patch-2.3.14 linux/drivers/scsi/oktagon_esp.c

Next file: linux/drivers/scsi/oktagon_esp.h
Previous file: linux/drivers/scsi/ncr53c8xx.c
Back to the patch index
Back to the overall index

diff -u --recursive --new-file v2.3.13/linux/drivers/scsi/oktagon_esp.c linux/drivers/scsi/oktagon_esp.c
@@ -0,0 +1,597 @@
+/*
+ * Oktagon_esp.c -- Driver for bsc Oktagon
+ *
+ * Written by Carsten Pluntke 1998
+ *
+ * Based on cyber_esp.c
+ */
+
+#include <linux/config.h>
+
+#if defined(CONFIG_AMIGA) || defined(CONFIG_APUS)
+#define USE_BOTTOM_HALF
+#endif
+
+#define __KERNEL_SYSCALLS__
+#include <linux/module.h>
+
+#include <linux/kernel.h>
+#include <linux/delay.h>
+#include <linux/types.h>
+#include <linux/string.h>
+#include <linux/malloc.h>
+#include <linux/blk.h>
+#include <linux/proc_fs.h>
+#include <linux/stat.h>
+#include <linux/reboot.h>
+#include <asm/system.h>
+#include <asm/ptrace.h>
+#include <asm/pgtable.h>
+#include <asm/io.h>
+
+#include "scsi.h"
+#include "hosts.h"
+#include "NCR53C9x.h"
+#include "oktagon_esp.h"
+
+#include <linux/zorro.h>
+#include <asm/irq.h>
+#include <asm/amigaints.h>
+#include <asm/amigahw.h>
+
+#ifdef USE_BOTTOM_HALF
+#include <linux/tqueue.h>
+#include <linux/interrupt.h>
+#endif
+
+#include <linux/unistd.h>
+
+static int  dma_bytes_sent(struct NCR_ESP *esp, int fifo_count);
+static int  dma_can_transfer(struct NCR_ESP *esp, Scsi_Cmnd *sp);
+static void dma_dump_state(struct NCR_ESP *esp);
+static void dma_init_read(struct NCR_ESP *esp, __u32 vaddress, int length);
+static void dma_init_write(struct NCR_ESP *esp, __u32 vaddress, int length);
+static void dma_ints_off(struct NCR_ESP *esp);
+static void dma_ints_on(struct NCR_ESP *esp);
+static int  dma_irq_p(struct NCR_ESP *esp);
+static void dma_led_off(struct NCR_ESP *esp);
+static void dma_led_on(struct NCR_ESP *esp);
+static int  dma_ports_p(struct NCR_ESP *esp);
+static void dma_setup(struct NCR_ESP *esp, __u32 addr, int count, int write);
+
+static void dma_irq_exit(struct NCR_ESP *esp);
+static void dma_invalidate(struct NCR_ESP *esp);
+
+static void dma_mmu_get_scsi_one(struct NCR_ESP *,Scsi_Cmnd *);
+static void dma_mmu_get_scsi_sgl(struct NCR_ESP *,Scsi_Cmnd *);
+static void dma_mmu_release_scsi_one(struct NCR_ESP *,Scsi_Cmnd *);
+static void dma_mmu_release_scsi_sgl(struct NCR_ESP *,Scsi_Cmnd *);
+static void dma_advance_sg(Scsi_Cmnd *);
+static int  oktagon_notify_reboot(struct notifier_block *this, unsigned long code, void *x);
+
+void esp_bootup_reset(struct NCR_ESP *esp,struct ESP_regs *eregs);
+
+#ifdef USE_BOTTOM_HALF
+static void dma_commit(void *opaque);
+
+long oktag_to_io(long *paddr, long *addr, long len);
+long oktag_from_io(long *addr, long *paddr, long len);
+
+static struct tq_struct tq_fake_dma = { NULL, 0, dma_commit, NULL };
+
+#define DMA_MAXTRANSFER 0x8000
+
+#else
+
+/*
+ * No bottom half. Use transfer directly from IRQ. Find a narrow path
+ * between too much IRQ overhead and clogging the IRQ for too long.
