1999-10-15  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vectors.S (not_window_overflow): Actually the dispatcher
	into exception VSRs.  Bugfix: wrong number was placed in %l3 for
	call to xvsr, so for *some* types of trap, including user traps
	other than those handled directly in-HAL, a vector way off the end
	of the array would be used.  Apparently those other types of trap
	never happen, but just in case...

1999-10-13  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vectors.S: Optionally implement Multiple (rather than
	Single) Vector Trapping.  The config options to control this are
	in the platform HAL because some SPARCs do not support SVT, where
	it is optional the platform says so.  This demands a variable
	sized region be reserved for trampoline code or the vectors in
	RAM, which is done via section .ram_vectors, defined herein.

	* src/sparclite.ld: Define SECTION_ram_vectors(...) to deal with
	the new .ram_vectors section.  This is to provide a variable sized
	area at the start of RAM to hold either trampoline code when SVT
	is selected, or a fill 4k + handlers of vectors when MVT is
	selected.  Previously this was bodged by pretending RAM starts at
	0x4000200 in ROM startup builds.
	(comment from the example .ldi files...)
	The ram_vectors section is to allow some free space to copy
	vectors into from the ROM.  This is required to be variable size
	to accomodate SVT or MVT; 80 bytes vs 4kB.  Copying is not
	necessary with MVT, but it is optional because it may offer
	performance gains.  Copying is required for SVT because the
	(aligned) start of ROM contains initialization instructions.  RAM
	copy is used rather than leave a big gap in the ROM to get an
	aligned address for the trampoline code.  For RAM startup,
	ram_vectors will usually be of size zero, unless MVT and copying
	are enabled for memory estimation reasons.

1999-09-29  Hugo Tyson  <hmt@cygnus.co.uk>

	[src/vectors.S: needed no change!]
	
	* include/vectors.h: Potentially accept a register window count
	from the platform HAL, or according to platform selection.  Make
	more of the dependent macros computed.

	* src/vec_ivsr.S (hal_default_interrupt_vsr): Handle other register
	window sizes correctly; particularly when handling underflow trap
	by hand.

	* src/vec_xvsr.S (hal_default_exception_vsr): Handle other register
	window sizes correctly; particularly when handling underflow trap
	by hand.

	* src/context.S (hal_thread_load_context): Handle other register
	window sizes correctly.

	* src/icontext.c (cyg_hal_sparc_get_gdb_regs): Handle other register
	window sizes correctly; set up initial context differently.

1999-06-07  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_arch.h (hal_jmp_buf): force this to be 64-bit
	aligned so that std ops in hal_setjmp(), hal_longjmp() work.

1999-06-03  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/hal_boot.c: Add a dummy __gccmain() to prevent ctors being
	called twice.  This seems to be a recent compiler change:
	[from gcc/ChangeLog]
	Sat May 22 18:18:43 1999  Jason Merrill  <jason@yorick.cygnus.com>
           * sparc/liteelf.h: Handle ctors like MIPS crosses.
           * sparc/lite.h (STARTFILE_SPEC, LIB_SPEC): Define to "".
           * configure.in: Don't use libgloss.h on sparclite-elf.
           Use collect2 on sparclite-aout.

1999-05-20  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_intr.h (HAL_INTERRUPT_STACK_CALL_PENDING_DSRS): 
	Define this if CYGIMP_HAL_COMMON_INTERRUPTS_USE_INTERRUPT_STACK;
	ie. we are running a separate interrupt stack.  Also define
	symbols for it so that we can monitor its usage easily.

	* src/vec_ivsr.S (hal_interrupt_stack_call_pending_DSRs): Add this
	function to run DSRs on the interrupt stack (if using it).  Switch
	over to executing on the interrupt stack when needed.

	* src/vectors.S (cyg_interrupt_stack_base): Define the interrupt
	stack either if we need it as an interrupt stack or if we need it
	as a bootup stack (simulator); if it's there, use it as a bootup
	stack anyway.

