xref: /linux/tools/testing/selftests/bpf/progs/verifier_subprog_precision.c (revision ae22a94997b8a03dcb3c922857c203246711f9d4)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2023 Meta Platforms, Inc. and affiliates. */
3 
4 #include <errno.h>
5 #include <string.h>
6 #include <linux/bpf.h>
7 #include <bpf/bpf_helpers.h>
8 #include "bpf_misc.h"
9 
10 #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
11 
12 int vals[] SEC(".data.vals") = {1, 2, 3, 4};
13 
14 __naked __noinline __used
15 static unsigned long identity_subprog()
16 {
17 	/* the simplest *static* 64-bit identity function */
18 	asm volatile (
19 		"r0 = r1;"
20 		"exit;"
21 	);
22 }
23 
24 __noinline __used
25 unsigned long global_identity_subprog(__u64 x)
26 {
27 	/* the simplest *global* 64-bit identity function */
28 	return x;
29 }
30 
31 __naked __noinline __used
32 static unsigned long callback_subprog()
33 {
34 	/* the simplest callback function */
35 	asm volatile (
36 		"r0 = 0;"
37 		"exit;"
38 	);
39 }
40 
41 SEC("?raw_tp")
42 __success __log_level(2)
43 __msg("7: (0f) r1 += r0")
44 __msg("mark_precise: frame0: regs=r0 stack= before 6: (bf) r1 = r7")
45 __msg("mark_precise: frame0: regs=r0 stack= before 5: (27) r0 *= 4")
46 __msg("mark_precise: frame0: regs=r0 stack= before 11: (95) exit")
47 __msg("mark_precise: frame1: regs=r0 stack= before 10: (bf) r0 = r1")
48 __msg("mark_precise: frame1: regs=r1 stack= before 4: (85) call pc+5")
49 __msg("mark_precise: frame0: regs=r1 stack= before 3: (bf) r1 = r6")
50 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
51 __naked int subprog_result_precise(void)
52 {
53 	asm volatile (
54 		"r6 = 3;"
55 		/* pass r6 through r1 into subprog to get it back as r0;
56 		 * this whole chain will have to be marked as precise later
57 		 */
58 		"r1 = r6;"
59 		"call identity_subprog;"
60 		/* now use subprog's returned value (which is a
61 		 * r6 -> r1 -> r0 chain), as index into vals array, forcing
62 		 * all of that to be known precisely
63 		 */
64 		"r0 *= 4;"
65 		"r1 = %[vals];"
66 		/* here r0->r1->r6 chain is forced to be precise and has to be
67 		 * propagated back to the beginning, including through the
68 		 * subprog call
69 		 */
70 		"r1 += r0;"
71 		"r0 = *(u32 *)(r1 + 0);"
72 		"exit;"
73 		:
74 		: __imm_ptr(vals)
75 		: __clobber_common, "r6"
76 	);
77 }
78 
79 SEC("?raw_tp")
80 __success __log_level(2)
81 __msg("9: (0f) r1 += r0")
82 __msg("mark_precise: frame0: last_idx 9 first_idx 0")
83 __msg("mark_precise: frame0: regs=r0 stack= before 8: (bf) r1 = r7")
84 __msg("mark_precise: frame0: regs=r0 stack= before 7: (27) r0 *= 4")
85 __msg("mark_precise: frame0: regs=r0 stack= before 5: (a5) if r0 < 0x4 goto pc+1")
86 __msg("mark_precise: frame0: regs=r0 stack= before 4: (85) call pc+7")
87 __naked int global_subprog_result_precise(void)
88 {
89 	asm volatile (
90 		"r6 = 3;"
91 		/* pass r6 through r1 into subprog to get it back as r0;
92 		 * given global_identity_subprog is global, precision won't
93 		 * propagate all the way back to r6
94 		 */
95 		"r1 = r6;"
96 		"call global_identity_subprog;"
97 		/* now use subprog's returned value (which is unknown now, so
98 		 * we need to clamp it), as index into vals array, forcing r0
99 		 * to be marked precise (with no effect on r6, though)
100 		 */
101 		"if r0 < %[vals_arr_sz] goto 1f;"
