patch-2.1.79 linux/drivers/fc4/soc.c
Next file: linux/drivers/fc4/soc.h
Previous file: linux/drivers/fc4/fcp_scsi.h
Back to the patch index
Back to the overall index
- Lines: 711
- Date:
Mon Jan 12 15:19:36 1998
- Orig file:
v2.1.78/linux/drivers/fc4/soc.c
- Orig date:
Wed Dec 31 16:00:00 1969
diff -u --recursive --new-file v2.1.78/linux/drivers/fc4/soc.c linux/drivers/fc4/soc.c
@@ -0,0 +1,710 @@
+/* soc.c: Sparc SUNW,soc (Serial Optical Channel) Fibre Channel Sbus adapter support.
+ *
+ * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
+ * Copyright (C) 1997 Jiri Hanika (geo@ff.cuni.cz)
+ *
+ * Sources:
+ * Fibre Channel Physical & Signaling Interface (FC-PH), dpANS, 1994
+ * dpANS Fibre Channel Protocol for SCSI (X3.269-199X), Rev. 012, 1995
+ */
+
+static char *version =
+ "soc.c:v1.0 24/Nov/97 Jakub Jelinek (jj@sunsite.mff.cuni.cz), Jiri Hanika (geo@ff.cuni.cz)\n";
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/fcntl.h>
+#include <linux/interrupt.h>
+#include <linux/ptrace.h>
+#include <linux/ioport.h>
+#include <linux/in.h>
+#include <linux/malloc.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <asm/system.h>
+#include <asm/bitops.h>
+#include <asm/io.h>
+#include <asm/dma.h>
+#include <linux/errno.h>
+#include <asm/byteorder.h>
+
+#include <asm/openprom.h>
+#include <asm/oplib.h>
+#include <asm/auxio.h>
+#include <asm/system.h>
+#include <asm/pgtable.h>
+#include <asm/irq.h>
+
+/* #define SOCDEBUG */
+/* #define HAVE_SOC_UCODE */
+
+#include "fcp_scsi.h"
+#include "soc.h"
+#ifdef HAVE_SOC_UCODE
+#include "soc_asm.c"
+#endif
+
+#define soc_printk printk ("soc%d: ", s->soc_no); printk
+
+#ifdef SOCDEBUG
+#define SOD(x) soc_printk x;
+#else
+#define SOD(x)
+#endif
+
+#define for_each_soc(s) for (s = socs; s; s = s->next)
+struct soc *socs = NULL;
+
+static inline void soc_disable(struct soc *s)
+{
+ s->regs->imask = 0; s->regs->cmd = SOC_CMD_SOFT_RESET;
+}
+
+static inline void soc_enable(struct soc *s)
+{
+ SOD(("enable %08x\n", s->cfg))
+ s->regs->sae = 0; s->regs->cfg = s->cfg;
+ s->regs->cmd = SOC_CMD_RSP_QALL;
+ SOC_SETIMASK(s, SOC_IMASK_RSP_QALL | SOC_IMASK_SAE);
+}
+
+static void soc_reset(fc_channel *fc)
+{
+ soc_port *port = (soc_port *)fc;
+ struct soc *s = port->s;
+
+ /* FIXME */
+ soc_disable(s);
+ s->req[0].seqno = 1;
+ s->req[1].seqno = 1;
+ s->rsp[0].seqno = 1;
+ s->rsp[1].seqno = 1;
+ s->req[0].in = 0;
+ s->req[1].in = 0;
+ s->rsp[0].in = 0;
+ s->rsp[1].in = 0;
+ s->req[0].out = 0;
+ s->req[1].out = 0;
+ s->rsp[0].out = 0;
+ s->rsp[1].out = 0;
+
+ /* FIXME */
