Linux-2.6.12-rc2
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
diff --git a/drivers/scsi/ultrastor.c b/drivers/scsi/ultrastor.c
new file mode 100644
index 0000000..7484916
--- /dev/null
+++ b/drivers/scsi/ultrastor.c
@@ -0,0 +1,1204 @@
+/*
+ * ultrastor.c Copyright (C) 1992 David B. Gentzel
+ * Low-level SCSI driver for UltraStor 14F, 24F, and 34F
+ * by David B. Gentzel, Whitfield Software Services, Carnegie, PA
+ * (gentzel@nova.enet.dec.com)
+ * scatter/gather added by Scott Taylor (n217cg@tamuts.tamu.edu)
+ * 24F and multiple command support by John F. Carr (jfc@athena.mit.edu)
+ * John's work modified by Caleb Epstein (cae@jpmorgan.com) and
+ * Eric Youngdale (ericy@cais.com).
+ * Thanks to UltraStor for providing the necessary documentation
+ *
+ * This is an old driver, for the 14F and 34F you should be using the
+ * u14-34f driver instead.
+ */
+
+/*
+ * TODO:
+ * 1. Find out why scatter/gather is limited to 16 requests per command.
+ * This is fixed, at least on the 24F, as of version 1.12 - CAE.
+ * 2. Look at command linking (mscp.command_link and
+ * mscp.command_link_id). (Does not work with many disks,
+ * and no performance increase. ERY).
+ * 3. Allow multiple adapters.
+ */
+
+/*
+ * NOTES:
+ * The UltraStor 14F, 24F, and 34F are a family of intelligent, high
+ * performance SCSI-2 host adapters. They all support command queueing
+ * and scatter/gather I/O. Some of them can also emulate the standard
+ * WD1003 interface for use with OS's which don't support SCSI. Here
+ * is the scoop on the various models:
+ * 14F - ISA first-party DMA HA with floppy support and WD1003 emulation.
+ * 14N - ISA HA with floppy support. I think that this is a non-DMA
+ * HA. Nothing further known.
+ * 24F - EISA Bus Master HA with floppy support and WD1003 emulation.
+ * 34F - VL-Bus Bus Master HA with floppy support (no WD1003 emulation).
+ *
+ * The 14F, 24F, and 34F are supported by this driver.
+ *
+ * Places flagged with a triple question-mark are things which are either
+ * unfinished, questionable, or wrong.
+ */
+
+/* Changes from version 1.11 alpha to 1.12
+ *
+ * Increased the size of the scatter-gather list to 33 entries for
+ * the 24F adapter (it was 16). I don't have the specs for the 14F
+ * or the 34F, so they may support larger s-g lists as well.
+ *
+ * Caleb Epstein <cae@jpmorgan.com>
+ */
+
+/* Changes from version 1.9 to 1.11
+ *
+ * Patches to bring this driver up to speed with the default kernel
+ * driver which supports only the 14F and 34F adapters. This version
+ * should compile cleanly into 0.99.13, 0.99.12 and probably 0.99.11.
+ *
+ * Fixes from Eric Youngdale to fix a few possible race conditions and
+ * several problems with bit testing operations (insufficient
+ * parentheses).
+ *
+ * Removed the ultrastor_abort() and ultrastor_reset() functions
+ * (enclosed them in #if 0 / #endif). These functions, at least on
+ * the 24F, cause the SCSI bus to do odd things and generally lead to
+ * kernel panics and machine hangs. This is like the Adaptec code.
+ *
+ * Use check/snarf_region for 14f, 34f to avoid I/O space address conflicts.
+ */
+
+/* Changes from version 1.8 to version 1.9
+ *
+ * 0.99.11 patches (cae@jpmorgan.com) */
+
+/* Changes from version 1.7 to version 1.8
+ *
+ * Better error reporting.
+ */
+
+/* Changes from version 1.6 to version 1.7
+ *
+ * Removed CSIR command code.
+ *
+ * Better race condition avoidance (xchgb function added).
+ *
+ * Set ICM and OGM status to zero at probe (24F)
+ *
+ * reset sends soft reset to UltraStor adapter
+ *
+ * reset adapter if adapter interrupts with an invalid MSCP address
+ *
+ * handle aborted command interrupt (24F)
+ *
+ */
+
+/* Changes from version 1.5 to version 1.6:
+ *
+ * Read MSCP address from ICM _before_ clearing the interrupt flag.
+ * This fixes a race condition.
+ */
+
+/* Changes from version 1.4 to version 1.5:
+ *
+ * Abort now calls done when multiple commands are enabled.
+ *
+ * Clear busy when aborted command finishes, not when abort is called.
+ *
+ * More debugging messages for aborts.
+ */
+
+/* Changes from version 1.3 to version 1.4:
+ *
+ * Enable automatic request of sense data on error (requires newer version
+ * of scsi.c to be useful).
+ *
+ * Fix PORT_OVERRIDE for 14F.
+ *
+ * Fix abort and reset to work properly (config.aborted wasn't cleared
+ * after it was tested, so after a command abort no further commands would
+ * work).
+ *
+ * Boot time test to enable SCSI bus reset (defaults to not allowing reset).
+ *
+ * Fix test for OGM busy -- the busy bit is in different places on the 24F.
+ *
+ * Release ICM slot by clearing first byte on 24F.
+ */
+
+#include <linux/module.h>
+#include <linux/blkdev.h>
+#include <linux/interrupt.h>
+#include <linux/stddef.h>
+#include <linux/string.h>
+#include <linux/kernel.h>
+#include <linux/ioport.h>
+#include <linux/proc_fs.h>
+#include <linux/spinlock.h>
+#include <linux/stat.h>
+#include <linux/bitops.h>
+
+#include <asm/io.h>
+#include <asm/system.h>
+#include <asm/dma.h>
+
+#define ULTRASTOR_PRIVATE /* Get the private stuff from ultrastor.h */
+#include "scsi.h"
+#include <scsi/scsi_host.h>
+#include "ultrastor.h"
+
+#define FALSE 0
+#define TRUE 1
+
+#ifndef ULTRASTOR_DEBUG
+#define ULTRASTOR_DEBUG (UD_ABORT|UD_CSIR|UD_RESET)
+#endif
+
+#define VERSION "1.12"
+
+#define PACKED __attribute__((packed))
+#define ALIGNED(x) __attribute__((aligned(x)))
+
+
+/* The 14F uses an array of 4-byte ints for its scatter/gather list.