102 		"r0 = %[vals_arr_sz] - 1;"
103 	"1:"
104 		"r0 *= 4;"
105 		"r1 = %[vals];"
106 		/* here r0 is forced to be precise and has to be
107 		 * propagated back to the global subprog call, but it
108 		 * shouldn't go all the way to mark r6 as precise
109 		 */
110 		"r1 += r0;"
111 		"r0 = *(u32 *)(r1 + 0);"
112 		"exit;"
113 		:
114 		: __imm_ptr(vals),
115 		  __imm_const(vals_arr_sz, ARRAY_SIZE(vals))
116 		: __clobber_common, "r6"
117 	);
118 }
119 
120 __naked __noinline __used
121 static unsigned long loop_callback_bad()
122 {
123 	/* bpf_loop() callback that can return values outside of [0, 1] range */
124 	asm volatile (
125 		"call %[bpf_get_prandom_u32];"
126 		"if r0 s> 1000 goto 1f;"
127 		"r0 = 0;"
128 	"1:"
129 		"goto +0;" /* checkpoint */
130 		/* bpf_loop() expects [0, 1] values, so branch above skipping
131 		 * r0 = 0; should lead to a failure, but if exit instruction
132 		 * doesn't enforce r0's precision, this callback will be
133 		 * successfully verified
134 		 */
135 		"exit;"
136 		:
137 		: __imm(bpf_get_prandom_u32)
138 		: __clobber_common
139 	);
140 }
141 
142 SEC("?raw_tp")
143 __failure __log_level(2)
144 __flag(BPF_F_TEST_STATE_FREQ)
145 /* check that fallthrough code path marks r0 as precise */
146 __msg("mark_precise: frame1: regs=r0 stack= before 11: (b7) r0 = 0")
147 /* check that we have branch code path doing its own validation */
148 __msg("from 10 to 12: frame1: R0=scalar(smin=umin=1001")
149 /* check that branch code path marks r0 as precise, before failing */
150 __msg("mark_precise: frame1: regs=r0 stack= before 9: (85) call bpf_get_prandom_u32#7")
151 __msg("At callback return the register R0 has smin=1001 should have been in [0, 1]")
152 __naked int callback_precise_return_fail(void)
153 {
154 	asm volatile (
155 		"r1 = 1;"			/* nr_loops */
156 		"r2 = %[loop_callback_bad];"	/* callback_fn */
157 		"r3 = 0;"			/* callback_ctx */
158 		"r4 = 0;"			/* flags */
159 		"call %[bpf_loop];"
160 
161 		"r0 = 0;"
162 		"exit;"
163 		:
164 		: __imm_ptr(loop_callback_bad),
165 		  __imm(bpf_loop)
166 		: __clobber_common
167 	);
168 }
169 
170 SEC("?raw_tp")
171 __success __log_level(2)
172 /* First simulated path does not include callback body,
173  * r1 and r4 are always precise for bpf_loop() calls.
174  */
175 __msg("9: (85) call bpf_loop#181")
176 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
177 __msg("mark_precise: frame0: parent state regs=r4 stack=:")
178 __msg("mark_precise: frame0: last_idx 8 first_idx 0 subseq_idx 9")
179 __msg("mark_precise: frame0: regs=r4 stack= before 8: (b7) r4 = 0")
180 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
181 __msg("mark_precise: frame0: parent state regs=r1 stack=:")
182 __msg("mark_precise: frame0: last_idx 8 first_idx 0 subseq_idx 9")
183 __msg("mark_precise: frame0: regs=r1 stack= before 8: (b7) r4 = 0")
184 __msg("mark_precise: frame0: regs=r1 stack= before 7: (b7) r3 = 0")
185 __msg("mark_precise: frame0: regs=r1 stack= before 6: (bf) r2 = r8")
186 __msg("mark_precise: frame0: regs=r1 stack= before 5: (bf) r1 = r6")
187 __msg("mark_precise: frame0: regs=r6 stack= before 4: (b7) r6 = 3")
188 /* r6 precision propagation */
189 __msg("14: (0f) r1 += r6")
190 __msg("mark_precise: frame0: last_idx 14 first_idx 9")
