linux/drivers/scsi/NCR53C9x.c

3655 lines
102 KiB
C
Raw Normal View History

/* NCR53C9x.c: Generic SCSI driver code for NCR53C9x chips.
*
* Originally esp.c : EnhancedScsiProcessor Sun SCSI driver code.
*
* Copyright (C) 1995, 1998 David S. Miller (davem@caip.rutgers.edu)
*
* Most DMA dependencies put in driver specific files by
* Jesper Skov (jskov@cygnus.co.uk)
*
* Set up to use esp_read/esp_write (preprocessor macros in NCR53c9x.h) by
* Tymm Twillman (tymm@coe.missouri.edu)
*/
/* TODO:
*
* 1) Maybe disable parity checking in config register one for SCSI1
* targets. (Gilmore says parity error on the SBus can lock up
* old sun4c's)
* 2) Add support for DMA2 pipelining.
* 3) Add tagged queueing.
* 4) Maybe change use of "esp" to something more "NCR"'ish.
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/delay.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/slab.h>
#include <linux/blkdev.h>
#include <linux/interrupt.h>
#include <linux/proc_fs.h>
#include <linux/stat.h>
#include <linux/init.h>
#include "scsi.h"
#include <scsi/scsi_host.h>
#include "NCR53C9x.h"
#include <asm/system.h>
#include <asm/ptrace.h>
#include <asm/pgtable.h>
#include <asm/io.h>
#include <asm/irq.h>
/* Command phase enumeration. */
enum {
not_issued = 0x00, /* Still in the issue_SC queue. */
/* Various forms of selecting a target. */
#define in_slct_mask 0x10
in_slct_norm = 0x10, /* ESP is arbitrating, normal selection */
in_slct_stop = 0x11, /* ESP will select, then stop with IRQ */
in_slct_msg = 0x12, /* select, then send a message */
in_slct_tag = 0x13, /* select and send tagged queue msg */
in_slct_sneg = 0x14, /* select and acquire sync capabilities */
/* Any post selection activity. */
#define in_phases_mask 0x20
in_datain = 0x20, /* Data is transferring from the bus */
in_dataout = 0x21, /* Data is transferring to the bus */
in_data_done = 0x22, /* Last DMA data operation done (maybe) */
in_msgin = 0x23, /* Eating message from target */
in_msgincont = 0x24, /* Eating more msg bytes from target */
in_msgindone = 0x25, /* Decide what to do with what we got */
in_msgout = 0x26, /* Sending message to target */
in_msgoutdone = 0x27, /* Done sending msg out */
in_cmdbegin = 0x28, /* Sending cmd after abnormal selection */
in_cmdend = 0x29, /* Done sending slow cmd */
in_status = 0x2a, /* Was in status phase, finishing cmd */
in_freeing = 0x2b, /* freeing the bus for cmd cmplt or disc */
in_the_dark = 0x2c, /* Don't know what bus phase we are in */
/* Special states, ie. not normal bus transitions... */
#define in_spec_mask 0x80
in_abortone = 0x80, /* Aborting one command currently */
in_abortall = 0x81, /* Blowing away all commands we have */
in_resetdev = 0x82, /* SCSI target reset in progress */
in_resetbus = 0x83, /* SCSI bus reset in progress */
in_tgterror = 0x84, /* Target did something stupid */
};
enum {
/* Zero has special meaning, see skipahead[12]. */
/*0*/ do_never,
/*1*/ do_phase_determine,
/*2*/ do_reset_bus,
/*3*/ do_reset_complete,
/*4*/ do_work_bus,
/*5*/ do_intr_end
};
/* The master ring of all esp hosts we are managing in this driver. */
static struct NCR_ESP *espchain;
int nesps = 0, esps_in_use = 0, esps_running = 0;
EXPORT_SYMBOL(nesps);
EXPORT_SYMBOL(esps_running);
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 21:55:46 +08:00
irqreturn_t esp_intr(int irq, void *dev_id);
/* Debugging routines */
static struct esp_cmdstrings {
unchar cmdchar;
char *text;
} esp_cmd_strings[] = {
/* Miscellaneous */
{ ESP_CMD_NULL, "ESP_NOP", },
{ ESP_CMD_FLUSH, "FIFO_FLUSH", },
{ ESP_CMD_RC, "RSTESP", },
{ ESP_CMD_RS, "RSTSCSI", },
/* Disconnected State Group */
{ ESP_CMD_RSEL, "RESLCTSEQ", },
{ ESP_CMD_SEL, "SLCTNATN", },
{ ESP_CMD_SELA, "SLCTATN", },
{ ESP_CMD_SELAS, "SLCTATNSTOP", },
{ ESP_CMD_ESEL, "ENSLCTRESEL", },
{ ESP_CMD_DSEL, "DISSELRESEL", },
{ ESP_CMD_SA3, "SLCTATN3", },
{ ESP_CMD_RSEL3, "RESLCTSEQ", },
/* Target State Group */
{ ESP_CMD_SMSG, "SNDMSG", },
{ ESP_CMD_SSTAT, "SNDSTATUS", },
{ ESP_CMD_SDATA, "SNDDATA", },
{ ESP_CMD_DSEQ, "DISCSEQ", },
{ ESP_CMD_TSEQ, "TERMSEQ", },
{ ESP_CMD_TCCSEQ, "TRGTCMDCOMPSEQ", },
{ ESP_CMD_DCNCT, "DISC", },
{ ESP_CMD_RMSG, "RCVMSG", },
{ ESP_CMD_RCMD, "RCVCMD", },
{ ESP_CMD_RDATA, "RCVDATA", },
{ ESP_CMD_RCSEQ, "RCVCMDSEQ", },
/* Initiator State Group */
{ ESP_CMD_TI, "TRANSINFO", },
{ ESP_CMD_ICCSEQ, "INICMDSEQCOMP", },
{ ESP_CMD_MOK, "MSGACCEPTED", },
{ ESP_CMD_TPAD, "TPAD", },
{ ESP_CMD_SATN, "SATN", },
{ ESP_CMD_RATN, "RATN", },
};
#define NUM_ESP_COMMANDS ((sizeof(esp_cmd_strings)) / (sizeof(struct esp_cmdstrings)))
/* Print textual representation of an ESP command */
static inline void esp_print_cmd(unchar espcmd)
{
unchar dma_bit = espcmd & ESP_CMD_DMA;
int i;
espcmd &= ~dma_bit;
for(i=0; i<NUM_ESP_COMMANDS; i++)
if(esp_cmd_strings[i].cmdchar == espcmd)
break;
if(i==NUM_ESP_COMMANDS)
printk("ESP_Unknown");
else
printk("%s%s", esp_cmd_strings[i].text,
((dma_bit) ? "+DMA" : ""));
}
/* Print the status register's value */
static inline void esp_print_statreg(unchar statreg)
{
unchar phase;
printk("STATUS<");
phase = statreg & ESP_STAT_PMASK;
printk("%s,", (phase == ESP_DOP ? "DATA-OUT" :
(phase == ESP_DIP ? "DATA-IN" :
(phase == ESP_CMDP ? "COMMAND" :
(phase == ESP_STATP ? "STATUS" :
(phase == ESP_MOP ? "MSG-OUT" :
(phase == ESP_MIP ? "MSG_IN" :
"unknown")))))));
if(statreg & ESP_STAT_TDONE)
printk("TRANS_DONE,");
if(statreg & ESP_STAT_TCNT)
printk("TCOUNT_ZERO,");
if(statreg & ESP_STAT_PERR)
printk("P_ERROR,");
if(statreg & ESP_STAT_SPAM)
printk("SPAM,");
if(statreg & ESP_STAT_INTR)
printk("IRQ,");
printk(">");
}
/* Print the interrupt register's value */
static inline void esp_print_ireg(unchar intreg)
{
printk("INTREG< ");
if(intreg & ESP_INTR_S)
printk("SLCT_NATN ");
if(intreg & ESP_INTR_SATN)
printk("SLCT_ATN ");
if(intreg & ESP_INTR_RSEL)
printk("RSLCT ");
if(intreg & ESP_INTR_FDONE)
printk("FDONE ");
if(intreg & ESP_INTR_BSERV)
printk("BSERV ");
if(intreg & ESP_INTR_DC)
printk("DISCNCT ");
if(intreg & ESP_INTR_IC)
printk("ILL_CMD ");
if(intreg & ESP_INTR_SR)
printk("SCSI_BUS_RESET ");
printk(">");
}
/* Print the sequence step registers contents */
static inline void esp_print_seqreg(unchar stepreg)
{
stepreg &= ESP_STEP_VBITS;
printk("STEP<%s>",
(stepreg == ESP_STEP_ASEL ? "SLCT_ARB_CMPLT" :
(stepreg == ESP_STEP_SID ? "1BYTE_MSG_SENT" :
(stepreg == ESP_STEP_NCMD ? "NOT_IN_CMD_PHASE" :
(stepreg == ESP_STEP_PPC ? "CMD_BYTES_LOST" :
(stepreg == ESP_STEP_FINI4 ? "CMD_SENT_OK" :
"UNKNOWN"))))));
}
static char *phase_string(int phase)
{
switch(phase) {
case not_issued:
return "UNISSUED";
case in_slct_norm:
return "SLCTNORM";
case in_slct_stop:
return "SLCTSTOP";
case in_slct_msg:
return "SLCTMSG";
case in_slct_tag:
return "SLCTTAG";
case in_slct_sneg:
return "SLCTSNEG";
case in_datain:
return "DATAIN";
case in_dataout:
return "DATAOUT";
case in_data_done:
return "DATADONE";
case in_msgin:
return "MSGIN";
case in_msgincont:
return "MSGINCONT";
case in_msgindone:
return "MSGINDONE";
case in_msgout:
return "MSGOUT";
case in_msgoutdone:
return "MSGOUTDONE";
case in_cmdbegin:
return "CMDBEGIN";
case in_cmdend:
return "CMDEND";
case in_status:
return "STATUS";
case in_freeing:
return "FREEING";
case in_the_dark:
return "CLUELESS";
case in_abortone:
return "ABORTONE";
case in_abortall:
return "ABORTALL";
case in_resetdev:
return "RESETDEV";
case in_resetbus:
return "RESETBUS";
case in_tgterror:
return "TGTERROR";
default:
return "UNKNOWN";
};
}
#ifdef DEBUG_STATE_MACHINE
static inline void esp_advance_phase(Scsi_Cmnd *s, int newphase)
{
ESPLOG(("<%s>", phase_string(newphase)));
s->SCp.sent_command = s->SCp.phase;
s->SCp.phase = newphase;
}
#else
#define esp_advance_phase(__s, __newphase) \
(__s)->SCp.sent_command = (__s)->SCp.phase; \
(__s)->SCp.phase = (__newphase);
#endif
#ifdef DEBUG_ESP_CMDS
static inline void esp_cmd(struct NCR_ESP *esp, struct ESP_regs *eregs,
unchar cmd)
{
esp->espcmdlog[esp->espcmdent] = cmd;
esp->espcmdent = (esp->espcmdent + 1) & 31;
esp_write(eregs->esp_cmnd, cmd);
}
#else
#define esp_cmd(__esp, __eregs, __cmd) esp_write((__eregs)->esp_cmnd, (__cmd))
#endif
/* How we use the various Linux SCSI data structures for operation.
*
* struct scsi_cmnd:
*
* We keep track of the syncronous capabilities of a target
* in the device member, using sync_min_period and
* sync_max_offset. These are the values we directly write
* into the ESP registers while running a command. If offset
* is zero the ESP will use asynchronous transfers.
* If the borken flag is set we assume we shouldn't even bother
* trying to negotiate for synchronous transfer as this target
* is really stupid. If we notice the target is dropping the
* bus, and we have been allowing it to disconnect, we clear
* the disconnect flag.
*/
/* Manipulation of the ESP command queues. Thanks to the aha152x driver
* and its author, Juergen E. Fischer, for the methods used here.
* Note that these are per-ESP queues, not global queues like
* the aha152x driver uses.
*/
static inline void append_SC(Scsi_Cmnd **SC, Scsi_Cmnd *new_SC)
{
Scsi_Cmnd *end;
new_SC->host_scribble = (unsigned char *) NULL;
if(!*SC)
*SC = new_SC;
else {
for(end=*SC;end->host_scribble;end=(Scsi_Cmnd *)end->host_scribble)
;
end->host_scribble = (unsigned char *) new_SC;
}
}
static inline void prepend_SC(Scsi_Cmnd **SC, Scsi_Cmnd *new_SC)
{
new_SC->host_scribble = (unsigned char *) *SC;
*SC = new_SC;
}
static inline Scsi_Cmnd *remove_first_SC(Scsi_Cmnd **SC)
{
Scsi_Cmnd *ptr;
ptr = *SC;
if(ptr)
*SC = (Scsi_Cmnd *) (*SC)->host_scribble;
return ptr;
}
static inline Scsi_Cmnd *remove_SC(Scsi_Cmnd **SC, int target, int lun)
{
Scsi_Cmnd *ptr, *prev;
for(ptr = *SC, prev = NULL;
ptr && ((ptr->device->id != target) || (ptr->device->lun != lun));
prev = ptr, ptr = (Scsi_Cmnd *) ptr->host_scribble)
;
if(ptr) {
if(prev)
prev->host_scribble=ptr->host_scribble;
else
*SC=(Scsi_Cmnd *)ptr->host_scribble;
}
return ptr;
}
/* Resetting various pieces of the ESP scsi driver chipset */
/* Reset the ESP chip, _not_ the SCSI bus. */
static void esp_reset_esp(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
int family_code, version, i;
volatile int trash;
/* Now reset the ESP chip */
esp_cmd(esp, eregs, ESP_CMD_RC);
esp_cmd(esp, eregs, ESP_CMD_NULL | ESP_CMD_DMA);
if(esp->erev == fast)
esp_write(eregs->esp_cfg2, ESP_CONFIG2_FENAB);
esp_cmd(esp, eregs, ESP_CMD_NULL | ESP_CMD_DMA);
/* This is the only point at which it is reliable to read
* the ID-code for a fast ESP chip variant.
*/
esp->max_period = ((35 * esp->ccycle) / 1000);
if(esp->erev == fast) {
char *erev2string[] = {
"Emulex FAS236",
"Emulex FPESP100A",
"fast",
"QLogic FAS366",
"Emulex FAS216",
"Symbios Logic 53CF9x-2",
"unknown!"
