1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Event entry/exit for Hexagon
4  *
5  * Copyright (c) 2010-2013, The Linux Foundation. All rights reserved.
6  */
7 
8 #include <asm/asm-offsets.h>  /*  assembly-safer versions of C defines */
9 #include <asm/mem-layout.h>   /*  sigh, except for page_offset  */
10 #include <asm/hexagon_vm.h>
11 #include <asm/thread_info.h>
12 
13 /*
14  * Entry into guest-mode Linux under Hexagon Virtual Machine.
15  * Stack pointer points to event record - build pt_regs on top of it,
16  * set up a plausible C stack frame, and dispatch to the C handler.
17  * On return, do vmrte virtual instruction with SP where we started.
18  *
19  * VM Spec 0.5 uses a trap to fetch HVM record now.
20  */
21 
22 /*
23  * Save full register state, while setting up thread_info struct
24  * pointer derived from kernel stack pointer in THREADINFO_REG
25  * register, putting prior thread_info.regs pointer in a callee-save
26  * register (R24, which had better not ever be assigned to THREADINFO_REG),
27  * and updating thread_info.regs to point to current stack frame,
28  * so as to support nested events in kernel mode.
29  *
30  * As this is common code, we set the pt_regs system call number
31  * to -1 for all events.  It will be replaced with the system call
32  * number in the case where we decode a system call (trap0(#1)).
33  */
34 
35 #if CONFIG_HEXAGON_ARCH_VERSION < 4
36 #define save_pt_regs()\
37  memd(R0 + #_PT_R3130) = R31:30; \
38  { memw(R0 + #_PT_R2928) = R28; \
39    R31 = memw(R0 + #_PT_ER_VMPSP); }\
40  { memw(R0 + #(_PT_R2928 + 4)) = R31; \
41    R31 = ugp; } \
42  { memd(R0 + #_PT_R2726) = R27:26; \
43    R30 = gp ; } \
44  memd(R0 + #_PT_R2524) = R25:24; \
45  memd(R0 + #_PT_R2322) = R23:22; \
46  memd(R0 + #_PT_R2120) = R21:20; \
47  memd(R0 + #_PT_R1918) = R19:18; \
48  memd(R0 + #_PT_R1716) = R17:16; \
49  memd(R0 + #_PT_R1514) = R15:14; \
50  memd(R0 + #_PT_R1312) = R13:12; \
51  { memd(R0 + #_PT_R1110) = R11:10; \
52    R15 = lc0; } \
53  { memd(R0 + #_PT_R0908) = R9:8; \
54    R14 = sa0; } \
55  { memd(R0 + #_PT_R0706) = R7:6; \
56    R13 = lc1; } \
57  { memd(R0 + #_PT_R0504) = R5:4; \
58    R12 = sa1; } \
59  { memd(R0 + #_PT_GPUGP) = R31:30; \
60    R11 = m1; \
61    R2.H = #HI(_THREAD_SIZE); } \
62  { memd(R0 + #_PT_LC0SA0) = R15:14; \
63    R10 = m0; \
64    R2.L = #LO(_THREAD_SIZE); } \
65  { memd(R0 + #_PT_LC1SA1) = R13:12; \
66    R15 = p3:0; \
67    R2 = neg(R2); } \
68  { memd(R0 + #_PT_M1M0) = R11:10; \
69    R14  = usr; \
70    R2 = and(R0,R2); } \
71  { memd(R0 + #_PT_PREDSUSR) =  R15:14; \
72    THREADINFO_REG = R2; } \
73  { r24 = memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS); \
74    memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R0; \
75    R2 = #-1; } \
76  { memw(R0 + #_PT_SYSCALL_NR) = R2; \
77    R30 = #0; }
78 #else
79 /* V4+ */