+ The data can be unaligned, but need not be. It's easier to give
+ the list normal alignment since it doesn't need to fit into a
+ packed structure. */
+
+typedef struct {
+ u32 address;
+ u32 num_bytes;
+} ultrastor_sg_list;
+
+
+/* MailBox SCSI Command Packet. Basic command structure for communicating
+ with controller. */
+struct mscp {
+ unsigned char opcode: 3; /* type of command */
+ unsigned char xdir: 2; /* data transfer direction */
+ unsigned char dcn: 1; /* disable disconnect */
+ unsigned char ca: 1; /* use cache (if available) */
+ unsigned char sg: 1; /* scatter/gather operation */
+ unsigned char target_id: 3; /* target SCSI id */
+ unsigned char ch_no: 2; /* SCSI channel (always 0 for 14f) */
+ unsigned char lun: 3; /* logical unit number */
+ unsigned int transfer_data PACKED; /* transfer data pointer */
+ unsigned int transfer_data_length PACKED; /* length in bytes */
+ unsigned int command_link PACKED; /* for linking command chains */
+ unsigned char scsi_command_link_id; /* identifies command in chain */
+ unsigned char number_of_sg_list; /* (if sg is set) 8 bytes per list */
+ unsigned char length_of_sense_byte;
+ unsigned char length_of_scsi_cdbs; /* 6, 10, or 12 */
+ unsigned char scsi_cdbs[12]; /* SCSI commands */
+ unsigned char adapter_status; /* non-zero indicates HA error */
+ unsigned char target_status; /* non-zero indicates target error */
+ u32 sense_data PACKED;
+ /* The following fields are for software only. They are included in
+ the MSCP structure because they are associated with SCSI requests. */
+ void (*done)(Scsi_Cmnd *);
+ Scsi_Cmnd *SCint;
+ ultrastor_sg_list sglist[ULTRASTOR_24F_MAX_SG]; /* use larger size for 24F */
+};
+
+
+/* Port addresses (relative to the base address) */
+#define U14F_PRODUCT_ID(port) ((port) + 0x4)
+#define CONFIG(port) ((port) + 0x6)
+
+/* Port addresses relative to the doorbell base address. */
+#define LCL_DOORBELL_MASK(port) ((port) + 0x0)
+#define LCL_DOORBELL_INTR(port) ((port) + 0x1)
+#define SYS_DOORBELL_MASK(port) ((port) + 0x2)
+#define SYS_DOORBELL_INTR(port) ((port) + 0x3)
+
+
+/* Used to store configuration info read from config i/o registers. Most of
+ this is not used yet, but might as well save it.
+
+ This structure also holds port addresses that are not at the same offset
+ on the 14F and 24F.
+
+ This structure holds all data that must be duplicated to support multiple
+ adapters. */
+
+static struct ultrastor_config
+{
+ unsigned short port_address; /* base address of card */
+ unsigned short doorbell_address; /* base address of doorbell CSRs */
+ unsigned short ogm_address; /* base address of OGM */
+ unsigned short icm_address; /* base address of ICM */
+ const void *bios_segment;
+ unsigned char interrupt: 4;
+ unsigned char dma_channel: 3;
+ unsigned char bios_drive_number: 1;
+ unsigned char heads;
+ unsigned char sectors;
+ unsigned char ha_scsi_id: 3;
+ unsigned char subversion: 4;
+ unsigned char revision;
+ /* The slot number is used to distinguish the 24F (slot != 0) from
+ the 14F and 34F (slot == 0). */
+ unsigned char slot;
+
+#ifdef PRINT_U24F_VERSION
+ volatile int csir_done;
+#endif
+
+ /* A pool of MSCP structures for this adapter, and a bitmask of
+ busy structures. (If ULTRASTOR_14F_MAX_CMDS == 1, a 1 byte
+ busy flag is used instead.) */
+
+#if ULTRASTOR_MAX_CMDS == 1
+ unsigned char mscp_busy;
+#else
+ unsigned long mscp_free;
+#endif
+ volatile unsigned char aborted[ULTRASTOR_MAX_CMDS];
+ struct mscp mscp[ULTRASTOR_MAX_CMDS];
+} config = {0};
+
+/* Set this to 1 to reset the SCSI bus on error. */
+static int ultrastor_bus_reset;
+
+
+/* Allowed BIOS base addresses (NULL indicates reserved) */
+static const void *const bios_segment_table[8] = {
+ NULL, (void *)0xC4000, (void *)0xC8000, (void *)0xCC000,
+ (void *)0xD0000, (void *)0xD4000, (void *)0xD8000, (void *)0xDC000,
+};
+
+/* Allowed IRQs for 14f */
+static const unsigned char interrupt_table_14f[4] = { 15, 14, 11, 10 };
+
+/* Allowed DMA channels for 14f (0 indicates reserved) */
+static const unsigned char dma_channel_table_14f[4] = { 5, 6, 7, 0 };
+
+/* Head/sector mappings allowed by 14f */
+static const struct {
+ unsigned char heads;
+ unsigned char sectors;
+} mapping_table[4] = { { 16, 63 }, { 64, 32 }, { 64, 63 }, { 64, 32 } };
+
+#ifndef PORT_OVERRIDE
+/* ??? A probe of address 0x310 screws up NE2000 cards */
+static const unsigned short ultrastor_ports_14f[] = {
+ 0x330, 0x340, /*0x310,*/ 0x230, 0x240, 0x210, 0x130, 0x140,
+};
+#endif
+
+static void ultrastor_interrupt(int, void *, struct pt_regs *);
+static irqreturn_t do_ultrastor_interrupt(int, void *, struct pt_regs *);
+static inline void build_sg_list(struct mscp *, Scsi_Cmnd *SCpnt);
+
+
+/* Always called with host lock held */
+
+static inline int find_and_clear_bit_16(unsigned long *field)
+{
+ int rv;
+
+ if (*field == 0) panic("No free mscp");
+ asm("xorl %0,%0\n0:\tbsfw %1,%w0\n\tbtr %0,%1\n\tjnc 0b"
+ : "=&r" (rv), "=m" (*field) : "1" (*field));
+ return rv;
+}
+
+/* This has been re-implemented with the help of Richard Earnshaw,
+ <rwe@pegasus.esprit.ec.org> and works with gcc-2.5.8 and gcc-2.6.0.
