226 lines
		
	
	
	
		
			7.2 KiB
			
		
	
	
	
		
			ArmAsm
		
	
	
	
	
	
		
		
			
		
	
	
			226 lines
		
	
	
	
		
			7.2 KiB
			
		
	
	
	
		
			ArmAsm
		
	
	
	
	
	
|   | /* | ||
|  |  * arch/alpha/lib/ev6-clear_user.S | ||
|  |  * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com>
 | ||
|  |  * | ||
|  |  * Zero user space, handling exceptions as we go. | ||
|  |  * | ||
|  |  * We have to make sure that $0 is always up-to-date and contains the | ||
|  |  * right "bytes left to zero" value (and that it is updated only _after_ | ||
|  |  * a successful copy).  There is also some rather minor exception setup | ||
|  |  * stuff. | ||
|  |  * | ||
|  |  * NOTE! This is not directly C-callable, because the calling semantics | ||
|  |  * are different: | ||
|  |  * | ||
|  |  * Inputs: | ||
|  |  *	length in $0 | ||
|  |  *	destination address in $6 | ||
|  |  *	exception pointer in $7 | ||
|  |  *	return address in $28 (exceptions expect it there) | ||
|  |  * | ||
|  |  * Outputs: | ||
|  |  *	bytes left to copy in $0 | ||
|  |  * | ||
|  |  * Clobbers: | ||
|  |  *	$1,$2,$3,$4,$5,$6 | ||
|  |  * | ||
|  |  * Much of the information about 21264 scheduling/coding comes from: | ||
|  |  *	Compiler Writer's Guide for the Alpha 21264 | ||
|  |  *	abbreviated as 'CWG' in other comments here | ||
|  |  *	ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html | ||
|  |  * Scheduling notation: | ||
|  |  *	E	- either cluster | ||
|  |  *	U	- upper subcluster; U0 - subcluster U0; U1 - subcluster U1
 | ||
|  |  *	L	- lower subcluster; L0 - subcluster L0; L1 - subcluster L1
 | ||
|  |  * Try not to change the actual algorithm if possible for consistency. | ||
|  |  * Determining actual stalls (other than slotting) doesn't appear to be easy to do. | ||
|  |  * From perusing the source code context where this routine is called, it is | ||
|  |  * a fair assumption that significant fractions of entire pages are zeroed, so | ||
|  |  * it's going to be worth the effort to hand-unroll a big loop, and use wh64. | ||
|  |  * ASSUMPTION: | ||
|  |  *	The believed purpose of only updating $0 after a store is that a signal | ||
|  |  *	may come along during the execution of this chunk of code, and we don't | ||
|  |  *	want to leave a hole (and we also want to avoid repeating lots of work) | ||
|  |  */ | ||
|  | 
 | ||
|  | /* Allow an exception for an insn; exit if we get one.  */ | ||
|  | #define EX(x,y...)			\ | ||
|  | 	99: x,##y;			\ | ||
|  | 	.section __ex_table,"a";	\
 | ||
|  | 	.long 99b - .;			\
 | ||
|  | 	lda $31, $exception-99b($31); 	\
 | ||
|  | 	.previous | ||
|  | 
 | ||
|  | 	.set noat
 | ||
|  | 	.set noreorder
 | ||
|  | 	.align 4
 | ||
|  | 
 | ||
|  | 	.globl __do_clear_user
 | ||
|  | 	.ent __do_clear_user
 | ||
|  | 	.frame	$30, 0, $28 | ||
|  | 	.prologue 0
 | ||
|  | 
 | ||
|  | 				# Pipeline info : Slotting & Comments | ||
|  | __do_clear_user: | ||
|  | 	and	$6, 7, $4	# .. E  .. ..	: find dest head misalignment | ||
|  | 	beq	$0, $zerolength # U  .. .. ..	:  U L U L | ||
|  | 
 | ||
|  | 	addq	$0, $4, $1	# .. .. .. E	: bias counter | ||
|  | 	and	$1, 7, $2	# .. .. E  ..	: number of misaligned bytes in tail | ||
|  | # Note - we never actually use $2, so this is a moot computation | ||
|  | # and we can rewrite this later... | ||