+ soc_enable(s);
+}
+
+static void inline soc_solicited (struct soc *s)
+{
+ fc_hdr fchdr;
+ soc_rsp *hwrsp;
+ soc_cq *sw_cq;
+ int token;
+ int status;
+ fc_channel *fc;
+
+ sw_cq = &s->rsp[SOC_SOLICITED_RSP_Q];
+
+ if (sw_cq->pool == NULL)
+ sw_cq->pool =
+ (soc_req *)(s->xram + (xram_get_32low ((xram_p)&sw_cq->hw_cq->address) / sizeof(u16)));
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ SOD (("soc_solicited, %d packets arrived\n", (sw_cq->in - sw_cq->out) & sw_cq->last))
+ for (;;) {
+ hwrsp = (soc_rsp *)sw_cq->pool + sw_cq->out;
+ token = xram_get_32low ((xram_p)&hwrsp->shdr.token);
+ status = xram_get_32low ((xram_p)&hwrsp->status);
+ fc = (fc_channel *)(&s->port[(token >> 11) & 1]);
+
+ if (status == SOC_OK)
+ fcp_receive_solicited(fc, token >> 12, token & ((1 << 11) - 1), FC_STATUS_OK, NULL);
+ else {
+ xram_copy_from(&fchdr, (xram_p)&hwrsp->fchdr, sizeof(fchdr));
+ /* We have intentionally defined FC_STATUS_* constants to match SOC_* constants, otherwise
+ we'd have to translate status */
+ fcp_receive_solicited(fc, token >> 12, token & ((1 << 11) - 1), status, &fchdr);
+ }
+
+ if (++sw_cq->out > sw_cq->last) {
+ sw_cq->seqno++;
+ sw_cq->out = 0;
+ }
+
+ if (sw_cq->out == sw_cq->in) {
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ if (sw_cq->out == sw_cq->in) {
+ /* Tell the hardware about it */
+ s->regs->cmd = (sw_cq->out << 24) | (SOC_CMD_RSP_QALL & ~(SOC_CMD_RSP_Q0 << SOC_SOLICITED_RSP_Q));
+ /* Read it, so that we're sure it has been updated */
+ s->regs->cmd;
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ if (sw_cq->out == sw_cq->in)
+ break;
+ }
+ }
+ }
+}
+
+static void inline soc_request (struct soc *s, u32 cmd)
+{
+ SOC_SETIMASK(s, s->imask & ~(cmd & SOC_CMD_REQ_QALL));
+
+ SOD(("Queues available %08x OUT %X %X\n", cmd, xram_get_8((xram_p)&s->req[0].hw_cq->out), xram_get_8((xram_p)&s->req[0].hw_cq->out)))
+ if (s->port[s->curr_port].fc.state != FC_STATE_OFFLINE) {
+ fcp_queue_empty ((fc_channel *)&(s->port[s->curr_port]));
+ if (((s->req[1].in + 1) & s->req[1].last) != (s->req[1].out))
+ fcp_queue_empty ((fc_channel *)&(s->port[1 - s->curr_port]));
+ } else
+ fcp_queue_empty ((fc_channel *)&(s->port[1 - s->curr_port]));
+ if (s->port[1 - s->curr_port].fc.state != FC_STATE_OFFLINE)
+ s->curr_port ^= 1;
+}
+
+static void inline soc_unsolicited (struct soc *s)
+{
+ soc_rsp *hwrsp, *hwrspc;
+ soc_cq *sw_cq;
+ int count;
+ int status;
+ int flags;
+ fc_channel *fc;
+
+ sw_cq = &s->rsp[SOC_UNSOLICITED_RSP_Q];
+ if (sw_cq->pool == NULL)
+ sw_cq->pool =