+ The instability noted by jfc below appears to be a bug in
+ gcc-2.5.x when compiling w/o optimization. --Caleb
+
+ This asm is fragile: it doesn't work without the casts and it may
+ not work without optimization. Maybe I should add a swap builtin
+ to gcc. --jfc */
+static inline unsigned char xchgb(unsigned char reg,
+ volatile unsigned char *mem)
+{
+ __asm__ ("xchgb %0,%1" : "=q" (reg), "=m" (*mem) : "0" (reg));
+ return reg;
+}
+
+#if ULTRASTOR_DEBUG & (UD_COMMAND | UD_ABORT)
+
+/* Always called with the host lock held */
+static void log_ultrastor_abort(struct ultrastor_config *config,
+ int command)
+{
+ static char fmt[80] = "abort %d (%x); MSCP free pool: %x;";
+ int i;
+
+ for (i = 0; i < ULTRASTOR_MAX_CMDS; i++)
+ {
+ fmt[20 + i*2] = ' ';
+ if (! (config->mscp_free & (1 << i)))
+ fmt[21 + i*2] = '0' + config->mscp[i].target_id;
+ else
+ fmt[21 + i*2] = '-';
+ }
+ fmt[20 + ULTRASTOR_MAX_CMDS * 2] = '\n';
+ fmt[21 + ULTRASTOR_MAX_CMDS * 2] = 0;
+ printk(fmt, command, &config->mscp[command], config->mscp_free);
+
+}
+#endif
+
+static int ultrastor_14f_detect(Scsi_Host_Template * tpnt)
+{
+ size_t i;
+ unsigned char in_byte, version_byte = 0;
+ struct config_1 {
+ unsigned char bios_segment: 3;
+ unsigned char removable_disks_as_fixed: 1;
+ unsigned char interrupt: 2;
+ unsigned char dma_channel: 2;
+ } config_1;
+ struct config_2 {
+ unsigned char ha_scsi_id: 3;
+ unsigned char mapping_mode: 2;
+ unsigned char bios_drive_number: 1;
+ unsigned char tfr_port: 2;
+ } config_2;
+
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: called\n");
+#endif
+
+ /* If a 24F has already been configured, don't look for a 14F. */
+ if (config.bios_segment)
+ return FALSE;
+
+#ifdef PORT_OVERRIDE
+ if(!request_region(PORT_OVERRIDE, 0xc, "ultrastor")) {
+ printk("Ultrastor I/O space already in use\n");
+ return FALSE;
+ };
+ config.port_address = PORT_OVERRIDE;
+#else
+ for (i = 0; i < ARRAY_SIZE(ultrastor_ports_14f); i++) {
+ if(!request_region(ultrastor_ports_14f[i], 0x0c, "ultrastor")) continue;
+ config.port_address = ultrastor_ports_14f[i];
+#endif
+
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: testing port address %03X\n", config.port_address);
+#endif
+
+ in_byte = inb(U14F_PRODUCT_ID(config.port_address));
+ if (in_byte != US14F_PRODUCT_ID_0) {
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+# ifdef PORT_OVERRIDE
+ printk("US14F: detect: wrong product ID 0 - %02X\n", in_byte);
+# else
+ printk("US14F: detect: no adapter at port %03X\n", config.port_address);
+# endif
+#endif
+#ifdef PORT_OVERRIDE
+ goto out_release_port;
+#else
+ release_region(config.port_address, 0x0c);
+ continue;
+#endif
+ }
+ in_byte = inb(U14F_PRODUCT_ID(config.port_address) + 1);
+ /* Only upper nibble is significant for Product ID 1 */
+ if ((in_byte & 0xF0) != US14F_PRODUCT_ID_1) {
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+# ifdef PORT_OVERRIDE
+ printk("US14F: detect: wrong product ID 1 - %02X\n", in_byte);
+# else
+ printk("US14F: detect: no adapter at port %03X\n", config.port_address);
+# endif
+#endif
+#ifdef PORT_OVERRIDE
+ goto out_release_port;
+#else
+ release_region(config.port_address, 0x0c);
+ continue;
+#endif
+ }
+ version_byte = in_byte;
+#ifndef PORT_OVERRIDE
+ break;
+ }
+ if (i == ARRAY_SIZE(ultrastor_ports_14f)) {
+# if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: no port address found!\n");
+# endif
+ /* all ports probed already released - we can just go straight out */
+ return FALSE;
+ }
+#endif
+
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: adapter found at port address %03X\n",
+ config.port_address);
+#endif
+
+ /* Set local doorbell mask to disallow bus reset unless
+ ultrastor_bus_reset is true. */
+ outb(ultrastor_bus_reset ? 0xc2 : 0x82, LCL_DOORBELL_MASK(config.port_address));
+
+ /* All above tests passed, must be the right thing. Get some useful
+ info. */
+
+ /* Register the I/O space that we use */
+
+ *(char *)&config_1 = inb(CONFIG(config.port_address + 0));
+ *(char *)&config_2 = inb(CONFIG(config.port_address + 1));
+ config.bios_segment = bios_segment_table[config_1.bios_segment];
+ config.doorbell_address = config.port_address;
+ config.ogm_address = config.port_address + 0x8;
+ config.icm_address = config.port_address + 0xC;
+ config.interrupt = interrupt_table_14f[config_1.interrupt];
+ config.ha_scsi_id = config_2.ha_scsi_id;
+ config.heads = mapping_table[config_2.mapping_mode].heads;