+ */
+
+#define DMA_MAXTRANSFER 0x1000
+
+#endif
+
+static struct notifier_block oktagon_notifier = { 
+	oktagon_notify_reboot,
+	NULL,
+	0
+};
+
+static long *paddress;
+static long *address;
+static long len;
+static long dma_on;
+static int direction;
+static struct NCR_ESP *current_esp;
+
+
+volatile unsigned char cmd_buffer[16];
+				/* This is where all commands are put
+				 * before they are trasfered to the ESP chip
+				 * via PIO.
+				 */
+
+/***************************************************************** Detection */
+int oktagon_esp_detect(Scsi_Host_Template *tpnt)
+{
+	struct NCR_ESP *esp;
+	const struct ConfigDev *esp_dev;
+	int key;
+	unsigned long address;
+	struct ESP_regs *eregs;
+
+	if((key = zorro_find(ZORRO_PROD_BSC_OKTAGON_2008, 0, 0))){
+		esp_dev = zorro_get_board(key);
+
+		/*
+		 * It is a SCSI controller.
+		 * Hardwire Host adapter to SCSI ID 7
+		 */
+		
+		address = (unsigned long)ZTWO_VADDR(esp_dev->cd_BoardAddr);
+		eregs = (struct ESP_regs *)(address + OKTAGON_ESP_ADDR);
+
+		/* This line was 5 lines lower */
+		esp = esp_allocate(tpnt, (void *) esp_dev);
+
+		/* we have to shift the registers only one bit for oktagon */
+		esp->shift = 1;
+
+		esp_write(eregs->esp_cfg1, (ESP_CONFIG1_PENABLE | 7));
+		udelay(5);
+		if (esp_read(eregs->esp_cfg1) != (ESP_CONFIG1_PENABLE | 7))
+			return 0; /* Bail out if address did not hold data */
+
+		/* Do command transfer with programmed I/O */
+		esp->do_pio_cmds = 1;
+
+		/* Required functions */
+		esp->dma_bytes_sent = &dma_bytes_sent;
+		esp->dma_can_transfer = &dma_can_transfer;
+		esp->dma_dump_state = &dma_dump_state;
+		esp->dma_init_read = &dma_init_read;
+		esp->dma_init_write = &dma_init_write;
+		esp->dma_ints_off = &dma_ints_off;
+		esp->dma_ints_on = &dma_ints_on;
+		esp->dma_irq_p = &dma_irq_p;
+		esp->dma_ports_p = &dma_ports_p;
+		esp->dma_setup = &dma_setup;
+
+		/* Optional functions */
+		esp->dma_barrier = 0;
+		esp->dma_drain = 0;
+		esp->dma_invalidate = &dma_invalidate;
+		esp->dma_irq_entry = 0;
+		esp->dma_irq_exit = &dma_irq_exit;
+		esp->dma_led_on = &dma_led_on;
+		esp->dma_led_off = &dma_led_off;
+		esp->dma_poll = 0;
+		esp->dma_reset = 0;
+
+		esp->dma_mmu_get_scsi_one = &dma_mmu_get_scsi_one;
+		esp->dma_mmu_get_scsi_sgl = &dma_mmu_get_scsi_sgl;
+		esp->dma_mmu_release_scsi_one = &dma_mmu_release_scsi_one;
+		esp->dma_mmu_release_scsi_sgl = &dma_mmu_release_scsi_sgl;
+		esp->dma_advance_sg = &dma_advance_sg;
+
+		/* SCSI chip speed */
+		/* Looking at the quartz of the SCSI board... */
+		esp->cfreq = 25000000;
+
+		/* The DMA registers on the CyberStorm are mapped
+		 * relative to the device (i.e. in the same Zorro
+		 * I/O block).
