diff options
author | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 15:20:36 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 15:20:36 -0700 |
commit | 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch) | |
tree | 0bba044c4ce775e45a88a51686b5d9f90697ea9d /drivers/scsi/53c7xx.scr |
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Diffstat (limited to 'drivers/scsi/53c7xx.scr')
-rw-r--r-- | drivers/scsi/53c7xx.scr | 1591 |
1 files changed, 1591 insertions, 0 deletions
diff --git a/drivers/scsi/53c7xx.scr b/drivers/scsi/53c7xx.scr new file mode 100644 index 00000000000..9c5694a2da8 --- /dev/null +++ b/drivers/scsi/53c7xx.scr @@ -0,0 +1,1591 @@ +#undef DEBUG +#undef EVENTS +#undef NO_SELECTION_TIMEOUT +#define BIG_ENDIAN + +; 53c710 driver. Modified from Drew Eckhardts driver +; for 53c810 by Richard Hirst [richard@sleepie.demon.co.uk] +; +; I have left the script for the 53c8xx family in here, as it is likely +; to be useful to see what I changed when bug hunting. + +; NCR 53c810 driver, main script +; Sponsored by +; iX Multiuser Multitasking Magazine +; hm@ix.de +; +; Copyright 1993, 1994, 1995 Drew Eckhardt +; Visionary Computing +; (Unix and Linux consulting and custom programming) +; drew@PoohSticks.ORG +; +1 (303) 786-7975 +; +; TolerANT and SCSI SCRIPTS are registered trademarks of NCR Corporation. +; +; PRE-ALPHA +; +; For more information, please consult +; +; NCR 53C810 +; PCI-SCSI I/O Processor +; Data Manual +; +; NCR 53C710 +; SCSI I/O Processor +; Programmers Guide +; +; NCR Microelectronics +; 1635 Aeroplaza Drive +; Colorado Springs, CO 80916 +; 1+ (719) 578-3400 +; +; Toll free literature number +; +1 (800) 334-5454 +; +; IMPORTANT : This code is self modifying due to the limitations of +; the NCR53c7,8xx series chips. Persons debugging this code with +; the remote debugger should take this into account, and NOT set +; breakpoints in modified instructions. +; +; Design: +; The NCR53c7,8xx family of SCSI chips are busmasters with an onboard +; microcontroller using a simple instruction set. +; +; So, to minimize the effects of interrupt latency, and to maximize +; throughput, this driver offloads the practical maximum amount +; of processing to the SCSI chip while still maintaining a common +; structure. +; +; Where tradeoffs were needed between efficiency on the older +; chips and the newer NCR53c800 series, the NCR53c800 series +; was chosen. +; +; While the NCR53c700 and NCR53c700-66 lacked the facilities to fully +; automate SCSI transfers without host processor intervention, this +; isn't the case with the NCR53c710 and newer chips which allow +; +; - reads and writes to the internal registers from within the SCSI +; scripts, allowing the SCSI SCRIPTS(tm) code to save processor +; state so that multiple threads of execution are possible, and also +; provide an ALU for loop control, etc. +; +; - table indirect addressing for some instructions. This allows +; pointers to be located relative to the DSA ((Data Structure +; Address) register. +; +; These features make it possible to implement a mailbox style interface, +; where the same piece of code is run to handle I/O for multiple threads +; at once minimizing our need to relocate code. Since the NCR53c700/ +; NCR53c800 series have a unique combination of features, making a +; a standard ingoing/outgoing mailbox system, costly, I've modified it. +; +; - Mailboxes are a mixture of code and data. This lets us greatly +; simplify the NCR53c810 code and do things that would otherwise +; not be possible. +; +; The saved data pointer is now implemented as follows : +; +; Control flow has been architected such that if control reaches +; munge_save_data_pointer, on a restore pointers message or +; reconnection, a jump to the address formerly in the TEMP register +; will allow the SCSI command to resume execution. +; + +; +; Note : the DSA structures must be aligned on 32 bit boundaries, +; since the source and destination of MOVE MEMORY instructions +; must share the same alignment and this is the alignment of the +; NCR registers. +; + +; For some systems (MVME166, for example) dmode is always the same, so don't +; waste time writing it + +#if 1 +#define DMODE_MEMORY_TO_NCR +#define DMODE_MEMORY_TO_MEMORY +#define DMODE_NCR_TO_MEMORY +#else +#define DMODE_MEMORY_TO_NCR MOVE dmode_memory_to_ncr TO DMODE +#define DMODE_MEMORY_TO_MEMORY MOVE dmode_memory_to_memory TO DMODE +#define DMODE_NCR_TO_MEMORY MOVE dmode_ncr_to_memory TO DMODE +#endif + +ABSOLUTE dsa_temp_lun = 0 ; Patch to lun for current dsa +ABSOLUTE dsa_temp_next = 0 ; Patch to dsa next for current dsa +ABSOLUTE dsa_temp_addr_next = 0 ; Patch to address of dsa next address + ; for current dsa +ABSOLUTE dsa_temp_sync = 0 ; Patch to address of per-target + ; sync routine +ABSOLUTE dsa_sscf_710 = 0 ; Patch to address of per-target + ; sscf value (53c710) +ABSOLUTE dsa_temp_target = 0 ; Patch to id for current dsa +ABSOLUTE dsa_temp_addr_saved_pointer = 0; Patch to address of per-command + ; saved data pointer +ABSOLUTE dsa_temp_addr_residual = 0 ; Patch to address of per-command + ; current residual code +ABSOLUTE dsa_temp_addr_saved_residual = 0; Patch to address of per-command + ; saved residual code +ABSOLUTE dsa_temp_addr_new_value = 0 ; Address of value for JUMP operand +ABSOLUTE dsa_temp_addr_array_value = 0 ; Address to copy to +ABSOLUTE dsa_temp_addr_dsa_value = 0 ; Address of this DSA value + +; +; Once a device has initiated reselection, we need to compare it +; against the singly linked list of commands which have disconnected +; and are pending reselection. These commands are maintained in +; an unordered singly linked list of DSA structures, through the +; DSA pointers at their 'centers' headed by the reconnect_dsa_head +; pointer. +; +; To avoid complications in removing commands from the list, +; I minimize the amount of expensive (at eight operations per +; addition @ 500-600ns