18c2ecf20Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0-only */
28c2ecf20Sopenharmony_ci/*
38c2ecf20Sopenharmony_ci * User string length functions for kernel
48c2ecf20Sopenharmony_ci *
58c2ecf20Sopenharmony_ci * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
68c2ecf20Sopenharmony_ci */
78c2ecf20Sopenharmony_ci
88c2ecf20Sopenharmony_ci#define isrc	r0
98c2ecf20Sopenharmony_ci#define max	r1	/*  Do not change!  */
108c2ecf20Sopenharmony_ci
118c2ecf20Sopenharmony_ci#define end	r2
128c2ecf20Sopenharmony_ci#define tmp1	r3
138c2ecf20Sopenharmony_ci
148c2ecf20Sopenharmony_ci#define obo	r6	/*  off-by-one  */
158c2ecf20Sopenharmony_ci#define start	r7
168c2ecf20Sopenharmony_ci#define mod8	r8
178c2ecf20Sopenharmony_ci#define dbuf    r15:14
188c2ecf20Sopenharmony_ci#define dcmp	r13:12
198c2ecf20Sopenharmony_ci
208c2ecf20Sopenharmony_ci/*
218c2ecf20Sopenharmony_ci * The vector mask version of this turned out *really* badly.
228c2ecf20Sopenharmony_ci * The hardware loop version also turned out *really* badly.
238c2ecf20Sopenharmony_ci * Seems straight pointer arithmetic basically wins here.
248c2ecf20Sopenharmony_ci */
258c2ecf20Sopenharmony_ci
268c2ecf20Sopenharmony_ci#define fname __strnlen_user
278c2ecf20Sopenharmony_ci
288c2ecf20Sopenharmony_ci	.text
298c2ecf20Sopenharmony_ci	.global fname
308c2ecf20Sopenharmony_ci	.type fname, @function
318c2ecf20Sopenharmony_ci	.p2align 5  /*  why?  */
328c2ecf20Sopenharmony_cifname:
338c2ecf20Sopenharmony_ci	{
348c2ecf20Sopenharmony_ci		mod8 = and(isrc,#7);
358c2ecf20Sopenharmony_ci		end = add(isrc,max);
368c2ecf20Sopenharmony_ci		start = isrc;
378c2ecf20Sopenharmony_ci	}
388c2ecf20Sopenharmony_ci	{
398c2ecf20Sopenharmony_ci		P0 = cmp.eq(mod8,#0);
408c2ecf20Sopenharmony_ci		mod8 = and(end,#7);
418c2ecf20Sopenharmony_ci		dcmp = #0;
428c2ecf20Sopenharmony_ci		if (P0.new) jump:t dw_loop;	/*  fire up the oven  */
438c2ecf20Sopenharmony_ci	}
448c2ecf20Sopenharmony_ci
458c2ecf20Sopenharmony_cialignment_loop:
468c2ecf20Sopenharmony_cifail_1:	{
478c2ecf20Sopenharmony_ci		tmp1 = memb(start++#1);
488c2ecf20Sopenharmony_ci	}
498c2ecf20Sopenharmony_ci	{
508c2ecf20Sopenharmony_ci		P0 = cmp.eq(tmp1,#0);
518c2ecf20Sopenharmony_ci		if (P0.new) jump:nt exit_found;
528c2ecf20Sopenharmony_ci		P1 = cmp.gtu(end,start);
538c2ecf20Sopenharmony_ci		mod8 = and(start,#7);
548c2ecf20Sopenharmony_ci	}
558c2ecf20Sopenharmony_ci	{
568c2ecf20Sopenharmony_ci		if (!P1) jump exit_error;  /*  hit the end  */
578c2ecf20Sopenharmony_ci		P0 = cmp.eq(mod8,#0);
588c2ecf20Sopenharmony_ci	}
598c2ecf20Sopenharmony_ci	{