};
version = esp_read(eregs->esp_uid);
family_code = (version & 0xf8) >> 3;
if(family_code == 0x02) {
if ((version & 7) == 2)
esp->erev = fas216;
else
esp->erev = fas236;
} else if(family_code == 0x0a)
esp->erev = fas366; /* Version is usually '5'. */
else if(family_code == 0x00) {
if ((version & 7) == 2)
esp->erev = fas100a; /* NCR53C9X */
else
esp->erev = espunknown;
} else if(family_code == 0x14) {
if ((version & 7) == 2)
esp->erev = fsc;
else
esp->erev = espunknown;
} else if(family_code == 0x00) {
if ((version & 7) == 2)
esp->erev = fas100a; /* NCR53C9X */
else
esp->erev = espunknown;
} else
esp->erev = espunknown;
ESPLOG(("esp%d: FAST chip is %s (family=%d, version=%d)\n",
esp->esp_id, erev2string[esp->erev - fas236],
family_code, (version & 7)));
esp->min_period = ((4 * esp->ccycle) / 1000);
} else {
esp->min_period = ((5 * esp->ccycle) / 1000);
}
/* Reload the configuration registers */
esp_write(eregs->esp_cfact, esp->cfact);
esp->prev_stp = 0;
esp_write(eregs->esp_stp, 0);
esp->prev_soff = 0;
esp_write(eregs->esp_soff, 0);
esp_write(eregs->esp_timeo, esp->neg_defp);
esp->max_period = (esp->max_period + 3)>>2;
esp->min_period = (esp->min_period + 3)>>2;
esp_write(eregs->esp_cfg1, esp->config1);
switch(esp->erev) {
case esp100:
/* nothing to do */
break;
case esp100a:
esp_write(eregs->esp_cfg2, esp->config2);
break;
case esp236:
/* Slow 236 */
esp_write(eregs->esp_cfg2, esp->config2);
esp->prev_cfg3 = esp->config3[0];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
break;
case fas366:
panic("esp: FAS366 support not present, please notify "
"jongk@cs.utwente.nl");
break;
case fas216:
case fas236:
case fsc:
/* Fast ESP variants */
esp_write(eregs->esp_cfg2, esp->config2);
for(i=0; i<8; i++)
esp->config3[i] |= ESP_CONFIG3_FCLK;
esp->prev_cfg3 = esp->config3[0];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
if(esp->diff)
esp->radelay = 0;
else
esp->radelay = 16;
/* Different timeout constant for these chips */
esp->neg_defp =
FSC_NEG_DEFP(esp->cfreq,
(esp->cfact == ESP_CCF_F0 ?
ESP_CCF_F7 + 1 : esp->cfact));
esp_write(eregs->esp_timeo, esp->neg_defp);
/* Enable Active Negotiation if possible */
if((esp->erev == fsc) && !esp->diff)
esp_write(eregs->esp_cfg4, ESP_CONFIG4_EAN);
break;
case fas100a:
/* Fast 100a */
esp_write(eregs->esp_cfg2, esp->config2);
for(i=0; i<8; i++)
esp->config3[i] |= ESP_CONFIG3_FCLOCK;
esp->prev_cfg3 = esp->config3[0];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
esp->radelay = 32;
break;
default:
panic("esp: what could it be... I wonder...");
break;
};
/* Eat any bitrot in the chip */
trash = esp_read(eregs->esp_intrpt);
udelay(100);
}
/* This places the ESP into a known state at boot time. */
void esp_bootup_reset(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
volatile unchar trash;
/* Reset the DMA */
if(esp->dma_reset)
esp->dma_reset(esp);
/* Reset the ESP */
esp_reset_esp(esp, eregs);
/* Reset the SCSI bus, but tell ESP not to generate an irq */
esp_write(eregs->esp_cfg1, (esp_read(eregs->esp_cfg1) | ESP_CONFIG1_SRRDISAB));
esp_cmd(esp, eregs, ESP_CMD_RS);
udelay(400);
esp_write(eregs->esp_cfg1, esp->config1);
/* Eat any bitrot in the chip and we are done... */
trash = esp_read(eregs->esp_intrpt);
}
EXPORT_SYMBOL(esp_bootup_reset);
/* Allocate structure and insert basic data such as SCSI chip frequency
* data and a pointer to the device
*/
struct NCR_ESP* esp_allocate(struct scsi_host_template *tpnt, void *esp_dev,
int hotplug)
{
struct NCR_ESP *esp, *elink;
struct Scsi_Host *esp_host;
if (hotplug)
esp_host = scsi_host_alloc(tpnt, sizeof(struct NCR_ESP));
else
esp_host = scsi_register(tpnt, sizeof(struct NCR_ESP));
if(!esp_host)
panic("Cannot register ESP SCSI host");
esp = (struct NCR_ESP *) esp_host->hostdata;
if(!esp)
panic("No esp in hostdata");
esp->ehost = esp_host;
esp->edev = esp_dev;
esp->esp_id = nesps++;
/* Set bitshift value (only used on Amiga with multiple ESPs) */
esp->shift = 2;
/* Put into the chain of esp chips detected */
if(espchain) {
elink = espchain;
while(elink->next) elink = elink->next;
elink->next = esp;
} else {
espchain = esp;
}
esp->next = NULL;
return esp;
}
void esp_deallocate(struct NCR_ESP *esp)
{
struct NCR_ESP *elink;
if(espchain == esp) {
espchain = NULL;
} else {
for(elink = espchain; elink && (elink->next != esp); elink = elink->next);
if(elink)
elink->next = esp->next;
}
nesps--;
}
/* Complete initialization of ESP structure and device
* Caller must have initialized appropriate parts of the ESP structure
* between the call to esp_allocate and this function.
*/
void esp_initialize(struct NCR_ESP *esp)
{
struct ESP_regs *eregs = esp->eregs;
unsigned int fmhz;
unchar ccf;
int i;
/* Check out the clock properties of the chip. */
/* This is getting messy but it has to be done
* correctly or else you get weird behavior all
* over the place. We are trying to basically
* figure out three pieces of information.
*
* a) Clock Conversion Factor
*
* This is a representation of the input
* crystal clock frequency going into the
* ESP on this machine. Any operation whose
* timing is longer than 400ns depends on this
* value being correct. For example, you'll
* get blips for arbitration/selection during
* high load or with multiple targets if this
* is not set correctly.
*
* b) Selection Time-Out
*
* The ESP isn't very bright and will arbitrate
* for the bus and try to select a target
* forever if you let it. This value tells
* the ESP when it has taken too long to
* negotiate and that it should interrupt
* the CPU so we can see what happened.
* The value is computed as follows (from
* NCR/Symbios chip docs).
*
* (Time Out Period) * (Input Clock)
* STO = ----------------------------------
* (8192) * (Clock Conversion Factor)
*
* You usually want the time out period to be
* around 250ms, I think we'll set it a little
* bit higher to account for fully loaded SCSI
* bus's and slow devices that don't respond so
* quickly to selection attempts. (yeah, I know
* this is out of spec. but there is a lot of
* buggy pieces of firmware out there so bite me)
*
* c) Imperical constants for synchronous offset
* and transfer period register values
*
* This entails the smallest and largest sync
* period we could ever handle on this ESP.
*/
fmhz = esp->cfreq;
if(fmhz <= (5000000))
ccf = 0;
else
ccf = (((5000000 - 1) + (fmhz))/(5000000));
if(!ccf || ccf > 8) {
/* If we can't find anything reasonable,
* just assume 20MHZ. This is the clock
* frequency of the older sun4c's where I've
* been unable to find the clock-frequency
* PROM property. All other machines provide
* useful values it seems.
*/
ccf = ESP_CCF_F4;
fmhz = (20000000);
}
if(ccf==(ESP_CCF_F7+1))
esp->cfact = ESP_CCF_F0;
else if(ccf == ESP_CCF_NEVER)
esp->cfact = ESP_CCF_F2;
else
esp->cfact = ccf;
esp->cfreq = fmhz;
esp->ccycle = ESP_MHZ_TO_CYCLE(fmhz);
esp->ctick = ESP_TICK(ccf, esp->ccycle);
esp->neg_defp = ESP_NEG_DEFP(fmhz, ccf);
esp->sync_defp = SYNC_DEFP_SLOW;
printk("SCSI ID %d Clk %dMHz CCF=%d TOut %d ",
esp->scsi_id, (esp->cfreq / 1000000),
ccf, (int) esp->neg_defp);
/* Fill in ehost data */
esp->ehost->base = (unsigned long)eregs;
esp->ehost->this_id = esp->scsi_id;
esp->ehost->irq = esp->irq;
/* SCSI id mask */
esp->scsi_id_mask = (1 << esp->scsi_id);
/* Probe the revision of this esp */
esp->config1 = (ESP_CONFIG1_PENABLE | (esp->scsi_id & 7));
esp->config2 = (ESP_CONFIG2_SCSI2ENAB | ESP_CONFIG2_REGPARITY);
esp_write(eregs->esp_cfg2, esp->config2);
if((esp_read(eregs->esp_cfg2) & ~(ESP_CONFIG2_MAGIC)) !=
(ESP_CONFIG2_SCSI2ENAB | ESP_CONFIG2_REGPARITY)) {
printk("NCR53C90(esp100)\n");
esp->erev = esp100;
} else {
esp->config2 = 0;
esp_write(eregs->esp_cfg2, 0);
esp_write(eregs->esp_cfg3, 5);
if(esp_read(eregs->esp_cfg3) != 5) {
printk("NCR53C90A(esp100a)\n");
esp->erev = esp100a;
} else {
int target;
for(target=0; target<8; target++)
esp->config3[target] = 0;
esp->prev_cfg3 = 0;
esp_write(eregs->esp_cfg3, 0);
if(ccf > ESP_CCF_F5) {
printk("NCR53C9XF(espfast)\n");
esp->erev = fast;
esp->sync_defp = SYNC_DEFP_FAST;
} else {
printk("NCR53C9x(esp236)\n");
esp->erev = esp236;
}
}
}
/* Initialize the command queues */
esp->current_SC = NULL;
esp->disconnected_SC = NULL;
esp->issue_SC = NULL;
/* Clear the state machines. */
esp->targets_present = 0;
esp->resetting_bus = 0;
esp->snip = 0;
init_waitqueue_head(&esp->reset_queue);
esp->fas_premature_intr_workaround = 0;
for(i = 0; i < 32; i++)
esp->espcmdlog[i] = 0;
esp->espcmdent = 0;
for(i = 0; i < 16; i++) {
esp->cur_msgout[i] = 0;
esp->cur_msgin[i] = 0;
}
esp->prevmsgout = esp->prevmsgin = 0;
esp->msgout_len = esp->msgin_len = 0;
/* Clear the one behind caches to hold unmatchable values. */
esp->prev_soff = esp->prev_stp = esp->prev_cfg3 = 0xff;
/* Reset the thing before we try anything... */
esp_bootup_reset(esp, eregs);
esps_in_use++;
}
/* The info function will return whatever useful
* information the developer sees fit. If not provided, then
* the name field will be used instead.
*/
const char *esp_info(struct Scsi_Host *host)
{
struct NCR_ESP *esp;
esp = (struct NCR_ESP *) host->hostdata;
switch(esp->erev) {
case esp100:
return "ESP100 (NCR53C90)";
case esp100a:
return "ESP100A (NCR53C90A)";
case esp236:
return "ESP236 (NCR53C9x)";
case fas216:
return "Emulex FAS216";
case fas236:
return "Emulex FAS236";
case fas366:
return "QLogic FAS366";
case fas100a:
return "FPESP100A";
case fsc:
return "Symbios Logic 53CF9x-2";
default:
panic("Bogon ESP revision");
};
}
EXPORT_SYMBOL(esp_info);
/* From Wolfgang Stanglmeier's NCR scsi driver. */
struct info_str
{
char *buffer;
int length;
int offset;
int pos;
};
static void copy_mem_info(struct info_str *info, char *data, int len)
{
if (info->pos + len > info->length)
len = info->length - info->pos;
if (info->pos + len < info->offset) {
info->pos += len;
return;
}
if (info->pos < info->offset) {
data += (info->offset - info->pos);
len -= (info->offset - info->pos);
}
if (len > 0) {
memcpy(info->buffer + info->pos, data, len);
info->pos += len;
}
}
static int copy_info(struct info_str *info, char *fmt, ...)
{
va_list args;
char buf[81];
int len;
va_start(args, fmt);
len = vsprintf(buf, fmt, args);
va_end(args);
copy_mem_info(info, buf, len);
return len;
}
static int esp_host_info(struct NCR_ESP *esp, char *ptr, off_t offset, int len)
{
struct scsi_device *sdev;
struct info_str info;
int i;
info.buffer = ptr;
info.length = len;
info.offset = offset;
info.pos = 0;
copy_info(&info, "ESP Host Adapter:\n");
copy_info(&info, "\tESP Model\t\t");
switch(esp->erev) {
case esp100:
copy_info(&info, "ESP100 (NCR53C90)\n");
break;
case esp100a:
copy_info(&info, "ESP100A (NCR53C90A)\n");
break;
case esp236:
copy_info(&info, "ESP236 (NCR53C9x)\n");
break;
case fas216:
copy_info(&info, "Emulex FAS216\n");
break;
case fas236:
copy_info(&info, "Emulex FAS236\n");
break;
case fas100a:
copy_info(&info, "FPESP100A\n");
break;
case fast:
copy_info(&info, "Generic FAST\n");
break;
case fas366:
copy_info(&info, "QLogic FAS366\n");
break;
case fsc:
copy_info(&info, "Symbios Logic 53C9x-2\n");
break;
case espunknown:
default:
copy_info(&info, "Unknown!\n");
break;
};
copy_info(&info, "\tLive Targets\t\t[ ");
for(i = 0; i < 15; i++) {
if(esp->targets_present & (1 << i))
copy_info(&info, "%d ", i);
}
copy_info(&info, "]\n\n");
/* Now describe the state of each existing target. */
copy_info(&info, "Target #\tconfig3\t\tSync Capabilities\tDisconnect\n");
shost_for_each_device(sdev, esp->ehost) {
struct esp_device *esp_dev = sdev->hostdata;
uint id = sdev->id;
if (!(esp->targets_present & (1 << id)))
continue;
copy_info(&info, "%d\t\t", id);
copy_info(&info, "%08lx\t", esp->config3[id]);
copy_info(&info, "[%02lx,%02lx]\t\t\t",
esp_dev->sync_max_offset,
esp_dev->sync_min_period);
copy_info(&info, "%s\n", esp_dev->disconnect ? "yes" : "no");
}
return info.pos > info.offset? info.pos - info.offset : 0;
}
/* ESP proc filesystem code. */
int esp_proc_info(struct Scsi_Host *shost, char *buffer, char **start, off_t offset, int length,
int inout)
{
struct NCR_ESP *esp = (struct NCR_ESP *)shost->hostdata;
if(inout)
return -EINVAL; /* not yet */
if(start)
*start = buffer;
return esp_host_info(esp, buffer, offset, length);
}
EXPORT_SYMBOL(esp_proc_info);
static void esp_get_dmabufs(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
if(sp->use_sg == 0) {
sp->SCp.this_residual = sp->request_bufflen;
sp->SCp.buffer = (struct scatterlist *) sp->request_buffer;
sp->SCp.buffers_residual = 0;
if (esp->dma_mmu_get_scsi_one)
esp->dma_mmu_get_scsi_one(esp, sp);
else
sp->SCp.ptr =
(char *) virt_to_phys(sp->request_buffer);
} else {
sp->SCp.buffer = (struct scatterlist *) sp->request_buffer;
sp->SCp.buffers_residual = sp->use_sg - 1;
sp->SCp.this_residual = sp->SCp.buffer->length;
if (esp->dma_mmu_get_scsi_sgl)
esp->dma_mmu_get_scsi_sgl(esp, sp);
else
sp->SCp.ptr =
(char *) virt_to_phys((page_address(sp->SCp.buffer->page) + sp->SCp.buffer->offset));
}
}
static void esp_release_dmabufs(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
if(sp->use_sg == 0) {
if (esp->dma_mmu_release_scsi_one)
esp->dma_mmu_release_scsi_one(esp, sp);
} else {
if (esp->dma_mmu_release_scsi_sgl)
esp->dma_mmu_release_scsi_sgl(esp, sp);
}
}
static void esp_restore_pointers(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
struct esp_pointers *ep = &esp->data_pointers[scmd_id(sp)];
sp->SCp.ptr = ep->saved_ptr;
sp->SCp.buffer = ep->saved_buffer;
sp->SCp.this_residual = ep->saved_this_residual;
sp->SCp.buffers_residual = ep->saved_buffers_residual;
}
static void esp_save_pointers(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
struct esp_pointers *ep = &esp->data_pointers[scmd_id(sp)];
ep->saved_ptr = sp->SCp.ptr;
ep->saved_buffer = sp->SCp.buffer;
ep->saved_this_residual = sp->SCp.this_residual;
ep->saved_buffers_residual = sp->SCp.buffers_residual;
}
/* Some rules:
*
* 1) Never ever panic while something is live on the bus.