	* src/context.S (hal_thread_load_context): Install the PSR that
	was saved in its entirety, including the interrupt and trap enable
	state.  This is needed for running a separate interrupt stack, but
	harmless when not so doing.

	* src/icontext.c (hal_thread_init_context): Install an initial PSR
	which includes ET, S and PS (Enable traps, Supermode, prevSuper)
	and a PIL (processor interrupt level) of 15 (maximum)
	so that a context switch which preserves interrupt mode will work.

1999-05-13  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_arch.h (HAL_THREAD_INIT_CONTEXT): Align stack
	(rather conservatively) before use.

1999-04-28  Bart Veer  <bartv@cygnus.co.uk>

	* src/PKGconf.mak:
	Remove the -n argument to tail, it does not appear to be required
	on any supported host and causes problems with some
	implementations of tail.

1999-04-15  Jonathan Larmour  <jlarmour@cygnus.co.uk>

	* src/sparclite.ld: Define __bss_end at the end of the BSS
	* src/hal_boot.c (hal_zero_bss): Stop at __bss_end rather than _end
	when clearing BSS
	
	These fix PR 19750

1999-04-14  Jonathan Larmour  <jlarmour@cygnus.co.uk>

	* src/PKGconf.mak (EXTRAS): Don't generate extras.o here any more
	But do define EXTRAS every time for the linker script.

1999-04-09  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_intr.h (HAL_RESTORE_INTERRUPTS et al): Put in 3 nops
	after any writes to the psr; TBH I don't think they're strictly
	necessary when not altering the window state, but for complete
	safety and adherence to the book of words....

1999-04-08  John Dallaway  <jld@cygnus.co.uk>

	* src/*.ld: Revised SECTION_* macro arguments to
	avoid padded output sections (PR 19787)

1999-04-07  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_io.h: If not the simulator, get hal_hwio.h too -
	which describes board-and-CPUvariant-specific IO needs.

1999-04-07  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_intr.h (HAL_VSR_SET_TO_ECOS_HANDLER): Define this
	macro for use in tests that are interested in exception diddling.
	Redefine CYGNUM_HAL_ISR_MIN to describe the number of interrupt
	sources; while there are XSRs too, this definition is better for
	the error checking in kernel interrupt stuff.  Changed the symbols
	used for defining tables sizes to match, since there are still 27
	VSRs and ISRs/XSRs and their data.

	* src/hal_intr.c: Use the correct symbol for the size of the
	isr/vsr/their-data tables.

1999-03-26  Hugo Tyson  <hmt@cygnus.co.uk>

	* tests/sparc_ex.c (cyg_[user_]start):
	If there is no kernel, cyg_user_start() is never called, main gets
	in and steals the CPU.  So we hang.  So if there is no kernel,
	define cyg_start() instead to wrest control.

1999-03-22  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_arch.h (CYGNUM_HAL_STACK_SIZE_TYPICAL): 
	And this time make the edit correctly ;-/

1999-03-22  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_arch.h: 
	Use CYGNUM_HAL_STACK_SIZE_TYPICAL for the stack size instead of
	CYGNUM_HAL_MINIMUM_STACK_SIZE.

1999-03-17  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vectors.S (rom_vectors_end, rom_vectors): Add two new
	symbols for copying trampoline code into RAM in ROM startup.  This
	keeps ROM startup as similar as possible to RAM, for simplicity;
	no re-ordering of vectors.S's contents.

	* src/hal_boot.c (hal_copy_data): Copy in long sized units rather
	than long longs, better to cope with ROM data alignment.
	Also add some LED output (if SLEB hardware) for startup to help
	with debugging ROM start.

1999-03-17  John Dallaway  <jld@cygnus.co.uk>

	* src/PKGconf.mak: Remove dependence on echo '-e' switch.

1999-03-16  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vectors.S (genuine_reset): Add a little extra debug to SLEB
	LEDs, and use the genuine top-of-memory stack when running on
	hardware; only define cyg_bootup_stack in the simulator case.

	* src/sparclite.ld: Define rom_data_start, regardless of start
	type, no harm is done.