each) pointer operations which must +; be done in the NCR driver by precomputing them on the +; host processor during dsa structure generation. +; +; The fixed-up per DSA code knows how to recognize the nexus +; associated with the corresponding SCSI command, and modifies +; the source and destination pointers for the MOVE MEMORY +; instruction which is executed when reselected_ok is called +; to remove the command from the list. Similarly, DSA is +; loaded with the address of the next DSA structure and +; reselected_check_next is called if a failure occurs. +; +; Perhaps more concisely, the net effect of the mess is +; +; for (dsa = reconnect_dsa_head, dest = &reconnect_dsa_head, +; src = NULL; dsa; dest = &dsa->next, dsa = dsa->next) { +; src = &dsa->next; +; if (target_id == dsa->id && target_lun == dsa->lun) { +; *dest = *src; +; break; +; } +; } +; +; if (!dsa) +; error (int_err_unexpected_reselect); +; else +; longjmp (dsa->jump_resume, 0); +; +; + +#if (CHIP != 700) && (CHIP != 70066) +; Define DSA structure used for mailboxes +ENTRY dsa_code_template +dsa_code_template: +ENTRY dsa_code_begin +dsa_code_begin: +; RGH: Don't care about TEMP and DSA here + DMODE_MEMORY_TO_NCR + MOVE MEMORY 4, dsa_temp_addr_dsa_value, addr_scratch + DMODE_MEMORY_TO_MEMORY +#if (CHIP == 710) + MOVE MEMORY 4, addr_scratch, saved_dsa + ; We are about to go and select the device, so must set SSCF bits + MOVE MEMORY 4, dsa_sscf_710, addr_scratch +#ifdef BIG_ENDIAN + MOVE SCRATCH3 TO SFBR +#else + MOVE SCRATCH0 TO SFBR +#endif + MOVE SFBR TO SBCL + MOVE MEMORY 4, saved_dsa, addr_dsa +#else + CALL scratch_to_dsa +#endif + CALL select +; Handle the phase mismatch which may have resulted from the +; MOVE FROM dsa_msgout if we returned here. The CLEAR ATN +; may or may not be necessary, and we should update script_asm.pl +; to handle multiple pieces. + CLEAR ATN + CLEAR ACK + +; Replace second operand with address of JUMP instruction dest operand +; in schedule table for this DSA. Becomes dsa_jump_dest in 53c7,8xx.c. +ENTRY dsa_code_fix_jump +dsa_code_fix_jump: + MOVE MEMORY 4, NOP_insn, 0 + JUMP select_done + +; wrong_dsa loads the DSA register with the value of the dsa_next +; field. +; +wrong_dsa: +#if (CHIP == 710) +; NOTE DSA is corrupt when we arrive here! +#endif +; Patch the MOVE MEMORY INSTRUCTION such that +; the destination address is the address of the OLD +; next pointer. +; + MOVE MEMORY 4, dsa_temp_addr_next, reselected_ok_patch + 8 + DMODE_MEMORY_TO_NCR +; +; Move the _contents_ of the next pointer into the DSA register as +; the next I_T_L or I_T_L_Q tupple to check against the established +; nexus. +; + MOVE MEMORY 4, dsa_temp_next, addr_scratch + DMODE_MEMORY_TO_MEMORY +#if (CHIP == 710) + MOVE MEMORY 4, addr_scratch, saved_dsa + MOVE MEMORY 4, saved_dsa, addr_dsa +#else + CALL scratch_to_dsa +#endif + JUMP reselected_check_next + +ABSOLUTE dsa_save_data_pointer = 0 +ENTRY dsa_code_save_data_pointer +dsa_code_save_data_pointer: +#if (CHIP == 710) + ; When we get here, TEMP has been saved in jump_temp+4, DSA is corrupt + ; We MUST return with DSA correct + MOVE MEMORY 4, jump_temp+4, dsa_temp_addr_saved_pointer +; HARD CODED : 24 bytes needs to agree with 53c7,8xx.h + MOVE MEMORY 24, dsa_temp_addr_residual, dsa_temp_addr_saved_residual + CLEAR ACK +#ifdef DEBUG + INT int_debug_saved +#endif + MOVE MEMORY 4, saved_dsa, addr_dsa + JUMP jump_temp +#else + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_temp, dsa_temp_addr_saved_pointer + DMODE_MEMORY_TO_MEMORY +; HARD CODED : 24 bytes needs to agree with 53c7,8xx.h + MOVE MEMORY 24, dsa_temp_addr_residual, dsa_temp_addr_saved_residual + CLEAR ACK +#ifdef DEBUG + INT int_debug_saved +#endif + RETURN +#endif +ABSOLUTE dsa_restore_pointers = 0 +ENTRY dsa_code_restore_pointers +dsa_code_restore_pointers: +#if (CHIP == 710) + ; TEMP and DSA are corrupt when we get here, but who cares! + MOVE MEMORY 4, dsa_temp_addr_saved_pointer, jump_temp + 4 +; HARD CODED : 24 bytes needs to agree with 53c7,8xx.h + MOVE MEMORY 24, dsa_temp_addr_saved_residual, dsa_temp_addr_residual + CLEAR ACK + ; Restore DSA, note we don't care about TEMP + MOVE MEMORY 4, saved_dsa, addr_dsa +#ifdef DEBUG + INT int_debug_restored +#endif + JUMP jump_temp +#else + DMODE_MEMORY_TO_NCR + MOVE MEMORY 4, dsa_temp_addr_saved_pointer, addr_temp + DMODE_MEMORY_TO_MEMORY +; HARD CODED : 24 bytes needs to agree with 53c7,8xx.h + MOVE MEMORY 24, dsa_temp_addr_saved_residual, dsa_temp_addr_residual + CLEAR ACK +#ifdef DEBUG + INT int_debug_restored +#endif + RETURN +#endif + +ABSOLUTE dsa_check_reselect = 0 +; dsa_check_reselect determines whether or not the current target and +; lun match the current DSA +ENTRY dsa_code_check_reselect +dsa_code_check_reselect: +#if (CHIP == 710) + /* Arrives here with DSA correct */ + /* Assumes we are always ID 7 */ + MOVE LCRC TO SFBR ; LCRC has our ID and his ID bits set + JUMP REL (wrong_dsa), IF NOT dsa_temp_target, AND MASK 0x80 +#else + MOVE SSID TO SFBR ; SSID contains 3 bit target ID +; FIXME : we need to accommodate bit fielded and binary here for '7xx/'8xx chips + JUMP REL (wrong_dsa), IF NOT dsa_temp_target, AND MASK 0xf8 +#endif +; +; Hack - move to scratch first, since SFBR is not writeable +; via the CPU and hence a MOVE MEMORY instruction. +; + DMODE_MEMORY_TO_NCR + MOVE MEMORY 1, reselected_identify, addr_scratch + DMODE_MEMORY_TO_MEMORY +#ifdef BIG_ENDIAN + ; BIG ENDIAN ON MVME16x + MOVE SCRATCH3 TO SFBR +#else + MOVE SCRATCH0 TO SFBR +#endif +; FIXME : we need to accommodate bit fielded and binary here for '7xx/'8xx chips +; Are you sure about that? richard@sleepie.demon.co.uk + JUMP REL (wrong_dsa), IF NOT dsa_temp_lun, AND MASK 0xf8 +; Patch the MOVE MEMORY INSTRUCTION such that +; the source address is the address of this dsa's +; next pointer. + MOVE MEMORY 4, dsa_temp_addr_next, reselected_ok_patch + 4 + CALL reselected_ok +#if (CHIP == 710) +; Restore DSA following memory