+ config.sectors = mapping_table[config_2.mapping_mode].sectors;
+ config.bios_drive_number = config_2.bios_drive_number;
+ config.subversion = (version_byte & 0x0F);
+ if (config.subversion == U34F)
+ config.dma_channel = 0;
+ else
+ config.dma_channel = dma_channel_table_14f[config_1.dma_channel];
+
+ if (!config.bios_segment) {
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: not detected.\n");
+#endif
+ goto out_release_port;
+ }
+
+ /* Final consistency check, verify previous info. */
+ if (config.subversion != U34F)
+ if (!config.dma_channel || !(config_2.tfr_port & 0x2)) {
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: consistency check failed\n");
+#endif
+ goto out_release_port;
+ }
+
+ /* If we were TRULY paranoid, we could issue a host adapter inquiry
+ command here and verify the data returned. But frankly, I'm
+ exhausted! */
+
+ /* Finally! Now I'm satisfied... */
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US14F: detect: detect succeeded\n"
+ " Port address: %03X\n"
+ " BIOS segment: %05X\n"
+ " Interrupt: %u\n"
+ " DMA channel: %u\n"
+ " H/A SCSI ID: %u\n"
+ " Subversion: %u\n",
+ config.port_address, config.bios_segment, config.interrupt,
+ config.dma_channel, config.ha_scsi_id, config.subversion);
+#endif
+ tpnt->this_id = config.ha_scsi_id;
+ tpnt->unchecked_isa_dma = (config.subversion != U34F);
+
+#if ULTRASTOR_MAX_CMDS > 1
+ config.mscp_free = ~0;
+#endif
+
+ /*
+ * Brrr, &config.mscp[0].SCint->host) it is something magical....
+ * XXX and FIXME
+ */
+ if (request_irq(config.interrupt, do_ultrastor_interrupt, 0, "Ultrastor", &config.mscp[0].SCint->device->host)) {
+ printk("Unable to allocate IRQ%u for UltraStor controller.\n",
+ config.interrupt);
+ goto out_release_port;
+ }
+ if (config.dma_channel && request_dma(config.dma_channel,"Ultrastor")) {
+ printk("Unable to allocate DMA channel %u for UltraStor controller.\n",
+ config.dma_channel);
+ free_irq(config.interrupt, NULL);
+ goto out_release_port;
+ }
+ tpnt->sg_tablesize = ULTRASTOR_14F_MAX_SG;
+ printk("UltraStor driver version" VERSION ". Using %d SG lists.\n",
+ ULTRASTOR_14F_MAX_SG);
+
+ return TRUE;
+out_release_port:
+ release_region(config.port_address, 0x0c);
+ return FALSE;
+}
+
+static int ultrastor_24f_detect(Scsi_Host_Template * tpnt)
+{
+ int i;
+ struct Scsi_Host * shpnt = NULL;
+
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US24F: detect");
+#endif
+
+ /* probe each EISA slot at slot address C80 */
+ for (i = 1; i < 15; i++)
+ {
+ unsigned char config_1, config_2;
+ unsigned short addr = (i << 12) | ULTRASTOR_24F_PORT;
+
+ if (inb(addr) != US24F_PRODUCT_ID_0 &&
+ inb(addr+1) != US24F_PRODUCT_ID_1 &&
+ inb(addr+2) != US24F_PRODUCT_ID_2)
+ continue;
+
+ config.revision = inb(addr+3);
+ config.slot = i;
+ if (! (inb(addr+4) & 1))
+ {
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("U24F: found disabled card in slot %u\n", i);
+#endif
+ continue;
+ }
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("U24F: found card in slot %u\n", i);
+#endif
+ config_1 = inb(addr + 5);
+ config.bios_segment = bios_segment_table[config_1 & 7];
+ switch(config_1 >> 4)
+ {
+ case 1:
+ config.interrupt = 15;
+ break;
+ case 2:
+ config.interrupt = 14;
+ break;
+ case 4:
+ config.interrupt = 11;
+ break;
+ case 8:
+ config.interrupt = 10;
+ break;
+ default:
+ printk("U24F: invalid IRQ\n");
+ return FALSE;
+ }
+
+ /* BIOS addr set */
+ /* base port set */
+ config.port_address = addr;
+ config.doorbell_address = addr + 12;
+ config.ogm_address = addr + 0x17;
+ config.icm_address = addr + 0x1C;
+ config_2 = inb(addr + 7);
+ config.ha_scsi_id = config_2 & 7;
+ config.heads = mapping_table[(config_2 >> 3) & 3].heads;
+ config.sectors = mapping_table[(config_2 >> 3) & 3].sectors;
+#if (ULTRASTOR_DEBUG & UD_DETECT)
+ printk("US24F: detect: detect succeeded\n"
+ " Port address: %03X\n"
+ " BIOS segment: %05X\n"
+ " Interrupt: %u\n"
+ " H/A SCSI ID: %u\n",
+ config.port_address, config.bios_segment,
+ config.interrupt, config.ha_scsi_id);
+#endif
+ tpnt->this_id = config.ha_scsi_id;
+ tpnt->unchecked_isa_dma = 0;
+ tpnt->sg_tablesize = ULTRASTOR_24F_MAX_SG;
+
+ shpnt = scsi_register(tpnt, 0);
+ if (!shpnt) {
+ printk(KERN_WARNING "(ultrastor:) Could not register scsi device. Aborting registration.\n");
+ free_irq(config.interrupt, do_ultrastor_interrupt);
+ return FALSE;
+ }
+
+ if (request_irq(config.interrupt, do_ultrastor_interrupt, 0, "Ultrastor", shpnt))
+ {
+ printk("Unable to allocate IRQ%u for UltraStor controller.\n",
