xref: /linux/tools/testing/selftests/bpf/progs/verifier_subprog_precision.c (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
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 #include <../../../tools/include/linux/filter.h>
10 
11 #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
12 
13 int vals[] SEC(".data.vals") = {1, 2, 3, 4};
14 
15 __naked __noinline __used
16 static unsigned long identity_subprog()
17 {
18 	/* the simplest *static* 64-bit identity function */
19 	asm volatile (
20 		"r0 = r1;"
21 		"exit;"
22 	);
23 }
24 
25 __noinline __used
26 unsigned long global_identity_subprog(__u64 x)
27 {
28 	/* the simplest *global* 64-bit identity function */
29 	return x;
30 }
31 
32 __naked __noinline __used
33 static unsigned long callback_subprog()
34 {
35 	/* the simplest callback function */
36 	asm volatile (
37 		"r0 = 0;"
38 		"exit;"
39 	);
40 }
41 
42 SEC("?raw_tp")
43 __success __log_level(2)
44 __msg("7: (0f) r1 += r0")
45 __msg("mark_precise: frame0: regs=r0 stack= before 6: (bf) r1 = r7")
46 __msg("mark_precise: frame0: regs=r0 stack= before 5: (27) r0 *= 4")
47 __msg("mark_precise: frame0: regs=r0 stack= before 11: (95) exit")
48 __msg("mark_precise: frame1: regs=r0 stack= before 10: (bf) r0 = r1")
49 __msg("mark_precise: frame1: regs=r1 stack= before 4: (85) call pc+5")
50 __msg("mark_precise: frame0: regs=r1 stack= before 3: (bf) r1 = r6")
51 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
52 __naked int subprog_result_precise(void)
53 {
54 	asm volatile (
55 		"r6 = 3;"
56 		/* pass r6 through r1 into subprog to get it back as r0;
57 		 * this whole chain will have to be marked as precise later
58 		 */
59 		"r1 = r6;"
60 		"call identity_subprog;"
61 		/* now use subprog's returned value (which is a
62 		 * r6 -> r1 -> r0 chain), as index into vals array, forcing
63 		 * all of that to be known precisely
64 		 */
65 		"r0 *= 4;"
66 		"r1 = %[vals];"
67 		/* here r0->r1->r6 chain is forced to be precise and has to be
68 		 * propagated back to the beginning, including through the
69 		 * subprog call
70 		 */
71 		"r1 += r0;"
72 		"r0 = *(u32 *)(r1 + 0);"
73 		"exit;"
74 		:
75 		: __imm_ptr(vals)
76 		: __clobber_common, "r6"
77 	);
78 }
79 
80 __naked __noinline __used
81 static unsigned long fp_leaking_subprog()
82 {
83 	asm volatile (
84 		".8byte %[r0_eq_r10_cast_s8];"
85 		"exit;"
86 		:: __imm_insn(r0_eq_r10_cast_s8, BPF_MOVSX64_REG(BPF_REG_0, BPF_REG_10, 8))
87 	);
88 }
89 
90 __naked __noinline __used
91 static unsigned long sneaky_fp_leaking_subprog()
92 {
93 	asm volatile (
94 		"r1 = r10;"
95 		".8byte %[r0_eq_r1_cast_s8];"
96 		"exit;"
97 		:: __imm_insn(r0_eq_r1_cast_s8, BPF_MOVSX64_REG(BPF_REG_0, BPF_REG_1, 8))
98 	);
99 }
100 
101 SEC("?raw_tp")
102 __success __log_level(2)
103 __msg("6: (0f) r1 += r0")
104 __msg("mark_precise: frame0: last_idx 6 first_idx 0 subseq_idx -1")
105 __msg("mark_precise: frame0: regs=r0 stack= before 5: (bf) r1 = r6")
106 __msg("mark_precise: frame0: regs=r0 stack= before 4: (27) r0 *= 4")
107 __msg("mark_precise: frame0: regs=r0 stack= before 3: (57) r0 &= 3")
108 __msg("mark_precise: frame0: regs=r0 stack= before 10: (95) exit")
109 __msg("mark_precise: frame1: regs=r0 stack= before 9: (bf) r0 = (s8)r10")
110 __msg("7: R0_w=scalar")
111 __naked int fp_precise_subprog_result(void)
112 {
113 	asm volatile (
114 		"call fp_leaking_subprog;"
115 		/* use subprog's returned value (which is derived from r10=fp
116 		 * register), as index into vals array, forcing all of that to
117 		 * be known precisely
118 		 */
119 		"r0 &= 3;"
120 		"r0 *= 4;"
121 		"r1 = %[vals];"
122 		/* force precision marking */
123 		"r1 += r0;"
124 		"r0 = *(u32 *)(r1 + 0);"