* If there is to be any chance of syncing the disks this
* rule is to be obeyed.
*
* 2) Any target that causes a foul condition will no longer
* have synchronous transfers done to it, no questions
* asked.
*
* 3) Keep register accesses to a minimum. Think about some
* day when we have Xbus machines this is running on and
* the ESP chip is on the other end of the machine on a
* different board from the cpu where this is running.
*/
/* Fire off a command. We assume the bus is free and that the only
* case where we could see an interrupt is where we have disconnected
* commands active and they are trying to reselect us.
*/
static inline void esp_check_cmd(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
switch(sp->cmd_len) {
case 6:
case 10:
case 12:
esp->esp_slowcmd = 0;
break;
default:
esp->esp_slowcmd = 1;
esp->esp_scmdleft = sp->cmd_len;
esp->esp_scmdp = &sp->cmnd[0];
break;
};
}
static inline void build_sync_nego_msg(struct NCR_ESP *esp, int period, int offset)
{
esp->cur_msgout[0] = EXTENDED_MESSAGE;
esp->cur_msgout[1] = 3;
esp->cur_msgout[2] = EXTENDED_SDTR;
esp->cur_msgout[3] = period;
esp->cur_msgout[4] = offset;
esp->msgout_len = 5;
}
static void esp_exec_cmd(struct NCR_ESP *esp)
{
struct ESP_regs *eregs = esp->eregs;
struct esp_device *esp_dev;
Scsi_Cmnd *SCptr;
struct scsi_device *SDptr;
volatile unchar *cmdp = esp->esp_command;
unsigned char the_esp_command;
int lun, target;
int i;
/* Hold off if we have disconnected commands and
* an IRQ is showing...
*/
if(esp->disconnected_SC && esp->dma_irq_p(esp))
return;
/* Grab first member of the issue queue. */
SCptr = esp->current_SC = remove_first_SC(&esp->issue_SC);
/* Safe to panic here because current_SC is null. */
if(!SCptr)
panic("esp: esp_exec_cmd and issue queue is NULL");
SDptr = SCptr->device;
esp_dev = SDptr->hostdata;
lun = SCptr->device->lun;
target = SCptr->device->id;
esp->snip = 0;
esp->msgout_len = 0;
/* Send it out whole, or piece by piece? The ESP
* only knows how to automatically send out 6, 10,
* and 12 byte commands. I used to think that the
* Linux SCSI code would never throw anything other
* than that to us, but then again there is the
* SCSI generic driver which can send us anything.
*/
esp_check_cmd(esp, SCptr);
/* If arbitration/selection is successful, the ESP will leave
* ATN asserted, causing the target to go into message out
* phase. The ESP will feed the target the identify and then
* the target can only legally go to one of command,
* datain/out, status, or message in phase, or stay in message
* out phase (should we be trying to send a sync negotiation
* message after the identify). It is not allowed to drop
* BSY, but some buggy targets do and we check for this
* condition in the selection complete code. Most of the time
* we'll make the command bytes available to the ESP and it
* will not interrupt us until it finishes command phase, we
* cannot do this for command sizes the ESP does not
* understand and in this case we'll get interrupted right
* when the target goes into command phase.
*
* It is absolutely _illegal_ in the presence of SCSI-2 devices
* to use the ESP select w/o ATN command. When SCSI-2 devices are
* present on the bus we _must_ always go straight to message out
* phase with an identify message for the target. Being that
* selection attempts in SCSI-1 w/o ATN was an option, doing SCSI-2
* selections should not confuse SCSI-1 we hope.
*/
if(esp_dev->sync) {
/* this targets sync is known */
#ifdef CONFIG_SCSI_MAC_ESP
do_sync_known:
#endif
if(esp_dev->disconnect)
*cmdp++ = IDENTIFY(1, lun);
else
*cmdp++ = IDENTIFY(0, lun);
if(esp->esp_slowcmd) {
the_esp_command = (ESP_CMD_SELAS | ESP_CMD_DMA);
esp_advance_phase(SCptr, in_slct_stop);
} else {
the_esp_command = (ESP_CMD_SELA | ESP_CMD_DMA);
esp_advance_phase(SCptr, in_slct_norm);
}
} else if(!(esp->targets_present & (1<<target)) || !(esp_dev->disconnect)) {
/* After the bootup SCSI code sends both the
* TEST_UNIT_READY and INQUIRY commands we want
* to at least attempt allowing the device to
* disconnect.
*/
ESPMISC(("esp: Selecting device for first time. target=%d "
"lun=%d\n", target, SCptr->device->lun));
if(!SDptr->borken && !esp_dev->disconnect)
esp_dev->disconnect = 1;
*cmdp++ = IDENTIFY(0, lun);
esp->prevmsgout = NOP;
esp_advance_phase(SCptr, in_slct_norm);
the_esp_command = (ESP_CMD_SELA | ESP_CMD_DMA);
/* Take no chances... */
esp_dev->sync_max_offset = 0;
esp_dev->sync_min_period = 0;
} else {
int toshiba_cdrom_hwbug_wkaround = 0;
#ifdef CONFIG_SCSI_MAC_ESP
/* Never allow synchronous transfers (disconnect OK) on
* Macintosh. Well, maybe later when we figured out how to
* do DMA on the machines that support it ...
*/
esp_dev->disconnect = 1;
esp_dev->sync_max_offset = 0;
esp_dev->sync_min_period = 0;
esp_dev->sync = 1;
esp->snip = 0;
goto do_sync_known;
#endif
/* We've talked to this guy before,
* but never negotiated. Let's try
* sync negotiation.
*/
if(!SDptr->borken) {
if((SDptr->type == TYPE_ROM) &&
(!strncmp(SDptr->vendor, "TOSHIBA", 7))) {
/* Nice try sucker... */
ESPMISC(("esp%d: Disabling sync for buggy "
"Toshiba CDROM.\n", esp->esp_id));
toshiba_cdrom_hwbug_wkaround = 1;
build_sync_nego_msg(esp, 0, 0);
} else {
build_sync_nego_msg(esp, esp->sync_defp, 15);
}
} else {
build_sync_nego_msg(esp, 0, 0);
}
esp_dev->sync = 1;
esp->snip = 1;
/* A fix for broken SCSI1 targets, when they disconnect
* they lock up the bus and confuse ESP. So disallow
* disconnects for SCSI1 targets for now until we
* find a better fix.
*
* Addendum: This is funny, I figured out what was going
* on. The blotzed SCSI1 target would disconnect,
* one of the other SCSI2 targets or both would be
* disconnected as well. The SCSI1 target would
* stay disconnected long enough that we start
* up a command on one of the SCSI2 targets. As
* the ESP is arbitrating for the bus the SCSI1
* target begins to arbitrate as well to reselect
* the ESP. The SCSI1 target refuses to drop it's
* ID bit on the data bus even though the ESP is
* at ID 7 and is the obvious winner for any
* arbitration. The ESP is a poor sport and refuses
* to lose arbitration, it will continue indefinitely
* trying to arbitrate for the bus and can only be
* stopped via a chip reset or SCSI bus reset.
* Therefore _no_ disconnects for SCSI1 targets
* thank you very much. ;-)
*/
if(((SDptr->scsi_level < 3) && (SDptr->type != TYPE_TAPE)) ||
toshiba_cdrom_hwbug_wkaround || SDptr->borken) {
ESPMISC((KERN_INFO "esp%d: Disabling DISCONNECT for target %d "
"lun %d\n", esp->esp_id, SCptr->device->id, SCptr->device->lun));
esp_dev->disconnect = 0;
*cmdp++ = IDENTIFY(0, lun);
} else {
*cmdp++ = IDENTIFY(1, lun);
}
/* ESP fifo is only so big...
* Make this look like a slow command.
*/
esp->esp_slowcmd = 1;
esp->esp_scmdleft = SCptr->cmd_len;
esp->esp_scmdp = &SCptr->cmnd[0];
the_esp_command = (ESP_CMD_SELAS | ESP_CMD_DMA);
esp_advance_phase(SCptr, in_slct_msg);
}
if(!esp->esp_slowcmd)
for(i = 0; i < SCptr->cmd_len; i++)
*cmdp++ = SCptr->cmnd[i];
esp_write(eregs->esp_busid, (target & 7));
if (esp->prev_soff != esp_dev->sync_max_offset ||
esp->prev_stp != esp_dev->sync_min_period ||
(esp->erev > esp100a &&
esp->prev_cfg3 != esp->config3[target])) {
esp->prev_soff = esp_dev->sync_max_offset;
esp_write(eregs->esp_soff, esp->prev_soff);
esp->prev_stp = esp_dev->sync_min_period;
esp_write(eregs->esp_stp, esp->prev_stp);
if(esp->erev > esp100a) {
esp->prev_cfg3 = esp->config3[target];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
}
}
i = (cmdp - esp->esp_command);
/* Set up the DMA and ESP counters */
if(esp->do_pio_cmds){
int j = 0;
/*
* XXX MSch:
*
* It seems this is required, at least to clean up
* after failed commands when using PIO mode ...
*/
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
for(;j<i;j++)
esp_write(eregs->esp_fdata, esp->esp_command[j]);
the_esp_command &= ~ESP_CMD_DMA;
/* Tell ESP to "go". */
esp_cmd(esp, eregs, the_esp_command);
} else {
/* Set up the ESP counters */
esp_write(eregs->esp_tclow, i);
esp_write(eregs->esp_tcmed, 0);
esp->dma_init_write(esp, esp->esp_command_dvma, i);
/* Tell ESP to "go". */
esp_cmd(esp, eregs, the_esp_command);
}
}
/* Queue a SCSI command delivered from the mid-level Linux SCSI code. */
int esp_queue(Scsi_Cmnd *SCpnt, void (*done)(Scsi_Cmnd *))
{
struct NCR_ESP *esp;
/* Set up func ptr and initial driver cmd-phase. */
SCpnt->scsi_done = done;
SCpnt->SCp.phase = not_issued;
esp = (struct NCR_ESP *) SCpnt->device->host->hostdata;
if(esp->dma_led_on)
esp->dma_led_on(esp);
/* We use the scratch area. */
ESPQUEUE(("esp_queue: target=%d lun=%d ", SCpnt->device->id, SCpnt->lun));
ESPDISC(("N<%02x,%02x>", SCpnt->device->id, SCpnt->lun));
esp_get_dmabufs(esp, SCpnt);
esp_save_pointers(esp, SCpnt); /* FIXME for tag queueing */
SCpnt->SCp.Status = CHECK_CONDITION;
SCpnt->SCp.Message = 0xff;
SCpnt->SCp.sent_command = 0;
/* Place into our queue. */
if(SCpnt->cmnd[0] == REQUEST_SENSE) {
ESPQUEUE(("RQSENSE\n"));
prepend_SC(&esp->issue_SC, SCpnt);
} else {
ESPQUEUE(("\n"));
append_SC(&esp->issue_SC, SCpnt);
}
/* Run it now if we can. */
if(!esp->current_SC && !esp->resetting_bus)
esp_exec_cmd(esp);
return 0;
}
/* Dump driver state. */
static void esp_dump_cmd(Scsi_Cmnd *SCptr)
{
ESPLOG(("[tgt<%02x> lun<%02x> "
"pphase<%s> cphase<%s>]",
SCptr->device->id, SCptr->device->lun,
phase_string(SCptr->SCp.sent_command),
phase_string(SCptr->SCp.phase)));
}
static void esp_dump_state(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
#ifdef DEBUG_ESP_CMDS
int i;
#endif
ESPLOG(("esp%d: dumping state\n", esp->esp_id));
/* Print DMA status */
esp->dma_dump_state(esp);
ESPLOG(("esp%d: SW [sreg<%02x> sstep<%02x> ireg<%02x>]\n",
esp->esp_id, esp->sreg, esp->seqreg, esp->ireg));
ESPLOG(("esp%d: HW reread [sreg<%02x> sstep<%02x> ireg<%02x>]\n",
esp->esp_id, esp_read(eregs->esp_status), esp_read(eregs->esp_sstep),
esp_read(eregs->esp_intrpt)));
#ifdef DEBUG_ESP_CMDS
printk("esp%d: last ESP cmds [", esp->esp_id);
i = (esp->espcmdent - 1) & 31;
printk("<");
esp_print_cmd(esp->espcmdlog[i]);
printk(">");
i = (i - 1) & 31;
printk("<");
esp_print_cmd(esp->espcmdlog[i]);
printk(">");
i = (i - 1) & 31;
printk("<");
esp_print_cmd(esp->espcmdlog[i]);
printk(">");
i = (i - 1) & 31;
printk("<");
esp_print_cmd(esp->espcmdlog[i]);
printk(">");
printk("]\n");
#endif /* (DEBUG_ESP_CMDS) */
if(SCptr) {
ESPLOG(("esp%d: current command ", esp->esp_id));
esp_dump_cmd(SCptr);
}
ESPLOG(("\n"));
SCptr = esp->disconnected_SC;
ESPLOG(("esp%d: disconnected ", esp->esp_id));
while(SCptr) {
esp_dump_cmd(SCptr);
SCptr = (Scsi_Cmnd *) SCptr->host_scribble;
}
ESPLOG(("\n"));
}
/* Abort a command. The host_lock is acquired by caller. */
int esp_abort(Scsi_Cmnd *SCptr)
{
struct NCR_ESP *esp = (struct NCR_ESP *) SCptr->device->host->hostdata;
struct ESP_regs *eregs = esp->eregs;
int don;
ESPLOG(("esp%d: Aborting command\n", esp->esp_id));
esp_dump_state(esp, eregs);
/* Wheee, if this is the current command on the bus, the
* best we can do is assert ATN and wait for msgout phase.
* This should even fix a hung SCSI bus when we lose state
* in the driver and timeout because the eventual phase change
* will cause the ESP to (eventually) give an interrupt.
*/
if(esp->current_SC == SCptr) {
esp->cur_msgout[0] = ABORT;
esp->msgout_len = 1;
esp->msgout_ctr = 0;
esp_cmd(esp, eregs, ESP_CMD_SATN);
return SUCCESS;
}
/* If it is still in the issue queue then we can safely
* call the completion routine and report abort success.