1999-03-12  Gary Thomas  <gthomas@cygnus.co.uk>

	* include/pkgconf/hal_sparclite.h: 
	* include/hal_arch.h: Moved definition of 'CYGNUM_HAL_MINIMUM_STACK_SIZE'

1999-03-11  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/icontext.c (cyg_hal_sparc_[gs]et_gdb_regs): 
	Two new routines to communicate our nice, optimized stack save
	states to a reg list for GDB.  My maximal register saves are not
	really suitable for passing to GDB in any case, they're for humans
	to debug the interrupt code more than anything else, and my
	minimal save sets naturally contain pointers and ways of finding
	the rest of the reg set, but it depends on whether it's an
	interrupt save or a synchronous one, and given the SPARClite's
	stack layout, it's a bit hairy.

	* src/vec_ivsr.S (hal_default_interrupt_vsr): 
	Do save %g0 even in a minimal context, so that the code for
	communicating register state with GDB can tell it's a minimum
	context.  Costs one store cycle per interrupt.

	* include/hal_arch.h (HAL_[GS]ET_GDB_REGISTERS): 
	Define and call routines cyg_hal_sparc_[gs]et_gdb_regs for 
	communicating register state of a thread to GDB.

	* src/vectors.S (real_vector): Reinstate the anullment of the
	instructions that were blamed for register corruption earlier;
	they are correct and safe.

1999-03-10  Jesper Skov  <jskov@cygnus.co.uk>

	* include/hal_intr.h (HAL_INTERRUPT_IN_USE): Added.

1999-03-09  Gary Thomas  <gthomas@cygnus.co.uk>

	* src/hal_boot.c: Added missing <cyg/infra/cyg_type.h> include file.

1999-03-05  Gary Thomas  <gthomas@cygnus.co.uk>

	* src/sparclite.ld: 
	* src/PKGconf.mak: Clean up I/O package changes.

1999-03-04  Jonathan Larmour  <jlarmour@cygnus.co.uk>

	* src/sparclite.ld: 
	Unconditionalize use of libextras.a and add to GROUP()

1999-03-04  Gary Thomas  <gthomas@cygnus.co.uk>

	* src/PKGconf.mak: 
	* src/sparclite.ld: New I/O package support.

	* src/vectors.S: Fix register corruption problem - stems from
	use of annulled instructions in window overflow/underflow code.

1999-02-25  Hugo Tyson  <hmt@cygnus.co.uk>

	These log entries from 1999-02-22 to 1999-02-24 were executed in
	a branch; this change is a merge of that branch into the trunk.

	(1999-02-24  Hugo Tyson  <hmt@cygnus.co.uk>)

	* tests/sparc_ex.c (sparc_ex_main): 
	For all traps except the USER_TRAP, install the default VSR
	hal_default_exception_vsr(); this is so that this test catches
	exception ie. is able to work, even when we are playing nice with
	CygMon and GDB.  The USER_TRAP is left so that breakpoints work,
	for the test infrastructure.

	(1999-02-22  Hugo Tyson  <hmt@cygnus.co.uk>)

	* src/vectors.S (real_vector): Remove old testing code.  Add delay
	NOPs after writes to %wim and %psr for safety.  Leave installing
	the trap vector to platform-specific code, along with RAM control
	and cache initialization.

	* src/vec_ivsr.S (hal_default_interrupt_vsr): Add delay NOPs after
	writing the %wim before doing a restore; that was a bug that was.

	* src/vec_xvsr.S (hal_default_exception_vsr): same change.

	* src/hal_intr.c: Remove platform-specific clock period variable; it
	happens that both platforms sim and sleb have one, but it's not a
	requirement of the target architecture.

	* src/hal_boot.c (cyg_hal_start): Move diddling the sim-specific
	watchdog into platform-specific startup code viz. new routines
	hal_board_{pre/post}start().

	* src/context.S: a couple of extra NOPs to ensure that %wim and
	%psr changes work on real hardware.