moves in reselected_ok +; dsa_temp_sync doesn't really care about DSA, but it has an +; optional debug INT so a valid DSA is a good idea. + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif + CALL dsa_temp_sync +; Release ACK on the IDENTIFY message _after_ we've set the synchronous +; transfer parameters! + CLEAR ACK +; Implicitly restore pointers on reselection, so a RETURN +; will transfer control back to the right spot. + CALL REL (dsa_code_restore_pointers) + RETURN +ENTRY dsa_zero +dsa_zero: +ENTRY dsa_code_template_end +dsa_code_template_end: + +; Perform sanity check for dsa_fields_start == dsa_code_template_end - +; dsa_zero, puke. + +ABSOLUTE dsa_fields_start = 0 ; Sanity marker + ; pad 48 bytes (fix this RSN) +ABSOLUTE dsa_next = 48 ; len 4 Next DSA + ; del 4 Previous DSA address +ABSOLUTE dsa_cmnd = 56 ; len 4 Scsi_Cmnd * for this thread. +ABSOLUTE dsa_select = 60 ; len 4 Device ID, Period, Offset for + ; table indirect select +ABSOLUTE dsa_msgout = 64 ; len 8 table indirect move parameter for + ; select message +ABSOLUTE dsa_cmdout = 72 ; len 8 table indirect move parameter for + ; command +ABSOLUTE dsa_dataout = 80 ; len 4 code pointer for dataout +ABSOLUTE dsa_datain = 84 ; len 4 code pointer for datain +ABSOLUTE dsa_msgin = 88 ; len 8 table indirect move for msgin +ABSOLUTE dsa_status = 96 ; len 8 table indirect move for status byte +ABSOLUTE dsa_msgout_other = 104 ; len 8 table indirect for normal message out + ; (Synchronous transfer negotiation, etc). +ABSOLUTE dsa_end = 112 + +ABSOLUTE schedule = 0 ; Array of JUMP dsa_begin or JUMP (next), + ; terminated by a call to JUMP wait_reselect + +; Linked lists of DSA structures +ABSOLUTE reconnect_dsa_head = 0 ; Link list of DSAs which can reconnect +ABSOLUTE addr_reconnect_dsa_head = 0 ; Address of variable containing + ; address of reconnect_dsa_head + +; These select the source and destination of a MOVE MEMORY instruction +ABSOLUTE dmode_memory_to_memory = 0x0 +ABSOLUTE dmode_memory_to_ncr = 0x0 +ABSOLUTE dmode_ncr_to_memory = 0x0 + +ABSOLUTE addr_scratch = 0x0 +ABSOLUTE addr_temp = 0x0 +#if (CHIP == 710) +ABSOLUTE saved_dsa = 0x0 +ABSOLUTE emulfly = 0x0 +ABSOLUTE addr_dsa = 0x0 +#endif +#endif /* CHIP != 700 && CHIP != 70066 */ + +; Interrupts - +; MSB indicates type +; 0 handle error condition +; 1 handle message +; 2 handle normal condition +; 3 debugging interrupt +; 4 testing interrupt +; Next byte indicates specific error + +; XXX not yet implemented, I'm not sure if I want to - +; Next byte indicates the routine the error occurred in +; The LSB indicates the specific place the error occurred + +ABSOLUTE int_err_unexpected_phase = 0x00000000 ; Unexpected phase encountered +ABSOLUTE int_err_selected = 0x00010000 ; SELECTED (nee RESELECTED) +ABSOLUTE int_err_unexpected_reselect = 0x00020000 +ABSOLUTE int_err_check_condition = 0x00030000 +ABSOLUTE int_err_no_phase = 0x00040000 +ABSOLUTE int_msg_wdtr = 0x01000000 ; WDTR message received +ABSOLUTE int_msg_sdtr = 0x01010000 ; SDTR received +ABSOLUTE int_msg_1 = 0x01020000 ; single byte special message + ; received + +ABSOLUTE int_norm_select_complete = 0x02000000 ; Select complete, reprogram + ; registers. +ABSOLUTE int_norm_reselect_complete = 0x02010000 ; Nexus established +ABSOLUTE int_norm_command_complete = 0x02020000 ; Command complete +ABSOLUTE int_norm_disconnected = 0x02030000 ; Disconnected +ABSOLUTE int_norm_aborted =0x02040000 ; Aborted *dsa +ABSOLUTE int_norm_reset = 0x02050000 ; Generated BUS reset. +ABSOLUTE int_norm_emulateintfly = 0x02060000 ; 53C710 Emulated intfly +ABSOLUTE int_debug_break = 0x03000000 ; Break point +#ifdef DEBUG +ABSOLUTE int_debug_scheduled = 0x03010000 ; new I/O scheduled +ABSOLUTE int_debug_idle = 0x03020000 ; scheduler is idle +ABSOLUTE int_debug_dsa_loaded = 0x03030000 ; dsa reloaded +ABSOLUTE int_debug_reselected = 0x03040000 ; NCR reselected +ABSOLUTE int_debug_head = 0x03050000 ; issue head overwritten +ABSOLUTE int_debug_disconnected = 0x03060000 ; disconnected +ABSOLUTE int_debug_disconnect_msg = 0x03070000 ; got message to disconnect +ABSOLUTE int_debug_dsa_schedule = 0x03080000 ; in dsa_schedule +ABSOLUTE int_debug_reselect_check = 0x03090000 ; Check for reselection of DSA +ABSOLUTE int_debug_reselected_ok = 0x030a0000 ; Reselection accepted +#endif +ABSOLUTE int_debug_panic = 0x030b0000 ; Panic driver +#ifdef DEBUG +ABSOLUTE int_debug_saved = 0x030c0000 ; save/restore pointers +ABSOLUTE int_debug_restored = 0x030d0000 +ABSOLUTE int_debug_sync = 0x030e0000 ; Sanity check synchronous + ; parameters. +ABSOLUTE int_debug_datain = 0x030f0000 ; going into data in phase + ; now. +ABSOLUTE int_debug_check_dsa = 0x03100000 ; Sanity check DSA against + ; SDID. +#endif + +ABSOLUTE int_test_1 = 0x04000000 ; Test 1 complete +ABSOLUTE int_test_2 = 0x04010000 ; Test 2 complete +ABSOLUTE int_test_3 = 0x04020000 ; Test 3 complete + + +; These should start with 0x05000000, with low bits incrementing for +; each one. + +#ifdef EVENTS +ABSOLUTE int_EVENT_SELECT = 0 +ABSOLUTE int_EVENT_DISCONNECT = 0 +ABSOLUTE int_EVENT_RESELECT = 0 +ABSOLUTE int_EVENT_COMPLETE = 0 +ABSOLUTE int_EVENT_IDLE = 0 +ABSOLUTE int_EVENT_SELECT_FAILED = 0 +ABSOLUTE int_EVENT_BEFORE_SELECT = 0 +ABSOLUTE int_EVENT_RESELECT_FAILED = 0 +#endif + +ABSOLUTE NCR53c7xx_msg_abort = 0 ; Pointer to abort message +ABSOLUTE NCR53c7xx_msg_reject = 0 ; Pointer to reject message +ABSOLUTE NCR53c7xx_zero = 0 ; long with zero in it, use for source +ABSOLUTE NCR53c7xx_sink = 0 ; long to dump worthless data in +ABSOLUTE NOP_insn = 0 ; NOP instruction + +; Pointer to message, potentially multi-byte +ABSOLUTE msg_buf = 0 + +; Pointer to holding area for reselection information +ABSOLUTE reselected_identify = 0 +ABSOLUTE reselected_tag = 0 + +; Request sense command pointer, it's a 6 byte command, should +; be constant for all commands since we always want 16 bytes of +; sense and we don't need to change any fields as we did under +; SCSI-I when we actually cared about the LUN field. +;EXTERNAL NCR53c7xx_sense ; Request sense command + +#if (CHIP != 700) && (CHIP != 70066) +; dsa_schedule +; PURPOSE : after a DISCONNECT message has been received, and pointers +; saved, insert the current DSA structure at the head of the +; disconnected queue and fall through to the scheduler. +; +; CALLS : OK +; +; INPUTS : dsa - current DSA structure, reconnect_dsa_head - list +; of disconnected commands +; +; MODIFIES : SCRATCH, reconnect_dsa_head +; +; EXITS : always passes control to schedule + +ENTRY dsa_schedule +dsa_schedule: +#ifdef DEBUG + INT int_debug_dsa_schedule +#endif + +; +; Calculate the address of the next pointer within the DSA +; structure of the command that is currently disconnecting +; +#if (CHIP == 710) + ; Read what should be the current DSA from memory - actual DSA + ; register is probably corrupt + MOVE MEMORY 4, saved_dsa, addr_scratch +#else + CALL dsa_to_scratch +#endif + MOVE SCRATCH0 + dsa_next TO SCRATCH0 + MOVE SCRATCH1 + 0 TO SCRATCH1 WITH CARRY + MOVE SCRATCH2 + 0 TO SCRATCH2 WITH CARRY + MOVE SCRATCH3 + 0 TO SCRATCH3 WITH CARRY + +; Point the next field of this DSA structure at the current disconnected +; list + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, dsa_schedule_insert + 8 + DMODE_MEMORY_TO_MEMORY +dsa_schedule_insert: + MOVE MEMORY 4, reconnect_dsa_head, 0 + +; And update the head pointer. +#if (CHIP == 710) + ; Read what should be the current DSA from memory - actual DSA + ; register is probably corrupt + MOVE MEMORY 4, saved_dsa, addr_scratch +#else + CALL dsa_to_scratch +#endif + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, reconnect_dsa_head + DMODE_MEMORY_TO_MEMORY +/* Temporarily, see what happens. */ +#ifndef ORIGINAL +#if (CHIP != 710) + MOVE SCNTL2 & 0x7f TO SCNTL2 +#endif + CLEAR ACK +#endif +#if (CHIP == 710) + ; Time to correct DSA following memory move + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif + WAIT DISCONNECT +#ifdef EVENTS + INT int_EVENT_DISCONNECT; +#endif +#ifdef DEBUG + INT int_debug_disconnected +#endif + JUMP schedule +#endif + +; +; select +; +; PURPOSE : establish a nexus for the SCSI command referenced by DSA. +; On success, the current DSA structure is removed from the issue +; queue. Usually, this is entered as a fall-through from schedule, +; although the contingent allegiance handling code will write +; the select entry address to the DSP to restart a command as a +; REQUEST SENSE. A message is sent (usually IDENTIFY, although +; additional SDTR or WDTR messages may be sent). COMMAND OUT +; is handled. +; +; INPUTS : DSA - SCSI command, issue_dsa_head +; +; CALLS : NOT OK +; +; MODIFIES : SCRATCH, issue_dsa_head +; +; EXITS : on reselection or selection, go to select_failed +; otherwise, RETURN so control is passed back to +; dsa_begin. +; + +ENTRY select +select: + +#ifdef EVENTS + INT int_EVENT_BEFORE_SELECT +#endif + +#ifdef DEBUG + INT int_debug_scheduled +#endif + CLEAR TARGET + +; XXX +; +; In effect, SELECTION operations are backgrounded, with execution +; continuing until code which waits for REQ or a fatal interrupt is +; encountered. +; +; So, for more performance, we could overlap the code which removes +; the command from the NCRs issue queue with the selection, but +; at this point I don't want to deal with the error recovery. +; + +#if (CHIP != 700) && (CHIP != 70066) +#if (CHIP == 710) + ; Enable selection timer +#ifdef NO_SELECTION_TIMEOUT + MOVE CTEST7 & 0xff TO CTEST7 +#else + MOVE CTEST7 & 0xef TO CTEST7 +#endif +#endif + SELECT ATN FROM dsa_select, select_failed + JUMP select_msgout, WHEN MSG_OUT +ENTRY select_msgout +select_msgout: +#if (CHIP == 710) + ; Disable selection timer + MOVE CTEST7 | 0x10 TO CTEST7 +#endif + MOVE FROM dsa_msgout, WHEN MSG_OUT +#else +ENTRY select_msgout + SELECT ATN 0, select_failed +select_msgout: + MOVE 0, 0, WHEN MSGOUT +#endif + +#ifdef EVENTS + INT int_EVENT_SELECT +#endif + RETURN + +; +; select_done +; +; PURPOSE: continue on to normal data transfer; called as the exit +; point from dsa_begin. +; +; INPUTS: dsa +; +; CALLS: OK +; +; + +select_done: +#if (CHIP == 710) +; NOTE DSA is corrupt when we arrive here! + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif + +#ifdef DEBUG +ENTRY select_check_dsa +select_check_dsa: + INT int_debug_check_dsa +#endif + +; After a successful selection, we should get either a CMD phase or +; some transfer request negotiation message. + + JUMP cmdout, WHEN CMD + INT int_err_unexpected_phase, WHEN NOT MSG_IN + +select_msg_in: + CALL msg_in, WHEN MSG_IN + JUMP select_msg_in, WHEN MSG_IN + +cmdout: + INT int_err_unexpected_phase, WHEN NOT CMD +#if (CHIP == 700) + INT int_norm_selected +#endif +ENTRY cmdout_cmdout +cmdout_cmdout: +#if (CHIP != 700) && (CHIP != 70066) + MOVE FROM dsa_cmdout, WHEN CMD +#else + MOVE 0, 0, WHEN CMD +#endif /* (CHIP != 700) && (CHIP != 70066) */ + +; +; data_transfer +; other_out +; other_in +; other_transfer +; +; PURPOSE : handle the main data transfer for a SCSI command in +; several parts. In the first part, data_transfer, DATA_IN +; and DATA_OUT phases are allowed, with the user provided +; code (usually dynamically generated based on the scatter/gather +; list associated with a SCSI command) called to handle these +; phases. +; +; After control has passed to one of the user provided +; DATA_IN or DATA_OUT routines, back calls are made to +; other_transfer_in or other_transfer_out to handle non-DATA IN +; and DATA OUT phases respectively, with the state of the active +; data pointer being preserved in TEMP. +; +; On completion, the user code passes control to other_transfer +; which causes DATA_IN and DATA_OUT to result in unexpected_phase +; interrupts so that data overruns may be trapped. +; +; INPUTS : DSA - SCSI command +; +; CALLS : OK in data_transfer_start, not ok in other_out and other_in, ok in +; other_transfer +; +; MODIFIES : SCRATCH +; +; EXITS : if STATUS IN is detected, signifying command completion, +; the NCR jumps to command_complete. If MSG IN occurs, a +; CALL is made to msg_in. Otherwise, other_transfer runs in +; an infinite loop. +; + +ENTRY data_transfer +data_transfer: + JUMP cmdout_cmdout, WHEN CMD + CALL msg_in, WHEN MSG_IN + INT int_err_unexpected_phase, WHEN MSG_OUT + JUMP do_dataout, WHEN DATA_OUT + JUMP do_datain, WHEN DATA_IN + JUMP command_complete, WHEN STATUS + JUMP data_transfer +ENTRY end_data_transfer +end_data_transfer: + +; +; FIXME: On NCR53c700 and NCR53c700-66 chips, do_dataout/do_datain +; should be fixed up whenever the nexus changes so it can point to the +; correct routine for that command. +; + +#if (CHIP != 700) && (CHIP != 70066) +; Nasty jump to dsa->dataout +do_dataout: +#if (CHIP == 710) + MOVE MEMORY 4, saved_dsa, addr_scratch +#else + CALL dsa_to_scratch +#endif + MOVE SCRATCH0 + dsa_dataout TO SCRATCH0 + MOVE SCRATCH1 + 0 TO SCRATCH1 WITH CARRY + MOVE SCRATCH2 + 0 TO SCRATCH2 WITH CARRY + MOVE SCRATCH3 + 0 TO SCRATCH3 WITH CARRY + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, dataout_to_jump + 4 + DMODE_MEMORY_TO_MEMORY +dataout_to_jump: + MOVE MEMORY 4, 0, dataout_jump + 4 +#if (CHIP == 710) + ; Time to correct DSA following memory move + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif +dataout_jump: + JUMP 0 + +; Nasty jump to dsa->dsain +do_datain: +#if (CHIP == 710) + MOVE MEMORY 4, saved_dsa, addr_scratch +#else + CALL dsa_to_scratch +#endif + MOVE SCRATCH0 + dsa_datain TO SCRATCH0 + MOVE SCRATCH1 + 0 TO SCRATCH1 WITH CARRY + MOVE SCRATCH2 + 0 TO SCRATCH2 WITH CARRY + MOVE SCRATCH3 + 0 TO SCRATCH3 WITH CARRY + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, datain_to_jump + 4 + DMODE_MEMORY_TO_MEMORY +ENTRY datain_to_jump +datain_to_jump: + MOVE MEMORY 4, 0, datain_jump + 4 +#if (CHIP == 710) + ; Time to correct DSA following memory move + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif +#ifdef DEBUG + INT int_debug_datain +#endif +datain_jump: + JUMP 0 +#endif /* (CHIP != 700) && (CHIP != 70066) */ + + +; Note that other_out and other_in loop until a non-data phase +; is discovered, so we only execute return statements when we +; can go on to the next data phase block move statement. + +ENTRY other_out +other_out: +#if 0 + INT 0x03ffdead +#endif + INT int_err_unexpected_phase, WHEN CMD + JUMP msg_in_restart, WHEN MSG_IN + INT int_err_unexpected_phase, WHEN MSG_OUT + INT int_err_unexpected_phase, WHEN DATA_IN + JUMP command_complete, WHEN STATUS + JUMP other_out, WHEN NOT DATA_OUT +#if (CHIP == 710) +; TEMP should be OK, as we got here from a call in the user dataout code. +#endif + RETURN + +ENTRY other_in +other_in: +#if 0 + INT 0x03ffdead +#endif + INT int_err_unexpected_phase, WHEN CMD + JUMP msg_in_restart, WHEN MSG_IN + INT int_err_unexpected_phase, WHEN MSG_OUT + INT int_err_unexpected_phase, WHEN DATA_OUT + JUMP command_complete, WHEN STATUS + JUMP other_in, WHEN NOT DATA_IN +#if (CHIP == 710) +; TEMP should be OK, as we got here from a call in the user datain code. +#endif + RETURN + + +ENTRY other_transfer +other_transfer: + INT int_err_unexpected_phase, WHEN CMD + CALL msg_in, WHEN MSG_IN + INT int_err_unexpected_phase, WHEN MSG_OUT + INT int_err_unexpected_phase, WHEN DATA_OUT + INT int_err_unexpected_phase, WHEN DATA_IN + JUMP command_complete, WHEN STATUS + JUMP other_transfer + +; +; msg_in_restart +; msg_in +; munge_msg +; +; PURPOSE : process messages from a target. msg_in is called when the +; caller hasn't read the first byte of the message. munge_message +; is called when the caller has read the first byte of the message, +; and left it in SFBR. msg_in_restart is called when the caller +; hasn't read the first byte of the message, and wishes RETURN +; to transfer control back to the address of the conditional +; CALL instruction rather than to the instruction after it. +; +; Various int_* interrupts are generated when the host system +; needs to intervene, as is the case with SDTR, WDTR, and +; INITIATE RECOVERY messages. +; +; When the host system handles one of these interrupts, +; it can respond by reentering at reject_message, +; which rejects the message and returns control to +; the caller of msg_in or munge_msg, accept_message +; which clears ACK and returns control, or reply_message +; which sends the message pointed to by the DSA +; msgout_other table indirect field. +; +; DISCONNECT messages are handled by moving the command +; to the reconnect_dsa_queue. +#if (CHIP == 710) +; NOTE: DSA should be valid when we get here - we cannot save both it +; and TEMP in this routine. +#endif +; +; INPUTS : DSA - SCSI COMMAND, SFBR - first byte of message (munge_msg +; only) +; +; CALLS : NO. The TEMP register isn't backed up to allow nested calls. +; +; MODIFIES : SCRATCH, DSA on DISCONNECT +; +; EXITS : On receipt of SAVE DATA POINTER, RESTORE POINTERS, +; and normal return from message handlers running under +; Linux, control is returned to the caller. Receipt +; of DISCONNECT messages pass control to dsa_schedule. +; +ENTRY msg_in_restart +msg_in_restart: +; XXX - hackish +; +; Since it's easier to debug changes to the statically +; compiled code, rather than the dynamically generated +; stuff, such as +; +; MOVE x, y, WHEN data_phase +; CALL other_z, WHEN NOT data_phase +; MOVE x, y, WHEN data_phase +; +; I'd like to have certain routines (notably the message handler) +; restart on the conditional call rather than the next instruction. +; +; So, subtract 8 from the return address + + MOVE TEMP0 + 0xf8 TO TEMP0 + MOVE TEMP1 + 0xff TO TEMP1 WITH CARRY + MOVE TEMP2 + 0xff TO TEMP2 WITH CARRY + MOVE TEMP3 + 0xff TO TEMP3 WITH CARRY + +ENTRY msg_in +msg_in: + MOVE 1, msg_buf, WHEN MSG_IN + +munge_msg: + JUMP munge_extended, IF 0x01 ; EXTENDED MESSAGE + JUMP munge_2, IF 0x20, AND MASK 0xdf ; two byte message +; +; XXX - I've seen a handful of broken SCSI devices which