*/
don = esp->dma_ports_p(esp);
if(don) {
esp->dma_ints_off(esp);
synchronize_irq(esp->irq);
}
if(esp->issue_SC) {
Scsi_Cmnd **prev, *this;
for(prev = (&esp->issue_SC), this = esp->issue_SC;
this;
prev = (Scsi_Cmnd **) &(this->host_scribble),
this = (Scsi_Cmnd *) this->host_scribble) {
if(this == SCptr) {
*prev = (Scsi_Cmnd *) this->host_scribble;
this->host_scribble = NULL;
esp_release_dmabufs(esp, this);
this->result = DID_ABORT << 16;
this->scsi_done(this);
if(don)
esp->dma_ints_on(esp);
return SUCCESS;
}
}
}
/* Yuck, the command to abort is disconnected, it is not
* worth trying to abort it now if something else is live
* on the bus at this time. So, we let the SCSI code wait
* a little bit and try again later.
*/
if(esp->current_SC) {
if(don)
esp->dma_ints_on(esp);
return FAILED;
}
/* It's disconnected, we have to reconnect to re-establish
* the nexus and tell the device to abort. However, we really
* cannot 'reconnect' per se. Don't try to be fancy, just
* indicate failure, which causes our caller to reset the whole
* bus.
*/
if(don)
esp->dma_ints_on(esp);
return FAILED;
}
/* We've sent ESP_CMD_RS to the ESP, the interrupt had just
* arrived indicating the end of the SCSI bus reset. Our job
* is to clean out the command queues and begin re-execution
* of SCSI commands once more.
*/
static int esp_finish_reset(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
Scsi_Cmnd *sp = esp->current_SC;
/* Clean up currently executing command, if any. */
if (sp != NULL) {
esp_release_dmabufs(esp, sp);
sp->result = (DID_RESET << 16);
sp->scsi_done(sp);
esp->current_SC = NULL;
}
/* Clean up disconnected queue, they have been invalidated
* by the bus reset.
*/
if (esp->disconnected_SC) {
while((sp = remove_first_SC(&esp->disconnected_SC)) != NULL) {
esp_release_dmabufs(esp, sp);
sp->result = (DID_RESET << 16);
sp->scsi_done(sp);
}
}
/* SCSI bus reset is complete. */
esp->resetting_bus = 0;
wake_up(&esp->reset_queue);
/* Ok, now it is safe to get commands going once more. */
if(esp->issue_SC)
esp_exec_cmd(esp);
return do_intr_end;
}
static int esp_do_resetbus(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
ESPLOG(("esp%d: Resetting scsi bus\n", esp->esp_id));
esp->resetting_bus = 1;
esp_cmd(esp, eregs, ESP_CMD_RS);
return do_intr_end;
}
/* Reset ESP chip, reset hanging bus, then kill active and
* disconnected commands for targets without soft reset.
*
* The host_lock is acquired by caller.
*/
int esp_reset(Scsi_Cmnd *SCptr)
{
struct NCR_ESP *esp = (struct NCR_ESP *) SCptr->device->host->hostdata;
spin_lock_irq(esp->ehost->host_lock);
(void) esp_do_resetbus(esp, esp->eregs);
spin_unlock_irq(esp->ehost->host_lock);
wait_event(esp->reset_queue, (esp->resetting_bus == 0));
return SUCCESS;
}
/* Internal ESP done function. */
static void esp_done(struct NCR_ESP *esp, int error)
{
Scsi_Cmnd *done_SC;
if(esp->current_SC) {
done_SC = esp->current_SC;
esp->current_SC = NULL;
esp_release_dmabufs(esp, done_SC);
done_SC->result = error;
done_SC->scsi_done(done_SC);
/* Bus is free, issue any commands in the queue. */
if(esp->issue_SC && !esp->current_SC)
esp_exec_cmd(esp);
} else {
/* Panic is safe as current_SC is null so we may still
* be able to accept more commands to sync disk buffers.
*/
ESPLOG(("panicing\n"));
panic("esp: done() called with NULL esp->current_SC");
}
}
/* Wheee, ESP interrupt engine. */
/* Forward declarations. */
static int esp_do_phase_determine(struct NCR_ESP *esp,
struct ESP_regs *eregs);
static int esp_do_data_finale(struct NCR_ESP *esp, struct ESP_regs *eregs);
static int esp_select_complete(struct NCR_ESP *esp, struct ESP_regs *eregs);
static int esp_do_status(struct NCR_ESP *esp, struct ESP_regs *eregs);
static int esp_do_msgin(struct NCR_ESP *esp, struct ESP_regs *eregs);
static int esp_do_msgindone(struct NCR_ESP *esp, struct ESP_regs *eregs);
static int esp_do_msgout(struct NCR_ESP *esp, struct ESP_regs *eregs);
static int esp_do_cmdbegin(struct NCR_ESP *esp, struct ESP_regs *eregs);
#define sreg_datainp(__sreg) (((__sreg) & ESP_STAT_PMASK) == ESP_DIP)
#define sreg_dataoutp(__sreg) (((__sreg) & ESP_STAT_PMASK) == ESP_DOP)
/* We try to avoid some interrupts by jumping ahead and see if the ESP
* has gotten far enough yet. Hence the following.
*/
static inline int skipahead1(struct NCR_ESP *esp, struct ESP_regs *eregs,
Scsi_Cmnd *scp, int prev_phase, int new_phase)
{
if(scp->SCp.sent_command != prev_phase)
return 0;
if(esp->dma_irq_p(esp)) {
/* Yes, we are able to save an interrupt. */
esp->sreg = (esp_read(eregs->esp_status) & ~(ESP_STAT_INTR));
esp->ireg = esp_read(eregs->esp_intrpt);
if(!(esp->ireg & ESP_INTR_SR))
return 0;
else
return do_reset_complete;
}
/* Ho hum, target is taking forever... */
scp->SCp.sent_command = new_phase; /* so we don't recurse... */
return do_intr_end;
}
static inline int skipahead2(struct NCR_ESP *esp,
struct ESP_regs *eregs,
Scsi_Cmnd *scp, int prev_phase1, int prev_phase2,
int new_phase)
{
if(scp->SCp.sent_command != prev_phase1 &&
scp->SCp.sent_command != prev_phase2)
return 0;
if(esp->dma_irq_p(esp)) {
/* Yes, we are able to save an interrupt. */
esp->sreg = (esp_read(eregs->esp_status) & ~(ESP_STAT_INTR));
esp->ireg = esp_read(eregs->esp_intrpt);
if(!(esp->ireg & ESP_INTR_SR))
return 0;
else
return do_reset_complete;
}
/* Ho hum, target is taking forever... */
scp->SCp.sent_command = new_phase; /* so we don't recurse... */
return do_intr_end;
}
/* Misc. esp helper macros. */
#define esp_setcount(__eregs, __cnt) \
esp_write((__eregs)->esp_tclow, ((__cnt) & 0xff)); \
esp_write((__eregs)->esp_tcmed, (((__cnt) >> 8) & 0xff))
#define esp_getcount(__eregs) \
((esp_read((__eregs)->esp_tclow)&0xff) | \
((esp_read((__eregs)->esp_tcmed)&0xff) << 8))
#define fcount(__esp, __eregs) \
(esp_read((__eregs)->esp_fflags) & ESP_FF_FBYTES)
#define fnzero(__esp, __eregs) \
(esp_read((__eregs)->esp_fflags) & ESP_FF_ONOTZERO)
/* XXX speculative nops unnecessary when continuing amidst a data phase
* XXX even on esp100!!! another case of flooding the bus with I/O reg
* XXX writes...
*/
#define esp_maybe_nop(__esp, __eregs) \
if((__esp)->erev == esp100) \
esp_cmd((__esp), (__eregs), ESP_CMD_NULL)
#define sreg_to_dataphase(__sreg) \
((((__sreg) & ESP_STAT_PMASK) == ESP_DOP) ? in_dataout : in_datain)
/* The ESP100 when in synchronous data phase, can mistake a long final
* REQ pulse from the target as an extra byte, it places whatever is on
* the data lines into the fifo. For now, we will assume when this
* happens that the target is a bit quirky and we don't want to
* be talking synchronously to it anyways. Regardless, we need to
* tell the ESP to eat the extraneous byte so that we can proceed
* to the next phase.
*/
static inline int esp100_sync_hwbug(struct NCR_ESP *esp, struct ESP_regs *eregs,
Scsi_Cmnd *sp, int fifocnt)
{
/* Do not touch this piece of code. */
if((!(esp->erev == esp100)) ||
(!(sreg_datainp((esp->sreg = esp_read(eregs->esp_status))) && !fifocnt) &&
!(sreg_dataoutp(esp->sreg) && !fnzero(esp, eregs)))) {
if(sp->SCp.phase == in_dataout)
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
return 0;
} else {
/* Async mode for this guy. */
build_sync_nego_msg(esp, 0, 0);
/* Ack the bogus byte, but set ATN first. */
esp_cmd(esp, eregs, ESP_CMD_SATN);
esp_cmd(esp, eregs, ESP_CMD_MOK);
return 1;
}
}
/* This closes the window during a selection with a reselect pending, because
* we use DMA for the selection process the FIFO should hold the correct
* contents if we get reselected during this process. So we just need to
* ack the possible illegal cmd interrupt pending on the esp100.
*/
static inline int esp100_reconnect_hwbug(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
volatile unchar junk;
if(esp->erev != esp100)
return 0;
junk = esp_read(eregs->esp_intrpt);
if(junk & ESP_INTR_SR)
return 1;
return 0;
}
/* This verifies the BUSID bits during a reselection so that we know which
* target is talking to us.
*/
static inline int reconnect_target(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
int it, me = esp->scsi_id_mask, targ = 0;
if(2 != fcount(esp, eregs))
return -1;
it = esp_read(eregs->esp_fdata);
if(!(it & me))
return -1;
it &= ~me;
if(it & (it - 1))
return -1;
while(!(it & 1))
targ++, it >>= 1;
return targ;
}
/* This verifies the identify from the target so that we know which lun is
* being reconnected.
*/
static inline int reconnect_lun(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
int lun;
if((esp->sreg & ESP_STAT_PMASK) != ESP_MIP)
return -1;
lun = esp_read(eregs->esp_fdata);
/* Yes, you read this correctly. We report lun of zero
* if we see parity error. ESP reports parity error for
* the lun byte, and this is the only way to hope to recover
* because the target is connected.
*/
if(esp->sreg & ESP_STAT_PERR)
return 0;
/* Check for illegal bits being set in the lun. */
if((lun & 0x40) || !(lun & 0x80))
return -1;
return lun & 7;
}
/* This puts the driver in a state where it can revitalize a command that
* is being continued due to reselection.
*/
static inline void esp_connect(struct NCR_ESP *esp, struct ESP_regs *eregs,
Scsi_Cmnd *sp)
{
struct scsi_device *dp = sp->device;
struct esp_device *esp_dev = dp->hostdata;
if(esp->prev_soff != esp_dev->sync_max_offset ||
esp->prev_stp != esp_dev->sync_min_period ||
(esp->erev > esp100a &&
esp->prev_cfg3 != esp->config3[scmd_id(sp)])) {
esp->prev_soff = esp_dev->sync_max_offset;
esp_write(eregs->esp_soff, esp->prev_soff);
esp->prev_stp = esp_dev->sync_min_period;
esp_write(eregs->esp_stp, esp->prev_stp);
if(esp->erev > esp100a) {
esp->prev_cfg3 = esp->config3[scmd_id(sp)];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
}
}
esp->current_SC = sp;
}
/* This will place the current working command back into the issue queue
* if we are to receive a reselection amidst a selection attempt.
*/
static inline void esp_reconnect(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
if(!esp->disconnected_SC)
ESPLOG(("esp%d: Weird, being reselected but disconnected "
"command queue is empty.\n", esp->esp_id));
esp->snip = 0;
esp->current_SC = NULL;
sp->SCp.phase = not_issued;
append_SC(&esp->issue_SC, sp);
}
/* Begin message in phase. */
static int esp_do_msgin(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
esp_maybe_nop(esp, eregs);
esp_cmd(esp, eregs, ESP_CMD_TI);
esp->msgin_len = 1;
esp->msgin_ctr = 0;
esp_advance_phase(esp->current_SC, in_msgindone);
return do_work_bus;
}
static inline void advance_sg(struct NCR_ESP *esp, Scsi_Cmnd *sp)
{
++sp->SCp.buffer;
--sp->SCp.buffers_residual;
sp->SCp.this_residual = sp->SCp.buffer->length;
if (esp->dma_advance_sg)
esp->dma_advance_sg (sp);
else
sp->SCp.ptr = (char *) virt_to_phys((page_address(sp->SCp.buffer->page) + sp->SCp.buffer->offset));
}
/* Please note that the way I've coded these routines is that I _always_
* check for a disconnect during any and all information transfer
* phases. The SCSI standard states that the target _can_ cause a BUS
* FREE condition by dropping all MSG/CD/IO/BSY signals. Also note
* that during information transfer phases the target controls every
* change in phase, the only thing the initiator can do is "ask" for
* a message out phase by driving ATN true. The target can, and sometimes
* will, completely ignore this request so we cannot assume anything when
* we try to force a message out phase to abort/reset a target. Most of
* the time the target will eventually be nice and go to message out, so
* we may have to hold on to our state about what we want to tell the target
* for some period of time.
*/
/* I think I have things working here correctly. Even partial transfers
* within a buffer or sub-buffer should not upset us at all no matter
* how bad the target and/or ESP fucks things up.
*/
static int esp_do_data(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
int thisphase, hmuch;
ESPDATA(("esp_do_data: "));
esp_maybe_nop(esp, eregs);
thisphase = sreg_to_dataphase(esp->sreg);
esp_advance_phase(SCptr, thisphase);
ESPDATA(("newphase<%s> ", (thisphase == in_datain) ? "DATAIN" : "DATAOUT"));
hmuch = esp->dma_can_transfer(esp, SCptr);
/*
* XXX MSch: cater for PIO transfer here; PIO used if hmuch == 0
*/
if (hmuch) { /* DMA */
/*
* DMA
*/
ESPDATA(("hmuch<%d> ", hmuch));
esp->current_transfer_size = hmuch;
esp_setcount(eregs, (esp->fas_premature_intr_workaround ?
(hmuch + 0x40) : hmuch));
esp->dma_setup(esp, (__u32)((unsigned long)SCptr->SCp.ptr),
hmuch, (thisphase == in_datain));
ESPDATA(("DMA|TI --> do_intr_end\n"));
esp_cmd(esp, eregs, ESP_CMD_DMA | ESP_CMD_TI);
return do_intr_end;
/*
* end DMA
*/
} else {
/*
* PIO
*/
int oldphase, i = 0; /* or where we left off last time ?? esp->current_data ?? */
int fifocnt = 0;
unsigned char *p = phys_to_virt((unsigned long)SCptr->SCp.ptr);
oldphase = esp_read(eregs->esp_status) & ESP_STAT_PMASK;
/*
* polled transfer; ugly, can we make this happen in a DRQ
* interrupt handler ??
* requires keeping track of state information in host or
* command struct!
* Problem: I've never seen a DRQ happen on Mac, not even
* with ESP_CMD_DMA ...