80 /* the # ## # syntax inserts a literal ## */
81 #define save_pt_regs()\
82 	{ memd(R0 + #_PT_R3130) = R31:30; \
83 		R30 = memw(R0 + #_PT_ER_VMPSP); }\
84 	{ memw(R0 + #_PT_R2928) = R28; \
85 		memw(R0 + #(_PT_R2928 + 4)) = R30; }\
86 	{ R31:30 = C11:10; \
87 		memd(R0 + #_PT_R2726) = R27:26; \
88 		memd(R0 + #_PT_R2524) = R25:24; }\
89 	{ memd(R0 + #_PT_R2322) = R23:22; \
90 		memd(R0 + #_PT_R2120) = R21:20; }\
91 	{ memd(R0 + #_PT_R1918) = R19:18; \
92 		memd(R0 + #_PT_R1716) = R17:16; }\
93 	{ memd(R0 + #_PT_R1514) = R15:14; \
94 		memd(R0 + #_PT_R1312) = R13:12; \
95 		R17:16 = C13:12; }\
96 	{ memd(R0 + #_PT_R1110) = R11:10; \
97 		memd(R0 + #_PT_R0908) = R9:8; \
98 	  R15:14 = C1:0; } \
99 	{ memd(R0 + #_PT_R0706) = R7:6; \
100 		memd(R0 + #_PT_R0504) = R5:4; \
101     R13:12 = C3:2; } \
102 	{ memd(R0 + #_PT_GPUGP) = R31:30; \
103 		memd(R0 + #_PT_LC0SA0) = R15:14; \
104 	  R11:10 = C7:6; }\
105 	{	THREADINFO_REG = and(R0, # ## #-_THREAD_SIZE); \
106 		memd(R0 + #_PT_LC1SA1) = R13:12; \
107 	  R15 = p3:0; }\
108 	{ memd(R0 + #_PT_M1M0) = R11:10; \
109 		memw(R0 + #_PT_PREDSUSR + 4) =  R15; }\
110 	{ r24 = memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS); \
111 	  memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R0; \
112 	  R2 = #-1; } \
113 	{ memw(R0 + #_PT_SYSCALL_NR) = R2; \
114 		memd(R0 + #_PT_CS1CS0) = R17:16; \
115 	  R30 = #0; }
116 #endif
117 
118 /*
119  * Restore registers and thread_info.regs state. THREADINFO_REG
120  * is assumed to still be sane, and R24 to have been correctly
121  * preserved. Don't restore R29 (SP) until later.
122  */
123 
124 #if CONFIG_HEXAGON_ARCH_VERSION < 4
125 #define restore_pt_regs() \
126 	{ memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R24; \
127 	  R15:14 = memd(R0 + #_PT_PREDSUSR); } \
128 	{ R11:10 = memd(R0 + #_PT_M1M0); \
129 	  p3:0 = R15; } \
130 	{ R13:12 = memd(R0 + #_PT_LC1SA1); \
131 	  usr = R14; } \
132 	{ R15:14 = memd(R0 + #_PT_LC0SA0); \
133 	  m1 = R11; } \
134 	{ R3:2 = memd(R0 + #_PT_R0302); \
135 	  m0 = R10; } \
136 	{ R5:4 = memd(R0 + #_PT_R0504); \
137 	  lc1 = R13; } \
138 	{ R7:6 = memd(R0 + #_PT_R0706); \
139 	  sa1 = R12; } \
140 	{ R9:8 = memd(R0 + #_PT_R0908); \
141 	  lc0 = R15; } \
142 	{ R11:10 = memd(R0 + #_PT_R1110); \
143 	  sa0 = R14; } \
144 	{ R13:12 = memd(R0 + #_PT_R1312); \
145 	  R15:14 = memd(R0 + #_PT_R1514); } \
146 	{ R17:16 = memd(R0 + #_PT_R1716); \
147 	  R19:18 = memd(R0 + #_PT_R1918); } \
148 	{ R21:20 = memd(R0 + #_PT_R2120); \
149 	  R23:22 = memd(R0 + #_PT_R2322); } \
150 	{ R25:24 = memd(R0 + #_PT_R2524); \
151 	  R27:26 = memd(R0 + #_PT_R2726); } \
152 	R31:30 = memd(R0 + #_PT_GPUGP); \
153 	{ R28 = memw(R0 + #_PT_R2928); \
154 	  ugp = R31; } \
155 	{ R31:30 = memd(R0 + #_PT_R3130); \
156 	  gp = R30; }
157 #else
158 /* V4+ */
159 #define restore_pt_regs() \
160 	{ memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R24; \