|  | 	srl	$1, 3, $1	# .. E  .. ..	: number of quadwords to clear | ||
|  | 	beq	$4, $headalign	# U  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | /* | ||
|  |  * Head is not aligned.  Write (8 - $4) bytes to head of destination | ||
|  |  * This means $6 is known to be misaligned | ||
|  |  */ | ||
|  | 	EX( ldq_u $5, 0($6) )	# .. .. .. L	: load dst word to mask back in | ||
|  | 	beq	$1, $onebyte	# .. .. U  ..	: sub-word store? | ||
|  | 	mskql	$5, $6, $5	# .. U  .. ..	: take care of misaligned head | ||
|  | 	addq	$6, 8, $6	# E  .. .. .. 	: L U U L | ||
|  | 
 | ||
|  | 	EX( stq_u $5, -8($6) )	# .. .. .. L	: | ||
|  | 	subq	$1, 1, $1	# .. .. E  ..	: | ||
|  | 	addq	$0, $4, $0	# .. E  .. ..	: bytes left -= 8 - misalignment | ||
|  | 	subq	$0, 8, $0	# E  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | 	.align	4
 | ||
|  | /* | ||
|  |  * (The .align directive ought to be a moot point) | ||
|  |  * values upon initial entry to the loop | ||
|  |  * $1 is number of quadwords to clear (zero is a valid value) | ||
|  |  * $2 is number of trailing bytes (0..7) ($2 never used...) | ||
|  |  * $6 is known to be aligned 0mod8 | ||
|  |  */ | ||
|  | $headalign: | ||
|  | 	subq	$1, 16, $4	# .. .. .. E	: If < 16, we can not use the huge loop | ||
|  | 	and	$6, 0x3f, $2	# .. .. E  ..	: Forward work for huge loop | ||
|  | 	subq	$2, 0x40, $3	# .. E  .. ..	: bias counter (huge loop) | ||
|  | 	blt	$4, $trailquad	# U  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | /* | ||
|  |  * We know that we're going to do at least 16 quads, which means we are | ||
|  |  * going to be able to use the large block clear loop at least once. | ||
|  |  * Figure out how many quads we need to clear before we are 0mod64 aligned | ||
|  |  * so we can use the wh64 instruction. | ||
|  |  */ | ||
|  | 
 | ||
|  | 	nop			# .. .. .. E | ||
|  | 	nop			# .. .. E  .. | ||
|  | 	nop			# .. E  .. .. | ||
|  | 	beq	$3, $bigalign	# U  .. .. ..	: U L U L : Aligned 0mod64 | ||
|  | 
 | ||
|  | $alignmod64: | ||
|  | 	EX( stq_u $31, 0($6) )	# .. .. .. L | ||
|  | 	addq	$3, 8, $3	# .. .. E  .. | ||
|  | 	subq	$0, 8, $0	# .. E  .. .. | ||
|  | 	nop			# E  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | 	nop			# .. .. .. E | ||
|  | 	subq	$1, 1, $1	# .. .. E  .. | ||
|  | 	addq	$6, 8, $6	# .. E  .. .. | ||
|  | 	blt	$3, $alignmod64	# U  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | $bigalign: | ||
|  | /* | ||
|  |  * $0 is the number of bytes left | ||
|  |  * $1 is the number of quads left | ||
|  |  * $6 is aligned 0mod64 | ||
|  |  * we know that we'll be taking a minimum of one trip through | ||
|  |  * CWG Section 3.7.6: do not expect a sustained store rate of > 1/cycle | ||
|  |  * We are _not_ going to update $0 after every single store.  That | ||
|  |  * would be silly, because there will be cross-cluster dependencies | ||
|  |  * no matter how the code is scheduled.  By doing it in slightly | ||
|  |  * staggered fashion, we can still do this loop in 5 fetches | ||
|  |  * The worse case will be doing two extra quads in some future execution, | ||
|  |  * in the event of an interrupted clear. | ||
|  |  * Assumes the wh64 needs to be for 2 trips through the loop in the future | ||
|  |  * The wh64 is issued on for the starting destination address for trip +2 | ||
