locore.s

<B>Digital的Unix操作系统VAX 4.2源码</B>

/*    @(#)locore.s	4.10     (ULTRIX)        4/11/91     */

/************************************************************************
 *									*
 *			Copyright (c) 1983 - 1989 by			*
 *		Digital Equipment Corporation, Maynard, MA		*
 *			All rights reserved.				*
 *									*
 *   This software is furnished under a license and may be used and	*
 *   copied  only  in accordance with the terms of such license and	*
 *   with the  inclusion  of  the  above  copyright  notice.   This	*
 *   software  or  any	other copies thereof may not be provided or	*
 *   otherwise made available to any other person.  No title to and	*
 *   ownership of the software is hereby transferred.			*
 *									*
 *   This software is  derived	from  software	received  from	the	*
 *   University    of	California,   Berkeley,   and	from   Bell	*
 *   Laboratories.  Use, duplication, or disclosure is	subject  to	*
 *   restrictions  under  license  agreements  with  University  of	*
 *   California and with AT&T.						*
 *									*
 *   The information in this software is subject to change  without	*
 *   notice  and should not be construed as a commitment by Digital	*
 *   Equipment Corporation.						*
 *									*
 *   Digital assumes no responsibility for the use  or	reliability	*
 *   of its software on equipment which is not supplied by Digital.	*
 *									*
 ************************************************************************/

/************************************************************************
 *
 * Modification history: /sys/vax/locore.s 
 *
 * 11-Apr-91	dlh
 *	add call to vp_reset() during slavestart; call will be made if this CPU
 *	is in vpmask.
 *
 * 30-Jan-91    jaw
 *	fix branch in stop cpu to properly save off stack pointer...
 *
 * 21-Jan-91	dlh
 *	In the case of the VAX9000, the routine cpuident() is accessing the
 *	wrong register to find out which cpu it is.  It was accessing the
 *	CPUCNF.  It now accesses CPUID.
 *
 * 10-Oct-90	paradis
 *	Modify Xprotflt handler to parse all the bits in the
 *	trap type word (including the new vector memory fault
 *	bits)
 *
 * 10-Oct-90	dlh
 *	add vector support for the Mariah (VAX6000-5xx)
 *
 * 4-Sep-90	dlh
 *	added support for vector processor support
 *		- entry point for vector disabled fault
 *		- Rigel - correct the way ACCS is dealt with - want to 
 *		  deal with FPA and vector present bit separately
 *		- a process which is a vector process needs to call
 *		  vp_contextlimbo()
 *		- added routine is_legal_vector_instruction()
 *
 * 31-Aug-90	paradis
 *	Removed debug scaffolding from call to ka9000_rxfct_isr().
 *	Removed call to ka9000_txfct_isr() entirely since we don't
 *	use it anymore.
 *	Set modify bit in PTEs for all kernel-writeable pages.  This
 *	(a) prevents crashes due to taking modify faults inside
 *	fault handlers, and (b) keeps us from taking superfluous 
 *	kernel modify faults (we don't do anything with them anyway).
 *
 * 03-Aug-90	rafiey (Ali Rafieymehr)
 *	Added support for VAX9000.
 *
 * 19-Jun-90 -- jmartin
 *	Fixes for "panic: vrelvm rssize"
 *
 * 07-Jun-90 jas
 *	Added code to Slavestart.  If VAX is 6400/6500, load ACCS with
 *	0.  This will put 6500 in 30-bit mode, and have no affect on 6400.
 *
 * 06-Jun-90 pmk
 *	Remove ci0int routine, ci vector interrupt routine now made by
 *	config.
 *
 * 01-May-90 jas
 *	added 6500 support.
 *
 * 30-Mar-90 jaw
 *	allow 6200 to recover from hard error.
 *
 * 03-Mar-90 jaw
 *	primitive change to optimize mips.
 *
 * 17-Nov-89    Paul Grist
 *      modified so that the pointer to the pc/psl will be passed for
 *      an nmifault for some 8800 error handling inmprovements.