fail to issue +; a SAVE POINTERS message before disconnecting in the middle of +; a transfer, assuming that the DATA POINTER will be implicitly +; restored. +; +; Historically, I've often done an implicit save when the DISCONNECT +; message is processed. We may want to consider having the option of +; doing that here. +; + JUMP munge_save_data_pointer, IF 0x02 ; SAVE DATA POINTER + JUMP munge_restore_pointers, IF 0x03 ; RESTORE POINTERS + JUMP munge_disconnect, IF 0x04 ; DISCONNECT + INT int_msg_1, IF 0x07 ; MESSAGE REJECT + INT int_msg_1, IF 0x0f ; INITIATE RECOVERY +#ifdef EVENTS + INT int_EVENT_SELECT_FAILED +#endif + JUMP reject_message + +munge_2: + JUMP reject_message +; +; The SCSI standard allows targets to recover from transient +; error conditions by backing up the data pointer with a +; RESTORE POINTERS message. +; +; So, we must save and restore the _residual_ code as well as +; the current instruction pointer. Because of this messiness, +; it is simpler to put dynamic code in the dsa for this and to +; just do a simple jump down there. +; + +munge_save_data_pointer: +#if (CHIP == 710) + ; We have something in TEMP here, so first we must save that + MOVE TEMP0 TO SFBR + MOVE SFBR TO SCRATCH0 + MOVE TEMP1 TO SFBR + MOVE SFBR TO SCRATCH1 + MOVE TEMP2 TO SFBR + MOVE SFBR TO SCRATCH2 + MOVE TEMP3 TO SFBR + MOVE SFBR TO SCRATCH3 + MOVE MEMORY 4, addr_scratch, jump_temp + 4 + ; Now restore DSA + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif + MOVE DSA0 + dsa_save_data_pointer TO SFBR + MOVE SFBR TO SCRATCH0 + MOVE DSA1 + 0xff TO SFBR WITH CARRY + MOVE SFBR TO SCRATCH1 + MOVE DSA2 + 0xff TO SFBR WITH CARRY + MOVE SFBR TO SCRATCH2 + MOVE DSA3 + 0xff TO SFBR WITH CARRY + MOVE SFBR TO SCRATCH3 + + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, jump_dsa_save + 4 + DMODE_MEMORY_TO_MEMORY +jump_dsa_save: + JUMP 0 + +munge_restore_pointers: +#if (CHIP == 710) + ; The code at dsa_restore_pointers will RETURN, but we don't care + ; about TEMP here, as it will overwrite it anyway. +#endif + MOVE DSA0 + dsa_restore_pointers TO SFBR + MOVE SFBR TO SCRATCH0 + MOVE DSA1 + 0xff TO SFBR WITH CARRY + MOVE SFBR TO SCRATCH1 + MOVE DSA2 + 0xff TO SFBR WITH CARRY + MOVE SFBR TO SCRATCH2 + MOVE DSA3 + 0xff TO SFBR WITH CARRY + MOVE SFBR TO SCRATCH3 + + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, jump_dsa_restore + 4 + DMODE_MEMORY_TO_MEMORY +jump_dsa_restore: + JUMP 0 + + +munge_disconnect: +#ifdef DEBUG + INT int_debug_disconnect_msg +#endif + +/* + * Before, we overlapped processing with waiting for disconnect, but + * debugging was beginning to appear messy. Temporarily move things + * to just before the WAIT DISCONNECT. + */ + +#ifdef ORIGINAL +#if (CHIP == 710) +; Following clears Unexpected Disconnect bit. What do we do? +#else + MOVE SCNTL2 & 0x7f TO SCNTL2 +#endif + CLEAR ACK +#endif + +#if (CHIP != 700) && (CHIP != 70066) + JUMP dsa_schedule +#else + WAIT DISCONNECT + INT int_norm_disconnected +#endif + +munge_extended: + CLEAR ACK + INT int_err_unexpected_phase, WHEN NOT MSG_IN + MOVE 1, msg_buf + 1, WHEN MSG_IN + JUMP munge_extended_2, IF 0x02 + JUMP munge_extended_3, IF 0x03 + JUMP reject_message + +munge_extended_2: + CLEAR ACK + MOVE 1, msg_buf + 2, WHEN MSG_IN + JUMP reject_message, IF NOT 0x02 ; Must be WDTR + CLEAR ACK + MOVE 1, msg_buf + 3, WHEN MSG_IN + INT int_msg_wdtr + +munge_extended_3: + CLEAR ACK + MOVE 1, msg_buf + 2, WHEN MSG_IN + JUMP reject_message, IF NOT 0x01 ; Must be SDTR + CLEAR ACK + MOVE 2, msg_buf + 3, WHEN MSG_IN + INT int_msg_sdtr + +ENTRY reject_message +reject_message: + SET ATN + CLEAR ACK + MOVE 1, NCR53c7xx_msg_reject, WHEN MSG_OUT + RETURN + +ENTRY accept_message +accept_message: + CLEAR ATN + CLEAR ACK + RETURN + +ENTRY respond_message +respond_message: + SET ATN + CLEAR ACK + MOVE FROM dsa_msgout_other, WHEN MSG_OUT + RETURN + +; +; command_complete +; +; PURPOSE : handle command termination when STATUS IN is detected by reading +; a status byte followed by a command termination message. +; +; Normal termination results in an INTFLY instruction, and +; the host system can pick out which command terminated by +; examining the MESSAGE and STATUS buffers of all currently +; executing commands; +; +; Abnormal (CHECK_CONDITION) termination results in an +; int_err_check_condition interrupt so that a REQUEST SENSE +; command can be issued out-of-order so that no other command +; clears the contingent allegiance condition. +; +; +; INPUTS : DSA - command +; +; CALLS : OK +; +; EXITS : On successful termination, control is passed to schedule. +; On abnormal termination, the user will usually modify the +; DSA fields and corresponding buffers and return control +; to select. +; + +ENTRY command_complete +command_complete: + MOVE FROM dsa_status, WHEN STATUS +#if (CHIP != 700) && (CHIP != 70066) + MOVE SFBR TO SCRATCH0 ; Save status +#endif /* (CHIP != 700) && (CHIP != 70066) */ +ENTRY command_complete_msgin +command_complete_msgin: + MOVE FROM dsa_msgin, WHEN MSG_IN +; Indicate that we should be expecting a disconnect +#if (CHIP != 710) + MOVE SCNTL2 & 0x7f TO SCNTL2 +#else + ; Above code cleared the Unexpected Disconnect bit, what do we do? +#endif + CLEAR ACK +#if (CHIP != 700) && (CHIP != 70066) + WAIT DISCONNECT + +; +; The SCSI specification states that when a UNIT ATTENTION condition +; is pending, as indicated by a CHECK CONDITION status message, +; the target shall revert to asynchronous transfers. Since +; synchronous transfers parameters are maintained on a per INITIATOR/TARGET +; basis, and returning control to our scheduler could work on a command +; running on another lun on that target using the old parameters, we must +; interrupt the host processor to get them changed, or change them ourselves. +; +; Once SCSI-II tagged queueing is implemented, things will be even more +; hairy, since contingent allegiance conditions exist on a per-target/lun +; basis, and issuing a new