+		 */
+		esp->dregs = (void *)(address + OKTAGON_DMA_ADDR);
+
+		paddress = (long *) esp->dregs;
+
+		/* ESP register base */
+		esp->eregs = eregs;
+		
+		/* Set the command buffer */
+		esp->esp_command = (volatile unsigned char*) cmd_buffer;
+
+		/* Yes, the virtual address. See below. */
+		esp->esp_command_dvma = (__u32) cmd_buffer;
+
+		esp->irq = IRQ_AMIGA_PORTS;
+		esp->slot = key;
+		request_irq(IRQ_AMIGA_PORTS, esp_intr, SA_SHIRQ,
+			    "BSC Oktagon SCSI", esp_intr);
+
+		/* Figure out our scsi ID on the bus */
+		esp->scsi_id = 7;
+		
+		/* We don't have a differential SCSI-bus. */
+		esp->diff = 0;
+
+		esp_initialize(esp);
+
+		zorro_config_board(key, 0);
+		printk("ESP_Oktagon Driver 1.1"
+#ifdef USE_BOTTOM_HALF
+		       " [BOTTOM_HALF]"
+#else
+		       " [IRQ]"
+#endif
+		       " registered.\n");
+		printk("ESP: Total of %d ESP hosts found, %d actually in use.\n", nesps,esps_in_use);
+		esps_running = esps_in_use;
+		current_esp = esp;
+		register_reboot_notifier(&oktagon_notifier);
+		return esps_in_use;
+	}
+	return 0;
+}
+
+
+/*
+ * On certain configurations the SCSI equipment gets confused on reboot,
+ * so we have to reset it then.
+ */
+
+static int
+oktagon_notify_reboot(struct notifier_block *this, unsigned long code, void *x)
+{
+  struct NCR_ESP *esp;
+  
+  if((code == SYS_DOWN || code == SYS_HALT) && (esp = current_esp))
+   {
+    esp_bootup_reset(esp,esp->eregs);
+    udelay(500); /* Settle time. Maybe unneccessary. */
+   }
+  return NOTIFY_DONE;
+}
+    
+
+	
+#ifdef USE_BOTTOM_HALF
+
+
+/*
+ * The bsc Oktagon controller has no real DMA, so we have to do the 'DMA
+ * transfer' in the interrupt (Yikes!) or use a bottom half to not to clutter
+ * IRQ's for longer-than-good.
+ *
+ * FIXME
+ * BIG PROBLEM: 'len' is usually the buffer length, not the expected length
+ * of the data. So DMA may finish prematurely, further reads lead to
+ * 'machine check' on APUS systems (don't know about m68k systems, AmigaOS
+ * deliberately ignores the bus faults) and a normal copy-loop can't
+ * be exited prematurely just at the right moment by the dma_invalidate IRQ.
+ * So do it the hard way, write an own copier in assembler and
+ * catch the exception.
+ *                                     -- Carsten
+ */
+ 
+ 
+static void dma_commit(void *opaque)
+{
+    long wait,len2,pos;
+    struct NCR_ESP *esp;
+
+    ESPDATA(("Transfer: %ld bytes, Address 0x%08lX, Direction: %d\n",
+         len,(long) address,direction));
+    dma_ints_off(current_esp);
+
+    pos = 0;
+    wait = 1;
+    if(direction) /* write? (memory to device) */
+     {
+      while(len > 0)
+       {
+        len2 = oktag_to_io(paddress, address+pos, len);
+	if(!len2)
+	 {
+	  if(wait > 1000)
+	   {
+	    printk("Expedited DMA exit (writing) %ld\n",len);
+	    break;
+	   }
+	  mdelay(wait);
+	  wait *= 2;
+	 }
+	pos += len2;
+	len -= len2*sizeof(long);
+       }
+     } else {
+      while(len > 0)
+       {
+        len2 = oktag_from_io(address+pos, paddress, len);
+	if(!len2)
+	 {
+	  if(wait > 1000)
+	   {
+	    printk("Expedited DMA exit (reading) %ld\n",len);
+	    break;
+	   }
+	  mdelay(wait);
+	  wait *= 2;
+	 }
+	pos += len2;
+	len -= len2*sizeof(long);
+       }
+     }
+
+    /* to make esp->shift work */
+    esp=current_esp;
+
+#if 0
+    len2 = (esp_read(current_esp->eregs->esp_tclow) & 0xff) |
+           ((esp_read(current_esp->eregs->esp_tcmed) & 0xff) << 8);
+
+    /*
+     * Uh uh. If you see this, len and transfer count registers were out of
+     * sync. That means really serious trouble.