	(1999-02-22  Hugo Tyson  <hmt@cygnus.co.uk>)

	* include/hal_arch.h (HAL_SET_GDB_REGISTERS): 
	Remove warning about unused var 'cos of the incompleteness of
	these macros; gdb interworking is not yet supported.

1999-02-25  Nick Garnett  <nickg@cygnus.co.uk>

	* include/vectors.h:
	Changed label used to access scheduler lock to one that is not
	mangled by C++. This is intended to make support for interrupt
	handling in non-kernel configurations easier.	

1999-02-20  Jonathan Larmour  <jlarmour@cygnus.co.uk>

	* include/hal_arch.h: 
	Rename deliver_exception() -> cyg_hal_deliver_exception()
	QA improvements
	
	* include/hal_intr.h:
	Reorganise vector/interrupt/exception names according to purpose
	Rename exception_handler() -> cyg_hal_exception_handler()
	QA improvements

	* src/hal_intr.c:
	Rename exception_handler() -> cyg_hal_exception_handler()
	Rename deliver_exception() -> cyg_hal_deliver_exception()
	Rename CYG_ISR_COUNT -> CYGNUM_HAL_ISR_COUNT
	QA improvements

	* src/vec_ivsr.S:
	* src/vec_xvsr.S:
	Rename CYG_ISR_COUNT -> CYGNUM_HAL_ISR_COUNT
	QA improvements

	* tests/sparc_ex.c:
	Rename CYG_EXCEPTION_* -> CYGNUM_HAL_EXCEPTION_*
	Rename CYG_VECTOR_UNALIGNED -> CYGNUM_HAL_VECTOR_UNALIGNED

1999-02-12  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/context.S (hal_thread_load_context, hal_setjmp):
	Disable interrupts (but not traps) while doing the flush of all my
	callers and their callers' callers register sets to the stack.
	There is evidence that the inconsistency in the stack during this
	is a bad thing, though I have to admit that I don't see exactly
	how it goes wrong.  Partly, it's impossible to debug, so if the
	problem I saw (a SEGV in instrumentation of interrupt_end()) is
	reproduced with this change, I'll have more to go on.  This is a
	conservative change in that sense.
	
	* include/hal_arch.h (HAL_FLUSH_REGISTERS_TO_STACK): 
	Disable interrupts but not traps while doing this.  Ditto.

1999-02-12  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/pkgconf/hal_sparclite.h (CYGNUM_HAL_MINIMUM_STACK_SIZE):
	Define a minimal sensible stack size for apps to use.  Note that
	this file is included by hal_arch.h and so is available to all HAL
	users, and in fact almost everyone by default.

1999-02-12  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/hal_intr.c (exception_handler): 
	Include <cyg/infra/cyg_ass.h> so that the CYG_FAIL() in some
	configurations actually compiles (actually: links).
	combo{22,18,10,6} were thus afflicted.

1999-02-11  Hugo Tyson  <hmt@cygnus.co.uk>

	* tests/sparc_ex.c (do_test): Make the test run for much less time
	when CYG_USE_TRACING and in any case bleat periodically about how
	successful it is to prevent testing system timeouts.  Also move
	the test routines (those which provoke the exception) to the end
	so that (ever helpful) -O3 cannot inline them and then "optimize"
	the result and thus spoil the whole damned test too.

1999-02-08  John Dallaway  <jld@cygnus.co.uk>

	* include/pkgconf/hal_sparclite.h: Correct capitalization of
	'SPARClite'.