125 		"exit;"
126 		:
127 		: __imm_ptr(vals)
128 		: __clobber_common
129 	);
130 }
131 
132 SEC("?raw_tp")
133 __success __log_level(2)
134 __msg("6: (0f) r1 += r0")
135 __msg("mark_precise: frame0: last_idx 6 first_idx 0 subseq_idx -1")
136 __msg("mark_precise: frame0: regs=r0 stack= before 5: (bf) r1 = r6")
137 __msg("mark_precise: frame0: regs=r0 stack= before 4: (27) r0 *= 4")
138 __msg("mark_precise: frame0: regs=r0 stack= before 3: (57) r0 &= 3")
139 __msg("mark_precise: frame0: regs=r0 stack= before 11: (95) exit")
140 __msg("mark_precise: frame1: regs=r0 stack= before 10: (bf) r0 = (s8)r1")
141 /* here r1 is marked precise, even though it's fp register, but that's fine
142  * because by the time we get out of subprogram it has to be derived from r10
143  * anyways, at which point we'll break precision chain
144  */
145 __msg("mark_precise: frame1: regs=r1 stack= before 9: (bf) r1 = r10")
146 __msg("7: R0_w=scalar")
147 __naked int sneaky_fp_precise_subprog_result(void)
148 {
149 	asm volatile (
150 		"call sneaky_fp_leaking_subprog;"
151 		/* use subprog's returned value (which is derived from r10=fp
152 		 * register), as index into vals array, forcing all of that to
153 		 * be known precisely
154 		 */
155 		"r0 &= 3;"
156 		"r0 *= 4;"
157 		"r1 = %[vals];"
158 		/* force precision marking */
159 		"r1 += r0;"
160 		"r0 = *(u32 *)(r1 + 0);"
161 		"exit;"
162 		:
163 		: __imm_ptr(vals)
164 		: __clobber_common
165 	);
166 }
167 
168 SEC("?raw_tp")
169 __success __log_level(2)
170 __msg("9: (0f) r1 += r0")
171 __msg("mark_precise: frame0: last_idx 9 first_idx 0")
172 __msg("mark_precise: frame0: regs=r0 stack= before 8: (bf) r1 = r7")
173 __msg("mark_precise: frame0: regs=r0 stack= before 7: (27) r0 *= 4")
174 __msg("mark_precise: frame0: regs=r0 stack= before 5: (a5) if r0 < 0x4 goto pc+1")
175 __msg("mark_precise: frame0: regs=r0 stack= before 4: (85) call pc+7")
176 __naked int global_subprog_result_precise(void)
177 {
178 	asm volatile (
179 		"r6 = 3;"
180 		/* pass r6 through r1 into subprog to get it back as r0;
181 		 * given global_identity_subprog is global, precision won't
182 		 * propagate all the way back to r6
183 		 */
184 		"r1 = r6;"
185 		"call global_identity_subprog;"
186 		/* now use subprog's returned value (which is unknown now, so
187 		 * we need to clamp it), as index into vals array, forcing r0
188 		 * to be marked precise (with no effect on r6, though)
189 		 */
190 		"if r0 < %[vals_arr_sz] goto 1f;"
191 		"r0 = %[vals_arr_sz] - 1;"
192 	"1:"
193 		"r0 *= 4;"
194 		"r1 = %[vals];"
195 		/* here r0 is forced to be precise and has to be
196 		 * propagated back to the global subprog call, but it
197 		 * shouldn't go all the way to mark r6 as precise
198 		 */
199 		"r1 += r0;"
200 		"r0 = *(u32 *)(r1 + 0);"
201 		"exit;"
202 		:
203 		: __imm_ptr(vals),
204 		  __imm_const(vals_arr_sz, ARRAY_SIZE(vals))
205 		: __clobber_common, "r6"
206 	);
207 }
208 
209 __naked __noinline __used
210 static unsigned long loop_callback_bad()
211 {
212 	/* bpf_loop() callback that can return values outside of [0, 1] range */
213 	asm volatile (
214 		"call %[bpf_get_prandom_u32];"
215 		"if r0 s> 1000 goto 1f;"
216 		"r0 = 0;"
217 	"1:"
218 		"goto +0;" /* checkpoint */
219 		/* bpf_loop() expects [0, 1] values, so branch above skipping
220 		 * r0 = 0; should lead to a failure, but if exit instruction
221 		 * doesn't enforce r0's precision, this callback will be
222 		 * successfully verified
223 		 */
224 		"exit;"
225 		:
226 		: __imm(bpf_get_prandom_u32)
227 		: __clobber_common
228 	);
229 }
230 
231 SEC("?raw_tp")
232 __failure __log_level(2)
233 __flag(BPF_F_TEST_STATE_FREQ)