+ (soc_req *)(s->xram + (xram_get_32low ((xram_p)&sw_cq->hw_cq->address) / sizeof(u16)));
+
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ SOD (("soc_unsolicited, %d packets arrived\n", (sw_cq->in - sw_cq->out) & sw_cq->last))
+ while (sw_cq->in != sw_cq->out) {
+ /* ...real work per entry here... */
+ hwrsp = (soc_rsp *)sw_cq->pool + sw_cq->out;
+ hwrspc = NULL;
+ flags = hwrsp->shdr.flags;
+ count = xram_get_8 ((xram_p)&hwrsp->count);
+ fc = (fc_channel *)&s->port[flags & SOC_PORT_B];
+ SOD(("FC %08lx fcp_state_change %08lx\n", (long)fc, (long)fc->fcp_state_change))
+
+ if (count != 1) {
+ /* Ugh, continuation entries */
+ u8 in;
+
+ if (count != 2) {
+ printk("%s: Too many continuations entries %d\n", fc->name, count);
+ goto update_out;
+ }
+
+ in = sw_cq->in;
+ if (in < sw_cq->out) in += sw_cq->last + 1;
+ if (in < sw_cq->out + 2) {
+ /* Ask the hardware about it if they haven't arrived yet */
+ s->regs->cmd = (sw_cq->out << 24) | (SOC_CMD_RSP_QALL & ~(SOC_CMD_RSP_Q0 << SOC_UNSOLICITED_RSP_Q));
+ /* Read it, so that we're sure it has been updated */
+ s->regs->cmd;
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ in = sw_cq->in;
+ if (in < sw_cq->out) in += sw_cq->last + 1;
+ if (in < sw_cq->out + 2) /* Nothing came, let us wait */
+ return;
+ }
+ if (sw_cq->out == sw_cq->last)
+ hwrspc = (soc_rsp *)sw_cq->pool;
+ else
+ hwrspc = hwrsp + 1;
+ }
+
+ switch (flags & ~SOC_PORT_B) {
+ case SOC_STATUS:
+ status = xram_get_32low ((xram_p)&hwrsp->status);
+ switch (status) {
+ case SOC_ONLINE:
+ fc->fcp_state_change(fc, FC_STATE_ONLINE);
+ break;
+ case SOC_OFFLINE:
+ fc->fcp_state_change(fc, FC_STATE_OFFLINE);
+ break;
+ default:
+ printk ("%s: Unknown STATUS no %d\n", fc->name, status);
+ break;
+ }
+ break;
+ case (SOC_UNSOLICITED|SOC_FC_HDR):
+ {
+ int r_ctl = xram_get_8 ((xram_p)&hwrsp->fchdr);
+ unsigned len;
+ char buf[64];
+
+ if ((r_ctl & 0xf0) == R_CTL_EXTENDED_SVC) {
+ len = xram_get_32 ((xram_p)&hwrsp->shdr.bytecnt);
+ if (len < 4 || !hwrspc)
+ printk ("%s: Invalid R_CTL %02x continuation entries\n", fc->name, r_ctl);
+ else {
+ if (len > 60) len = 60;
+ xram_copy_from (buf, (xram_p)hwrspc, (len + 3) & ~3);
+ if (*(u32 *)buf == LS_DISPLAY) {
+ int i;
+
+ for (i = 4; i < len; i++)
+ if (buf[i] == '\n') buf[i] = ' ';
+ buf[len] = 0;
+ printk ("%s message: %s\n", fc->name, buf + 4);
+ } else
+ printk ("%s: Unknown LS_CMD %02x\n", fc->name, buf[0]);
+ }
+ } else
+ printk ("%s: Unsolicited R_CTL %02x not handled\n", fc->name, r_ctl);
+ }
+ break;
+ default:
+ printk ("%s: Unexpected flags %08x\n", fc->name, flags);