191 __msg("mark_precise: frame0: regs=r6 stack= before 13: (bf) r1 = r7")
192 __msg("mark_precise: frame0: regs=r6 stack= before 12: (27) r6 *= 4")
193 __msg("mark_precise: frame0: regs=r6 stack= before 11: (25) if r6 > 0x3 goto pc+4")
194 __msg("mark_precise: frame0: regs=r6 stack= before 10: (bf) r6 = r0")
195 __msg("mark_precise: frame0: regs=r0 stack= before 9: (85) call bpf_loop")
196 /* State entering callback body popped from states stack */
197 __msg("from 9 to 17: frame1:")
198 __msg("17: frame1: R1=scalar() R2=0 R10=fp0 cb")
199 __msg("17: (b7) r0 = 0")
200 __msg("18: (95) exit")
201 __msg("returning from callee:")
202 __msg("to caller at 9:")
203 __msg("frame 0: propagating r1,r4")
204 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
205 __msg("mark_precise: frame0: regs=r1,r4 stack= before 18: (95) exit")
206 __msg("from 18 to 9: safe")
207 __naked int callback_result_precise(void)
208 {
209 	asm volatile (
210 		"r6 = 3;"
211 
212 		/* call subprog and use result; r0 shouldn't propagate back to
213 		 * callback_subprog
214 		 */
215 		"r1 = r6;"			/* nr_loops */
216 		"r2 = %[callback_subprog];"	/* callback_fn */
217 		"r3 = 0;"			/* callback_ctx */
218 		"r4 = 0;"			/* flags */
219 		"call %[bpf_loop];"
220 
221 		"r6 = r0;"
222 		"if r6 > 3 goto 1f;"
223 		"r6 *= 4;"
224 		"r1 = %[vals];"
225 		/* here r6 is forced to be precise and has to be propagated
226 		 * back to the bpf_loop() call, but not beyond
227 		 */
228 		"r1 += r6;"
229 		"r0 = *(u32 *)(r1 + 0);"
230 	"1:"
231 		"exit;"
232 		:
233 		: __imm_ptr(vals),
234 		  __imm_ptr(callback_subprog),
235 		  __imm(bpf_loop)
236 		: __clobber_common, "r6"
237 	);
238 }
239 
240 SEC("?raw_tp")
241 __success __log_level(2)
242 __msg("7: (0f) r1 += r6")
243 __msg("mark_precise: frame0: last_idx 7 first_idx 0")
244 __msg("mark_precise: frame0: regs=r6 stack= before 6: (bf) r1 = r7")
245 __msg("mark_precise: frame0: regs=r6 stack= before 5: (27) r6 *= 4")
246 __msg("mark_precise: frame0: regs=r6 stack= before 11: (95) exit")
247 __msg("mark_precise: frame1: regs= stack= before 10: (bf) r0 = r1")
248 __msg("mark_precise: frame1: regs= stack= before 4: (85) call pc+5")
249 __msg("mark_precise: frame0: regs=r6 stack= before 3: (b7) r1 = 0")
250 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
251 __naked int parent_callee_saved_reg_precise(void)
252 {
253 	asm volatile (
254 		"r6 = 3;"
255 
256 		/* call subprog and ignore result; we need this call only to
257 		 * complicate jump history
258 		 */
259 		"r1 = 0;"
260 		"call identity_subprog;"
261 
262 		"r6 *= 4;"
263 		"r1 = %[vals];"
264 		/* here r6 is forced to be precise and has to be propagated
265 		 * back to the beginning, handling (and ignoring) subprog call
266 		 */
267 		"r1 += r6;"
268 		"r0 = *(u32 *)(r1 + 0);"
269 		"exit;"
270 		:
271 		: __imm_ptr(vals)
272 		: __clobber_common, "r6"
273 	);
274 }
275 
276 SEC("?raw_tp")
277 __success __log_level(2)
278 __msg("7: (0f) r1 += r6")
279 __msg("mark_precise: frame0: last_idx 7 first_idx 0")
280 __msg("mark_precise: frame0: regs=r6 stack= before 6: (bf) r1 = r7")
281 __msg("mark_precise: frame0: regs=r6 stack= before 5: (27) r6 *= 4")
282 __msg("mark_precise: frame0: regs=r6 stack= before 4: (85) call pc+5")
283 __msg("mark_precise: frame0: regs=r6 stack= before 3: (b7) r1 = 0")
