xtensa/lib/usercopy.S

8  *  of the Xtensa link-time HAL, and those files may differ per
11  *  could lose the special functionality for accessing user-space
31  *     do 16B chunks with a loop, and then finish up with
32  *     8B, 4B, 2B, and 1B copies conditional on the length.
36  *     copying 1B and 2B and then retest.
37  *   This code tries to use fall-through braches for the common
73 	bbsi.l	a2, 0, .Ldst1mod2 # if dst is 1 mod 2
74 	bbsi.l	a2, 1, .Ldst2mod4 # if dst is 2 mod 4
76 	srli	a7, a4, 4	# number of loop iterations with 16B
97 	addi	a4, a4, -1
100 .Ldst2mod4:	# dst 16-bit aligned
109 	addi	a4, a4, -2
116 	.byte	0		# 1 mod 4 alignment for LOOPNEZ
117 				# (0 mod 4 alignment for LBEG)
138  * Destination and source are word-aligned.
140 	# copy 16 bytes per iteration for word-aligned dst and word-aligned src
141 	.align	4		# 1 mod 4 alignment for LOOPNEZ
142 	.byte	0		# (0 mod 4 alignment for LBEG)
149 	add	a8, a8, a3	# a8 = end of last 16B source chunk
157 EX(10f)	l32i	a7, a3, 12
160 EX(10f)	s32i	a7, a5, 12
202 	.byte	0		# 1 mod 4 alignement for LOOPNEZ
203 				# (0 mod 4 alignment for LBEG)
205 	# copy 16 bytes per iteration for word-aligned dst and unaligned src
216 	add	a10, a10, a3	# a10 = end of last 16B source chunk
219 	add	a12, a12, a3	# a12 = end of last 16B source chunk
227 EX(10f)	l32i	a9, a3, 12
235 EX(10f)	s32i	a9, a5, 12
292  * bytes_copied = a5 - a2
293  * retval = bytes_not_copied = original len - bytes_copied
294  * retval = a11 - (a5 - a2)
299 	sub	a2, a5, a2	/* a2 <-- bytes copied */
300 	sub	a2, a11, a2	/* a2 <-- bytes not copied */