command with a different tag would clear it. +; In these cases, we must interrupt the host processor to get a request +; added to the HEAD of the queue with the request sense command, or we +; must automatically issue the request sense command. + +#if 0 + MOVE SCRATCH0 TO SFBR + JUMP command_failed, IF 0x02 +#endif +#if (CHIP == 710) +#if defined(MVME16x_INTFLY) +; For MVME16x (ie CHIP=710) we will force an INTFLY by triggering a software +; interrupt (SW7). We can use SCRATCH, as we are about to jump to +; schedule, which corrupts it anyway. Will probably remove this later, +; but want to check performance effects first. + +#define INTFLY_ADDR 0xfff40070 + + MOVE 0 TO SCRATCH0 + MOVE 0x80 TO SCRATCH1 + MOVE 0 TO SCRATCH2 + MOVE 0 TO SCRATCH3 + MOVE MEMORY 4, addr_scratch, INTFLY_ADDR +#else + INT int_norm_emulateintfly +#endif +#else + INTFLY +#endif +#endif /* (CHIP != 700) && (CHIP != 70066) */ +#if (CHIP == 710) + ; Time to correct DSA following memory move + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif +#ifdef EVENTS + INT int_EVENT_COMPLETE +#endif +#if (CHIP != 700) && (CHIP != 70066) + JUMP schedule +command_failed: + INT int_err_check_condition +#else + INT int_norm_command_complete +#endif + +; +; wait_reselect +; +; PURPOSE : This is essentially the idle routine, where control lands +; when there are no new processes to schedule. wait_reselect +; waits for reselection, selection, and new commands. +; +; When a successful reselection occurs, with the aid +; of fixed up code in each DSA, wait_reselect walks the +; reconnect_dsa_queue, asking each dsa if the target ID +; and LUN match its. +; +; If a match is found, a call is made back to reselected_ok, +; which through the miracles of self modifying code, extracts +; the found DSA from the reconnect_dsa_queue and then +; returns control to the DSAs thread of execution. +; +; INPUTS : NONE +; +; CALLS : OK +; +; MODIFIES : DSA, +; +; EXITS : On successful reselection, control is returned to the +; DSA which called reselected_ok. If the WAIT RESELECT +; was interrupted by a new commands arrival signaled by +; SIG_P, control is passed to schedule. If the NCR is +; selected, the host system is interrupted with an +; int_err_selected which is usually responded to by +; setting DSP to the target_abort address. + +ENTRY wait_reselect +wait_reselect: +#ifdef EVENTS + int int_EVENT_IDLE +#endif +#ifdef DEBUG + int int_debug_idle +#endif + WAIT RESELECT wait_reselect_failed + +reselected: +#ifdef EVENTS + int int_EVENT_RESELECT +#endif + CLEAR TARGET + DMODE_MEMORY_TO_MEMORY + ; Read all data needed to reestablish the nexus - + MOVE 1, reselected_identify, WHEN MSG_IN + ; We used to CLEAR ACK here. +#if (CHIP != 700) && (CHIP != 70066) +#ifdef DEBUG + int int_debug_reselected +#endif + + ; Point DSA at the current head of the disconnected queue. + DMODE_MEMORY_TO_NCR + MOVE MEMORY 4, reconnect_dsa_head, addr_scratch + DMODE_MEMORY_TO_MEMORY +#if (CHIP == 710) + MOVE MEMORY 4, addr_scratch, saved_dsa +#else + CALL scratch_to_dsa +#endif + + ; Fix the update-next pointer so that the reconnect_dsa_head + ; pointer is the one that will be updated if this DSA is a hit + ; and we remove it from the queue. + + MOVE MEMORY 4, addr_reconnect_dsa_head, reselected_ok_patch + 8 +#if (CHIP == 710) + ; Time to correct DSA following memory move + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif + +ENTRY reselected_check_next +reselected_check_next: +#ifdef DEBUG + INT int_debug_reselect_check +#endif + ; Check for a NULL pointer. + MOVE DSA0 TO SFBR + JUMP reselected_not_end, IF NOT 0 + MOVE DSA1 TO SFBR + JUMP reselected_not_end, IF NOT 0 + MOVE DSA2 TO SFBR + JUMP reselected_not_end, IF NOT 0 + MOVE DSA3 TO SFBR + JUMP reselected_not_end, IF NOT 0 + INT int_err_unexpected_reselect + +reselected_not_end: + ; + ; XXX the ALU is only eight bits wide, and the assembler + ; wont do the dirt work for us. As long as dsa_check_reselect + ; is negative, we need to sign extend with 1 bits to the full + ; 32 bit width of the address. + ; + ; A potential work around would be to have a known alignment + ; of the DSA structure such that the base address plus + ; dsa_check_reselect doesn't require carrying from bytes + ; higher than the LSB. + ; + + MOVE DSA0 TO SFBR + MOVE SFBR + dsa_check_reselect TO SCRATCH0 + MOVE DSA1 TO SFBR + MOVE SFBR + 0xff TO SCRATCH1 WITH CARRY + MOVE DSA2 TO SFBR + MOVE SFBR + 0xff TO SCRATCH2 WITH CARRY + MOVE DSA3 TO SFBR + MOVE SFBR + 0xff TO SCRATCH3 WITH CARRY + + DMODE_NCR_TO_MEMORY + MOVE MEMORY 4, addr_scratch, reselected_check + 4 + DMODE_MEMORY_TO_MEMORY +#if (CHIP == 710) + ; Time to correct DSA following memory move + MOVE MEMORY 4, saved_dsa, addr_dsa +#endif +reselected_check: + JUMP 0 + + +; +; +#if (CHIP == 710) +; We have problems here - the memory move corrupts TEMP and DSA. This +; routine is called from DSA code, and patched from many places. Scratch +; is probably free when it is called. +; We have to: +; copy temp to scratch, one byte at a time +; write scratch to patch a jump in place of the return +; do the move memory +; jump to the patched in return address +; DSA is corrupt when we get here, and can be left corrupt + +ENTRY reselected_ok +reselected_ok: + MOVE TEMP0 TO SFBR + MOVE SFBR TO SCRATCH0 + MOVE TEMP1 TO SFBR + MOVE SFBR TO SCRATCH1 + MOVE TEMP2 TO SFBR + MOVE SFBR TO SCRATCH2 + MOVE TEMP3 TO SFBR + MOVE SFBR TO SCRATCH3 + MOVE MEMORY 4, addr_scratch, reselected_ok_jump + 4 +reselected_ok_patch: + MOVE MEMORY 4, 0, 0 +reselected_ok_jump: + JUMP 0 +#else +ENTRY reselected_ok +reselected_ok: +reselected_ok_patch: + MOVE MEMORY 4, 0, 0 ; Patched : first word + ; is address of + ; successful dsa_next + ; Second word is last + ; unsuccessful dsa_next, + ; starting with + ; dsa_reconnect_head + ; We used to CLEAR ACK here. +#ifdef DEBUG + INT int_debug_reselected_ok +#endif +#ifdef DEBUG + INT int_debug_check_dsa +#endif + RETURN ; Return control to where +#endif +#else + INT int_norm_reselected +#endif /* (CHIP != 700) && (CHIP != 70066) */ + +selected: + INT int_err_selected; + +; +; A select or reselect failure can be caused by one of two conditions : +; 1. SIG_P was set. This will be the case if the user has written +; a new value to a previously NULL head of the issue queue. +; +; 2. The NCR53c810 was selected or reselected by another device. +; +; 3. The bus was already busy since we were selected or reselected +; before starting the command. + +wait_reselect_failed: +#ifdef EVENTS + INT int_EVENT_RESELECT_FAILED +#endif +; Check selected bit. +#if (CHIP == 710) + ; Must work out how to tell if we are selected.... +#else + MOVE SIST0 & 0x20 TO SFBR + JUMP selected, IF 0x20 +#endif +; Reading CTEST2 clears the SIG_P bit in the ISTAT register. + MOVE CTEST2 & 0x40 TO SFBR + JUMP schedule, IF 0x40 +; Check connected bit. +; FIXME: this needs to change if we support target mode + MOVE ISTAT & 0x08 TO SFBR + JUMP reselected, IF 0x08 +; FIXME : Something bogus happened, and we shouldn't fail silently. +#if 0 + JUMP schedule +#else + INT int_debug_panic +#endif + + +select_failed: +#if (CHIP == 710) + ; Disable selection timer + MOVE CTEST7 | 0x10 TO CTEST7 +#endif +#ifdef EVENTS + int int_EVENT_SELECT_FAILED +#endif +; Otherwise, mask the selected and reselected bits off SIST0 +#if (CHIP ==710) + ; Let's assume we don't get selected for now + MOVE SSTAT0 & 0x10 TO SFBR +#else + MOVE SIST0 & 0x30 TO SFBR + JUMP selected, IF 0x20 +#endif + JUMP reselected, IF 0x10 +; If SIGP is set, the user just gave us another command, and +; we should restart or return to the scheduler. +; Reading CTEST2 clears the SIG_P bit in the ISTAT register. + MOVE CTEST2 & 0x40 TO SFBR + JUMP select, IF 0x40 +; Check connected bit. +; FIXME: this needs to change if we support target mode +; FIXME: is this really necessary? + MOVE ISTAT & 0x08 TO SFBR + JUMP reselected, IF 0x08 +; FIXME : Something bogus happened, and we shouldn't fail silently. +#if 0 + JUMP schedule +#else + INT int_debug_panic +#endif + +; +; test_1 +; test_2 +; +; PURPOSE : run some verification tests on the NCR. test_1 +; copies test_src to test_dest and interrupts the host +; processor, testing for cache coherency and interrupt +; problems in the processes. +; +; test_2 runs a command with offsets relative to the +; DSA on entry, and is useful for miscellaneous experimentation. +; + +; Verify that interrupts are working correctly and that we don't +; have a cache invalidation problem. + +ABSOLUTE test_src = 0, test_dest = 0 +ENTRY test_1 +test_1: + MOVE MEMORY 4, test_src, test_dest + INT int_test_1 + +; +; Run arbitrary commands, with test code establishing a DSA +; + +ENTRY test_2 +test_2: + CLEAR TARGET +#if (CHIP == 710) + ; Enable selection timer +#ifdef NO_SELECTION_TIMEOUT + MOVE CTEST7 & 0xff TO CTEST7 +#else + MOVE CTEST7 & 0xef TO CTEST7 +#endif +#endif + SELECT ATN FROM 0, test_2_fail + JUMP test_2_msgout, WHEN MSG_OUT +ENTRY test_2_msgout +test_2_msgout: +#if (CHIP == 710) + ; Disable selection timer + MOVE CTEST7 | 0x10 TO CTEST7 +#endif + MOVE FROM 8, WHEN MSG_OUT + MOVE FROM 16, WHEN CMD + MOVE FROM 24, WHEN DATA_IN + MOVE FROM 32, WHEN STATUS + MOVE FROM 40, WHEN MSG_IN +#if (CHIP != 710) + MOVE SCNTL2 & 0x7f TO SCNTL2 +#endif + CLEAR ACK + WAIT DISCONNECT +test_2_fail: +#if (CHIP == 710) + ; Disable selection timer + MOVE CTEST7 | 0x10 TO CTEST7 +#endif + INT int_test_2 + +ENTRY debug_break +debug_break: + INT int_debug_break + +; +; initiator_abort +; target_abort +; +; PURPOSE : Abort the currently established nexus from with initiator +; or target mode. +; +; + +ENTRY target_abort +target_abort: + SET TARGET + DISCONNECT + CLEAR TARGET + JUMP schedule + +ENTRY initiator_abort +initiator_abort: + SET ATN +; +; The SCSI-I specification says that targets may go into MSG out at +; their leisure upon receipt of the ATN single. On all versions of the +; specification, we can't change phases until REQ transitions true->false, +; so we need to sink/source one byte of data to allow the transition. +; +; For the sake of safety, we'll only source one byte of data in all +; cases, but to accommodate the SCSI-I dain bramage, we'll sink an +; arbitrary number of bytes. + JUMP spew_cmd, WHEN CMD + JUMP eat_msgin, WHEN MSG_IN + JUMP eat_datain, WHEN DATA_IN + JUMP eat_status, WHEN STATUS + JUMP spew_dataout, WHEN DATA_OUT + JUMP sated +spew_cmd: + MOVE 1, NCR53c7xx_zero, WHEN CMD + JUMP sated +eat_msgin: + MOVE 1, NCR53c7xx_sink, WHEN MSG_IN + JUMP eat_msgin, WHEN MSG_IN + JUMP sated +eat_status: + MOVE 1, NCR53c7xx_sink, WHEN STATUS + JUMP eat_status, WHEN STATUS + JUMP sated +eat_datain: + MOVE 1, NCR53c7xx_sink, WHEN DATA_IN + JUMP eat_datain, WHEN DATA_IN + JUMP sated +spew_dataout: + MOVE 1, NCR53c7xx_zero, WHEN DATA_OUT +sated: +#if (CHIP != 710) + MOVE SCNTL2 & 0x7f TO SCNTL2 +#endif + MOVE 1, NCR53c7xx_msg_abort, WHEN MSG_OUT + WAIT DISCONNECT + INT int_norm_aborted + +#if (CHIP != 710) +; +; dsa_to_scratch +; scratch_to_dsa +; +; PURPOSE : +; The NCR chips cannot do a move memory instruction with the DSA register +; as the source or destination. So, we provide a couple of subroutines +; that let us switch between the DSA register and scratch register. +; +; Memory moves to/from the DSPS register also don't work, but we +; don't use them. +; +; + + +dsa_to_scratch: + MOVE DSA0 TO SFBR + MOVE SFBR TO SCRATCH0 + MOVE DSA1 TO SFBR + MOVE SFBR TO SCRATCH1 + MOVE DSA2 TO SFBR + MOVE SFBR TO SCRATCH2 + MOVE DSA3 TO SFBR + MOVE SFBR TO SCRATCH3 + RETURN + +scratch_to_dsa: + MOVE SCRATCH0 TO SFBR + MOVE SFBR TO DSA0 + MOVE SCRATCH1 TO SFBR + MOVE SFBR TO DSA1 + MOVE SCRATCH2 TO SFBR + MOVE SFBR TO DSA2 + MOVE SCRATCH3 TO SFBR + MOVE SFBR TO DSA3 + RETURN +#endif + +#if (CHIP == 710) +; Little patched jump, used to overcome problems with TEMP getting +; corrupted on memory moves. + +jump_temp: + JUMP 0 +#endif |