284 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
285 __naked int parent_callee_saved_reg_precise_global(void)
286 {
287 	asm volatile (
288 		"r6 = 3;"
289 
290 		/* call subprog and ignore result; we need this call only to
291 		 * complicate jump history
292 		 */
293 		"r1 = 0;"
294 		"call global_identity_subprog;"
295 
296 		"r6 *= 4;"
297 		"r1 = %[vals];"
298 		/* here r6 is forced to be precise and has to be propagated
299 		 * back to the beginning, handling (and ignoring) subprog call
300 		 */
301 		"r1 += r6;"
302 		"r0 = *(u32 *)(r1 + 0);"
303 		"exit;"
304 		:
305 		: __imm_ptr(vals)
306 		: __clobber_common, "r6"
307 	);
308 }
309 
310 SEC("?raw_tp")
311 __success __log_level(2)
312 /* First simulated path does not include callback body */
313 __msg("12: (0f) r1 += r6")
314 __msg("mark_precise: frame0: last_idx 12 first_idx 9")
315 __msg("mark_precise: frame0: regs=r6 stack= before 11: (bf) r1 = r7")
316 __msg("mark_precise: frame0: regs=r6 stack= before 10: (27) r6 *= 4")
317 __msg("mark_precise: frame0: regs=r6 stack= before 9: (85) call bpf_loop")
318 __msg("mark_precise: frame0: parent state regs=r6 stack=:")
319 __msg("mark_precise: frame0: last_idx 8 first_idx 0 subseq_idx 9")
320 __msg("mark_precise: frame0: regs=r6 stack= before 8: (b7) r4 = 0")
321 __msg("mark_precise: frame0: regs=r6 stack= before 7: (b7) r3 = 0")
322 __msg("mark_precise: frame0: regs=r6 stack= before 6: (bf) r2 = r8")
323 __msg("mark_precise: frame0: regs=r6 stack= before 5: (b7) r1 = 1")
324 __msg("mark_precise: frame0: regs=r6 stack= before 4: (b7) r6 = 3")
325 /* State entering callback body popped from states stack */
326 __msg("from 9 to 15: frame1:")
327 __msg("15: frame1: R1=scalar() R2=0 R10=fp0 cb")
328 __msg("15: (b7) r0 = 0")
329 __msg("16: (95) exit")
330 __msg("returning from callee:")
331 __msg("to caller at 9:")
332 /* r1, r4 are always precise for bpf_loop(),
333  * r6 was marked before backtracking to callback body.
334  */
335 __msg("frame 0: propagating r1,r4,r6")
336 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
337 __msg("mark_precise: frame0: regs=r1,r4,r6 stack= before 16: (95) exit")
338 __msg("mark_precise: frame1: regs= stack= before 15: (b7) r0 = 0")
339 __msg("mark_precise: frame1: regs= stack= before 9: (85) call bpf_loop")
340 __msg("mark_precise: frame0: parent state regs= stack=:")
341 __msg("from 16 to 9: safe")
342 __naked int parent_callee_saved_reg_precise_with_callback(void)
343 {
344 	asm volatile (
345 		"r6 = 3;"
346 
347 		/* call subprog and ignore result; we need this call only to
348 		 * complicate jump history
349 		 */
350 		"r1 = 1;"			/* nr_loops */
351 		"r2 = %[callback_subprog];"	/* callback_fn */
352 		"r3 = 0;"			/* callback_ctx */
353 		"r4 = 0;"			/* flags */
354 		"call %[bpf_loop];"
355 
356 		"r6 *= 4;"
357 		"r1 = %[vals];"
358 		/* here r6 is forced to be precise and has to be propagated
359 		 * back to the beginning, handling (and ignoring) callback call
360 		 */
361 		"r1 += r6;"
362 		"r0 = *(u32 *)(r1 + 0);"
363 		"exit;"
364 		:
365 		: __imm_ptr(vals),
366 		  __imm_ptr(callback_subprog),
367 		  __imm(bpf_loop)
368 		: __clobber_common, "r6"
369 	);
370 }
371 
372 SEC("?raw_tp")
373 __success __log_level(2)
374 __msg("9: (0f) r1 += r6")
375 __msg("mark_precise: frame0: last_idx 9 first_idx 6")