+ break;
+ }
+update_out:
+ if (++sw_cq->out > sw_cq->last) {
+ sw_cq->seqno++;
+ sw_cq->out = 0;
+ }
+
+ if (hwrspc) {
+ if (++sw_cq->out > sw_cq->last) {
+ sw_cq->seqno++;
+ sw_cq->out = 0;
+ }
+ }
+
+ if (sw_cq->out == sw_cq->in) {
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ if (sw_cq->out == sw_cq->in) {
+ /* Tell the hardware about it */
+ s->regs->cmd = (sw_cq->out << 24) | (SOC_CMD_RSP_QALL & ~(SOC_CMD_RSP_Q0 << SOC_UNSOLICITED_RSP_Q));
+ /* Read it, so that we're sure it has been updated */
+ s->regs->cmd;
+ sw_cq->in = xram_get_8 ((xram_p)&sw_cq->hw_cq->in);
+ }
+ }
+ }
+}
+
+static void soc_intr(int irq, void *dev_id, struct pt_regs *regs)
+{
+ u32 cmd;
+ register struct soc *s = (struct soc *)dev_id;
+
+ cmd = s->regs->cmd;
+ for (; (cmd = SOC_INTR (s, cmd)); cmd = s->regs->cmd) {
+ if (cmd & SOC_CMD_RSP_Q1) soc_unsolicited (s);
+ if (cmd & SOC_CMD_RSP_Q0) soc_solicited (s);
+ if (cmd & SOC_CMD_REQ_QALL) soc_request (s, cmd);
+ }
+}
+
+#define TOKEN(proto, port, token) (((proto)<<12)|(token)|(port))
+
+static int soc_hw_enque (fc_channel *fc, fcp_cmnd *fcmd)
+{
+ soc_port *port = (soc_port *)fc;
+ struct soc *s = port->s;
+ int qno;
+ soc_cq *sw_cq;
+ int cq_next_in;
+ soc_req *request;
+ fc_hdr *fch;
+ int i;
+
+ if (fcmd->proto == TYPE_SCSI_FCP)
+ qno = 1;
+ else
+ qno = 0;
+ SOD(("Putting a FCP packet type %d into hw queue %d\n", fcmd->proto, qno))
+ if (s->imask & (SOC_IMASK_REQ_Q0 << qno))
+ return -EIO;
+ sw_cq = s->req + qno;
+ cq_next_in = (sw_cq->in + 1) & sw_cq->last;
+
+ if (cq_next_in == sw_cq->out
+ && cq_next_in == (sw_cq->out = xram_get_8((xram_p)&sw_cq->hw_cq->out))) {
+ SOD(("%d IN %d OUT %d LAST %d\n", qno, sw_cq->in, sw_cq->out, sw_cq->last))
+ SOC_SETIMASK(s, s->imask | (SOC_IMASK_REQ_Q0 << qno));
+ return -EBUSY; // If queue full, just say NO
+ }
+
+ request = sw_cq->pool + sw_cq->in;
+ fch = &request->fchdr;
+
+ switch (fcmd->proto) {
+ case TYPE_SCSI_FCP:
+ request->shdr.token = TOKEN(TYPE_SCSI_FCP, port->mask, fcmd->token);
+ request->data[0].base = fc->dma_scsi_cmd + fcmd->token * sizeof(fcp_cmd);
+ request->data[0].count = sizeof(fcp_cmd);
+ request->data[1].base = fc->dma_scsi_rsp + fcmd->token * fc->rsp_size;
+ request->data[1].count = fc->rsp_size;
+ if (fcmd->data) {
+ request->shdr.segcnt = 3;
+ i = fc->scsi_cmd_pool[fcmd->token].fcp_data_len;
+ request->shdr.bytecnt = i;
+ request->data[2].base = fcmd->data;
+ request->data[2].count = i;
+ request->type = (fc->scsi_cmd_pool[fcmd->token].fcp_cntl & FCP_CNTL_WRITE) ?