234 /* check that fallthrough code path marks r0 as precise */
235 __msg("mark_precise: frame1: regs=r0 stack= before 11: (b7) r0 = 0")
236 /* check that we have branch code path doing its own validation */
237 __msg("from 10 to 12: frame1: R0=scalar(smin=umin=1001")
238 /* check that branch code path marks r0 as precise, before failing */
239 __msg("mark_precise: frame1: regs=r0 stack= before 9: (85) call bpf_get_prandom_u32#7")
240 __msg("At callback return the register R0 has smin=1001 should have been in [0, 1]")
241 __naked int callback_precise_return_fail(void)
242 {
243 	asm volatile (
244 		"r1 = 1;"			/* nr_loops */
245 		"r2 = %[loop_callback_bad];"	/* callback_fn */
246 		"r3 = 0;"			/* callback_ctx */
247 		"r4 = 0;"			/* flags */
248 		"call %[bpf_loop];"
249 
250 		"r0 = 0;"
251 		"exit;"
252 		:
253 		: __imm_ptr(loop_callback_bad),
254 		  __imm(bpf_loop)
255 		: __clobber_common
256 	);
257 }
258 
259 SEC("?raw_tp")
260 __success __log_level(2)
261 /* First simulated path does not include callback body,
262  * r1 and r4 are always precise for bpf_loop() calls.
263  */
264 __msg("9: (85) call bpf_loop#181")
265 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
266 __msg("mark_precise: frame0: parent state regs=r4 stack=:")
267 __msg("mark_precise: frame0: last_idx 8 first_idx 0 subseq_idx 9")
268 __msg("mark_precise: frame0: regs=r4 stack= before 8: (b7) r4 = 0")
269 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
270 __msg("mark_precise: frame0: parent state regs=r1 stack=:")
271 __msg("mark_precise: frame0: last_idx 8 first_idx 0 subseq_idx 9")
272 __msg("mark_precise: frame0: regs=r1 stack= before 8: (b7) r4 = 0")
273 __msg("mark_precise: frame0: regs=r1 stack= before 7: (b7) r3 = 0")
274 __msg("mark_precise: frame0: regs=r1 stack= before 6: (bf) r2 = r8")
275 __msg("mark_precise: frame0: regs=r1 stack= before 5: (bf) r1 = r6")
276 __msg("mark_precise: frame0: regs=r6 stack= before 4: (b7) r6 = 3")
277 /* r6 precision propagation */
278 __msg("14: (0f) r1 += r6")
279 __msg("mark_precise: frame0: last_idx 14 first_idx 9")
280 __msg("mark_precise: frame0: regs=r6 stack= before 13: (bf) r1 = r7")
281 __msg("mark_precise: frame0: regs=r6 stack= before 12: (27) r6 *= 4")
282 __msg("mark_precise: frame0: regs=r6 stack= before 11: (25) if r6 > 0x3 goto pc+4")
283 __msg("mark_precise: frame0: regs=r6 stack= before 10: (bf) r6 = r0")
284 __msg("mark_precise: frame0: regs=r0 stack= before 9: (85) call bpf_loop")
285 /* State entering callback body popped from states stack */
286 __msg("from 9 to 17: frame1:")
287 __msg("17: frame1: R1=scalar() R2=0 R10=fp0 cb")
288 __msg("17: (b7) r0 = 0")
289 __msg("18: (95) exit")
290 __msg("returning from callee:")
291 __msg("to caller at 9:")
292 __msg("frame 0: propagating r1,r4")
293 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
294 __msg("mark_precise: frame0: regs=r1,r4 stack= before 18: (95) exit")
295 __msg("from 18 to 9: safe")
296 __naked int callback_result_precise(void)
297 {
298 	asm volatile (
299 		"r6 = 3;"
300 
301 		/* call subprog and use result; r0 shouldn't propagate back to
302 		 * callback_subprog
303 		 */
304 		"r1 = r6;"			/* nr_loops */
305 		"r2 = %[callback_subprog];"	/* callback_fn */
306 		"r3 = 0;"			/* callback_ctx */
307 		"r4 = 0;"			/* flags */
308 		"call %[bpf_loop];"
309 
310 		"r6 = r0;"
311 		"if r6 > 3 goto 1f;"
312 		"r6 *= 4;"
313 		"r1 = %[vals];"
314 		/* here r6 is forced to be precise and has to be propagated
315 		 * back to the bpf_loop() call, but not beyond
316 		 */
317 		"r1 += r6;"
318 		"r0 = *(u32 *)(r1 + 0);"
319 	"1:"
320 		"exit;"
321 		:
322 		: __imm_ptr(vals),
323 		  __imm_ptr(callback_subprog),