 *
 * 09-Nov-89 	bp
 *	Checked that physical memory doesn't exceed 511.5 Mb and
 *	when it does set it to 511.5 Mb
 *	optimized distcl-jmartin request
 *	have __km_alloc return null when index < MINBUCKET
 *
 * 20-Jul-89	rafiey (Ali Rafieymehr)
 *	Deleted the stub routine xmisst(). The real routine is now
 *	defined in xmiinit.c.
 *
 * 20-Jun-89 -- jaw
 *	fix old 780 bug where it falsely claims to have a protection
 *	fault.  fix stolen from VMS.
 *
 * 13-Jun-89 -- jmartin
 *	Change calculation of P0LR for physical-to-virtual transition
 *	to accomodate SVS kernel.  Before each LDPCTX instruction
 *	write 255 and 0 respectively into PCB_PRVCPU and PCB_ASN.
 *	This is minimal support for ECO 119.
 *
 * 30-May-89	darrell
 *	Added include of ../../machine/common/cpuconf.h -- cpu types
 *	were moved there.  Changed the values for processor types,
 *	and other things for the new cpusw.
 *
 * 22-May-89 -- darrell
 *	Changed tocons to cons_putc -- as part of the new cpusw.
 *
 * 18-May-89	Jaw
 *	fix MSI/CI interrupt vectors to count interrupts in percpu structure.
 *	fix baddaddr for 6400.
 *
 * 12-May-89	Todd M. Katz		TMK0002
 *	1. Increment counter on all CI interrupts.  Also, change the name of
 *	   the SCB vector from Xcia0int -> Xci0int.  Note: The second CI SCB
 *	   vector, Xci1int() has been deleted because only a single CI port
 *	   is currently supported.
 *	2. Completely change MSI interrupt code.
 *	3. Include header file "msi.h".
 *
 * 12-May-89 -- darrell
 *	Merged V3.1 changes.
 *
 * 5-May-89 --	Adrian Thoms
 *	Moved VAX_6200 to cpu #14 for VVAX
 *	Added VVAX dump routines v_qio and v_qioinit
 *
 * 07-March-89 -- gmm
 *	release_uarea_noreturn uses u_procp, instead of an argument to get
 *	the current proc pointer. 
 *
 * 15-Feb-89 -- jaw
 *	Add check to copyin/copyout to make sure not spin locks are held.
 *
 * 09-Feb-89 -- jaw
 *	Move CPU_TBI state flag to seperate longword.  This was done
 *	because the state flag field can only be written by the cpu
 *	who owns it.
 *
 * 1-Feb-89	Tom Kong
 *	Added workaround for Rigel 2nd pass bug.  When probing NXM
 *	that causes machine checks, clear Rigel error registers.
 *
 * 10 Jan 89 -- kong
 *	Added Rigel (VAX6400) support.  Added comments about CPU/system
 *	identification.
 *
 *  02-Feb-89	jaw
 *	Make longjmp mpsafe.  move newpc and newfp global to per-cpudata base.
 *
 * 17-Nov-88	darrell
 *	Added ka60memerr SCBVEC entry for Firefox bus error handling
 *
 *  26-Jan-89	jaw
 *	fix up start/stop cpu.
 *
 * 23-Jan-89 -- jmartin
 *	Introduce a new trap type NOACCESS for protection fault on read.
 *
 * 12-Jan-89 - jaw
 *	merge Xe changes for FFox drivers
 *
 * 03 Jan 89 -- jaw
 *	fix branch in stop_secondary_cpu
 *
 * 29 Dec 88 -- jmartin
 *	Zero physical memory from _edata to (vmbinfo).
 *
 * 08 Dec 88 -- jaw 
 *	When panicing the system put the cpu's into tight loop
 * 	after flushing out context.
 *
 * 22 Nov 88 -- jaw 
 *	fix case where we were sending signal to our parent before we
 *	were REALLY stopped....  more magic.
 *
 * 10 Oct 88 -- jaw
 *	test for lock held after swtch.