1999-02-08  John Dallaway  <jld@cygnus.co.uk>

	* src/sparclite.ld: New MLT-compatible linker script
	* src/PKGconf.mak: Add build rules for sparclite.ld

1999-02-04  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vec_ivsr.S (hal_default_interrupt_vsr): Preserve global
	registers over interrupts, and implement support for
	CYGDBG_HAL_COMMON_INTERRUPTS_SAVE_MINIMUM_CONTEXT ie. save a
	minimal register set rather than a maximal.  Create a saved
	register set whose address is passed to the interrupt handler
	(though it may well be mostly empty).
	[also tidied up trailing tabs and long blank lines]

	* src/vec_xvsr.S (hal_default_exception_vsr): Preserve all
	registers over interrupts; create a saved register set whose
	address is passed to the exception handler so it can modify it.
	[also tidied up trailing tabs and long blank lines]

	* src/context.S( hal_setjmp, hal_longjmp):
	(hal_thread_switch_context,hal_thread_load_context): Implement
	CYGDBG_HAL_COMMON_CONTEXT_SAVE_MINIMUM support; ie. save a minimal
	register set rather than a maximal.
	
	* include/hal_arch.h (HAL_FLUSH_REGISTERS_TO_STACK): New macro
	useful for writing exception handlers, for example.
	(HAL_GET_GDB_REGISTERS, HAL_SET_GDB_REGISTERS): provided initial
	dummy versions of these.

	* tests/sparc_ex.c: A test of the exception mechanism:
	deliberately cause unaligned accesses of various sizes and check
	we get exactly one trap and correctly step over it by modifying
	the saved register set whose address the exception ISR is handed.
	This test should be regarded as documentation for the exception
	handler API/ABI.
	
	* tests/PKGconf.mak (TESTS): Build sparc_ex.

1999-02-01  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vectors.S (genuine_reset): 
	Include <cyg/hal/halboot.si> to set up RAM controllers, cache and
	the like - this is rather platform dependant.

1999-02-01  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_intr.h (HAL_TRANSLATE_VECTOR): 
	Include a definition of this important macro.  Not.
	It's used when interrupt chaining is selected, BUT we the HAL make
	no distinction, it's up to the kernel to look after things.  So
	this macro is an assignment.

1999-01-28  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_intr.h: 
	Get interrupt controller definitions from a board-specific file:
	#include <cyg/hal/hal_xpic.h>
	Get clock control definitions from a board-specific file:
	#include <cyg/hal/hal_clock.h>
	Delete the original definitions from this central file.
	
	* include/hal_cache.h: REMOVED to the simulator tree in the first
	instance; more generally to various board-specific files.


1999-01-26  Hugo Tyson  <hmt@cygnus.co.uk>

	* include/hal_cache.h (HAL_DCACHE_ENABLE) (and its friends):
	Define initial empty placeholder versions of these so that cache
	tests build.  Well it's a start.
	
1999-01-21  Jonathan Larmour  <jlarmour@cygnus.co.uk>

	* src/hal_boot.c (cyg_hal_invoke_constructors): 
	Add code to deal with CYGSEM_HAL_STOP_CONSTRUCTORS_ON_FLAG
	Tidy up and update description header
	Shorten needlessly long lines
	Remove all traces of non-CYG_KERNEL_USE_INIT_PRIORITY code

1999-01-21  Hugo Tyson  <hmt@cygnus.co.uk>

	* src/vectors.S (genuine_reset): Set the RAM size to 4MB in the
	erc32 simulator.

	* include/hal_io.h: New file; macros to access IO space, really a
	place holder at present.

	* include/hal_intr.h: Do not define WATCHDOG() macro.

	* src/hal_intr.c (hal_default_isr): Do not mess with the watchdog
	in the the default interrupt handler.

	* src/hal_boot.c (hal_zero_bss): use an efficient 8-byte zero.
	(hal_copy_data): use an efficient 8-byte copy.
	(cyg_hal_start): turn off the ERC32 watchdog timer before we start
	anything very time consuming.

	NB: the 8-byte copy changes require alignment in the linker
	script: see the Changelog in hal/sparclite/sim/... for details.

1999-01-19  Hugo Tyson  <hmt@cygnus.co.uk>

	Modified files in arch:
	* include/hal_arch.h
	* include/hal_intr.h
	* src/PKGconf.mak
	* src/context.S
	* src/hal_intr.c
	* src/vec_ivsr.S
	* src/vec_xvsr.S
	* src/vectors.S

	New files in arch:
	* src/hal_boot.c	(bss initialization, constructors &c)
	* include/hal_cache.h	(empty but required header)

	Modified files in sim:
	* include/hal_diag.h

	Lots more progress; essentially, everything now works.