324 		  __imm(bpf_loop)
325 		: __clobber_common, "r6"
326 	);
327 }
328 
329 SEC("?raw_tp")
330 __success __log_level(2)
331 __msg("7: (0f) r1 += r6")
332 __msg("mark_precise: frame0: last_idx 7 first_idx 0")
333 __msg("mark_precise: frame0: regs=r6 stack= before 6: (bf) r1 = r7")
334 __msg("mark_precise: frame0: regs=r6 stack= before 5: (27) r6 *= 4")
335 __msg("mark_precise: frame0: regs=r6 stack= before 11: (95) exit")
336 __msg("mark_precise: frame1: regs= stack= before 10: (bf) r0 = r1")
337 __msg("mark_precise: frame1: regs= stack= before 4: (85) call pc+5")
338 __msg("mark_precise: frame0: regs=r6 stack= before 3: (b7) r1 = 0")
339 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
340 __naked int parent_callee_saved_reg_precise(void)
341 {
342 	asm volatile (
343 		"r6 = 3;"
344 
345 		/* call subprog and ignore result; we need this call only to
346 		 * complicate jump history
347 		 */
348 		"r1 = 0;"
349 		"call identity_subprog;"
350 
351 		"r6 *= 4;"
352 		"r1 = %[vals];"
353 		/* here r6 is forced to be precise and has to be propagated
354 		 * back to the beginning, handling (and ignoring) subprog call
355 		 */
356 		"r1 += r6;"
357 		"r0 = *(u32 *)(r1 + 0);"
358 		"exit;"
359 		:
360 		: __imm_ptr(vals)
361 		: __clobber_common, "r6"
362 	);
363 }
364 
365 SEC("?raw_tp")
366 __success __log_level(2)
367 __msg("7: (0f) r1 += r6")
368 __msg("mark_precise: frame0: last_idx 7 first_idx 0")
369 __msg("mark_precise: frame0: regs=r6 stack= before 6: (bf) r1 = r7")
370 __msg("mark_precise: frame0: regs=r6 stack= before 5: (27) r6 *= 4")
371 __msg("mark_precise: frame0: regs=r6 stack= before 4: (85) call pc+5")
372 __msg("mark_precise: frame0: regs=r6 stack= before 3: (b7) r1 = 0")
373 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
374 __naked int parent_callee_saved_reg_precise_global(void)
375 {
376 	asm volatile (
377 		"r6 = 3;"
378 
379 		/* call subprog and ignore result; we need this call only to
380 		 * complicate jump history
381 		 */
382 		"r1 = 0;"
383 		"call global_identity_subprog;"
384 
385 		"r6 *= 4;"
386 		"r1 = %[vals];"
387 		/* here r6 is forced to be precise and has to be propagated
388 		 * back to the beginning, handling (and ignoring) subprog call
389 		 */
390 		"r1 += r6;"
391 		"r0 = *(u32 *)(r1 + 0);"
392 		"exit;"
393 		:
394 		: __imm_ptr(vals)
395 		: __clobber_common, "r6"
396 	);
397 }
398 
399 SEC("?raw_tp")
400 __success __log_level(2)
401 /* First simulated path does not include callback body */
402 __msg("12: (0f) r1 += r6")
403 __msg("mark_precise: frame0: last_idx 12 first_idx 9")
404 __msg("mark_precise: frame0: regs=r6 stack= before 11: (bf) r1 = r7")
405 __msg("mark_precise: frame0: regs=r6 stack= before 10: (27) r6 *= 4")
406 __msg("mark_precise: frame0: regs=r6 stack= before 9: (85) call bpf_loop")
407 __msg("mark_precise: frame0: parent state regs=r6 stack=:")
408 __msg("mark_precise: frame0: last_idx 8 first_idx 0 subseq_idx 9")
409 __msg("mark_precise: frame0: regs=r6 stack= before 8: (b7) r4 = 0")
410 __msg("mark_precise: frame0: regs=r6 stack= before 7: (b7) r3 = 0")
411 __msg("mark_precise: frame0: regs=r6 stack= before 6: (bf) r2 = r8")
412 __msg("mark_precise: frame0: regs=r6 stack= before 5: (b7) r1 = 1")
413 __msg("mark_precise: frame0: regs=r6 stack= before 4: (b7) r6 = 3")
414 /* State entering callback body popped from states stack */
415 __msg("from 9 to 15: frame1:")
416 __msg("15: frame1: R1=scalar() R2=0 R10=fp0 cb")
417 __msg("15: (b7) r0 = 0")
418 __msg("16: (95) exit")
419 __msg("returning from callee:")
420 __msg("to caller at 9:")
421 /* r1, r4 are always precise for bpf_loop(),
422  * r6 was marked before backtracking to callback body.