376 __msg("mark_precise: frame0: regs=r6 stack= before 8: (bf) r1 = r7")
377 __msg("mark_precise: frame0: regs=r6 stack= before 7: (27) r6 *= 4")
378 __msg("mark_precise: frame0: regs=r6 stack= before 6: (79) r6 = *(u64 *)(r10 -8)")
379 __msg("mark_precise: frame0: parent state regs= stack=-8:")
380 __msg("mark_precise: frame0: last_idx 13 first_idx 0")
381 __msg("mark_precise: frame0: regs= stack=-8 before 13: (95) exit")
382 __msg("mark_precise: frame1: regs= stack= before 12: (bf) r0 = r1")
383 __msg("mark_precise: frame1: regs= stack= before 5: (85) call pc+6")
384 __msg("mark_precise: frame0: regs= stack=-8 before 4: (b7) r1 = 0")
385 __msg("mark_precise: frame0: regs= stack=-8 before 3: (7b) *(u64 *)(r10 -8) = r6")
386 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
387 __naked int parent_stack_slot_precise(void)
388 {
389 	asm volatile (
390 		/* spill reg */
391 		"r6 = 3;"
392 		"*(u64 *)(r10 - 8) = r6;"
393 
394 		/* call subprog and ignore result; we need this call only to
395 		 * complicate jump history
396 		 */
397 		"r1 = 0;"
398 		"call identity_subprog;"
399 
400 		/* restore reg from stack; in this case we'll be carrying
401 		 * stack mask when going back into subprog through jump
402 		 * history
403 		 */
404 		"r6 = *(u64 *)(r10 - 8);"
405 
406 		"r6 *= 4;"
407 		"r1 = %[vals];"
408 		/* here r6 is forced to be precise and has to be propagated
409 		 * back to the beginning, handling (and ignoring) subprog call
410 		 */
411 		"r1 += r6;"
412 		"r0 = *(u32 *)(r1 + 0);"
413 		"exit;"
414 		:
415 		: __imm_ptr(vals)
416 		: __clobber_common, "r6"
417 	);
418 }
419 
420 SEC("?raw_tp")
421 __success __log_level(2)
422 __msg("9: (0f) r1 += r6")
423 __msg("mark_precise: frame0: last_idx 9 first_idx 0")
424 __msg("mark_precise: frame0: regs=r6 stack= before 8: (bf) r1 = r7")
425 __msg("mark_precise: frame0: regs=r6 stack= before 7: (27) r6 *= 4")
426 __msg("mark_precise: frame0: regs=r6 stack= before 6: (79) r6 = *(u64 *)(r10 -8)")
427 __msg("mark_precise: frame0: regs= stack=-8 before 5: (85) call pc+6")
428 __msg("mark_precise: frame0: regs= stack=-8 before 4: (b7) r1 = 0")
429 __msg("mark_precise: frame0: regs= stack=-8 before 3: (7b) *(u64 *)(r10 -8) = r6")
430 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
431 __naked int parent_stack_slot_precise_global(void)
432 {
433 	asm volatile (
434 		/* spill reg */
435 		"r6 = 3;"
436 		"*(u64 *)(r10 - 8) = r6;"
437 
438 		/* call subprog and ignore result; we need this call only to
439 		 * complicate jump history
440 		 */
441 		"r1 = 0;"
442 		"call global_identity_subprog;"
443 
444 		/* restore reg from stack; in this case we'll be carrying
445 		 * stack mask when going back into subprog through jump
446 		 * history
447 		 */
448 		"r6 = *(u64 *)(r10 - 8);"
449 
450 		"r6 *= 4;"
451 		"r1 = %[vals];"
452 		/* here r6 is forced to be precise and has to be propagated
453 		 * back to the beginning, handling (and ignoring) subprog call
454 		 */
455 		"r1 += r6;"
456 		"r0 = *(u32 *)(r1 + 0);"
457 		"exit;"
458 		:
459 		: __imm_ptr(vals)
460 		: __clobber_common, "r6"
461 	);
462 }
463 
464 SEC("?raw_tp")
465 __success __log_level(2)
466 /* First simulated path does not include callback body */
467 __msg("14: (0f) r1 += r6")
468 __msg("mark_precise: frame0: last_idx 14 first_idx 10")