608c2ecf20Sopenharmony_ci		if (!P0) jump alignment_loop;
618c2ecf20Sopenharmony_ci	}
628c2ecf20Sopenharmony_ci
638c2ecf20Sopenharmony_ci
648c2ecf20Sopenharmony_ci
658c2ecf20Sopenharmony_cidw_loop:
668c2ecf20Sopenharmony_cifail_2:	{
678c2ecf20Sopenharmony_ci		dbuf = memd(start);
688c2ecf20Sopenharmony_ci		obo = add(start,#1);
698c2ecf20Sopenharmony_ci	}
708c2ecf20Sopenharmony_ci	{
718c2ecf20Sopenharmony_ci		P0 = vcmpb.eq(dbuf,dcmp);
728c2ecf20Sopenharmony_ci	}
738c2ecf20Sopenharmony_ci	{
748c2ecf20Sopenharmony_ci		tmp1 = P0;
758c2ecf20Sopenharmony_ci		P0 = cmp.gtu(end,start);
768c2ecf20Sopenharmony_ci	}
778c2ecf20Sopenharmony_ci	{
788c2ecf20Sopenharmony_ci		tmp1 = ct0(tmp1);
798c2ecf20Sopenharmony_ci		mod8 = and(end,#7);
808c2ecf20Sopenharmony_ci		if (!P0) jump end_check;
818c2ecf20Sopenharmony_ci	}
828c2ecf20Sopenharmony_ci	{
838c2ecf20Sopenharmony_ci		P0 = cmp.eq(tmp1,#32);
848c2ecf20Sopenharmony_ci		if (!P0.new) jump:nt exit_found;
858c2ecf20Sopenharmony_ci		if (!P0.new) start = add(obo,tmp1);
868c2ecf20Sopenharmony_ci	}
878c2ecf20Sopenharmony_ci	{
888c2ecf20Sopenharmony_ci		start = add(start,#8);
898c2ecf20Sopenharmony_ci		jump dw_loop;
908c2ecf20Sopenharmony_ci	}	/*  might be nice to combine these jumps...   */
918c2ecf20Sopenharmony_ci
928c2ecf20Sopenharmony_ci
938c2ecf20Sopenharmony_ciend_check:
948c2ecf20Sopenharmony_ci	{
958c2ecf20Sopenharmony_ci		P0 = cmp.gt(tmp1,mod8);
968c2ecf20Sopenharmony_ci		if (P0.new) jump:nt exit_error;	/*  neverfound!  */
978c2ecf20Sopenharmony_ci		start = add(obo,tmp1);
988c2ecf20Sopenharmony_ci	}
998c2ecf20Sopenharmony_ci
1008c2ecf20Sopenharmony_ciexit_found:
1018c2ecf20Sopenharmony_ci	{
1028c2ecf20Sopenharmony_ci		R0 = sub(start,isrc);
1038c2ecf20Sopenharmony_ci		jumpr R31;
1048c2ecf20Sopenharmony_ci	}
1058c2ecf20Sopenharmony_ci
1068c2ecf20Sopenharmony_ciexit_error:
1078c2ecf20Sopenharmony_ci	{
1088c2ecf20Sopenharmony_ci		R0 = add(max,#1);
1098c2ecf20Sopenharmony_ci		jumpr R31;
1108c2ecf20Sopenharmony_ci	}
1118c2ecf20Sopenharmony_ci
1128c2ecf20Sopenharmony_ci	/*  Uh, what does the "fixup" return here?  */
1138c2ecf20Sopenharmony_ci	.falign
1148c2ecf20Sopenharmony_cifix_1:
1158c2ecf20Sopenharmony_ci	{
1168c2ecf20Sopenharmony_ci		R0 = #0;
1178c2ecf20Sopenharmony_ci		jumpr R31;
1188c2ecf20Sopenharmony_ci	}
1198c2ecf20Sopenharmony_ci
1208c2ecf20Sopenharmony_ci	.size fname,.-fname
1218c2ecf20Sopenharmony_ci
1228c2ecf20Sopenharmony_ci
1238c2ecf20Sopenharmony_ci.section __ex_table,"a"
1248c2ecf20Sopenharmony_ci.long fail_1,fix_1
1258c2ecf20Sopenharmony_ci.long fail_2,fix_1
1268c2ecf20Sopenharmony_ci.previous
127