*/
/* figure out how much needs to be transferred */
hmuch = SCptr->SCp.this_residual;
ESPDATA(("hmuch<%d> pio ", hmuch));
esp->current_transfer_size = hmuch;
/* tell the ESP ... */
esp_setcount(eregs, hmuch);
/* loop */
while (hmuch) {
int j, fifo_stuck = 0, newphase;
unsigned long timeout;
#if 0
unsigned long flags;
#endif
#if 0
if ( i % 10 )
ESPDATA(("\r"));
else
ESPDATA(( /*"\n"*/ "\r"));
#endif
#if 0
local_irq_save(flags);
#endif
if(thisphase == in_datain) {
/* 'go' ... */
esp_cmd(esp, eregs, ESP_CMD_TI);
/* wait for data */
timeout = 1000000;
while (!((esp->sreg=esp_read(eregs->esp_status)) & ESP_STAT_INTR) && --timeout)
udelay(2);
if (timeout == 0)
printk("DRQ datain timeout! \n");
newphase = esp->sreg & ESP_STAT_PMASK;
/* see how much we got ... */
fifocnt = (esp_read(eregs->esp_fflags) & ESP_FF_FBYTES);
if (!fifocnt)
fifo_stuck++;
else
fifo_stuck = 0;
ESPDATA(("\rgot %d st %x ph %x", fifocnt, esp->sreg, newphase));
/* read fifo */
for(j=0;j<fifocnt;j++)
p[i++] = esp_read(eregs->esp_fdata);
ESPDATA(("(%d) ", i));
/* how many to go ?? */
hmuch -= fifocnt;
/* break if status phase !! */
if(newphase == ESP_STATP) {
/* clear int. */
esp->ireg = esp_read(eregs->esp_intrpt);
break;
}
} else {
#define MAX_FIFO 8
/* how much will fit ? */
int this_count = MAX_FIFO - fifocnt;
if (this_count > hmuch)
this_count = hmuch;
/* fill fifo */
for(j=0;j<this_count;j++)
esp_write(eregs->esp_fdata, p[i++]);
/* how many left if this goes out ?? */
hmuch -= this_count;
/* 'go' ... */
esp_cmd(esp, eregs, ESP_CMD_TI);
/* wait for 'got it' */
timeout = 1000000;
while (!((esp->sreg=esp_read(eregs->esp_status)) & ESP_STAT_INTR) && --timeout)
udelay(2);
if (timeout == 0)
printk("DRQ dataout timeout! \n");
newphase = esp->sreg & ESP_STAT_PMASK;
/* need to check how much was sent ?? */
fifocnt = (esp_read(eregs->esp_fflags) & ESP_FF_FBYTES);
ESPDATA(("\rsent %d st %x ph %x", this_count - fifocnt, esp->sreg, newphase));
ESPDATA(("(%d) ", i));
/* break if status phase !! */
if(newphase == ESP_STATP) {
/* clear int. */
esp->ireg = esp_read(eregs->esp_intrpt);
break;
}
}
/* clear int. */
esp->ireg = esp_read(eregs->esp_intrpt);
ESPDATA(("ir %x ... ", esp->ireg));
if (hmuch == 0)
ESPDATA(("done! \n"));
#if 0
local_irq_restore(flags);
#endif
/* check new bus phase */
if (newphase != oldphase && i < esp->current_transfer_size) {
/* something happened; disconnect ?? */
ESPDATA(("phase change, dropped out with %d done ... ", i));
break;
}
/* check int. status */
if (esp->ireg & ESP_INTR_DC) {
/* disconnect */
ESPDATA(("disconnect; %d transferred ... ", i));
break;
} else if (esp->ireg & ESP_INTR_FDONE) {
/* function done */
ESPDATA(("function done; %d transferred ... ", i));
break;
}
/* XXX fixme: bail out on stall */
if (fifo_stuck > 10) {
/* we're stuck */
ESPDATA(("fifo stall; %d transferred ... ", i));
break;
}
}
ESPDATA(("\n"));
/* check successful completion ?? */
if (thisphase == in_dataout)
hmuch += fifocnt; /* stuck?? adjust data pointer ...*/
/* tell do_data_finale how much was transferred */
esp->current_transfer_size -= hmuch;
/* still not completely sure on this one ... */
return /*do_intr_end*/ do_work_bus /*do_phase_determine*/ ;
/*
* end PIO
*/
}
return do_intr_end;
}
/* See how successful the data transfer was. */
static int esp_do_data_finale(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
struct esp_device *esp_dev = SCptr->device->hostdata;
int bogus_data = 0, bytes_sent = 0, fifocnt, ecount = 0;
if(esp->dma_led_off)
esp->dma_led_off(esp);
ESPDATA(("esp_do_data_finale: "));
if(SCptr->SCp.phase == in_datain) {
if(esp->sreg & ESP_STAT_PERR) {
/* Yuck, parity error. The ESP asserts ATN
* so that we can go to message out phase
* immediately and inform the target that
* something bad happened.
*/
ESPLOG(("esp%d: data bad parity detected.\n",
esp->esp_id));
esp->cur_msgout[0] = INITIATOR_ERROR;
esp->msgout_len = 1;
}
if(esp->dma_drain)
esp->dma_drain(esp);
}
if(esp->dma_invalidate)
esp->dma_invalidate(esp);
/* This could happen for the above parity error case. */
if(!(esp->ireg == ESP_INTR_BSERV)) {
/* Please go to msgout phase, please please please... */
ESPLOG(("esp%d: !BSERV after data, probably to msgout\n",
esp->esp_id));
return esp_do_phase_determine(esp, eregs);
}
/* Check for partial transfers and other horrible events. */
fifocnt = (esp_read(eregs->esp_fflags) & ESP_FF_FBYTES);
ecount = esp_getcount(eregs);
if(esp->fas_premature_intr_workaround)
ecount -= 0x40;
bytes_sent = esp->current_transfer_size;
ESPDATA(("trans_sz=%d, ", bytes_sent));
if(!(esp->sreg & ESP_STAT_TCNT))
bytes_sent -= ecount;
if(SCptr->SCp.phase == in_dataout)
bytes_sent -= fifocnt;
ESPDATA(("bytes_sent=%d (ecount=%d, fifocnt=%d), ", bytes_sent,
ecount, fifocnt));
/* If we were in synchronous mode, check for peculiarities. */
if(esp_dev->sync_max_offset)
bogus_data = esp100_sync_hwbug(esp, eregs, SCptr, fifocnt);
else
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
/* Until we are sure of what has happened, we are certainly
* in the dark.
*/
esp_advance_phase(SCptr, in_the_dark);
/* Check for premature interrupt condition. Can happen on FAS2x6
* chips. QLogic recommends a workaround by overprogramming the
* transfer counters, but this makes doing scatter-gather impossible.
* Until there is a way to disable scatter-gather for a single target,
* and not only for the entire host adapter as it is now, the workaround
* is way to expensive performance wise.
* Instead, it turns out that when this happens the target has disconnected
* already but it doesn't show in the interrupt register. Compensate for
* that here to try and avoid a SCSI bus reset.
*/
if(!esp->fas_premature_intr_workaround && (fifocnt == 1) &&
sreg_dataoutp(esp->sreg)) {
ESPLOG(("esp%d: Premature interrupt, enabling workaround\n",
esp->esp_id));
#if 0
/* Disable scatter-gather operations, they are not possible
* when using this workaround.
*/
esp->ehost->sg_tablesize = 0;
esp->ehost->use_clustering = ENABLE_CLUSTERING;
esp->fas_premature_intr_workaround = 1;
bytes_sent = 0;
if(SCptr->use_sg) {
ESPLOG(("esp%d: Aborting scatter-gather operation\n",
esp->esp_id));
esp->cur_msgout[0] = ABORT;
esp->msgout_len = 1;
esp->msgout_ctr = 0;
esp_cmd(esp, eregs, ESP_CMD_SATN);
esp_setcount(eregs, 0xffff);
esp_cmd(esp, eregs, ESP_CMD_NULL);
esp_cmd(esp, eregs, ESP_CMD_TPAD | ESP_CMD_DMA);
return do_intr_end;
}
#else
/* Just set the disconnected bit. That's what appears to
* happen anyway. The state machine will pick it up when
* we return.
*/
esp->ireg |= ESP_INTR_DC;
#endif
}
if(bytes_sent < 0) {
/* I've seen this happen due to lost state in this
* driver. No idea why it happened, but allowing
* this value to be negative caused things to
* lock up. This allows greater chance of recovery.
* In fact every time I've seen this, it has been
* a driver bug without question.
*/
ESPLOG(("esp%d: yieee, bytes_sent < 0!\n", esp->esp_id));
ESPLOG(("esp%d: csz=%d fifocount=%d ecount=%d\n",
esp->esp_id,
esp->current_transfer_size, fifocnt, ecount));
ESPLOG(("esp%d: use_sg=%d ptr=%p this_residual=%d\n",
esp->esp_id,
SCptr->use_sg, SCptr->SCp.ptr, SCptr->SCp.this_residual));
ESPLOG(("esp%d: Forcing async for target %d\n", esp->esp_id,
SCptr->device->id));
SCptr->device->borken = 1;
esp_dev->sync = 0;
bytes_sent = 0;
}
/* Update the state of our transfer. */
SCptr->SCp.ptr += bytes_sent;
SCptr->SCp.this_residual -= bytes_sent;
if(SCptr->SCp.this_residual < 0) {
/* shit */
ESPLOG(("esp%d: Data transfer overrun.\n", esp->esp_id));
SCptr->SCp.this_residual = 0;
}
/* Maybe continue. */
if(!bogus_data) {
ESPDATA(("!bogus_data, "));
/* NO MATTER WHAT, we advance the scatterlist,
* if the target should decide to disconnect
* in between scatter chunks (which is common)
* we could die horribly! I used to have the sg
* advance occur only if we are going back into
* (or are staying in) a data phase, you can
* imagine the hell I went through trying to
* figure this out.
*/
if(!SCptr->SCp.this_residual && SCptr->SCp.buffers_residual)
advance_sg(esp, SCptr);
#ifdef DEBUG_ESP_DATA
if(sreg_datainp(esp->sreg) || sreg_dataoutp(esp->sreg)) {
ESPDATA(("to more data\n"));
} else {
ESPDATA(("to new phase\n"));
}
#endif
return esp_do_phase_determine(esp, eregs);
}
/* Bogus data, just wait for next interrupt. */
ESPLOG(("esp%d: bogus_data during end of data phase\n",
esp->esp_id));
return do_intr_end;
}
/* We received a non-good status return at the end of
* running a SCSI command. This is used to decide if
* we should clear our synchronous transfer state for
* such a device when that happens.
*
* The idea is that when spinning up a disk or rewinding
* a tape, we don't want to go into a loop re-negotiating
* synchronous capabilities over and over.
*/
static int esp_should_clear_sync(Scsi_Cmnd *sp)
{
unchar cmd = sp->cmnd[0];
/* These cases are for spinning up a disk and
* waiting for that spinup to complete.
*/
if(cmd == START_STOP)
return 0;
if(cmd == TEST_UNIT_READY)
return 0;
/* One more special case for SCSI tape drives,
* this is what is used to probe the device for
* completion of a rewind or tape load operation.
*/
if(sp->device->type == TYPE_TAPE && cmd == MODE_SENSE)
return 0;
return 1;
}
/* Either a command is completing or a target is dropping off the bus
* to continue the command in the background so we can do other work.
*/
static int esp_do_freebus(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
int rval;
rval = skipahead2(esp, eregs, SCptr, in_status, in_msgindone, in_freeing);
if(rval)
return rval;
if(esp->ireg != ESP_INTR_DC) {
ESPLOG(("esp%d: Target will not disconnect\n", esp->esp_id));
return do_reset_bus; /* target will not drop BSY... */
}
esp->msgout_len = 0;
esp->prevmsgout = NOP;
if(esp->prevmsgin == COMMAND_COMPLETE) {
struct esp_device *esp_dev = SCptr->device->hostdata;
/* Normal end of nexus. */
if(esp->disconnected_SC)
esp_cmd(esp, eregs, ESP_CMD_ESEL);
if(SCptr->SCp.Status != GOOD &&
SCptr->SCp.Status != CONDITION_GOOD &&
((1<<scmd_id(SCptr)) & esp->targets_present) &&
esp_dev->sync && esp_dev->sync_max_offset) {
/* SCSI standard says that the synchronous capabilities
* should be renegotiated at this point. Most likely
* we are about to request sense from this target
* in which case we want to avoid using sync
* transfers until we are sure of the current target
* state.
*/
ESPMISC(("esp: Status <%d> for target %d lun %d\n",
SCptr->SCp.Status, SCptr->device->id, SCptr->device->lun));
/* But don't do this when spinning up a disk at
* boot time while we poll for completion as it
* fills up the console with messages. Also, tapes
* can report not ready many times right after
* loading up a tape.
*/
if(esp_should_clear_sync(SCptr) != 0)
esp_dev->sync = 0;
}
ESPDISC(("F<%02x,%02x>", SCptr->device->id, SCptr->device->lun));
esp_done(esp, ((SCptr->SCp.Status & 0xff) |
((SCptr->SCp.Message & 0xff)<<8) |
(DID_OK << 16)));
} else if(esp->prevmsgin == DISCONNECT) {
/* Normal disconnect. */
esp_cmd(esp, eregs, ESP_CMD_ESEL);
ESPDISC(("D<%02x,%02x>", SCptr->device->id, SCptr->device->lun));
append_SC(&esp->disconnected_SC, SCptr);
esp->current_SC = NULL;
if(esp->issue_SC)
esp_exec_cmd(esp);
} else {
/* Driver bug, we do not expect a disconnect here
* and should not have advanced the state engine
* to in_freeing.
*/
ESPLOG(("esp%d: last msg not disc and not cmd cmplt.\n",
esp->esp_id));
return do_reset_bus;
}
return do_intr_end;
}
/* When a reselect occurs, and we cannot find the command to
* reconnect to in our queues, we do this.
*/
static int esp_bad_reconnect(struct NCR_ESP *esp)
{
Scsi_Cmnd *sp;
ESPLOG(("esp%d: Eieeee, reconnecting unknown command!\n",
esp->esp_id));
ESPLOG(("QUEUE DUMP\n"));
sp = esp->issue_SC;
ESPLOG(("esp%d: issue_SC[", esp->esp_id));
while(sp) {
ESPLOG(("<%02x,%02x>", sp->device->id, sp->device->lun));
sp = (Scsi_Cmnd *) sp->host_scribble;
}
ESPLOG(("]\n"));
sp = esp->current_SC;
ESPLOG(("esp%d: current_SC[", esp->esp_id));
while(sp) {
ESPLOG(("<%02x,%02x>", sp->device->id, sp->device->lun));
sp = (Scsi_Cmnd *) sp->host_scribble;
}
ESPLOG(("]\n"));
sp = esp->disconnected_SC;
ESPLOG(("esp%d: disconnected_SC[", esp->esp_id));
while(sp) {
ESPLOG(("<%02x,%02x>", sp->device->id, sp->device->lun));
sp = (Scsi_Cmnd *) sp->host_scribble;
}
ESPLOG(("]\n"));
return do_reset_bus;
}
/* Do the needy when a target tries to reconnect to us. */
static int esp_do_reconnect(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
int lun, target;
Scsi_Cmnd *SCptr;
/* Check for all bogus conditions first. */
target = reconnect_target(esp, eregs);
if(target < 0) {
ESPDISC(("bad bus bits\n"));
return do_reset_bus;
}
lun = reconnect_lun(esp, eregs);
if(lun < 0) {
ESPDISC(("target=%2x, bad identify msg\n", target));
return do_reset_bus;
}
/* Things look ok... */
ESPDISC(("R<%02x,%02x>", target, lun));
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
if(esp100_reconnect_hwbug(esp, eregs))
return do_reset_bus;
esp_cmd(esp, eregs, ESP_CMD_NULL);
SCptr = remove_SC(&esp->disconnected_SC, (unchar) target, (unchar) lun);
if(!SCptr)
return esp_bad_reconnect(esp);
esp_connect(esp, eregs, SCptr);
esp_cmd(esp, eregs, ESP_CMD_MOK);
/* Reconnect implies a restore pointers operation. */
esp_restore_pointers(esp, SCptr);
esp->snip = 0;
esp_advance_phase(SCptr, in_the_dark);
return do_intr_end;
}
/* End of NEXUS (hopefully), pick up status + message byte then leave if
* all goes well.