469 __msg("mark_precise: frame0: regs=r6 stack= before 13: (bf) r1 = r7")
470 __msg("mark_precise: frame0: regs=r6 stack= before 12: (27) r6 *= 4")
471 __msg("mark_precise: frame0: regs=r6 stack= before 11: (79) r6 = *(u64 *)(r10 -8)")
472 __msg("mark_precise: frame0: regs= stack=-8 before 10: (85) call bpf_loop")
473 __msg("mark_precise: frame0: parent state regs= stack=-8:")
474 __msg("mark_precise: frame0: last_idx 9 first_idx 0 subseq_idx 10")
475 __msg("mark_precise: frame0: regs= stack=-8 before 9: (b7) r4 = 0")
476 __msg("mark_precise: frame0: regs= stack=-8 before 8: (b7) r3 = 0")
477 __msg("mark_precise: frame0: regs= stack=-8 before 7: (bf) r2 = r8")
478 __msg("mark_precise: frame0: regs= stack=-8 before 6: (bf) r1 = r6")
479 __msg("mark_precise: frame0: regs= stack=-8 before 5: (7b) *(u64 *)(r10 -8) = r6")
480 __msg("mark_precise: frame0: regs=r6 stack= before 4: (b7) r6 = 3")
481 /* State entering callback body popped from states stack */
482 __msg("from 10 to 17: frame1:")
483 __msg("17: frame1: R1=scalar() R2=0 R10=fp0 cb")
484 __msg("17: (b7) r0 = 0")
485 __msg("18: (95) exit")
486 __msg("returning from callee:")
487 __msg("to caller at 10:")
488 /* r1, r4 are always precise for bpf_loop(),
489  * fp-8 was marked before backtracking to callback body.
490  */
491 __msg("frame 0: propagating r1,r4,fp-8")
492 __msg("mark_precise: frame0: last_idx 10 first_idx 10 subseq_idx -1")
493 __msg("mark_precise: frame0: regs=r1,r4 stack=-8 before 18: (95) exit")
494 __msg("mark_precise: frame1: regs= stack= before 17: (b7) r0 = 0")
495 __msg("mark_precise: frame1: regs= stack= before 10: (85) call bpf_loop#181")
496 __msg("mark_precise: frame0: parent state regs= stack=:")
497 __msg("from 18 to 10: safe")
498 __naked int parent_stack_slot_precise_with_callback(void)
499 {
500 	asm volatile (
501 		/* spill reg */
502 		"r6 = 3;"
503 		"*(u64 *)(r10 - 8) = r6;"
504 
505 		/* ensure we have callback frame in jump history */
506 		"r1 = r6;"			/* nr_loops */
507 		"r2 = %[callback_subprog];"	/* callback_fn */
508 		"r3 = 0;"			/* callback_ctx */
509 		"r4 = 0;"			/* flags */
510 		"call %[bpf_loop];"
511 
512 		/* restore reg from stack; in this case we'll be carrying
513 		 * stack mask when going back into subprog through jump
514 		 * history
515 		 */
516 		"r6 = *(u64 *)(r10 - 8);"
517 
518 		"r6 *= 4;"
519 		"r1 = %[vals];"
520 		/* here r6 is forced to be precise and has to be propagated
521 		 * back to the beginning, handling (and ignoring) subprog call
522 		 */
523 		"r1 += r6;"
524 		"r0 = *(u32 *)(r1 + 0);"
525 		"exit;"
526 		:
527 		: __imm_ptr(vals),
528 		  __imm_ptr(callback_subprog),
529 		  __imm(bpf_loop)
530 		: __clobber_common, "r6"
531 	);
532 }
533 
534 __noinline __used
535 static __u64 subprog_with_precise_arg(__u64 x)
536 {
537 	return vals[x]; /* x is forced to be precise */
538 }
539 
540 SEC("?raw_tp")
541 __success __log_level(2)
542 __msg("8: (0f) r2 += r1")
543 __msg("mark_precise: frame1: last_idx 8 first_idx 0")
544 __msg("mark_precise: frame1: regs=r1 stack= before 6: (18) r2 = ")
545 __msg("mark_precise: frame1: regs=r1 stack= before 5: (67) r1 <<= 2")
546 __msg("mark_precise: frame1: regs=r1 stack= before 2: (85) call pc+2")
547 __msg("mark_precise: frame0: regs=r1 stack= before 1: (bf) r1 = r6")