161 	  R15:14 = memd(R0 + #_PT_PREDSUSR); } \
162 	{ R11:10 = memd(R0 + #_PT_M1M0); \
163 		R13:12 = memd(R0 + #_PT_LC1SA1); \
164 		p3:0 = R15; } \
165 	{ R15:14 = memd(R0 + #_PT_LC0SA0); \
166 		R3:2 = memd(R0 + #_PT_R0302); \
167 		usr = R14; } \
168 	{ R5:4 = memd(R0 + #_PT_R0504); \
169 		R7:6 = memd(R0 + #_PT_R0706); \
170 		C7:6 = R11:10; }\
171 	{ R9:8 = memd(R0 + #_PT_R0908); \
172 		R11:10 = memd(R0 + #_PT_R1110); \
173     C3:2 = R13:12; }\
174 	{ R13:12 = memd(R0 + #_PT_R1312); \
175 	  R15:14 = memd(R0 + #_PT_R1514); \
176 		C1:0 = R15:14; }\
177 	{ R17:16 = memd(R0 + #_PT_R1716); \
178 	  R19:18 = memd(R0 + #_PT_R1918); } \
179 	{ R21:20 = memd(R0 + #_PT_R2120); \
180 	  R23:22 = memd(R0 + #_PT_R2322); } \
181 	{ R25:24 = memd(R0 + #_PT_R2524); \
182 	  R27:26 = memd(R0 + #_PT_R2726); } \
183 	R31:30 = memd(R0 + #_PT_CS1CS0); \
184 	{ C13:12 = R31:30; \
185 		R31:30 = memd(R0 + #_PT_GPUGP) ; \
186 		R28 = memw(R0 + #_PT_R2928); }\
187 	{ C11:10 = R31:30; \
188 		R31:30 = memd(R0 + #_PT_R3130); }
189 #endif
190 
191 	/*
192 	 * Clears off enough space for the rest of pt_regs; evrec is a part
193 	 * of pt_regs in HVM mode.  Save R0/R1, set handler's address in R1.
194 	 * R0 is the address of pt_regs and is the parameter to save_pt_regs.
195 	 */
196 
197 /*
198  * Since the HVM isn't automagically pushing the EVREC onto the stack anymore,
199  * we'll subract the entire size out and then fill it in ourselves.
200  * Need to save off R0, R1, R2, R3 immediately.
201  */
202 
203 #if CONFIG_HEXAGON_ARCH_VERSION < 4
204 #define	vm_event_entry(CHandler) \
205 	{ \
206 		R29 = add(R29, #-(_PT_REGS_SIZE)); \
207 		memd(R29 + #(_PT_R0100 + -_PT_REGS_SIZE)) = R1:0; \
208 	} \
209 	{ \
210 		memd(R29 +#_PT_R0302) = R3:2; \
211 	} \
212 	trap1(#HVM_TRAP1_VMGETREGS); \
213 	{ \
214 		memd(R29 + #_PT_ER_VMEL) = R1:0; \
215 		R0 = R29; \
216 		R1.L = #LO(CHandler); \
217 	} \
218 	{ \
219 		memd(R29 + #_PT_ER_VMPSP) = R3:2; \
220 		R1.H = #HI(CHandler); \
221 		jump event_dispatch; \
222 	}
223 #else
224 /* V4+ */
225 /* turn on I$ prefetch early */
226 /* the # ## # syntax inserts a literal ## */
227 #define	vm_event_entry(CHandler) \
228 	{ \
229 		R29 = add(R29, #-(_PT_REGS_SIZE)); \
230 		memd(R29 + #(_PT_R0100 + -_PT_REGS_SIZE)) = R1:0; \
231 		memd(R29 + #(_PT_R0302 + -_PT_REGS_SIZE)) = R3:2; \
232 		R0 = usr; \
233 	} \
234 	{ \
235 		memw(R29 + #_PT_PREDSUSR) = R0; \
236 		R0 = setbit(R0, #16); \
237 	} \
238 	usr = R0; \
239 	R1:0 = G1:0; \
240 	{ \
241 		memd(R29 + #_PT_ER_VMEL) = R1:0; \
242 		R1 = # ## #(CHandler); \
243 		R3:2 = G3:2; \
244 	} \
245 	{ \
246 		R0 = R29; \
247 		memd(R29 + #_PT_ER_VMPSP) = R3:2; \
248 		jump event_dispatch; \
249 	}
250 #endif
251 
252 .text
253 	/*
254 	 * Do bulk save/restore in one place.
255 	 * Adds a jump to dispatch latency, but
256 	 * saves hundreds of bytes.