*/
static int esp_do_status(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
int intr, rval;
rval = skipahead1(esp, eregs, SCptr, in_the_dark, in_status);
if(rval)
return rval;
intr = esp->ireg;
ESPSTAT(("esp_do_status: "));
if(intr != ESP_INTR_DC) {
int message_out = 0; /* for parity problems */
/* Ack the message. */
ESPSTAT(("ack msg, "));
esp_cmd(esp, eregs, ESP_CMD_MOK);
if(esp->dma_poll)
esp->dma_poll(esp, (unsigned char *) esp->esp_command);
ESPSTAT(("got something, "));
/* ESP chimes in with one of
*
* 1) function done interrupt:
* both status and message in bytes
* are available
*
* 2) bus service interrupt:
* only status byte was acquired
*
* 3) Anything else:
* can't happen, but we test for it
* anyways
*
* ALSO: If bad parity was detected on either
* the status _or_ the message byte then
* the ESP has asserted ATN on the bus
* and we must therefore wait for the
* next phase change.
*/
if(intr & ESP_INTR_FDONE) {
/* We got it all, hallejulia. */
ESPSTAT(("got both, "));
SCptr->SCp.Status = esp->esp_command[0];
SCptr->SCp.Message = esp->esp_command[1];
esp->prevmsgin = SCptr->SCp.Message;
esp->cur_msgin[0] = SCptr->SCp.Message;
if(esp->sreg & ESP_STAT_PERR) {
/* There was bad parity for the
* message byte, the status byte
* was ok.
*/
message_out = MSG_PARITY_ERROR;
}
} else if(intr == ESP_INTR_BSERV) {
/* Only got status byte. */
ESPLOG(("esp%d: got status only, ", esp->esp_id));
if(!(esp->sreg & ESP_STAT_PERR)) {
SCptr->SCp.Status = esp->esp_command[0];
SCptr->SCp.Message = 0xff;
} else {
/* The status byte had bad parity.
* we leave the scsi_pointer Status
* field alone as we set it to a default
* of CHECK_CONDITION in esp_queue.
*/
message_out = INITIATOR_ERROR;
}
} else {
/* This shouldn't happen ever. */
ESPSTAT(("got bolixed\n"));
esp_advance_phase(SCptr, in_the_dark);
return esp_do_phase_determine(esp, eregs);
}
if(!message_out) {
ESPSTAT(("status=%2x msg=%2x, ", SCptr->SCp.Status,
SCptr->SCp.Message));
if(SCptr->SCp.Message == COMMAND_COMPLETE) {
ESPSTAT(("and was COMMAND_COMPLETE\n"));
esp_advance_phase(SCptr, in_freeing);
return esp_do_freebus(esp, eregs);
} else {
ESPLOG(("esp%d: and _not_ COMMAND_COMPLETE\n",
esp->esp_id));
esp->msgin_len = esp->msgin_ctr = 1;
esp_advance_phase(SCptr, in_msgindone);
return esp_do_msgindone(esp, eregs);
}
} else {
/* With luck we'll be able to let the target
* know that bad parity happened, it will know
* which byte caused the problems and send it
* again. For the case where the status byte
* receives bad parity, I do not believe most
* targets recover very well. We'll see.
*/
ESPLOG(("esp%d: bad parity somewhere mout=%2x\n",
esp->esp_id, message_out));
esp->cur_msgout[0] = message_out;
esp->msgout_len = esp->msgout_ctr = 1;
esp_advance_phase(SCptr, in_the_dark);
return esp_do_phase_determine(esp, eregs);
}
} else {
/* If we disconnect now, all hell breaks loose. */
ESPLOG(("esp%d: whoops, disconnect\n", esp->esp_id));
esp_advance_phase(SCptr, in_the_dark);
return esp_do_phase_determine(esp, eregs);
}
}
static int esp_enter_status(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
unchar thecmd = ESP_CMD_ICCSEQ;
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
if(esp->do_pio_cmds) {
esp_advance_phase(esp->current_SC, in_status);
esp_cmd(esp, eregs, thecmd);
while(!(esp_read(esp->eregs->esp_status) & ESP_STAT_INTR));
esp->esp_command[0] = esp_read(eregs->esp_fdata);
while(!(esp_read(esp->eregs->esp_status) & ESP_STAT_INTR));
esp->esp_command[1] = esp_read(eregs->esp_fdata);
} else {
esp->esp_command[0] = esp->esp_command[1] = 0xff;
esp_write(eregs->esp_tclow, 2);
esp_write(eregs->esp_tcmed, 0);
esp->dma_init_read(esp, esp->esp_command_dvma, 2);
thecmd |= ESP_CMD_DMA;
esp_cmd(esp, eregs, thecmd);
esp_advance_phase(esp->current_SC, in_status);
}
return esp_do_status(esp, eregs);
}
static int esp_disconnect_amidst_phases(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
Scsi_Cmnd *sp = esp->current_SC;
struct esp_device *esp_dev = sp->device->hostdata;
/* This means real problems if we see this
* here. Unless we were actually trying
* to force the device to abort/reset.
*/
ESPLOG(("esp%d: Disconnect amidst phases, ", esp->esp_id));
ESPLOG(("pphase<%s> cphase<%s>, ",
phase_string(sp->SCp.phase),
phase_string(sp->SCp.sent_command)));
if(esp->disconnected_SC)
esp_cmd(esp, eregs, ESP_CMD_ESEL);
switch(esp->cur_msgout[0]) {
default:
/* We didn't expect this to happen at all. */
ESPLOG(("device is bolixed\n"));
esp_advance_phase(sp, in_tgterror);
esp_done(esp, (DID_ERROR << 16));
break;
case BUS_DEVICE_RESET:
ESPLOG(("device reset successful\n"));
esp_dev->sync_max_offset = 0;
esp_dev->sync_min_period = 0;
esp_dev->sync = 0;
esp_advance_phase(sp, in_resetdev);
esp_done(esp, (DID_RESET << 16));
break;
case ABORT:
ESPLOG(("device abort successful\n"));
esp_advance_phase(sp, in_abortone);
esp_done(esp, (DID_ABORT << 16));
break;
};
return do_intr_end;
}
static int esp_enter_msgout(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
esp_advance_phase(esp->current_SC, in_msgout);
return esp_do_msgout(esp, eregs);
}
static int esp_enter_msgin(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
esp_advance_phase(esp->current_SC, in_msgin);
return esp_do_msgin(esp, eregs);
}
static int esp_enter_cmd(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
esp_advance_phase(esp->current_SC, in_cmdbegin);
return esp_do_cmdbegin(esp, eregs);
}
static int esp_enter_badphase(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
ESPLOG(("esp%d: Bizarre bus phase %2x.\n", esp->esp_id,
esp->sreg & ESP_STAT_PMASK));
return do_reset_bus;
}
typedef int (*espfunc_t)(struct NCR_ESP *,
struct ESP_regs *);
static espfunc_t phase_vector[] = {
esp_do_data, /* ESP_DOP */
esp_do_data, /* ESP_DIP */
esp_enter_cmd, /* ESP_CMDP */
esp_enter_status, /* ESP_STATP */
esp_enter_badphase, /* ESP_STAT_PMSG */
esp_enter_badphase, /* ESP_STAT_PMSG | ESP_STAT_PIO */
esp_enter_msgout, /* ESP_MOP */
esp_enter_msgin, /* ESP_MIP */
};
/* The target has control of the bus and we have to see where it has
* taken us.
*/
static int esp_do_phase_determine(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
if ((esp->ireg & ESP_INTR_DC) != 0)
return esp_disconnect_amidst_phases(esp, eregs);
return phase_vector[esp->sreg & ESP_STAT_PMASK](esp, eregs);
}
/* First interrupt after exec'ing a cmd comes here. */
static int esp_select_complete(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
struct esp_device *esp_dev = SCptr->device->hostdata;
int cmd_bytes_sent, fcnt;
fcnt = (esp_read(eregs->esp_fflags) & ESP_FF_FBYTES);
cmd_bytes_sent = esp->dma_bytes_sent(esp, fcnt);
if(esp->dma_invalidate)
esp->dma_invalidate(esp);
/* Let's check to see if a reselect happened
* while we we're trying to select. This must
* be checked first.
*/
if(esp->ireg == (ESP_INTR_RSEL | ESP_INTR_FDONE)) {
esp_reconnect(esp, SCptr);
return esp_do_reconnect(esp, eregs);
}
/* Looks like things worked, we should see a bus service &
* a function complete interrupt at this point. Note we
* are doing a direct comparison because we don't want to
* be fooled into thinking selection was successful if
* ESP_INTR_DC is set, see below.
*/
if(esp->ireg == (ESP_INTR_FDONE | ESP_INTR_BSERV)) {
/* target speaks... */
esp->targets_present |= (1<<scmd_id(SCptr));
/* What if the target ignores the sdtr? */
if(esp->snip)
esp_dev->sync = 1;
/* See how far, if at all, we got in getting
* the information out to the target.
*/
switch(esp->seqreg) {
default:
case ESP_STEP_ASEL:
/* Arbitration won, target selected, but
* we are in some phase which is not command
* phase nor is it message out phase.
*
* XXX We've confused the target, obviously.
* XXX So clear it's state, but we also end
* XXX up clearing everyone elses. That isn't
* XXX so nice. I'd like to just reset this
* XXX target, but if I cannot even get it's
* XXX attention and finish selection to talk
* XXX to it, there is not much more I can do.
* XXX If we have a loaded bus we're going to
* XXX spend the next second or so renegotiating
* XXX for synchronous transfers.
*/
ESPLOG(("esp%d: STEP_ASEL for tgt %d\n",
esp->esp_id, SCptr->device->id));
case ESP_STEP_SID:
/* Arbitration won, target selected, went
* to message out phase, sent one message
* byte, then we stopped. ATN is asserted
* on the SCSI bus and the target is still
* there hanging on. This is a legal
* sequence step if we gave the ESP a select
* and stop command.
*
* XXX See above, I could set the borken flag
* XXX in the device struct and retry the
* XXX command. But would that help for
* XXX tagged capable targets?
*/
case ESP_STEP_NCMD:
/* Arbitration won, target selected, maybe
* sent the one message byte in message out
* phase, but we did not go to command phase
* in the end. Actually, we could have sent
* only some of the message bytes if we tried
* to send out the entire identify and tag
* message using ESP_CMD_SA3.
*/
cmd_bytes_sent = 0;
break;
case ESP_STEP_PPC:
/* No, not the powerPC pinhead. Arbitration
* won, all message bytes sent if we went to
* message out phase, went to command phase
* but only part of the command was sent.
*
* XXX I've seen this, but usually in conjunction
* XXX with a gross error which appears to have
* XXX occurred between the time I told the
* XXX ESP to arbitrate and when I got the
* XXX interrupt. Could I have misloaded the
* XXX command bytes into the fifo? Actually,
* XXX I most likely missed a phase, and therefore
* XXX went into never never land and didn't even
* XXX know it. That was the old driver though.
* XXX What is even more peculiar is that the ESP
* XXX showed the proper function complete and
* XXX bus service bits in the interrupt register.
*/
case ESP_STEP_FINI4:
case ESP_STEP_FINI5:
case ESP_STEP_FINI6:
case ESP_STEP_FINI7:
/* Account for the identify message */
if(SCptr->SCp.phase == in_slct_norm)
cmd_bytes_sent -= 1;
};
esp_cmd(esp, eregs, ESP_CMD_NULL);
/* Be careful, we could really get fucked during synchronous
* data transfers if we try to flush the fifo now.
*/
if(!fcnt && /* Fifo is empty and... */
/* either we are not doing synchronous transfers or... */
(!esp_dev->sync_max_offset ||
/* We are not going into data in phase. */
((esp->sreg & ESP_STAT_PMASK) != ESP_DIP)))
esp_cmd(esp, eregs, ESP_CMD_FLUSH); /* flush is safe */
/* See how far we got if this is not a slow command. */
if(!esp->esp_slowcmd) {
if(cmd_bytes_sent < 0)
cmd_bytes_sent = 0;
if(cmd_bytes_sent != SCptr->cmd_len) {
/* Crapola, mark it as a slowcmd
* so that we have some chance of
* keeping the command alive with
* good luck.
*
* XXX Actually, if we didn't send it all
* XXX this means either we didn't set things
* XXX up properly (driver bug) or the target
* XXX or the ESP detected parity on one of
* XXX the command bytes. This makes much
* XXX more sense, and therefore this code
* XXX should be changed to send out a
* XXX parity error message or if the status
* XXX register shows no parity error then
* XXX just expect the target to bring the
* XXX bus into message in phase so that it
* XXX can send us the parity error message.
* XXX SCSI sucks...
*/
esp->esp_slowcmd = 1;
esp->esp_scmdp = &(SCptr->cmnd[cmd_bytes_sent]);
esp->esp_scmdleft = (SCptr->cmd_len - cmd_bytes_sent);
}
}
/* Now figure out where we went. */
esp_advance_phase(SCptr, in_the_dark);
return esp_do_phase_determine(esp, eregs);
}
/* Did the target even make it? */
if(esp->ireg == ESP_INTR_DC) {
/* wheee... nobody there or they didn't like
* what we told it to do, clean up.
*/
/* If anyone is off the bus, but working on
* a command in the background for us, tell
* the ESP to listen for them.
*/
if(esp->disconnected_SC)
esp_cmd(esp, eregs, ESP_CMD_ESEL);
if(((1<<SCptr->device->id) & esp->targets_present) &&
esp->seqreg && esp->cur_msgout[0] == EXTENDED_MESSAGE &&
(SCptr->SCp.phase == in_slct_msg ||
SCptr->SCp.phase == in_slct_stop)) {
/* shit */
esp->snip = 0;
ESPLOG(("esp%d: Failed synchronous negotiation for target %d "
"lun %d\n", esp->esp_id, SCptr->device->id, SCptr->device->lun));
esp_dev->sync_max_offset = 0;
esp_dev->sync_min_period = 0;
esp_dev->sync = 1; /* so we don't negotiate again */
/* Run the command again, this time though we
* won't try to negotiate for synchronous transfers.