+     */
+
+    if(len2)
+      printk("Eeeek!! Transfer count still %ld!\n",len2);
+#endif
+
+    /*
+     * Normally we just need to exit and wait for the interrupt to come.
+     * But at least one device (my Microtek ScanMaker 630) regularly mis-
+     * calculates the bytes it should send which is really ugly because
+     * it locks up the SCSI bus if not accounted for.
+     */
+
+    if(!(esp_read(current_esp->eregs->esp_status) & ESP_STAT_INTR))
+     {
+      long len = 100;
+      long trash[10];
+
+      /*
+       * Interrupt bit was not set. Either the device is just plain lazy
+       * so we give it a 10 ms chance or...
+       */
+      while(len-- && (!(esp_read(current_esp->eregs->esp_status) & ESP_STAT_INTR)))
+        udelay(100);
+
+
+      if(!(esp_read(current_esp->eregs->esp_status) & ESP_STAT_INTR))
+       {
+        /*
+	 * So we think that the transfer count is out of sync. Since we
+	 * have all we want we are happy and can ditch the trash.
+	 */
+	 
+        len = DMA_MAXTRANSFER;
+
+        while(len-- && (!(esp_read(current_esp->eregs->esp_status) & ESP_STAT_INTR)))
+          oktag_from_io(trash,paddress,2);
+
+        if(!(esp_read(current_esp->eregs->esp_status) & ESP_STAT_INTR))
+         {
+          /*
+           * Things really have gone wrong. If we leave the system in that
+           * state, the SCSI bus is locked forever. I hope that this will
+           * turn the system in a more or less running state.
+           */
+          printk("Device is bolixed, trying bus reset...\n");
+	  esp_bootup_reset(current_esp,current_esp->eregs);
+         }
+       }
+     }
+
+    ESPDATA(("Transfer_finale: do_data_finale should come\n"));
+
+    len = 0;
+    dma_on = 0;
+    dma_ints_on(current_esp);
+}
+
+#endif
+
+/************************************************************* DMA Functions */
+static int dma_bytes_sent(struct NCR_ESP *esp, int fifo_count)
+{
+	/* Since the CyberStorm DMA is fully dedicated to the ESP chip,
+	 * the number of bytes sent (to the ESP chip) equals the number
+	 * of bytes in the FIFO - there is no buffering in the DMA controller.
+	 * XXXX Do I read this right? It is from host to ESP, right?
+	 */
+	return fifo_count;
+}
+
+static int dma_can_transfer(struct NCR_ESP *esp, Scsi_Cmnd *sp)
+{
+	unsigned long sz = sp->SCp.this_residual;
+	if(sz > DMA_MAXTRANSFER)
+		sz = DMA_MAXTRANSFER;
+	return sz;
+}
+
+static void dma_dump_state(struct NCR_ESP *esp)
+{
+}
+
+/*
+ * What the f$@& is this?
+ *
+ * Some SCSI devices (like my Microtek ScanMaker 630 scanner) want to transfer
+ * more data than requested. How much? Dunno. So ditch the bogus data into
+ * the sink, hoping the device will advance to the next phase sooner or later.