+ SOC_CQTYPE_IO_WRITE : SOC_CQTYPE_IO_READ;
+ } else {
+ request->shdr.segcnt = 2;
+ request->shdr.bytecnt = 0;
+ request->data[2].base = 0;
+ request->data[2].count = 0;
+ request->type = SOC_CQTYPE_SIMPLE;
+ }
+ FILL_FCHDR_RCTL_DID(fch, R_CTL_COMMAND, fc->did);
+ FILL_FCHDR_SID(fch, fc->sid);
+ FILL_FCHDR_TYPE_FCTL(fch, TYPE_SCSI_FCP, F_CTL_FIRST_SEQ | F_CTL_SEQ_INITIATIVE);
+ FILL_FCHDR_SEQ_DF_SEQ(fch, 0, 0, 0);
+ FILL_FCHDR_OXRX(fch, 0xffff, 0xffff);
+ fch->param = 0;
+ request->shdr.flags = port->flags;
+ request->shdr.class = 2;
+ break;
+
+ case PROTO_OFFLINE:
+ memset (request, 0, sizeof(*request));
+ request->shdr.token = TOKEN(PROTO_OFFLINE, port->mask, fcmd->token);
+ request->type = SOC_CQTYPE_OFFLINE;
+ FILL_FCHDR_RCTL_DID(fch, R_CTL_COMMAND, fc->did);
+ FILL_FCHDR_SID(fch, fc->sid);
+ FILL_FCHDR_TYPE_FCTL(fch, TYPE_SCSI_FCP, F_CTL_FIRST_SEQ | F_CTL_SEQ_INITIATIVE);
+ FILL_FCHDR_SEQ_DF_SEQ(fch, 0, 0, 0);
+ FILL_FCHDR_OXRX(fch, 0xffff, 0xffff);
+ request->shdr.flags = port->flags;
+ break;
+
+ default:
+ request->shdr.token = TOKEN(fcmd->proto, port->mask, fcmd->token);
+ request->shdr.class = 2;
+ request->shdr.flags = port->flags;
+ memcpy (fch, &fcmd->fch, sizeof(fc_hdr));
+ request->data[0].count = fcmd->cmdlen;
+ request->data[1].count = fcmd->rsplen;
+ request->type = fcmd->class;
+ switch (fcmd->class) {
+ case FC_CLASS_OUTBOUND:
+ request->data[0].base = fcmd->cmd;
+ request->data[0].count = fcmd->cmdlen;
+ request->type = SOC_CQTYPE_OUTBOUND;
+ request->shdr.bytecnt = fcmd->cmdlen;
+ request->shdr.segcnt = 1;
+ break;
+ case FC_CLASS_INBOUND:
+ request->data[0].base = fcmd->rsp;
+ request->data[0].count = fcmd->rsplen;
+ request->type = SOC_CQTYPE_INBOUND;
+ request->shdr.bytecnt = 0;
+ request->shdr.segcnt = 1;
+ break;
+ case FC_CLASS_SIMPLE:
+ request->data[0].base = fcmd->cmd;
+ request->data[1].base = fcmd->rsp;
+ request->data[0].count = fcmd->cmdlen;
+ request->data[1].count = fcmd->rsplen;
+ request->type = SOC_CQTYPE_SIMPLE;
+ request->shdr.bytecnt = fcmd->cmdlen;
+ request->shdr.segcnt = 2;
+ break;
+ case FC_CLASS_IO_READ:
+ case FC_CLASS_IO_WRITE:
+ request->data[0].base = fcmd->cmd;
+ request->data[1].base = fcmd->rsp;
+ request->data[0].count = fcmd->cmdlen;
+ request->data[1].count = fcmd->rsplen;
+ request->type = (fcmd->class == FC_CLASS_IO_READ) ? SOC_CQTYPE_IO_READ : SOC_CQTYPE_IO_WRITE;
+ if (fcmd->data) {
+ request->data[2].base = fcmd->data;
+ request->data[2].count = fcmd->datalen;
+ request->shdr.bytecnt = fcmd->datalen;
+ request->shdr.segcnt = 3;
+ } else {
+ request->shdr.bytecnt = 0;
+ request->shdr.segcnt = 2;
+ }
+ break;
+ }
+ break;
+ }
+
+ request->count = 1;
+ request->flags = 0;
+ request->seqno = sw_cq->seqno;
+
+ /* And now tell the SOC about it */
+
+ if (++sw_cq->in > sw_cq->last) {
+ sw_cq->in = 0;
+ sw_cq->seqno++;
+ }
+
+ SOD(("Putting %08x into cmd\n", SOC_CMD_RSP_QALL | (sw_cq->in << 24) | (SOC_CMD_REQ_Q0 << qno)))
+
+ s->regs->cmd = SOC_CMD_RSP_QALL | (sw_cq->in << 24) | (SOC_CMD_REQ_Q0 << qno);
+ /* Read so that command is completed */
+ s->regs->cmd;
+
+ return 0;
+}
+
+static inline void soc_download_fw(struct soc *s)
+{
+#ifdef HAVE_SOC_UCODE
+ xram_copy_to (s->xram, soc_ucode, sizeof(soc_ucode));
+ xram_bzero (s->xram + (sizeof(soc_ucode)/sizeof(u16)), 32768 - sizeof(soc_ucode));
+#endif
+}
+
+/* Check for what the best SBUS burst we can use happens
+ * to be on this machine.
