changelog

eCos1.31版

2000-02-16  Jesper Skov  <[e-mail]>

	* src/vectors.S: 
	* src/hal_boot.c: 
	CYG_HAL_SPARCLITE_ > CYGPKG_HAL_SPARCELITE_

2000-02-16  Jesper Skov  <[e-mail]>

	* cdl/hal_sparclite.cdl: removed fix me

2000-01-19  Hugo Tyson  <[e-mail]>

	* cdl/*.cdl: Add descriptions to a number of options &c which were
	lacking same, also tidied up other typos as noticed en passant.

1999-12-02  John Dallaway  <[e-mail]>

	* cdl/hal_sparclite.cdl:

	Use the <PACKAGE> token in custom rules.

1999-12-01  John Dallaway  <[e-mail]>

	* cdl/hal_sparclite.cdl:

	Use the <PREFIX> token in custom rules.

1999-11-04  John Dallaway  <[e-mail]>

	* cdl/hal_sparclite.cdl:

	Output custom rule dependency information to .deps files in
	the current directory.

	Dispense with the need to create a 'src' sub-directory.

1999-11-03  John Dallaway  <[e-mail]>

	* cdl/hal_sparclite.cdl: Define build options.

1999-11-01  Jesper Skov  <[e-mail]>

	* cdl/hal_sparclite.cdl: Addded.

1999-10-15  Hugo Tyson  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

	[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  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>
           * 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  <[e-mail]>

	* 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  <[e-mail]>

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

1999-04-28  Bart Veer  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

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

1999-04-07  Hugo Tyson  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

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

1999-03-22  Hugo Tyson  <[e-mail]>

	* 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  <[e-mail]>

	* 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  <[e-mail]>

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

1999-03-16  Hugo Tyson  <[e-mail]>

	* 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  <[e-mail]>

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

1999-03-11  Hugo Tyson  <[e-mail]>

	* 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  <[e-mail]>

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

1999-03-09  Gary Thomas  <[e-mail]>

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

1999-03-05  Gary Thomas  <[e-mail]>

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

1999-03-04  Jonathan Larmour  <[e-mail]>

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

1999-03-04  Gary Thomas  <[e-mail]>

	* 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  <[e-mail]>

	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  <[e-mail]>)

	* 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  <[e-mail]>)

	* 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.