	Simulator's treatment of stdio is poor though, it doesn't work
	under pkgtest.  This has been slowing me up.

1999-01-14  Hugo Tyson  <hmt@cygnus.co.uk>

	* ecc/ecc/hal/sparclite/...:

	Initial checkin of SPARClite HAL; it doesn't work yet, except the
	most trivial of tests - timer and interrupt manipulation stuff is
	entirely absent.

	Here's some documentation of the initial checkin tree; this will
	change, as platform dependent stuff moves correctly into the sim
	or <board> directories.

	arch/current/include:
		basetype.h	nothing much here.
		vectors.h	definitions of symbols shared between the
			various vector code assembler files, can be
			included in both C and assembler.
		hal_intr.h	various macros for clock and interrupt
			control plus the eCos vector numbering scheme; we
			map thus:
			what	SPARC trap type	    eCos vector numbers
			-----------------------------------------------
			(not used)   ---                0	
			interrupts (17-31)		1-15
			user traps (128-255)		16
			exceptions (1-9,36==4)		17-25
			others	   (10,?)		26
			-----------------------------------------------
			Trap types 5 and 6 which would be eCos vectors 21
			and 22 are handled in the trampoline code in
			file src/vectors.S
		hal_arch.h	definition of a thread context, plus
			macros for thread switching and the like.  Some
			ancillary bit twiddling macros and cache barriers
			too.

	arch/current/include/pkgconf:
		hal_sparclite.h 	pkgconf config/header file.
	
	arch/current/src:
		vectors.S	boot-up stuff, trampoline code that hangs
			on the interrupt/trap vector, handlers for window
			under/overflow.  Saves some state in registers
			then jumps to the interrupt or exception VSR
	   		respectively.
			Also instantiates the VSR table.
		vec_ivsr.S	the default interrupt VSR; it establishes
			a C calling environment (ie. anticipates a window
			overflow) then locks the scheduler, calls the ISR
			with appropriate arguments, and then interrupt_end()
			before undoing that environment, anticipating
			window underflow and returning to the interrupted
			instruction.
		vec_xvsr.S	the default exception AKA trap VSR; does
			the same as the interrupt one but without the
			scheduler lock or interrupt end.
		hal_intr.c	instantiates the ISR table used by the
			code above.
		context.S	context switch code, saves and loads up a
			whole register state, coroutine or longjump-like.
		icontext.c	initializes a context to "jump" to in
			context.S at the birth of a new thread.

	sim/current/include:
		hal_diag.h	macros to deal with debugging output via a
			fake serial device of some kind; initially empty,
			but definitely board/sim dependent.

	sim/current/include/pkgconf:
		hal_sparclite_sim.h 	pkgconf config/header file.

	sim/current/src:
		sim.ld		linker script bits for "sim" target.

//===========================================================================
//####COPYRIGHTBEGIN####
//                                                                          
// -------------------------------------------                              
// The contents of this file are subject to the Red Hat eCos Public License 
// Version 1.0 (the "License"); you may not use this file except in         
// compliance with the License.  You may obtain a copy of the License at    
// http://sourceware.cygnus.com/ecos                                        
//                                                                          
// Software distributed under the License is distributed on an       
// basis, WITHOUT WARRANTY OF ANY KIND, either express or implied.  See the 
// License for the specific language governing rights and limitations under 
// the License.                                                             
//                                                                          
// The Original Code is eCos - Embedded Configurable Operating System,      
// released September 30, 1998.                                             
//                                                                          
// The Initial Developer of the Original Code is Red Hat.                   
// Portions created by Red Hat are                                          
// Copyright (C) 1998, 1999, 2000 Red Hat, Inc.                             
// All Rights Reserved.                                                     
// -------------------------------------------                              
//                                                                          
//####COPYRIGHTEND####
//===========================================================================

//EOF ChangeLog