+ config.interrupt);
+ return FALSE;
+ }
+
+ shpnt->irq = config.interrupt;
+ shpnt->dma_channel = config.dma_channel;
+ shpnt->io_port = config.port_address;
+
+#if ULTRASTOR_MAX_CMDS > 1
+ config.mscp_free = ~0;
+#endif
+ /* Mark ICM and OGM free */
+ outb(0, addr + 0x16);
+ outb(0, addr + 0x1B);
+
+ /* Set local doorbell mask to disallow bus reset unless
+ ultrastor_bus_reset is true. */
+ outb(ultrastor_bus_reset ? 0xc2 : 0x82, LCL_DOORBELL_MASK(addr+12));
+ outb(0x02, SYS_DOORBELL_MASK(addr+12));
+ printk("UltraStor driver version " VERSION ". Using %d SG lists.\n",
+ tpnt->sg_tablesize);
+ return TRUE;
+ }
+ return FALSE;
+}
+
+static int ultrastor_detect(Scsi_Host_Template * tpnt)
+{
+ tpnt->proc_name = "ultrastor";
+ return ultrastor_14f_detect(tpnt) || ultrastor_24f_detect(tpnt);
+}
+
+static int ultrastor_release(struct Scsi_Host *shost)
+{
+ if (shost->irq)
+ free_irq(shost->irq, NULL);
+ if (shost->dma_channel != 0xff)
+ free_dma(shost->dma_channel);
+ if (shost->io_port && shost->n_io_port)
+ release_region(shost->io_port, shost->n_io_port);
+ scsi_unregister(shost);
+ return 0;
+}
+
+static const char *ultrastor_info(struct Scsi_Host * shpnt)
+{
+ static char buf[64];
+
+ if (config.slot)
+ sprintf(buf, "UltraStor 24F SCSI @ Slot %u IRQ%u",
+ config.slot, config.interrupt);
+ else if (config.subversion)
+ sprintf(buf, "UltraStor 34F SCSI @ Port %03X BIOS %05X IRQ%u",
+ config.port_address, (int)config.bios_segment,
+ config.interrupt);
+ else
+ sprintf(buf, "UltraStor 14F SCSI @ Port %03X BIOS %05X IRQ%u DMA%u",
+ config.port_address, (int)config.bios_segment,
+ config.interrupt, config.dma_channel);
+ return buf;
+}
+
+static inline void build_sg_list(struct mscp *mscp, Scsi_Cmnd *SCpnt)
+{
+ struct scatterlist *sl;
+ long transfer_length = 0;
+ int i, max;
+
+ sl = (struct scatterlist *) SCpnt->request_buffer;
+ max = SCpnt->use_sg;
+ for (i = 0; i < max; i++) {
+ mscp->sglist[i].address = isa_page_to_bus(sl[i].page) + sl[i].offset;
+ mscp->sglist[i].num_bytes = sl[i].length;
+ transfer_length += sl[i].length;
+ }
+ mscp->number_of_sg_list = max;
+ mscp->transfer_data = isa_virt_to_bus(mscp->sglist);
+ /* ??? May not be necessary. Docs are unclear as to whether transfer
+ length field is ignored or whether it should be set to the total
+ number of bytes of the transfer. */
+ mscp->transfer_data_length = transfer_length;
+}
+
+static int ultrastor_queuecommand(Scsi_Cmnd *SCpnt, void (*done)(Scsi_Cmnd *))
+{
+ struct mscp *my_mscp;
+#if ULTRASTOR_MAX_CMDS > 1
+ int mscp_index;
+#endif
+ unsigned int status;
+
+ /* Next test is for debugging; "can't happen" */
+ if ((config.mscp_free & ((1U << ULTRASTOR_MAX_CMDS) - 1)) == 0)
+ panic("ultrastor_queuecommand: no free MSCP\n");
+ mscp_index = find_and_clear_bit_16(&config.mscp_free);
+
+ /* Has the command been aborted? */
+ if (xchgb(0xff, &config.aborted[mscp_index]) != 0)
+ {
+ status = DID_ABORT << 16;
+ goto aborted;
+ }
+
+ my_mscp = &config.mscp[mscp_index];
+
+ *(unsigned char *)my_mscp = OP_SCSI | (DTD_SCSI << 3);
+
+ /* Tape drives don't work properly if the cache is used. The SCSI
+ READ command for a tape doesn't have a block offset, and the adapter
+ incorrectly assumes that all reads from the tape read the same
+ blocks. Results will depend on read buffer size and other disk
+ activity.
+
+ ??? Which other device types should never use the cache? */
+ my_mscp->ca = SCpnt->device->type != TYPE_TAPE;
+ my_mscp->target_id = SCpnt->device->id;
+ my_mscp->ch_no = 0;
+ my_mscp->lun = SCpnt->device->lun;
+ if (SCpnt->use_sg) {
+ /* Set scatter/gather flag in SCSI command packet */
+ my_mscp->sg = TRUE;
+ build_sg_list(my_mscp, SCpnt);
+ } else {
+ /* Unset scatter/gather flag in SCSI command packet */
+ my_mscp->sg = FALSE;
+ my_mscp->transfer_data = isa_virt_to_bus(SCpnt->request_buffer);
+ my_mscp->transfer_data_length = SCpnt->request_bufflen;
+ }
+ my_mscp->command_link = 0; /*???*/
+ my_mscp->scsi_command_link_id = 0; /*???*/
+ my_mscp->length_of_sense_byte = sizeof SCpnt->sense_buffer;
+ my_mscp->length_of_scsi_cdbs = SCpnt->cmd_len;
+ memcpy(my_mscp->scsi_cdbs, SCpnt->cmnd, my_mscp->length_of_scsi_cdbs);
+ my_mscp->adapter_status = 0;
+ my_mscp->target_status = 0;
+ my_mscp->sense_data = isa_virt_to_bus(&SCpnt->sense_buffer);
+ my_mscp->done = done;
+ my_mscp->SCint = SCpnt;
+ SCpnt->host_scribble = (unsigned char *)my_mscp;
+
+ /* Find free OGM slot. On 24F, look for OGM status byte == 0.