548 __msg("mark_precise: frame0: regs=r6 stack= before 0: (b7) r6 = 3")
549 __naked int subprog_arg_precise(void)
550 {
551 	asm volatile (
552 		"r6 = 3;"
553 		"r1 = r6;"
554 		/* subprog_with_precise_arg expects its argument to be
555 		 * precise, so r1->r6 will be marked precise from inside the
556 		 * subprog
557 		 */
558 		"call subprog_with_precise_arg;"
559 		"r0 += r6;"
560 		"exit;"
561 		:
562 		:
563 		: __clobber_common, "r6"
564 	);
565 }
566 
567 /* r1 is pointer to stack slot;
568  * r2 is a register to spill into that slot
569  * subprog also spills r2 into its own stack slot
570  */
571 __naked __noinline __used
572 static __u64 subprog_spill_reg_precise(void)
573 {
574 	asm volatile (
575 		/* spill to parent stack */
576 		"*(u64 *)(r1 + 0) = r2;"
577 		/* spill to subprog stack (we use -16 offset to avoid
578 		 * accidental confusion with parent's -8 stack slot in
579 		 * verifier log output)
580 		 */
581 		"*(u64 *)(r10 - 16) = r2;"
582 		/* use both spills as return result to propagete precision everywhere */
583 		"r0 = *(u64 *)(r10 - 16);"
584 		"r2 = *(u64 *)(r1 + 0);"
585 		"r0 += r2;"
586 		"exit;"
587 	);
588 }
589 
590 SEC("?raw_tp")
591 __success __log_level(2)
592 __msg("10: (0f) r1 += r7")
593 __msg("mark_precise: frame0: last_idx 10 first_idx 7 subseq_idx -1")
594 __msg("mark_precise: frame0: regs=r7 stack= before 9: (bf) r1 = r8")
595 __msg("mark_precise: frame0: regs=r7 stack= before 8: (27) r7 *= 4")
596 __msg("mark_precise: frame0: regs=r7 stack= before 7: (79) r7 = *(u64 *)(r10 -8)")
597 __msg("mark_precise: frame0: parent state regs= stack=-8:  R0_w=2 R6_w=1 R8_rw=map_value(map=.data.vals,ks=4,vs=16) R10=fp0 fp-8_rw=P1")
598 __msg("mark_precise: frame0: last_idx 18 first_idx 0 subseq_idx 7")
599 __msg("mark_precise: frame0: regs= stack=-8 before 18: (95) exit")
600 __msg("mark_precise: frame1: regs= stack= before 17: (0f) r0 += r2")
601 __msg("mark_precise: frame1: regs= stack= before 16: (79) r2 = *(u64 *)(r1 +0)")
602 __msg("mark_precise: frame1: regs= stack= before 15: (79) r0 = *(u64 *)(r10 -16)")
603 __msg("mark_precise: frame1: regs= stack= before 14: (7b) *(u64 *)(r10 -16) = r2")
604 __msg("mark_precise: frame1: regs= stack= before 13: (7b) *(u64 *)(r1 +0) = r2")
605 __msg("mark_precise: frame1: regs=r2 stack= before 6: (85) call pc+6")
606 __msg("mark_precise: frame0: regs=r2 stack= before 5: (bf) r2 = r6")
607 __msg("mark_precise: frame0: regs=r6 stack= before 4: (07) r1 += -8")
608 __msg("mark_precise: frame0: regs=r6 stack= before 3: (bf) r1 = r10")
609 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 1")
610 __naked int subprog_spill_into_parent_stack_slot_precise(void)
611 {
612 	asm volatile (
613 		"r6 = 1;"
614 
615 		/* pass pointer to stack slot and r6 to subprog;
616 		 * r6 will be marked precise and spilled into fp-8 slot, which
617 		 * also should be marked precise
618 		 */
619 		"r1 = r10;"
620 		"r1 += -8;"
621 		"r2 = r6;"
622 		"call subprog_spill_reg_precise;"
623 
624 		/* restore reg from stack; in this case we'll be carrying
625 		 * stack mask when going back into subprog through jump
626 		 * history
627 		 */
628 		"r7 = *(u64 *)(r10 - 8);"
629 
630 		"r7 *= 4;"
631 		"r1 = %[vals];"
632 		/* here r7 is forced to be precise and has to be propagated
633 		 * back to the beginning, handling subprog call and logic
