#include <asm/arch/hardware.h>
+#include <asm/io.h>
+
+enum soc_type_t {
+       k2hk
+};
+
+#define QM_OK                    0
+#define QM_ERR                  -1
+#define QM_DESC_TYPE_HOST        0
+#define QM_DESC_PSINFO_IN_DESCR  0
+#define QM_DESC_DEFAULT_DESCINFO    (QM_DESC_TYPE_HOST << 30) | \
+                                       (QM_DESC_PSINFO_IN_DESCR << 22)
+
+/* Packet Info */
+#define QM_DESC_PINFO_EPIB              1
+#define QM_DESC_PINFO_RETURN_OWN        1
+#define QM_DESC_DEFAULT_PINFO           (QM_DESC_PINFO_EPIB << 31) | \
+                                       (QM_DESC_PINFO_RETURN_OWN << 15)
+
+struct qm_cfg_reg {
+       u32     revision;
+       u32     __pad1;
+       u32     divert;
+       u32     link_ram_base0;
+       u32     link_ram_size0;
+       u32     link_ram_base1;
+       u32     link_ram_size1;
+       u32     link_ram_base2;
+       u32     starvation[0];
+};
+
+struct descr_mem_setup_reg {
+       u32     base_addr;
+       u32     start_idx;
+       u32     desc_reg_size;
+       u32     _res0;
+};
+
+struct qm_reg_queue {
+       u32     entry_count;
+       u32     byte_count;
+       u32     packet_size;
+       u32     ptr_size_thresh;
+};
+
+struct qm_config {
+       /* QM module addresses */
+       u32     stat_cfg;       /* status and config            */
+       struct qm_reg_queue *queue;     /* management region    */
+       u32     mngr_vbusm;     /* management region (VBUSM)    */
+       u32     i_lram;         /* internal linking RAM         */
+       struct qm_reg_queue *proxy;
+       u32     status_ram;
+       struct qm_cfg_reg *mngr_cfg;
+                               /* Queue manager config region  */
+       u32     intd_cfg;       /* QMSS INTD config region      */
+       struct  descr_mem_setup_reg *desc_mem;
+                               /* descritor memory setup region*/
+       u32     region_num;
+       u32     pdsp_cmd;       /* PDSP1 command interface      */
+       u32     pdsp_ctl;       /* PDSP1 control registers      */
+       u32     pdsp_iram;
+       /* QM configuration parameters */
+
+       u32     qpool_num;      /* */
+};
+
+struct qm_host_desc {
+       u32 desc_info;
+       u32 tag_info;
+       u32 packet_info;
+       u32 buff_len;
+       u32 buff_ptr;
+       u32 next_bdptr;
+       u32 orig_buff_len;
+       u32 orig_buff_ptr;
+       u32 timestamp;
+       u32 swinfo[3];
+       u32 ps_data[20];
+};
+
+#define HDESC_NUM        256
+
+int    qm_init(void);
+void   qm_close(void);
+void   qm_push(struct qm_host_desc *hd, u32 qnum);
+struct qm_host_desc *qm_pop(u32 qnum);
+
+void   qm_buff_push(struct qm_host_desc *hd, u32 qnum,
+                    void *buff_ptr, u32 buff_len);
+
+struct qm_host_desc *qm_pop_from_free_pool(void);
+void   queue_close(u32 qnum);
+
+/*
+ * DMA API
+ */
+#define CPDMA_REG_VAL_MAKE_RX_FLOW_A(einfo, psinfo, rxerr, desc, \
+                                    psloc, sopoff, qmgr, qnum) \
+       (((einfo & 1) << 30)  | \
+        ((psinfo & 1) << 29) | \
+        ((rxerr & 1) << 28)  | \
+        ((desc & 3) << 26)   | \
+        ((psloc & 1) << 25)  | \
+        ((sopoff & 0x1ff) << 16) | \
+        ((qmgr & 3) << 12)   | \
+        ((qnum & 0xfff) << 0))
+
+#define CPDMA_REG_VAL_MAKE_RX_FLOW_D(fd0qm, fd0qnum, fd1qm, fd1qnum) \
+       (((fd0qm & 3) << 28)  | \
+        ((fd0qnum & 0xfff) << 16) | \
+        ((fd1qm & 3) << 12)  | \
+        ((fd1qnum & 0xfff) <<  0))
+
+#define CPDMA_CHAN_A_ENABLE ((u32)1 << 31)
+#define CPDMA_CHAN_A_TDOWN  (1 << 30)
+#define TDOWN_TIMEOUT_COUNT  100
+
+struct global_ctl_regs {
+       u32     revision;
+       u32     perf_control;
+       u32     emulation_control;
+       u32     priority_control;
+       u32     qm_base_addr[4];
+};
+
+struct tx_chan_regs {
+       u32     cfg_a;
+       u32     cfg_b;
+       u32     res[6];
+};
+
+struct rx_chan_regs {
+       u32     cfg_a;
+       u32     res[7];
+};
+
+struct rx_flow_regs {
+       u32     control;
+       u32     tags;
+       u32     tag_sel;
+       u32     fdq_sel[2];
+       u32     thresh[3];
+};
+
+struct pktdma_cfg {
+       struct global_ctl_regs  *global;
+       struct tx_chan_regs     *tx_ch;
+       u32                     tx_ch_num;
+       struct rx_chan_regs     *rx_ch;
+       u32                     rx_ch_num;
+       u32                     *tx_sched;
+       struct rx_flow_regs     *rx_flows;
+       u32                     rx_flow_num;
+
+       u32                     rx_free_q;
+       u32                     rx_rcv_q;
+       u32                     tx_snd_q;
+
+       u32                     rx_flow; /* flow that is used for RX */
+};
+
+/*
+ * packet dma user allocates memory for rx buffers
+ * and describe it in the following structure
+ */
+struct rx_buff_desc {
+       u8      *buff_ptr;
+       u32     num_buffs;
+       u32     buff_len;
+       u32     rx_flow;
+};
+
+int netcp_close(void);
+int netcp_init(struct rx_buff_desc *rx_buffers);
+int netcp_send(u32 *pkt, int num_bytes, u32 swinfo2);
+void *netcp_recv(u32 **pkt, int *num_bytes);
+void netcp_release_rxhd(void *hd);
+
+#endif  /* _KEYSTONE_NAV_H_ */