 *
 * 31 Aug 88 -- jmartin
 *	Use the lock lk_text in Fastreclaim.  Rationalize Fastreclaim
 *	to use longword operands and to avoid branching on the primary
 *	code path.
 *
 * 19 Aug 88 -- miche
 *	cleaned up context switch code for SMP
 *
 * 25 Jul 88 -- jmartin
 *	Use the lock lk_cmap in Fastreclaim.  Also don't write the bit
 *	SPTECHG to (struct proc *)foo->p_vm, because VAXen don't care.
 *
 * 18-Jun-88 -- jaw  change to new cpu data format.
 *
 * 22 Apr 88 -- mbh
 *	Slave checks for work to do on all queues.  This section
 *	should be consolidated with that of the primary processor.
 *
 * 24 Mar 88 -- prs
 *	Changed copy*str routines to return ENAMETOOLONG, instead
 *	of ENOENT for POSIX conformance.
 *
 * 21 Jan 88 -- jmartin
 *	Replace calls to the (inline) functions clearseg and copyseg
 *	respectively with blkclr (or bzero) and blkcpy (or bcopy).
 *	Establish a window in process memory through which a parent
 *	can write to (and read from) the memory of the child.  This
 *	window is UPAGES*NBPG bytes located between the u-area and the
 *	user stack.  Remove the following entities: CMAP1, CADDR1,
 *	CMAP2, CADDR2, Vfmap, vfutl, clearseg, copyseg.  Redefine
 *	Forkmap and forkutl.  Change the computation for the location
 *	of USRSTACK and the size of the process page table.
 *
 ***********************SMP rev history above
 * 17-Oct-88	jaw
 *	Prefetch istream in bisst routine so that 8200 won't machinecheck.
 *
 * 01-Sep-88	darrell
 *	Added a call to ka60initslave to turn on the slave processors
 *	internal cache.
 *
 * 30-Aug-88	jaw
 *	add proc field p_master to fix ASMP bug.
 *
 * 18-Aug-88	darrell
 *	Added multiprocessor support for VAX60, and a call to chrqueue
 *	in the fcpdma code to get around the fact that the VAX60 console
 *	DZ interrupts come in at IPL 16.
 *
 * 08-Aug-88	Fred Canter
 *	Temporary fix for PVAX SCSI/SCSI controller hardware hack.
 *	Do an I/O space access each time thru the idle loop to
 *	generate address strobe needed to allow the SCSI/SCSI
 *	controller temporary hardware fix to work.
 *
 * 15-Jul-88	Todd M. Katz		TMK0001
 *	Add stub xmisst() routine for resetting XMI nodes.
 *
 * 15-July-88	robin
 *	Added cpu_sub_subtype because cpu_systype was being misused.  This
 *	allows sizer and uerf to both work for the KA640.
 *
 * 26-Apr-88    jaw
 *	Add VAX8820 support.
 *
 * 24-Mar-88 -- robin
 *	added code to set up the interupt vec. for ka640 lance chips (NI)
 *	
 * 07-Mar-88 -- prs
 *	Changed copy*str routines to return ENAMETOOLONG, instead
 *	of ENOENT for POSIX conformance.
 *
 * 12-Feb-88 -- fred (Fred Canter)
 *	Changes for VAX420 (CVAXstar/PVAX) support.
 *	Added case 11 to all casel statements.
 *	Modified MVAX class cpu ident code.
 *	Save arch_ident bits in SID EXT register.
 *
 * 25-Jan-88 -- rsp (Ricky Palmer)
 *      Added interrupt code for MSI.
 *
 * 19-Jan-88 -- jaw
 *	added cpu support for Calypso.  Also moved intrcpu into machdep 
 *	because the routine shouldn't be written in assembly code.
 *
 * 12-11-87	Robin L. and Larry C.
 *	Added portclass/kmalloc support to the system.
 *
 * 2-Oct-87	robin
 *	fixed a problem which stoped the AIO init process from working.
 *	The arg to qioinit is indirect and the code was direct.
 *
 * 20-Apr-87 -- afd
 *	Removed work-around for bugs in the P1 CVAX chips.