+ On 14F and 34F, wait for local interrupt pending flag to clear.
+
+ FIXME: now we are using new_eh we should punt here and let the
+ midlayer sort it out */
+
+retry:
+ if (config.slot)
+ while (inb(config.ogm_address - 1) != 0 && config.aborted[mscp_index] == 0xff)
+ barrier();
+
+ /* else??? */
+
+ while ((inb(LCL_DOORBELL_INTR(config.doorbell_address)) & (config.slot ? 2 : 1)) && config.aborted[mscp_index] == 0xff)
+ barrier();
+
+ /* To avoid race conditions, keep the code to write to the adapter
+ atomic. This simplifies the abort code. Right now the
+ scsi mid layer has the host_lock already held
+ */
+
+ if (inb(LCL_DOORBELL_INTR(config.doorbell_address)) & (config.slot ? 2 : 1))
+ goto retry;
+
+ status = xchgb(0, &config.aborted[mscp_index]);
+ if (status != 0xff) {
+
+#if ULTRASTOR_DEBUG & (UD_COMMAND | UD_ABORT)
+ printk("USx4F: queuecommand: aborted\n");
+#if ULTRASTOR_MAX_CMDS > 1
+ log_ultrastor_abort(&config, mscp_index);
+#endif
+#endif
+ status <<= 16;
+
+ aborted:
+ set_bit(mscp_index, &config.mscp_free);
+ /* If the driver queues commands, call the done proc here. Otherwise
+ return an error. */
+#if ULTRASTOR_MAX_CMDS > 1
+ SCpnt->result = status;
+ done(SCpnt);
+ return 0;
+#else
+ return status;
+#endif
+ }
+
+ /* Store pointer in OGM address bytes */
+ outl(isa_virt_to_bus(my_mscp), config.ogm_address);
+
+ /* Issue OGM interrupt */
+ if (config.slot) {
+ /* Write OGM command register on 24F */
+ outb(1, config.ogm_address - 1);
+ outb(0x2, LCL_DOORBELL_INTR(config.doorbell_address));
+ } else {
+ outb(0x1, LCL_DOORBELL_INTR(config.doorbell_address));
+ }
+
+#if (ULTRASTOR_DEBUG & UD_COMMAND)
+ printk("USx4F: queuecommand: returning\n");
+#endif
+
+ return 0;
+}
+
+/* This code must deal with 2 cases:
+
+ 1. The command has not been written to the OGM. In this case, set
+ the abort flag and return.
+
+ 2. The command has been written to the OGM and is stuck somewhere in
+ the adapter.
+
+ 2a. On a 24F, ask the adapter to abort the command. It will interrupt
+ when it does.
+
+ 2b. Call the command's done procedure.
+
+ */
+
+static int ultrastor_abort(Scsi_Cmnd *SCpnt)
+{
+#if ULTRASTOR_DEBUG & UD_ABORT
+ char out[108];
+ unsigned char icm_status = 0, ogm_status = 0;
+ unsigned int icm_addr = 0, ogm_addr = 0;
+#endif
+ unsigned int mscp_index;
+ unsigned char old_aborted;
+ unsigned long flags;
+ void (*done)(Scsi_Cmnd *);
+ struct Scsi_Host *host = SCpnt->device->host;
+
+ if(config.slot)
+ return FAILED; /* Do not attempt an abort for the 24f */
+
+ /* Simple consistency checking */
+ if(!SCpnt->host_scribble)
+ return FAILED;
+
+ mscp_index = ((struct mscp *)SCpnt->host_scribble) - config.mscp;
+ if (mscp_index >= ULTRASTOR_MAX_CMDS)
+ panic("Ux4F aborting invalid MSCP");
+
+#if ULTRASTOR_DEBUG & UD_ABORT
+ if (config.slot)
+ {
+ int port0 = (config.slot << 12) | 0xc80;
+ int i;
+ unsigned long flags;
+
+ spin_lock_irqsave(host->host_lock, flags);
+ strcpy(out, "OGM %d:%x ICM %d:%x ports: ");
+ for (i = 0; i < 16; i++)
+ {
+ unsigned char p = inb(port0 + i);
+ out[28 + i * 3] = "0123456789abcdef"[p >> 4];
+ out[29 + i * 3] = "0123456789abcdef"[p & 15];
+ out[30 + i * 3] = ' ';
+ }
+ out[28 + i * 3] = '\n';
+ out[29 + i * 3] = 0;
+ ogm_status = inb(port0 + 22);
+ ogm_addr = (unsigned int)isa_bus_to_virt(inl(port0 + 23));
+ icm_status = inb(port0 + 27);
+ icm_addr = (unsigned int)isa_bus_to_virt(inl(port0 + 28));
+ spin_lock_irqsave(host->host_lock, flags);
+ }
+
+ /* First check to see if an interrupt is pending. I suspect the SiS
+ chipset loses interrupts. (I also suspect is mangles data, but
+ one bug at a time... */
+ if (config.slot ? inb(config.icm_address - 1) == 2 :
+ (inb(SYS_DOORBELL_INTR(config.doorbell_address)) & 1))
+ {
+ printk("Ux4F: abort while completed command pending\n");