257 	 */
258 
259 event_dispatch:
260 	save_pt_regs()
261 	callr	r1
262 
263 	/*
264 	 * Coming back from the C-world, our thread info pointer
265 	 * should be in the designated register (usually R19)
266 	 *
267 	 * If we were in kernel mode, we don't need to check scheduler
268 	 * or signals if CONFIG_PREEMPTION is not set.  If set, then it has
269 	 * to jump to a need_resched kind of block.
270 	 * BTW, CONFIG_PREEMPTION is not supported yet.
271 	 */
272 
273 #ifdef CONFIG_PREEMPTION
274 	R0 = #VM_INT_DISABLE
275 	trap1(#HVM_TRAP1_VMSETIE)
276 #endif
277 
278 	/*  "Nested control path" -- if the previous mode was kernel  */
279 	{
280 		R0 = memw(R29 + #_PT_ER_VMEST);
281 		R26.L = #LO(do_work_pending);
282 	}
283 	{
284 		P0 = tstbit(R0, #HVM_VMEST_UM_SFT);
285 		if (!P0.new) jump:nt restore_all;
286 		R26.H = #HI(do_work_pending);
287 		R0 = #VM_INT_DISABLE;
288 	}
289 
290 	/*
291 	 * Check also the return from fork/system call, normally coming back from
292 	 * user mode
293 	 *
294 	 * R26 needs to have do_work_pending, and R0 should have VM_INT_DISABLE
295 	 */
296 
297 check_work_pending:
298 	/*  Disable interrupts while checking TIF  */
299 	trap1(#HVM_TRAP1_VMSETIE)
300 	{
301 		R0 = R29;  /*  regs should still be at top of stack  */
302 		R1 = memw(THREADINFO_REG + #_THREAD_INFO_FLAGS);
303 		callr R26;
304 	}
305 
306 	{
307 		P0 = cmp.eq(R0, #0); if (!P0.new) jump:nt check_work_pending;
308 		R0 = #VM_INT_DISABLE;
309 	}
310 
311 restore_all:
312 	/*
313 	 * Disable interrupts, if they weren't already, before reg restore.
314 	 * R0 gets preloaded with #VM_INT_DISABLE before we get here.
315 	 */
316 	trap1(#HVM_TRAP1_VMSETIE)
317 
318 	/*  do the setregs here for VM 0.5  */
319 	/*  R29 here should already be pointing at pt_regs  */
320 	{
321 		R1:0 = memd(R29 + #_PT_ER_VMEL);
322 		R3:2 = memd(R29 + #_PT_ER_VMPSP);
323 	}
324 #if CONFIG_HEXAGON_ARCH_VERSION < 4
325 	trap1(#HVM_TRAP1_VMSETREGS);
326 #else
327 	G1:0 = R1:0;
328 	G3:2 = R3:2;
329 #endif
330 
331 	R0 = R29
332 	restore_pt_regs()
333 	{
334 		R1:0 = memd(R29 + #_PT_R0100);
335 		R29 = add(R29, #_PT_REGS_SIZE);
336 	}
337 	trap1(#HVM_TRAP1_VMRTE)
338 	/* Notreached */
339 
340 
341 	.globl _K_enter_genex
342 _K_enter_genex:
343 	vm_event_entry(do_genex)
344 
345 	.globl _K_enter_interrupt
346 _K_enter_interrupt:
347 	vm_event_entry(arch_do_IRQ)
348 
349 	.globl _K_enter_trap0
350 _K_enter_trap0:
351 	vm_event_entry(do_trap0)
352 
353 	.globl _K_enter_machcheck
354 _K_enter_machcheck:
355 	vm_event_entry(do_machcheck)
356 
357 	.globl _K_enter_debug
358 _K_enter_debug:
359 	vm_event_entry(do_debug_exception)
360 
361 	.globl ret_from_fork
362 ret_from_fork:
363 	{
364 		call schedule_tail
365 		R26.H = #HI(do_work_pending);
366 	}
367 	{
368 		P0 = cmp.eq(R24, #0);
369 		R26.L = #LO(do_work_pending);
370 		R0 = #VM_INT_DISABLE;
371 	}
372 	if (P0) jump check_work_pending
373 	{
374 		R0 = R25;
375 		callr R24
376 	}
377 	{
378 		jump check_work_pending
379 		R0 = #VM_INT_DISABLE;
380 	}
381