*
* XXX I'd like to do something like send an
* XXX INITIATOR_ERROR or ABORT message to the
* XXX target to tell it, "Sorry I confused you,
* XXX please come back and I will be nicer next
* XXX time". But that requires having the target
* XXX on the bus, and it has dropped BSY on us.
*/
esp->current_SC = NULL;
esp_advance_phase(SCptr, not_issued);
prepend_SC(&esp->issue_SC, SCptr);
esp_exec_cmd(esp);
return do_intr_end;
}
/* Ok, this is normal, this is what we see during boot
* or whenever when we are scanning the bus for targets.
* But first make sure that is really what is happening.
*/
if(((1<<SCptr->device->id) & esp->targets_present)) {
ESPLOG(("esp%d: Warning, live target %d not responding to "
"selection.\n", esp->esp_id, SCptr->device->id));
/* This _CAN_ happen. The SCSI standard states that
* the target is to _not_ respond to selection if
* _it_ detects bad parity on the bus for any reason.
* Therefore, we assume that if we've talked successfully
* to this target before, bad parity is the problem.
*/
esp_done(esp, (DID_PARITY << 16));
} else {
/* Else, there really isn't anyone there. */
ESPMISC(("esp: selection failure, maybe nobody there?\n"));
ESPMISC(("esp: target %d lun %d\n",
SCptr->device->id, SCptr->device->lun));
esp_done(esp, (DID_BAD_TARGET << 16));
}
return do_intr_end;
}
ESPLOG(("esp%d: Selection failure.\n", esp->esp_id));
printk("esp%d: Currently -- ", esp->esp_id);
esp_print_ireg(esp->ireg);
printk(" ");
esp_print_statreg(esp->sreg);
printk(" ");
esp_print_seqreg(esp->seqreg);
printk("\n");
printk("esp%d: New -- ", esp->esp_id);
esp->sreg = esp_read(eregs->esp_status);
esp->seqreg = esp_read(eregs->esp_sstep);
esp->ireg = esp_read(eregs->esp_intrpt);
esp_print_ireg(esp->ireg);
printk(" ");
esp_print_statreg(esp->sreg);
printk(" ");
esp_print_seqreg(esp->seqreg);
printk("\n");
ESPLOG(("esp%d: resetting bus\n", esp->esp_id));
return do_reset_bus; /* ugh... */
}
/* Continue reading bytes for msgin phase. */
static int esp_do_msgincont(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
if(esp->ireg & ESP_INTR_BSERV) {
/* in the right phase too? */
if((esp->sreg & ESP_STAT_PMASK) == ESP_MIP) {
/* phew... */
esp_cmd(esp, eregs, ESP_CMD_TI);
esp_advance_phase(esp->current_SC, in_msgindone);
return do_intr_end;
}
/* We changed phase but ESP shows bus service,
* in this case it is most likely that we, the
* hacker who has been up for 20hrs straight
* staring at the screen, drowned in coffee
* smelling like retched cigarette ashes
* have miscoded something..... so, try to
* recover as best we can.
*/
ESPLOG(("esp%d: message in mis-carriage.\n", esp->esp_id));
}
esp_advance_phase(esp->current_SC, in_the_dark);
return do_phase_determine;
}
static int check_singlebyte_msg(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
esp->prevmsgin = esp->cur_msgin[0];
if(esp->cur_msgin[0] & 0x80) {
/* wheee... */
ESPLOG(("esp%d: target sends identify amidst phases\n",
esp->esp_id));
esp_advance_phase(esp->current_SC, in_the_dark);
return 0;
} else if(((esp->cur_msgin[0] & 0xf0) == 0x20) ||
(esp->cur_msgin[0] == EXTENDED_MESSAGE)) {
esp->msgin_len = 2;
esp_advance_phase(esp->current_SC, in_msgincont);
return 0;
}
esp_advance_phase(esp->current_SC, in_the_dark);
switch(esp->cur_msgin[0]) {
default:
/* We don't want to hear about it. */
ESPLOG(("esp%d: msg %02x which we don't know about\n", esp->esp_id,
esp->cur_msgin[0]));
return MESSAGE_REJECT;
case NOP:
ESPLOG(("esp%d: target %d sends a nop\n", esp->esp_id,
esp->current_SC->device->id));
return 0;
case RESTORE_POINTERS:
/* In this case we might also have to backup the
* "slow command" pointer. It is rare to get such
* a save/restore pointer sequence so early in the
* bus transition sequences, but cover it.
*/
if(esp->esp_slowcmd) {
esp->esp_scmdleft = esp->current_SC->cmd_len;
esp->esp_scmdp = &esp->current_SC->cmnd[0];
}
esp_restore_pointers(esp, esp->current_SC);
return 0;
case SAVE_POINTERS:
esp_save_pointers(esp, esp->current_SC);
return 0;
case COMMAND_COMPLETE:
case DISCONNECT:
/* Freeing the bus, let it go. */
esp->current_SC->SCp.phase = in_freeing;
return 0;
case MESSAGE_REJECT:
ESPMISC(("msg reject, "));
if(esp->prevmsgout == EXTENDED_MESSAGE) {
struct esp_device *esp_dev = esp->current_SC->device->hostdata;
/* Doesn't look like this target can
* do synchronous or WIDE transfers.
*/
ESPSDTR(("got reject, was trying nego, clearing sync/WIDE\n"));
esp_dev->sync = 1;
esp_dev->wide = 1;
esp_dev->sync_min_period = 0;
esp_dev->sync_max_offset = 0;
return 0;
} else {
ESPMISC(("not sync nego, sending ABORT\n"));
return ABORT;
}
};
}
/* Target negotiates for synchronous transfers before we do, this
* is legal although very strange. What is even funnier is that
* the SCSI2 standard specifically recommends against targets doing
* this because so many initiators cannot cope with this occurring.
*/
static int target_with_ants_in_pants(struct NCR_ESP *esp,
Scsi_Cmnd *SCptr,
struct esp_device *esp_dev)
{
if(esp_dev->sync || SCptr->device->borken) {
/* sorry, no can do */
ESPSDTR(("forcing to async, "));
build_sync_nego_msg(esp, 0, 0);
esp_dev->sync = 1;
esp->snip = 1;
ESPLOG(("esp%d: hoping for msgout\n", esp->esp_id));
esp_advance_phase(SCptr, in_the_dark);
return EXTENDED_MESSAGE;
}
/* Ok, we'll check them out... */
return 0;
}
static void sync_report(struct NCR_ESP *esp)
{
int msg3, msg4;
char *type;
msg3 = esp->cur_msgin[3];
msg4 = esp->cur_msgin[4];
if(msg4) {
int hz = 1000000000 / (msg3 * 4);
int integer = hz / 1000000;
int fraction = (hz - (integer * 1000000)) / 10000;
if((msg3 * 4) < 200) {
type = "FAST";
} else {
type = "synchronous";
}
/* Do not transform this back into one big printk
* again, it triggers a bug in our sparc64-gcc272
* sibling call optimization. -DaveM
*/
ESPLOG((KERN_INFO "esp%d: target %d ",
esp->esp_id, esp->current_SC->device->id));
ESPLOG(("[period %dns offset %d %d.%02dMHz ",
(int) msg3 * 4, (int) msg4,
integer, fraction));
ESPLOG(("%s SCSI%s]\n", type,
(((msg3 * 4) < 200) ? "-II" : "")));
} else {
ESPLOG((KERN_INFO "esp%d: target %d asynchronous\n",
esp->esp_id, esp->current_SC->device->id));
}
}
static int check_multibyte_msg(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
struct esp_device *esp_dev = SCptr->device->hostdata;
unchar regval = 0;
int message_out = 0;
ESPSDTR(("chk multibyte msg: "));
if(esp->cur_msgin[2] == EXTENDED_SDTR) {
int period = esp->cur_msgin[3];
int offset = esp->cur_msgin[4];
ESPSDTR(("is sync nego response, "));
if(!esp->snip) {
int rval;
/* Target negotiates first! */
ESPSDTR(("target jumps the gun, "));
message_out = EXTENDED_MESSAGE; /* we must respond */
rval = target_with_ants_in_pants(esp, SCptr, esp_dev);
if(rval)
return rval;
}
ESPSDTR(("examining sdtr, "));
/* Offset cannot be larger than ESP fifo size. */
if(offset > 15) {
ESPSDTR(("offset too big %2x, ", offset));
offset = 15;
ESPSDTR(("sending back new offset\n"));
build_sync_nego_msg(esp, period, offset);
return EXTENDED_MESSAGE;
}
if(offset && period > esp->max_period) {
/* Yeee, async for this slow device. */
ESPSDTR(("period too long %2x, ", period));
build_sync_nego_msg(esp, 0, 0);
ESPSDTR(("hoping for msgout\n"));
esp_advance_phase(esp->current_SC, in_the_dark);
return EXTENDED_MESSAGE;
} else if (offset && period < esp->min_period) {
ESPSDTR(("period too short %2x, ", period));
period = esp->min_period;
if(esp->erev > esp236)
regval = 4;
else
regval = 5;
} else if(offset) {
int tmp;
ESPSDTR(("period is ok, "));
tmp = esp->ccycle / 1000;
regval = (((period << 2) + tmp - 1) / tmp);
if(regval && (esp->erev > esp236)) {
if(period >= 50)
regval--;
}
}
if(offset) {
unchar bit;
esp_dev->sync_min_period = (regval & 0x1f);
esp_dev->sync_max_offset = (offset | esp->radelay);
if(esp->erev > esp236) {
if(esp->erev == fas100a)
bit = ESP_CONFIG3_FAST;
else
bit = ESP_CONFIG3_FSCSI;
if(period < 50)
esp->config3[SCptr->device->id] |= bit;
else
esp->config3[SCptr->device->id] &= ~bit;
esp->prev_cfg3 = esp->config3[SCptr->device->id];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
}
esp->prev_soff = esp_dev->sync_min_period;
esp_write(eregs->esp_soff, esp->prev_soff);
esp->prev_stp = esp_dev->sync_max_offset;
esp_write(eregs->esp_stp, esp->prev_stp);
ESPSDTR(("soff=%2x stp=%2x cfg3=%2x\n",
esp_dev->sync_max_offset,
esp_dev->sync_min_period,
esp->config3[scmd_id(SCptr)]));
esp->snip = 0;
} else if(esp_dev->sync_max_offset) {
unchar bit;
/* back to async mode */
ESPSDTR(("unaccaptable sync nego, forcing async\n"));
esp_dev->sync_max_offset = 0;
esp_dev->sync_min_period = 0;
esp->prev_soff = 0;
esp_write(eregs->esp_soff, 0);
esp->prev_stp = 0;
esp_write(eregs->esp_stp, 0);
if(esp->erev > esp236) {
if(esp->erev == fas100a)
bit = ESP_CONFIG3_FAST;
else
bit = ESP_CONFIG3_FSCSI;
esp->config3[SCptr->device->id] &= ~bit;
esp->prev_cfg3 = esp->config3[SCptr->device->id];
esp_write(eregs->esp_cfg3, esp->prev_cfg3);
}
}
sync_report(esp);
ESPSDTR(("chk multibyte msg: sync is known, "));
esp_dev->sync = 1;
if(message_out) {
ESPLOG(("esp%d: sending sdtr back, hoping for msgout\n",
esp->esp_id));
build_sync_nego_msg(esp, period, offset);
esp_advance_phase(SCptr, in_the_dark);
return EXTENDED_MESSAGE;
}
ESPSDTR(("returning zero\n"));
esp_advance_phase(SCptr, in_the_dark); /* ...or else! */
return 0;
} else if(esp->cur_msgin[2] == EXTENDED_WDTR) {
ESPLOG(("esp%d: AIEEE wide msg received\n", esp->esp_id));
message_out = MESSAGE_REJECT;
} else if(esp->cur_msgin[2] == EXTENDED_MODIFY_DATA_POINTER) {
ESPLOG(("esp%d: rejecting modify data ptr msg\n", esp->esp_id));
message_out = MESSAGE_REJECT;
}
esp_advance_phase(SCptr, in_the_dark);
return message_out;
}
static int esp_do_msgindone(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
int message_out = 0, it = 0, rval;
rval = skipahead1(esp, eregs, SCptr, in_msgin, in_msgindone);
if(rval)
return rval;
if(SCptr->SCp.sent_command != in_status) {
if(!(esp->ireg & ESP_INTR_DC)) {
if(esp->msgin_len && (esp->sreg & ESP_STAT_PERR)) {
message_out = MSG_PARITY_ERROR;
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
} else if((it = (esp_read(eregs->esp_fflags) & ESP_FF_FBYTES))!=1) {
/* We certainly dropped the ball somewhere. */
message_out = INITIATOR_ERROR;
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
} else if(!esp->msgin_len) {
it = esp_read(eregs->esp_fdata);
esp_advance_phase(SCptr, in_msgincont);
} else {
/* it is ok and we want it */
it = esp->cur_msgin[esp->msgin_ctr] =
esp_read(eregs->esp_fdata);
esp->msgin_ctr++;
}
} else {
esp_advance_phase(SCptr, in_the_dark);
return do_work_bus;
}
} else {
it = esp->cur_msgin[0];
}
if(!message_out && esp->msgin_len) {
if(esp->msgin_ctr < esp->msgin_len) {
esp_advance_phase(SCptr, in_msgincont);
} else if(esp->msgin_len == 1) {
message_out = check_singlebyte_msg(esp, eregs);
} else if(esp->msgin_len == 2) {
if(esp->cur_msgin[0] == EXTENDED_MESSAGE) {
if((it+2) >= 15) {
message_out = MESSAGE_REJECT;
} else {
esp->msgin_len = (it + 2);
esp_advance_phase(SCptr, in_msgincont);
}
} else {
message_out = MESSAGE_REJECT; /* foo on you */
}
} else {
message_out = check_multibyte_msg(esp, eregs);
}
}
if(message_out < 0) {
return -message_out;
} else if(message_out) {
if(((message_out != 1) &&
((message_out < 0x20) || (message_out & 0x80))))
esp->msgout_len = 1;
esp->cur_msgout[0] = message_out;
esp_cmd(esp, eregs, ESP_CMD_SATN);
esp_advance_phase(SCptr, in_the_dark);
esp->msgin_len = 0;
}
esp->sreg = esp_read(eregs->esp_status);
esp->sreg &= ~(ESP_STAT_INTR);
if((esp->sreg & (ESP_STAT_PMSG|ESP_STAT_PCD)) == (ESP_STAT_PMSG|ESP_STAT_PCD))
esp_cmd(esp, eregs, ESP_CMD_MOK);
if((SCptr->SCp.sent_command == in_msgindone) &&
(SCptr->SCp.phase == in_freeing))
return esp_do_freebus(esp, eregs);
return do_intr_end;
}
static int esp_do_cmdbegin(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