|  |  * through the loop, and if there are less than two trips left, the target | ||
|  |  * address will be for the current trip. | ||
|  |  */ | ||
|  | 	nop			# E : | ||
|  | 	nop			# E : | ||
|  | 	nop			# E : | ||
|  | 	bis	$6,$6,$3	# E : U L U L : Initial wh64 address is dest | ||
|  | 	/* This might actually help for the current trip... */ | ||
|  | 
 | ||
|  | $do_wh64: | ||
|  | 	wh64	($3)		# .. .. .. L1	: memory subsystem hint | ||
|  | 	subq	$1, 16, $4	# .. .. E  ..	: Forward calculation - repeat the loop? | ||
|  | 	EX( stq_u $31, 0($6) )	# .. L  .. .. | ||
|  | 	subq	$0, 8, $0	# E  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | 	addq	$6, 128, $3	# E : Target address of wh64 | ||
|  | 	EX( stq_u $31, 8($6) )	# L : | ||
|  | 	EX( stq_u $31, 16($6) )	# L : | ||
|  | 	subq	$0, 16, $0	# E : U L L U | ||
|  | 
 | ||
|  | 	nop			# E : | ||
|  | 	EX( stq_u $31, 24($6) )	# L : | ||
|  | 	EX( stq_u $31, 32($6) )	# L : | ||
|  | 	subq	$0, 168, $5	# E : U L L U : two trips through the loop left? | ||
|  | 	/* 168 = 192 - 24, since we've already completed some stores */ | ||
|  | 
 | ||
|  | 	subq	$0, 16, $0	# E : | ||
|  | 	EX( stq_u $31, 40($6) )	# L : | ||
|  | 	EX( stq_u $31, 48($6) )	# L : | ||
|  | 	cmovlt	$5, $6, $3	# E : U L L U : Latency 2, extra mapping cycle | ||
|  | 
 | ||
|  | 	subq	$1, 8, $1	# E : | ||
|  | 	subq	$0, 16, $0	# E : | ||
|  | 	EX( stq_u $31, 56($6) )	# L : | ||
|  | 	nop			# E : U L U L | ||
|  | 
 | ||
|  | 	nop			# E : | ||
|  | 	subq	$0, 8, $0	# E : | ||
|  | 	addq	$6, 64, $6	# E : | ||
|  | 	bge	$4, $do_wh64	# U : U L U L | ||
|  | 
 | ||
|  | $trailquad: | ||
|  | 	# zero to 16 quadwords left to store, plus any trailing bytes | ||
|  | 	# $1 is the number of quadwords left to go. | ||
|  | 	#  | ||
|  | 	nop			# .. .. .. E | ||
|  | 	nop			# .. .. E  .. | ||
|  | 	nop			# .. E  .. .. | ||
|  | 	beq	$1, $trailbytes	# U  .. .. ..	: U L U L : Only 0..7 bytes to go | ||
|  | 
 | ||
|  | $onequad: | ||
|  | 	EX( stq_u $31, 0($6) )	# .. .. .. L | ||
|  | 	subq	$1, 1, $1	# .. .. E  .. | ||
|  | 	subq	$0, 8, $0	# .. E  .. .. | ||
|  | 	nop			# E  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | 	nop			# .. .. .. E | ||
|  | 	nop			# .. .. E  .. | ||
|  | 	addq	$6, 8, $6	# .. E  .. .. | ||
|  | 	bgt	$1, $onequad	# U  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | 	# We have an unknown number of bytes left to go. | ||
|  | $trailbytes: | ||
|  | 	nop			# .. .. .. E | ||
|  | 	nop			# .. .. E  .. | ||
|  | 	nop			# .. E  .. .. | ||
|  | 	beq	$0, $zerolength	# U  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | 	# $0 contains the number of bytes left to copy (0..31) | ||
|  | 	# so we will use $0 as the loop counter | ||
|  | 	# We know for a fact that $0 > 0 zero due to previous context | ||
|  | $onebyte: | ||
|  | 	EX( stb $31, 0($6) )	# .. .. .. L | ||
|  | 	subq	$0, 1, $0	# .. .. E  ..	: | ||
|  | 	addq	$6, 1, $6	# .. E  .. ..	: | ||
|  | 	bgt	$0, $onebyte	# U  .. .. ..	: U L U L | ||
|  | 
 | ||
|  | $zerolength: | ||
|  | $exception:			# Destination for exception recovery(?) | ||
|  | 	nop			# .. .. .. E	: | ||
|  | 	nop			# .. .. E  ..	: | ||
|  | 	nop			# .. E  .. ..	: | ||
|  | 	ret	$31, ($28), 1	# L0 .. .. ..	: L U L U | ||
|  | 	.end __do_clear_user
 | ||
|  | 
 |