+ */
+static inline void soc_init_bursts(struct soc *s, struct linux_sbus_device *sdev)
+{
+ int bsizes, bsizes_more;
+
+ bsizes = (prom_getintdefault(sdev->prom_node,"burst-sizes",0xff) & 0xff);
+ bsizes_more = (prom_getintdefault(sdev->my_bus->prom_node, "burst-sizes", 0xff) & 0xff);
+ bsizes &= bsizes_more;
+ if ((bsizes & 0x7f) == 0x7f)
+ s->cfg = SOC_CFG_BURST_64;
+ else if ((bsizes & 0x3f) == 0x3f)
+ s->cfg = SOC_CFG_BURST_32;
+ else if ((bsizes & 0x1f) == 0x1f)
+ s->cfg = SOC_CFG_BURST_16;
+ else
+ s->cfg = SOC_CFG_BURST_4;
+}
+
+static inline void soc_init(struct linux_sbus_device *sdev, int no)
+{
+#ifdef __sparc_v9__
+ struct devid_cookie dcookie;
+#endif
+ unsigned char tmp[60];
+ int propl;
+ struct soc *s;
+ static unsigned version_printed = 0;
+ soc_hw_cq cq[8];
+ int size, i;
+ int irq;
+
+ s = kmalloc (sizeof (struct soc), GFP_KERNEL);
+ if (!s) return;
+ memset (s, 0, sizeof(struct soc));
+ s->soc_no = no;
+
+ SOD(("socs %08lx soc_intr %08lx soc_hw_enque %08x\n", (long)socs, (long)soc_intr, (long)soc_hw_enque))
+ if (version_printed++ == 0)
+ printk (version);
+
+ s->next = socs;
+ socs = s;
+ s->port[0].fc.dev = sdev;
+ s->port[1].fc.dev = sdev;
+ s->port[0].s = s;
+ s->port[1].s = s;
+
+ s->port[0].fc.next = &s->port[1].fc;
+
+ /* World Wide Name of SOC */
+ propl = prom_getproperty (sdev->prom_node, "soc-wwn", tmp, sizeof(tmp));
+ if (propl != sizeof (fc_wwn)) {
+ s->wwn.naaid = NAAID_IEEE;
+ s->wwn.lo = 0x12345678;
+ } else
+ memcpy (&s->wwn, tmp, sizeof (fc_wwn));
+
+ propl = prom_getproperty (sdev->prom_node, "port-wwns", tmp, sizeof(tmp));
+ if (propl != 2 * sizeof (fc_wwn)) {
+ s->port[0].fc.wwn_nport.naaid = NAAID_IEEE_EXT;
+ s->port[0].fc.wwn_nport.hi = s->wwn.hi;
+ s->port[0].fc.wwn_nport.lo = s->wwn.lo;
+ s->port[1].fc.wwn_nport.naaid = NAAID_IEEE_EXT;
+ s->port[1].fc.wwn_nport.nportid = 1;
+ s->port[1].fc.wwn_nport.hi = s->wwn.hi;
+ s->port[1].fc.wwn_nport.lo = s->wwn.lo;
+ } else {
+ memcpy (&s->port[0].fc.wwn_nport, tmp, sizeof (fc_wwn));
+ memcpy (&s->port[1].fc.wwn_nport, tmp + sizeof (fc_wwn), sizeof (fc_wwn));
+ }
+ memcpy (&s->port[0].fc.wwn_node, &s->wwn, sizeof (fc_wwn));
+ memcpy (&s->port[1].fc.wwn_node, &s->wwn, sizeof (fc_wwn));
+ SOD(("Got wwns %08x%08x ports %08x%08x and %08x%08x\n",
+ *(u32 *)&s->port[0].fc.wwn_nport, s->port[0].fc.wwn_nport.lo,
+ *(u32 *)&s->port[0].fc.wwn_nport, s->port[0].fc.wwn_nport.lo,
+ *(u32 *)&s->port[1].fc.wwn_nport, s->port[1].fc.wwn_nport.lo))