 *
 * 19-Mar-87 -- fred (Fred Canter for Brian Stevens)
 *	Schedule X from the idle loop.
 *
 * 06-Mar-87 -- afd
 *	Added CPU specific support for Mayfair/CVAX.
 * 
 * 03 Feb 87 -- depp and woodward
 *	Reinstated the "yech" code in Swtch()/Resume().  It was removed
 *	during the multiprocessing work.  It will only be invoked on the
 *	master CPU.
 *
 * 21 Jan 87 -- jaw
 *	performance fixes to syscall.
 *
 *  16-Dec-86 -- pmk
 *	Changed 750 console tu58 intr. branch to turintr instead of tudma.
 *	750 now uses MRSP protocol
 *
 *  27-Aug-86 -- fred (Fred Canter)
 *	Removed kludge used to identify pass 1 standard cell VAXstar CPUs.
 *
 *  5-Aug-86 -- fred (Fred Canter)
 *	Minor changes to VAXstar SLU pseudo DMA code (sspdma).
 *
 * 18-Jun-86 -- jaw
 *	bisst fix to set NOARB before hitting a node over the head.
 *
 * 18-Jun-86 -- fred (Fred Canter)
 *	Changes for VAXstar kernel support.
 *
 * 22-May-86 -- prs
 *	Added time saving measure to the qioinit routine for
 *	the generic dump driver.
 *
 * 18-Apr-86 -- jaw	hooks for nmi faults and fixes to bierrors.
 *
 * 16-Apr-86 -- darrell
 *	changed badaddr to bbadaddr -- to allow the routine 
 *	badaddr for backward comaptibility.
 *
 * 15-Apr-86 -- jf
 *	Add support for system processes.
 *
 * 14-Apr-86 -- afd
 *	Re-did mutually exclusive MicroVAX areas.  Mainly this is
 *	putting MicroVAX into the case stmts with other cpus.
 *
 * 09-Apr-86 -- prs
 *	Added two qio calls to use the vmb boot driver. These routines
 *	are called from the generic dump driver.
 *
 * 09-Apr-86 -- jaw  add dispatchs for multiple bi error routines.
 *
 * 02-Apr-86 -- jrs
 *	Rewrite scheduler to eliminate special master queue.
 *	Add code in inter-processor interrupts for panic handling.
 *
 * 18-Mar-86 -- jrs
 *	Change intrslave to intrcpu to do one at a time.  Changed
 *	entry conditions for cpuindex and merged w/ cpuident.
 *	Added Jim's change to keep ipl at 1f until machine specific
 *	autoconfig called.
 *
 * 18-mar-86  -- jaw     br/cvec changed to NOT use registers.
 *
 * 10-Mar-86 -- pmk
 *	change panic() entry mask to 0x0fff, save r0-r11 for errlogging.
 *
 * 06-Mar-86 -- tresvik
 *	clear RPB rp_flag so dump will work after copying in VMB's RPB
 *
 * 03-Mar-86 -- jrs
 *	Added code to initialize secondary processor start address in rpb
 *
 * 25-Feb-86 -- jrs
 *	Once again removed cold start check for rxcd routine that was
 *	accidentally added back in previous edit
 *
 * 19-Feb-86 -- bjg
 *	Added sbi error logging, uba error logging
 *	Changed doadump to do tbia before accessing rbp to prevent 
 *	protection fault on 8600
 *
 * 18-Feb-86 -- jrs
 *	Change slavestart to _slavestart so we can pick it up from c.
 *	Also get slave into virtual mode a little sooner.
 *	Cold start check moves up to rxcd() routine.
 *
 * 18-Mar-86 -- jaw  add routines to cpu switch for nexus/unibus addreses
 *		     also got rid of some globals like nexnum.
 *		     ka8800 cleanup.