unsigned char tmp;
Scsi_Cmnd *SCptr = esp->current_SC;
esp_advance_phase(SCptr, in_cmdend);
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
tmp = *esp->esp_scmdp++;
esp->esp_scmdleft--;
esp_write(eregs->esp_fdata, tmp);
esp_cmd(esp, eregs, ESP_CMD_TI);
return do_intr_end;
}
static int esp_do_cmddone(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
esp_cmd(esp, eregs, ESP_CMD_NULL);
if(esp->ireg & ESP_INTR_BSERV) {
esp_advance_phase(esp->current_SC, in_the_dark);
return esp_do_phase_determine(esp, eregs);
}
ESPLOG(("esp%d: in do_cmddone() but didn't get BSERV interrupt.\n",
esp->esp_id));
return do_reset_bus;
}
static int esp_do_msgout(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
switch(esp->msgout_len) {
case 1:
esp_write(eregs->esp_fdata, esp->cur_msgout[0]);
esp_cmd(esp, eregs, ESP_CMD_TI);
break;
case 2:
if(esp->do_pio_cmds){
esp_write(eregs->esp_fdata, esp->cur_msgout[0]);
esp_write(eregs->esp_fdata, esp->cur_msgout[1]);
esp_cmd(esp, eregs, ESP_CMD_TI);
} else {
esp->esp_command[0] = esp->cur_msgout[0];
esp->esp_command[1] = esp->cur_msgout[1];
esp->dma_setup(esp, esp->esp_command_dvma, 2, 0);
esp_setcount(eregs, 2);
esp_cmd(esp, eregs, ESP_CMD_DMA | ESP_CMD_TI);
}
break;
case 4:
esp->snip = 1;
if(esp->do_pio_cmds){
esp_write(eregs->esp_fdata, esp->cur_msgout[0]);
esp_write(eregs->esp_fdata, esp->cur_msgout[1]);
esp_write(eregs->esp_fdata, esp->cur_msgout[2]);
esp_write(eregs->esp_fdata, esp->cur_msgout[3]);
esp_cmd(esp, eregs, ESP_CMD_TI);
} else {
esp->esp_command[0] = esp->cur_msgout[0];
esp->esp_command[1] = esp->cur_msgout[1];
esp->esp_command[2] = esp->cur_msgout[2];
esp->esp_command[3] = esp->cur_msgout[3];
esp->dma_setup(esp, esp->esp_command_dvma, 4, 0);
esp_setcount(eregs, 4);
esp_cmd(esp, eregs, ESP_CMD_DMA | ESP_CMD_TI);
}
break;
case 5:
esp->snip = 1;
if(esp->do_pio_cmds){
esp_write(eregs->esp_fdata, esp->cur_msgout[0]);
esp_write(eregs->esp_fdata, esp->cur_msgout[1]);
esp_write(eregs->esp_fdata, esp->cur_msgout[2]);
esp_write(eregs->esp_fdata, esp->cur_msgout[3]);
esp_write(eregs->esp_fdata, esp->cur_msgout[4]);
esp_cmd(esp, eregs, ESP_CMD_TI);
} else {
esp->esp_command[0] = esp->cur_msgout[0];
esp->esp_command[1] = esp->cur_msgout[1];
esp->esp_command[2] = esp->cur_msgout[2];
esp->esp_command[3] = esp->cur_msgout[3];
esp->esp_command[4] = esp->cur_msgout[4];
esp->dma_setup(esp, esp->esp_command_dvma, 5, 0);
esp_setcount(eregs, 5);
esp_cmd(esp, eregs, ESP_CMD_DMA | ESP_CMD_TI);
}
break;
default:
/* whoops */
ESPMISC(("bogus msgout sending NOP\n"));
esp->cur_msgout[0] = NOP;
esp_write(eregs->esp_fdata, esp->cur_msgout[0]);
esp->msgout_len = 1;
esp_cmd(esp, eregs, ESP_CMD_TI);
break;
}
esp_advance_phase(esp->current_SC, in_msgoutdone);
return do_intr_end;
}
static int esp_do_msgoutdone(struct NCR_ESP *esp,
struct ESP_regs *eregs)
{
if((esp->msgout_len > 1) && esp->dma_barrier)
esp->dma_barrier(esp);
if(!(esp->ireg & ESP_INTR_DC)) {
esp_cmd(esp, eregs, ESP_CMD_NULL);
switch(esp->sreg & ESP_STAT_PMASK) {
case ESP_MOP:
/* whoops, parity error */
ESPLOG(("esp%d: still in msgout, parity error assumed\n",
esp->esp_id));
if(esp->msgout_len > 1)
esp_cmd(esp, eregs, ESP_CMD_SATN);
esp_advance_phase(esp->current_SC, in_msgout);
return do_work_bus;
case ESP_DIP:
break;
default:
if(!fcount(esp, eregs) &&
!(((struct esp_device *)esp->current_SC->device->hostdata)->sync_max_offset))
esp_cmd(esp, eregs, ESP_CMD_FLUSH);
break;
};
}
/* If we sent out a synchronous negotiation message, update
* our state.
*/
if(esp->cur_msgout[2] == EXTENDED_MESSAGE &&
esp->cur_msgout[4] == EXTENDED_SDTR) {
esp->snip = 1; /* anal retentiveness... */
}
esp->prevmsgout = esp->cur_msgout[0];
esp->msgout_len = 0;
esp_advance_phase(esp->current_SC, in_the_dark);
return esp_do_phase_determine(esp, eregs);
}
static int esp_bus_unexpected(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
ESPLOG(("esp%d: command in weird state %2x\n",
esp->esp_id, esp->current_SC->SCp.phase));
return do_reset_bus;
}
static espfunc_t bus_vector[] = {
esp_do_data_finale,
esp_do_data_finale,
esp_bus_unexpected,
esp_do_msgin,
esp_do_msgincont,
esp_do_msgindone,
esp_do_msgout,
esp_do_msgoutdone,
esp_do_cmdbegin,
esp_do_cmddone,
esp_do_status,
esp_do_freebus,
esp_do_phase_determine,
esp_bus_unexpected,
esp_bus_unexpected,
esp_bus_unexpected,
};
/* This is the second tier in our dual-level SCSI state machine. */
static int esp_work_bus(struct NCR_ESP *esp, struct ESP_regs *eregs)
{
Scsi_Cmnd *SCptr = esp->current_SC;
unsigned int phase;
ESPBUS(("esp_work_bus: "));
if(!SCptr) {
ESPBUS(("reconnect\n"));
return esp_do_reconnect(esp, eregs);
}
phase = SCptr->SCp.phase;
if ((phase & 0xf0) == in_phases_mask)
return bus_vector[(phase & 0x0f)](esp, eregs);
else if((phase & 0xf0) == in_slct_mask)
return esp_select_complete(esp, eregs);
else
return esp_bus_unexpected(esp, eregs);
}
static espfunc_t isvc_vector[] = {
NULL,
esp_do_phase_determine,
esp_do_resetbus,
esp_finish_reset,
esp_work_bus
};
/* Main interrupt handler for an esp adapter. */
void esp_handle(struct NCR_ESP *esp)
{
struct ESP_regs *eregs;
Scsi_Cmnd *SCptr;
int what_next = do_intr_end;
eregs = esp->eregs;
SCptr = esp->current_SC;
if(esp->dma_irq_entry)
esp->dma_irq_entry(esp);
/* Check for errors. */
esp->sreg = esp_read(eregs->esp_status);
esp->sreg &= (~ESP_STAT_INTR);
esp->seqreg = (esp_read(eregs->esp_sstep) & ESP_STEP_VBITS);
esp->ireg = esp_read(eregs->esp_intrpt); /* Unlatch intr and stat regs */
ESPIRQ(("handle_irq: [sreg<%02x> sstep<%02x> ireg<%02x>]\n",
esp->sreg, esp->seqreg, esp->ireg));
if(esp->sreg & (ESP_STAT_SPAM)) {
/* Gross error, could be due to one of:
*
* - top of fifo overwritten, could be because
* we tried to do a synchronous transfer with
* an offset greater than ESP fifo size
*
* - top of command register overwritten
*
* - DMA setup to go in one direction, SCSI
* bus points in the other, whoops
*
* - weird phase change during asynchronous
* data phase while we are initiator
*/
ESPLOG(("esp%d: Gross error sreg=%2x\n", esp->esp_id, esp->sreg));
/* If a command is live on the bus we cannot safely
* reset the bus, so we'll just let the pieces fall
* where they may. Here we are hoping that the
* target will be able to cleanly go away soon
* so we can safely reset things.
*/
if(!SCptr) {
ESPLOG(("esp%d: No current cmd during gross error, "
"resetting bus\n", esp->esp_id));
what_next = do_reset_bus;
goto state_machine;
}
}
/* No current cmd is only valid at this point when there are
* commands off the bus or we are trying a reset.
*/
if(!SCptr && !esp->disconnected_SC && !(esp->ireg & ESP_INTR_SR)) {
/* Panic is safe, since current_SC is null. */
ESPLOG(("esp%d: no command in esp_handle()\n", esp->esp_id));
panic("esp_handle: current_SC == penguin within interrupt!");
}
if(esp->ireg & (ESP_INTR_IC)) {
/* Illegal command fed to ESP. Outside of obvious
* software bugs that could cause this, there is
* a condition with ESP100 where we can confuse the
* ESP into an erroneous illegal command interrupt
* because it does not scrape the FIFO properly
* for reselection. See esp100_reconnect_hwbug()
* to see how we try very hard to avoid this.
*/
ESPLOG(("esp%d: invalid command\n", esp->esp_id));
esp_dump_state(esp, eregs);
if(SCptr) {
/* Devices with very buggy firmware can drop BSY
* during a scatter list interrupt when using sync
* mode transfers. We continue the transfer as
* expected, the target drops the bus, the ESP
* gets confused, and we get a illegal command
* interrupt because the bus is in the disconnected
* state now and ESP_CMD_TI is only allowed when
* a nexus is alive on the bus.
*/
ESPLOG(("esp%d: Forcing async and disabling disconnect for "
"target %d\n", esp->esp_id, SCptr->device->id));
SCptr->device->borken = 1; /* foo on you */
}
what_next = do_reset_bus;
} else if(!(esp->ireg & ~(ESP_INTR_FDONE | ESP_INTR_BSERV | ESP_INTR_DC))) {
int phase;
if(SCptr) {
phase = SCptr->SCp.phase;
if(phase & in_phases_mask) {
what_next = esp_work_bus(esp, eregs);
} else if(phase & in_slct_mask) {
what_next = esp_select_complete(esp, eregs);
} else {
ESPLOG(("esp%d: interrupt for no good reason...\n",
esp->esp_id));
what_next = do_intr_end;
}
} else {
ESPLOG(("esp%d: BSERV or FDONE or DC while SCptr==NULL\n",
esp->esp_id));
what_next = do_reset_bus;
}
} else if(esp->ireg & ESP_INTR_SR) {
ESPLOG(("esp%d: SCSI bus reset interrupt\n", esp->esp_id));
what_next = do_reset_complete;
} else if(esp->ireg & (ESP_INTR_S | ESP_INTR_SATN)) {
ESPLOG(("esp%d: AIEEE we have been selected by another initiator!\n",
esp->esp_id));
what_next = do_reset_bus;
} else if(esp->ireg & ESP_INTR_RSEL) {
if(!SCptr) {
/* This is ok. */
what_next = esp_do_reconnect(esp, eregs);
} else if(SCptr->SCp.phase & in_slct_mask) {
/* Only selection code knows how to clean
* up properly.
*/
ESPDISC(("Reselected during selection attempt\n"));
what_next = esp_select_complete(esp, eregs);
} else {
ESPLOG(("esp%d: Reselected while bus is busy\n",
esp->esp_id));
what_next = do_reset_bus;
}
}
/* This is tier-one in our dual level SCSI state machine. */
state_machine:
while(what_next != do_intr_end) {
if (what_next >= do_phase_determine &&
what_next < do_intr_end)
what_next = isvc_vector[what_next](esp, eregs);
else {
/* state is completely lost ;-( */
ESPLOG(("esp%d: interrupt engine loses state, resetting bus\n",
esp->esp_id));
what_next = do_reset_bus;
}
}
if(esp->dma_irq_exit)
esp->dma_irq_exit(esp);
}
EXPORT_SYMBOL(esp_handle);
#ifndef CONFIG_SMP
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 21:55:46 +08:00
irqreturn_t esp_intr(int irq, void *dev_id)
{
struct NCR_ESP *esp;
unsigned long flags;
int again;
struct Scsi_Host *dev = dev_id;
/* Handle all ESP interrupts showing at this IRQ level. */
spin_lock_irqsave(dev->host_lock, flags);
repeat:
again = 0;
for_each_esp(esp) {
#ifndef __mips__
if(((esp)->irq & 0xff) == irq) {
#endif
if(esp->dma_irq_p(esp)) {
again = 1;
esp->dma_ints_off(esp);
ESPIRQ(("I%d(", esp->esp_id));
esp_handle(esp);
ESPIRQ((")"));
esp->dma_ints_on(esp);
}
#ifndef __mips__
}
#endif
}
if(again)
goto repeat;
spin_unlock_irqrestore(dev->host_lock, flags);
return IRQ_HANDLED;
}
#else
/* For SMP we only service one ESP on the list list at our IRQ level! */
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 21:55:46 +08:00
irqreturn_t esp_intr(int irq, void *dev_id)
{
struct NCR_ESP *esp;
unsigned long flags;
struct Scsi_Host *dev = dev_id;
/* Handle all ESP interrupts showing at this IRQ level. */
spin_lock_irqsave(dev->host_lock, flags);
for_each_esp(esp) {
if(((esp)->irq & 0xf) == irq) {
if(esp->dma_irq_p(esp)) {
esp->dma_ints_off(esp);
ESPIRQ(("I[%d:%d](",
smp_processor_id(), esp->esp_id));
esp_handle(esp);
ESPIRQ((")"));
esp->dma_ints_on(esp);
goto out;
}
}
}
out:
spin_unlock_irqrestore(dev->host_lock, flags);
return IRQ_HANDLED;
}
#endif
int esp_slave_alloc(struct scsi_device *SDptr)
{
struct esp_device *esp_dev =
2007-07-19 16:49:03 +08:00
kzalloc(sizeof(struct esp_device), GFP_ATOMIC);
if (!esp_dev)
return -ENOMEM;
SDptr->hostdata = esp_dev;
return 0;
}
void esp_slave_destroy(struct scsi_device *SDptr)
{
struct NCR_ESP *esp = (struct NCR_ESP *) SDptr->host->hostdata;
esp->targets_present &= ~(1 << sdev_id(SDptr));
kfree(SDptr->hostdata);
SDptr->hostdata = NULL;
}
#ifdef MODULE
int init_module(void) { return 0; }
void cleanup_module(void) {}
void esp_release(void)
{
esps_in_use--;
esps_running = esps_in_use;
}
EXPORT_SYMBOL(esp_release);
#endif
EXPORT_SYMBOL(esp_abort);
EXPORT_SYMBOL(esp_allocate);
EXPORT_SYMBOL(esp_deallocate);
EXPORT_SYMBOL(esp_initialize);
EXPORT_SYMBOL(esp_intr);
EXPORT_SYMBOL(esp_queue);
EXPORT_SYMBOL(esp_reset);
EXPORT_SYMBOL(esp_slave_alloc);
EXPORT_SYMBOL(esp_slave_destroy);
EXPORT_SYMBOL(esps_in_use);
MODULE_LICENSE("GPL");