+
+ s->port[0].fc.sid = 1;
+ s->port[1].fc.sid = 17;
+ s->port[0].fc.did = 2;
+ s->port[1].fc.did = 18;
+
+ s->port[0].fc.reset = soc_reset;
+ s->port[1].fc.reset = soc_reset;
+
+ /* Setup the reg property for this device. */
+ prom_apply_sbus_ranges(sdev->my_bus, sdev->reg_addrs, sdev->num_registers, sdev);
+
+ if (sdev->num_registers == 1) {
+ /* Probably SunFire onboard SOC */
+ s->xram = (xram_p)
+ sparc_alloc_io (sdev->reg_addrs [0].phys_addr, 0,
+ sdev->reg_addrs [0].reg_size, "soc_xram",
+ sdev->reg_addrs [0].which_io, 0);
+ s->regs = (struct soc_regs *)((char *)s->xram + 0x10000);
+ } else {
+ /* Probably SOC sbus card */
+ s->xram = (xram_p)
+ sparc_alloc_io (sdev->reg_addrs [1].phys_addr, 0,
+ sdev->reg_addrs [1].reg_size, "soc_xram",
+ sdev->reg_addrs [1].which_io, 0);
+ s->regs = (struct soc_regs *)
+ sparc_alloc_io (sdev->reg_addrs [2].phys_addr, 0,
+ sdev->reg_addrs [2].reg_size, "soc_regs",
+ sdev->reg_addrs [2].which_io, 0);
+ }
+
+ soc_init_bursts(s, sdev);
+
+ SOD(("Disabling SOC\n"))
+
+ soc_disable (s);
+
+ irq = sdev->irqs[0].pri;
+
+#ifndef __sparc_v9__
+ if (request_irq (irq, soc_intr, SA_SHIRQ, "SOC", (void *)s)) {
+ soc_printk ("Cannot order irq %d to go\n", irq);
+ return;
+ }
+#else
+ dcookie.real_dev_id = s;
+ dcookie.imap = dcookie.iclr = 0;
+ dcookie.pil = -1;
+ dcookie.bus_cookie = sdev->my_bus;
+ if (request_irq (irq, soc_intr, (SA_SHIRQ | SA_SBUS | SA_DCOOKIE), "SOC", &dcookie)) {
+ soc_printk ("Cannot order irq %d to go\n", irq);
+ return;
+ }
+ irq = dcookie.ret_ino;
+#endif
+
+ SOD(("SOC uses IRQ%d\n", irq))
+
+ s->port[0].fc.irq = irq;
+ s->port[1].fc.irq = irq;
+
+ sprintf (s->port[0].fc.name, "soc%d port A", no);
+ sprintf (s->port[1].fc.name, "soc%d port B", no);
+ s->port[0].flags = SOC_FC_HDR | SOC_PORT_A;
+ s->port[1].flags = SOC_FC_HDR | SOC_PORT_B;
+ s->port[1].mask = (1 << 11);
+
+ s->port[0].fc.hw_enque = soc_hw_enque;
+ s->port[1].fc.hw_enque = soc_hw_enque;
+
+ soc_download_fw (s);
+
+ SOD(("Downloaded firmware\n"))
+
+ /* Now setup xram circular queues */
+ memset (cq, 0, sizeof(cq));
+
+ size = (SOC_CQ_REQ0_SIZE + SOC_CQ_REQ1_SIZE) * sizeof(soc_req);
+ s->req[0].pool = (soc_req *) sparc_dvma_malloc (size, "SOC request queues", &cq[0].address);
+ s->req[1].pool = s->req[0].pool + SOC_CQ_REQ0_SIZE;
+
+ s->req[0].hw_cq = (soc_hw_cq *)(s->xram + SOC_CQ_REQ_OFFSET);
+ s->req[1].hw_cq = (soc_hw_cq *)(s->xram + SOC_CQ_REQ_OFFSET + sizeof(soc_hw_cq) / sizeof(u16));