 *
 * 12-Feb-86 -- pmk
 *	Added save pc,psl to bierror
 *
 * 12-Feb-86 -- jrs
 *	Changed scheduling loop to only do TBIA on demand
 *
 * 05-Feb-86 -- jrs
 *	Added save of pcbb address for adb backtrace
 *
 * 04-Feb-86 -- tresvik
 *	added VMB boot path support
 *
 * 03-Feb-86 -- jaw  Set ingnorebi error bit so to ignore errors VMB cause.
 *
 * 20 Jan 86 -- bjg
 *	Added stray interrupt error logging (logstray)
 *
 * 17-Jul-85 -- jrs
 *	Add run queue locking
 *
 * 26-Oct-85 -- jaw
 * 	Fix to bisst() to reset BDA properly once system is up.
 *
 * 25-Sep-85 -- jaw
 *	Fix to bisst to mask problem with BIIC SST bit.
 *	After a dump...halt a 8200 so memory is preserved.
 *
 * 09 Sep 85 -- Reilly
 *	Modified to handle the new 4.3BSD namei code.
 * 03-Sep-85 -- jaw
 *	Added BI error interrupt code.
 *
 * 09-AUG-85 -- darrell dunnuck
 *	Unibus zero vectors are now counted here in locore.  We set a
 *	timer in ubaerror on the first zero vector that comes in,
 *	and count them here in locore until the timer has expired.
 *	We never do a unibus reset no matter how many zero vectors we get.
 *
 * 26-Jul-85 -- jaw
 *	fixed up Violation to SRM...TBIA data MBZ.
 *
 * 18-Jul-85 -- tresvik
 *	Change the way the end of doadump causes a reboot.  Call tocons
 *	instead of duplicating C code here in assembly.  VAX 8600 requires
 *	special handling already defined in tocons.
 *
 * 11-Jul-85 -- tresvik
 *	Fix bug in handling write timeout.  Branching around the PANIC
 *	macro requires local label other than 1, since 1 is already used
 *	within the macro.
 *
 * 19-Jun-85 -- jaw
 *	VAX8200 name change.
 *
 * 06-Jun-85 -- Jaw
 *	got rid of badwrite and replaced it with bisst(). 
 *
 * 05-May-85 -- Larry Cohen
 * 	network interrupt vectors are now configurable
 *
 * 16-Apr-85 -- lp
 *      Added cpu serial line support for VAX8200. Changed tu58 
 *      interface for non-MRSP machines. 
 *
 *  9 Apr 85 -- depp
 *	Added check for Allocated shared memory page in Fastreclaim
 *
 * 22-Mar-85 -- reilly
 *	Added support for floating point emulation for non microvaxes
 *
 * 13-Mar-85 -jaw
 *	Changes for support of the VAX8200 were merged in.
 *
 * 3-MAR-85 -- darrell
 *    handle sbi interupts.
 *
 * 27-Feb-85 -tresvik
 *	Changes for support of the VAX8600 were merged in.
 *
 * 17 Dec 84 -- jrs
 *	Change setjmp, resume to use new regs to avoid swap problems.
 *	Also add savectx for same reason.  Update code in copyin/copyout
 *	so it can be used with new inline program.
 *
 * 06 Nov 84 -- jrs
 *	Reroll code in swtch/resume to avoid double context switch
 *	and handle most likely case linearly.
 *
 * 31 Oct 84 -- jrs
 *	Remove auto calls to dh and dz timer routines in softclock.
 *	These are now done on a demand basis at a higher level
 *
 * 20 Aug 84 -- larry
 *	removed system page table and made it a seperate file - spt.s
 *	so that SYSPTSIZE could be configurable on a binary kernel.
 *
 *  5 May 84 -- rjl
 *	Added QVSS support. The bit map looks like regular memory so some
 *	of the normal autoconfiguration stuff has to be done here.
 *
 * 12 Apr 84 -- rjl
 *	Added float type check for MicroVAX
 *
 * 23 Mar 84 -- slr
 *	Added emulation support for MicroVAX
 *
 * 15 Feb 84 -- rjl
 *	Added support for MicroVAX intelligent boot. Now saving
 *	r10 and r11 as bootdevice and boothowto for later use by
 *	autoconfig.