+ *
+ *                         -- Carsten
+ */
+
+static long oktag_eva_buffer[16]; /* The data sink */
+
+static void oktag_check_dma(void)
+{
+  struct NCR_ESP *esp;
+
+  esp=current_esp;
+  if(!len)
+   {
+    address = oktag_eva_buffer;
+    len = 2;
+    /* esp_do_data sets them to zero like len */
+    esp_write(current_esp->eregs->esp_tclow,2);
+    esp_write(current_esp->eregs->esp_tcmed,0);
+   }
+}
+
+static void dma_init_read(struct NCR_ESP *esp, __u32 vaddress, int length)
+{
+	/* Zorro is noncached, everything else done using processor. */
+	/* cache_clear(addr, length); */
+	
+	if(dma_on)
+	  panic("dma_init_read while dma process is initialized/running!\n");
+	direction = 0;
+	address = (long *) vaddress;
+	current_esp = esp;
+	len = length;
+	oktag_check_dma();
+        dma_on = 1;
+}
+
+static void dma_init_write(struct NCR_ESP *esp, __u32 vaddress, int length)
+{
+	/* cache_push(addr, length); */
+
+	if(dma_on)
+	  panic("dma_init_write while dma process is initialized/running!\n");
+	direction = 1;
+	address = (long *) vaddress;
+	current_esp = esp;
+	len = length;
+	oktag_check_dma();
+	dma_on = 1;
+}
+
+static void dma_ints_off(struct NCR_ESP *esp)
+{
+	disable_irq(esp->irq);
+}
+
+static void dma_ints_on(struct NCR_ESP *esp)
+{
+	enable_irq(esp->irq);
+}
+
+static int dma_irq_p(struct NCR_ESP *esp)
+{
+	/* It's important to check the DMA IRQ bit in the correct way! */
+	return (esp_read(esp->eregs->esp_status) & ESP_STAT_INTR);
+}
+
+static void dma_led_off(struct NCR_ESP *esp)
+{
+}
+
+static void dma_led_on(struct NCR_ESP *esp)
+{
+}
+
+static int dma_ports_p(struct NCR_ESP *esp)
+{
+	return ((custom.intenar) & IF_PORTS);
+}
+
+static void dma_setup(struct NCR_ESP *esp, __u32 addr, int count, int write)
+{
+	/* On the Sparc, DMA_ST_WRITE means "move data from device to memory"
+	 * so when (write) is true, it actually means READ!
+	 */
+	if(write){
+		dma_init_read(esp, addr, count);
+	} else {
+		dma_init_write(esp, addr, count);
+	}
+}
+
+/*
+ * IRQ entry when DMA transfer is ready to be started
+ */
+
+static void dma_irq_exit(struct NCR_ESP *esp)
+{
+#ifdef USE_BOTTOM_HALF
+	if(dma_on)
+	 {
+	  tq_fake_dma.sync = 0;
+	  queue_task(&tq_fake_dma,&tq_immediate);
+	  mark_bh(IMMEDIATE_BH);
+	 }
+#else
+	while(len && !dma_irq_p(esp))
+	 {
+	  if(direction)
+	    *paddress = *address++;
+	   else
+	    *address++ = *paddress;
+	  len -= (sizeof(long));
+	 }
+	len = 0;
+        dma_on = 0;
+#endif
+}
+
+/*
+ * IRQ entry when DMA has just finished
+ */
+
+static void dma_invalidate(struct NCR_ESP *esp)
+{
+}
+
+/*
+ * Since the processor does the data transfer we have to use the custom
+ * mmu interface to pass the virtual address, not the physical.
+ */
+
+void dma_mmu_get_scsi_one(struct NCR_ESP *esp, Scsi_Cmnd *sp)
+{
+        sp->SCp.have_data_in = (int) sp->SCp.ptr =
+                sp->request_buffer;
+}
+
+void dma_mmu_get_scsi_sgl(struct NCR_ESP *esp, Scsi_Cmnd *sp)
+{
+        sp->SCp.ptr = 
+                sp->SCp.buffer->address;
+}
+
+void dma_mmu_release_scsi_one(struct NCR_ESP *esp, Scsi_Cmnd *sp)
+{
+}
+
+void dma_mmu_release_scsi_sgl(struct NCR_ESP *esp, Scsi_Cmnd *sp)
+{
+}
+
+void dma_advance_sg(Scsi_Cmnd *sp)
+{
+  sp->SCp.ptr = sp->SCp.buffer->address;
+}
+
+#ifdef MODULE
+
+#define HOSTS_C
+
+#include "oktagon_esp.h"
+
+Scsi_Host_Template driver_template = SCSI_OKTAGON_ESP;
+
+#include "scsi_module.c"
+
+#endif
+
+int oktagon_esp_release(struct Scsi_Host *instance)
+{
+#ifdef MODULE
+	unsigned int key;
+
+	key = ((struct NCR_ESP *)instance->hostdata)->slot;
+	esp_release();
+	zorro_unconfig_board(key, 0);
+	free_irq(IRQ_AMIGA_PORTS, esp_intr);
+	unregister_reboot_notifier(&oktagon_notifier);
+#endif
+	return 1;
+}

FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)