634 		 */
635 		"r1 += r7;"
636 		"r0 = *(u32 *)(r1 + 0);"
637 		"exit;"
638 		:
639 		: __imm_ptr(vals)
640 		: __clobber_common, "r6", "r7"
641 	);
642 }
643 
644 SEC("?raw_tp")
645 __success __log_level(2)
646 __msg("17: (0f) r1 += r0")
647 __msg("mark_precise: frame0: last_idx 17 first_idx 0 subseq_idx -1")
648 __msg("mark_precise: frame0: regs=r0 stack= before 16: (bf) r1 = r7")
649 __msg("mark_precise: frame0: regs=r0 stack= before 15: (27) r0 *= 4")
650 __msg("mark_precise: frame0: regs=r0 stack= before 14: (79) r0 = *(u64 *)(r10 -16)")
651 __msg("mark_precise: frame0: regs= stack=-16 before 13: (7b) *(u64 *)(r7 -8) = r0")
652 __msg("mark_precise: frame0: regs=r0 stack= before 12: (79) r0 = *(u64 *)(r8 +16)")
653 __msg("mark_precise: frame0: regs= stack=-16 before 11: (7b) *(u64 *)(r8 +16) = r0")
654 __msg("mark_precise: frame0: regs=r0 stack= before 10: (79) r0 = *(u64 *)(r7 -8)")
655 __msg("mark_precise: frame0: regs= stack=-16 before 9: (7b) *(u64 *)(r10 -16) = r0")
656 __msg("mark_precise: frame0: regs=r0 stack= before 8: (07) r8 += -32")
657 __msg("mark_precise: frame0: regs=r0 stack= before 7: (bf) r8 = r10")
658 __msg("mark_precise: frame0: regs=r0 stack= before 6: (07) r7 += -8")
659 __msg("mark_precise: frame0: regs=r0 stack= before 5: (bf) r7 = r10")
660 __msg("mark_precise: frame0: regs=r0 stack= before 21: (95) exit")
661 __msg("mark_precise: frame1: regs=r0 stack= before 20: (bf) r0 = r1")
662 __msg("mark_precise: frame1: regs=r1 stack= before 4: (85) call pc+15")
663 __msg("mark_precise: frame0: regs=r1 stack= before 3: (bf) r1 = r6")
664 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 1")
665 __naked int stack_slot_aliases_precision(void)
666 {
667 	asm volatile (
668 		"r6 = 1;"
669 		/* pass r6 through r1 into subprog to get it back as r0;
670 		 * this whole chain will have to be marked as precise later
671 		 */
672 		"r1 = r6;"
673 		"call identity_subprog;"
674 		/* let's setup two registers that are aliased to r10 */
675 		"r7 = r10;"
676 		"r7 += -8;"			/* r7 = r10 - 8 */
677 		"r8 = r10;"
678 		"r8 += -32;"			/* r8 = r10 - 32 */
679 		/* now spill subprog's return value (a r6 -> r1 -> r0 chain)
680 		 * a few times through different stack pointer regs, making
681 		 * sure to use r10, r7, and r8 both in LDX and STX insns, and
682 		 * *importantly* also using a combination of const var_off and
683 		 * insn->off to validate that we record final stack slot
684 		 * correctly, instead of relying on just insn->off derivation,
685 		 * which is only valid for r10-based stack offset
686 		 */
687 		"*(u64 *)(r10 - 16) = r0;"
688 		"r0 = *(u64 *)(r7 - 8);"	/* r7 - 8 == r10 - 16 */
689 		"*(u64 *)(r8 + 16) = r0;"	/* r8 + 16 = r10 - 16 */
690 		"r0 = *(u64 *)(r8 + 16);"
691 		"*(u64 *)(r7 - 8) = r0;"
692 		"r0 = *(u64 *)(r10 - 16);"
693 		/* get ready to use r0 as an index into array to force precision */
694 		"r0 *= 4;"
695 		"r1 = %[vals];"
696 		/* here r0->r1->r6 chain is forced to be precise and has to be
697 		 * propagated back to the beginning, including through the
698 		 * subprog call and all the stack spills and loads
699 		 */
700 		"r1 += r0;"
701 		"r0 = *(u32 *)(r1 + 0);"
702 		"exit;"
703 		:
704 		: __imm_ptr(vals)
705 		: __clobber_common, "r6"
706 	);
707 }
708 
709 char _license[] SEC("license") = "GPL";
710