+
+ spin_lock_irqsave(host->host_lock, flags);
+ /* FIXME: Ewww... need to think about passing host around properly */
+ ultrastor_interrupt(0, NULL, NULL);
+ spin_unlock_irqrestore(host->host_lock, flags);
+ return SUCCESS;
+ }
+#endif
+
+ old_aborted = xchgb(DID_ABORT, &config.aborted[mscp_index]);
+
+ /* aborted == 0xff is the signal that queuecommand has not yet sent
+ the command. It will notice the new abort flag and fail. */
+ if (old_aborted == 0xff)
+ return SUCCESS;
+
+ /* On 24F, send an abort MSCP request. The adapter will interrupt
+ and the interrupt handler will call done. */
+ if (config.slot && inb(config.ogm_address - 1) == 0)
+ {
+ unsigned long flags;
+
+ spin_lock_irqsave(host->host_lock, flags);
+ outl(isa_virt_to_bus(&config.mscp[mscp_index]), config.ogm_address);
+ udelay(8);
+ outb(0x80, config.ogm_address - 1);
+ outb(0x2, LCL_DOORBELL_INTR(config.doorbell_address));
+#if ULTRASTOR_DEBUG & UD_ABORT
+ log_ultrastor_abort(&config, mscp_index);
+ printk(out, ogm_status, ogm_addr, icm_status, icm_addr);
+#endif
+ spin_unlock_irqrestore(host->host_lock, flags);
+ /* FIXME: add a wait for the abort to complete */
+ return SUCCESS;
+ }
+
+#if ULTRASTOR_DEBUG & UD_ABORT
+ log_ultrastor_abort(&config, mscp_index);
+#endif
+
+ /* Can't request a graceful abort. Either this is not a 24F or
+ the OGM is busy. Don't free the command -- the adapter might
+ still be using it. Setting SCint = 0 causes the interrupt
+ handler to ignore the command. */
+
+ /* FIXME - devices that implement soft resets will still be running
+ the command after a bus reset. We would probably rather leave
+ the command in the queue. The upper level code will automatically
+ leave the command in the active state instead of requeueing it. ERY */
+
+#if ULTRASTOR_DEBUG & UD_ABORT
+ if (config.mscp[mscp_index].SCint != SCpnt)
+ printk("abort: command mismatch, %p != %p\n",
+ config.mscp[mscp_index].SCint, SCpnt);
+#endif
+ if (config.mscp[mscp_index].SCint == 0)
+ return SCSI_ABORT_NOT_RUNNING;
+
+ if (config.mscp[mscp_index].SCint != SCpnt) panic("Bad abort");
+ config.mscp[mscp_index].SCint = NULL;
+ done = config.mscp[mscp_index].done;
+ config.mscp[mscp_index].done = NULL;
+ SCpnt->result = DID_ABORT << 16;
+
+ /* Take the host lock to guard against scsi layer re-entry */
+ spin_lock_irqsave(host->host_lock, flags);
+ done(SCpnt);
+ spin_unlock_irqrestore(host->host_lock, flags);
+
+ /* Need to set a timeout here in case command never completes. */
+ return SUCCESS;
+}
+
+static int ultrastor_host_reset(Scsi_Cmnd * SCpnt)
+{
+ unsigned long flags;
+ int i;
+ struct Scsi_Host *host = SCpnt->device->host;
+
+#if (ULTRASTOR_DEBUG & UD_RESET)
+ printk("US14F: reset: called\n");
+#endif
+
+ if(config.slot)
+ return FAILED;
+
+ spin_lock_irqsave(host->host_lock, flags);
+ /* Reset the adapter and SCSI bus. The SCSI bus reset can be
+ inhibited by clearing ultrastor_bus_reset before probe. */
+ outb(0xc0, LCL_DOORBELL_INTR(config.doorbell_address));
+ if (config.slot)
+ {
+ outb(0, config.ogm_address - 1);
+ outb(0, config.icm_address - 1);
+ }
+
+#if ULTRASTOR_MAX_CMDS == 1
+ if (config.mscp_busy && config.mscp->done && config.mscp->SCint)
+ {
+ config.mscp->SCint->result = DID_RESET << 16;
+ config.mscp->done(config.mscp->SCint);
+ }
+ config.mscp->SCint = 0;
+#else
+ for (i = 0; i < ULTRASTOR_MAX_CMDS; i++)
+ {
+ if (! (config.mscp_free & (1 << i)) &&
+ config.mscp[i].done && config.mscp[i].SCint)
+ {
+ config.mscp[i].SCint->result = DID_RESET << 16;
+ config.mscp[i].done(config.mscp[i].SCint);
+ config.mscp[i].done = NULL;
+ }
+ config.mscp[i].SCint = NULL;
+ }
+#endif
+
+ /* FIXME - if the device implements soft resets, then the command
+ will still be running. ERY
+
+ Even bigger deal with new_eh!