+ s->rsp[0].hw_cq = (soc_hw_cq *)(s->xram + SOC_CQ_RSP_OFFSET);
+ s->rsp[1].hw_cq = (soc_hw_cq *)(s->xram + SOC_CQ_RSP_OFFSET + sizeof(soc_hw_cq) / sizeof(u16));
+
+ cq[1].address = cq[0].address + (SOC_CQ_REQ0_SIZE * sizeof(soc_req));
+ cq[4].address = 1;
+ cq[5].address = 1;
+ cq[0].last = SOC_CQ_REQ0_SIZE - 1;
+ cq[1].last = SOC_CQ_REQ1_SIZE - 1;
+ cq[4].last = SOC_CQ_RSP0_SIZE - 1;
+ cq[5].last = SOC_CQ_RSP1_SIZE - 1;
+ for (i = 0; i < 8; i++)
+ cq[i].seqno = 1;
+
+ s->req[0].last = SOC_CQ_REQ0_SIZE - 1;
+ s->req[1].last = SOC_CQ_REQ1_SIZE - 1;
+ s->rsp[0].last = SOC_CQ_RSP0_SIZE - 1;
+ s->rsp[1].last = SOC_CQ_RSP1_SIZE - 1;
+
+ s->req[0].seqno = 1;
+ s->req[1].seqno = 1;
+ s->rsp[0].seqno = 1;
+ s->rsp[1].seqno = 1;
+
+ xram_copy_to (s->xram + SOC_CQ_REQ_OFFSET, cq, sizeof(cq));
+
+ /* Make our sw copy of SOC service parameters */
+ xram_copy_from (s->serv_params, s->xram + 0x140, sizeof (s->serv_params));
+
+ s->port[0].fc.common_svc = (common_svc_parm *)s->serv_params;
+ s->port[0].fc.class_svcs = (svc_parm *)(s->serv_params + 0x20);
+ s->port[1].fc.common_svc = (common_svc_parm *)&s->serv_params;
+ s->port[1].fc.class_svcs = (svc_parm *)(s->serv_params + 0x20);
+
+ soc_enable (s);
+
+ SOD(("Enabled SOC\n"))
+}
+
+#ifndef MODULE
+__initfunc(int soc_probe(void))
+#else
+int init_module(void)
+#endif
+{
+ struct linux_sbus *bus;
+ struct linux_sbus_device *sdev = 0;
+ struct soc *s;
+ int cards = 0;
+
+ for_each_sbus(bus) {
+ for_each_sbusdev(sdev, bus) {
+ if(!strcmp(sdev->prom_name, "SUNW,soc")) {
+ soc_init(sdev, cards);
+ cards++;
+ }
+ }
+ }
+ if (!cards) return -EIO;
+
+ for_each_soc(s)
+ if (s->next)
+ s->port[1].fc.next = &s->next->port[0].fc;
+ fcp_init (&socs->port[0].fc);
+ return 0;
+}
+
+EXPORT_NO_SYMBOLS;
+
+#ifdef MODULE
+void cleanup_module(void)
+{
+ struct soc *s;
+ int irq;
+ struct linux_sbus_device *sdev;
+
+ for_each_soc(s) {
+ /* FIXME: Tell FC layer we say good bye */
+ irq = s->port[0].fc.irq;
+ disable_irq (irq);
+ free_irq (irq, s);
+ sdev = s->port[0].fc.dev;
+ if (sdev->num_registers == 1)
+ sparc_free_io ((char *)s->xram, sdev->reg_addrs [0].reg_size);
+ else {
+ sparc_free_io ((char *)s->xram, sdev->reg_addrs [1].reg_size);
+ sparc_free_io ((char *)s->regs, sdev->reg_addrs [2].reg_size);
+ }
+ /* FIXME: sparc_dvma_free() ??? */
+ }
+}
+#endif
FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen, slshen@lbl.gov