+ */
+
+ memset((unsigned char *)config.aborted, 0, sizeof config.aborted);
+#if ULTRASTOR_MAX_CMDS == 1
+ config.mscp_busy = 0;
+#else
+ config.mscp_free = ~0;
+#endif
+
+ spin_unlock_irqrestore(host->host_lock, flags);
+ return SCSI_RESET_SUCCESS;
+
+}
+
+int ultrastor_biosparam(struct scsi_device *sdev, struct block_device *bdev,
+ sector_t capacity, int * dkinfo)
+{
+ int size = capacity;
+ unsigned int s = config.heads * config.sectors;
+
+ dkinfo[0] = config.heads;
+ dkinfo[1] = config.sectors;
+ dkinfo[2] = size / s; /* Ignore partial cylinders */
+#if 0
+ if (dkinfo[2] > 1024)
+ dkinfo[2] = 1024;
+#endif
+ return 0;
+}
+
+static void ultrastor_interrupt(int irq, void *dev_id, struct pt_regs *regs)
+{
+ unsigned int status;
+#if ULTRASTOR_MAX_CMDS > 1
+ unsigned int mscp_index;
+#endif
+ struct mscp *mscp;
+ void (*done)(Scsi_Cmnd *);
+ Scsi_Cmnd *SCtmp;
+
+#if ULTRASTOR_MAX_CMDS == 1
+ mscp = &config.mscp[0];
+#else
+ mscp = (struct mscp *)isa_bus_to_virt(inl(config.icm_address));
+ mscp_index = mscp - config.mscp;
+ if (mscp_index >= ULTRASTOR_MAX_CMDS) {
+ printk("Ux4F interrupt: bad MSCP address %x\n", (unsigned int) mscp);
+ /* A command has been lost. Reset and report an error
+ for all commands. */
+ ultrastor_host_reset(dev_id);
+ return;
+ }
+#endif
+
+ /* Clean ICM slot (set ICMINT bit to 0) */
+ if (config.slot) {
+ unsigned char icm_status = inb(config.icm_address - 1);
+#if ULTRASTOR_DEBUG & (UD_INTERRUPT|UD_ERROR|UD_ABORT)
+ if (icm_status != 1 && icm_status != 2)
+ printk("US24F: ICM status %x for MSCP %d (%x)\n", icm_status,
+ mscp_index, (unsigned int) mscp);
+#endif
+ /* The manual says clear interrupt then write 0 to ICM status.
+ This seems backwards, but I'll do it anyway. --jfc */
+ outb(2, SYS_DOORBELL_INTR(config.doorbell_address));
+ outb(0, config.icm_address - 1);
+ if (icm_status == 4) {
+ printk("UltraStor abort command failed\n");
+ return;
+ }
+ if (icm_status == 3) {
+ void (*done)(Scsi_Cmnd *) = mscp->done;
+ if (done) {
+ mscp->done = NULL;
+ mscp->SCint->result = DID_ABORT << 16;
+ done(mscp->SCint);
+ }
+ return;
+ }
+ } else {
+ outb(1, SYS_DOORBELL_INTR(config.doorbell_address));
+ }
+
+ SCtmp = mscp->SCint;
+ mscp->SCint = NULL;
+
+ if (SCtmp == 0)
+ {
+#if ULTRASTOR_DEBUG & (UD_ABORT|UD_INTERRUPT)
+ printk("MSCP %d (%x): no command\n", mscp_index, (unsigned int) mscp);
+#endif
+#if ULTRASTOR_MAX_CMDS == 1
+ config.mscp_busy = FALSE;
+#else
+ set_bit(mscp_index, &config.mscp_free);
+#endif
+ config.aborted[mscp_index] = 0;
+ return;
+ }
+
+ /* Save done locally and zero before calling. This is needed as
+ once we call done, we may get another command queued before this
+ interrupt service routine can return. */
+ done = mscp->done;
+ mscp->done = NULL;
+
+ /* Let the higher levels know that we're done */
+ switch (mscp->adapter_status)
+ {
+ case 0:
+ status = DID_OK << 16;
+ break;
+ case 0x01: /* invalid command */
+ case 0x02: /* invalid parameters */
+ case 0x03: /* invalid data list */
+ default:
+ status = DID_ERROR << 16;
+ break;
+ case 0x84: /* SCSI bus abort */
+ status = DID_ABORT << 16;
+ break;
+ case 0x91:
+ status = DID_TIME_OUT << 16;
+ break;
+ }
+
+ SCtmp->result = status | mscp->target_status;
+
+ SCtmp->host_scribble = NULL;
+
+ /* Free up mscp block for next command */
+#if ULTRASTOR_MAX_CMDS == 1
+ config.mscp_busy = FALSE;
+#else
+ set_bit(mscp_index, &config.mscp_free);
+#endif
+
+#if ULTRASTOR_DEBUG & (UD_ABORT|UD_INTERRUPT)
+ if (config.aborted[mscp_index])
+ printk("Ux4 interrupt: MSCP %d (%x) aborted = %d\n",
+ mscp_index, (unsigned int) mscp, config.aborted[mscp_index]);
+#endif
+ config.aborted[mscp_index] = 0;
+
+ if (done)
+ done(SCtmp);
+ else
+ printk("US14F: interrupt: unexpected interrupt\n");
+
+ if (config.slot ? inb(config.icm_address - 1) :
+ (inb(SYS_DOORBELL_INTR(config.doorbell_address)) & 1))
+#if (ULTRASTOR_DEBUG & UD_MULTI_CMD)
+ printk("Ux4F: multiple commands completed\n");
+#else
+ ;
+#endif
+
+#if (ULTRASTOR_DEBUG & UD_INTERRUPT)
+ printk("USx4F: interrupt: returning\n");
+#endif
+}
+
+static irqreturn_t do_ultrastor_interrupt(int irq, void *dev_id,
+ struct pt_regs *regs)
+{
+ unsigned long flags;
+ struct Scsi_Host *dev = dev_id;
+
+ spin_lock_irqsave(dev->host_lock, flags);
+ ultrastor_interrupt(irq, dev_id, regs);
+ spin_unlock_irqrestore(dev->host_lock, flags);
+ return IRQ_HANDLED;
+}
+
+MODULE_LICENSE("GPL");
+
+static Scsi_Host_Template driver_template = {
+ .name = "UltraStor 14F/24F/34F",
+ .detect = ultrastor_detect,
+ .release = ultrastor_release,
+ .info = ultrastor_info,
+ .queuecommand = ultrastor_queuecommand,
+ .eh_abort_handler = ultrastor_abort,
+ .eh_host_reset_handler = ultrastor_host_reset,
+ .bios_param = ultrastor_biosparam,
+ .can_queue = ULTRASTOR_MAX_CMDS,
+ .sg_tablesize = ULTRASTOR_14F_MAX_SG,
+ .cmd_per_lun = ULTRASTOR_MAX_CMDS_PER_LUN,
+ .unchecked_isa_dma = 1,
+ .use_clustering = ENABLE_CLUSTERING,
+};
+#include "scsi_module.c"