423  */
424 __msg("frame 0: propagating r1,r4,r6")
425 __msg("mark_precise: frame0: last_idx 9 first_idx 9 subseq_idx -1")
426 __msg("mark_precise: frame0: regs=r1,r4,r6 stack= before 16: (95) exit")
427 __msg("mark_precise: frame1: regs= stack= before 15: (b7) r0 = 0")
428 __msg("mark_precise: frame1: regs= stack= before 9: (85) call bpf_loop")
429 __msg("mark_precise: frame0: parent state regs= stack=:")
430 __msg("from 16 to 9: safe")
431 __naked int parent_callee_saved_reg_precise_with_callback(void)
432 {
433 	asm volatile (
434 		"r6 = 3;"
435 
436 		/* call subprog and ignore result; we need this call only to
437 		 * complicate jump history
438 		 */
439 		"r1 = 1;"			/* nr_loops */
440 		"r2 = %[callback_subprog];"	/* callback_fn */
441 		"r3 = 0;"			/* callback_ctx */
442 		"r4 = 0;"			/* flags */
443 		"call %[bpf_loop];"
444 
445 		"r6 *= 4;"
446 		"r1 = %[vals];"
447 		/* here r6 is forced to be precise and has to be propagated
448 		 * back to the beginning, handling (and ignoring) callback call
449 		 */
450 		"r1 += r6;"
451 		"r0 = *(u32 *)(r1 + 0);"
452 		"exit;"
453 		:
454 		: __imm_ptr(vals),
455 		  __imm_ptr(callback_subprog),
456 		  __imm(bpf_loop)
457 		: __clobber_common, "r6"
458 	);
459 }
460 
461 SEC("?raw_tp")
462 __success __log_level(2)
463 __msg("9: (0f) r1 += r6")
464 __msg("mark_precise: frame0: last_idx 9 first_idx 6")
465 __msg("mark_precise: frame0: regs=r6 stack= before 8: (bf) r1 = r7")
466 __msg("mark_precise: frame0: regs=r6 stack= before 7: (27) r6 *= 4")
467 __msg("mark_precise: frame0: regs=r6 stack= before 6: (79) r6 = *(u64 *)(r10 -8)")
468 __msg("mark_precise: frame0: parent state regs= stack=-8:")
469 __msg("mark_precise: frame0: last_idx 13 first_idx 0")
470 __msg("mark_precise: frame0: regs= stack=-8 before 13: (95) exit")
471 __msg("mark_precise: frame1: regs= stack= before 12: (bf) r0 = r1")
472 __msg("mark_precise: frame1: regs= stack= before 5: (85) call pc+6")
473 __msg("mark_precise: frame0: regs= stack=-8 before 4: (b7) r1 = 0")
474 __msg("mark_precise: frame0: regs= stack=-8 before 3: (7b) *(u64 *)(r10 -8) = r6")
475 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
476 __naked int parent_stack_slot_precise(void)
477 {
478 	asm volatile (
479 		/* spill reg */
480 		"r6 = 3;"
481 		"*(u64 *)(r10 - 8) = r6;"
482 
483 		/* call subprog and ignore result; we need this call only to
484 		 * complicate jump history
485 		 */
486 		"r1 = 0;"
487 		"call identity_subprog;"
488 
489 		/* restore reg from stack; in this case we'll be carrying
490 		 * stack mask when going back into subprog through jump
491 		 * history
492 		 */
493 		"r6 = *(u64 *)(r10 - 8);"
494 
495 		"r6 *= 4;"
496 		"r1 = %[vals];"
497 		/* here r6 is forced to be precise and has to be propagated
498 		 * back to the beginning, handling (and ignoring) subprog call
499 		 */
500 		"r1 += r6;"
501 		"r0 = *(u32 *)(r1 + 0);"
502 		"exit;"
503 		:
504 		: __imm_ptr(vals)
505 		: __clobber_common, "r6"
506 	);
507 }
508 
509 SEC("?raw_tp")
510 __success __log_level(2)
511 __msg("9: (0f) r1 += r6")
512 __msg("mark_precise: frame0: last_idx 9 first_idx 0")
513 __msg("mark_precise: frame0: regs=r6 stack= before 8: (bf) r1 = r7")
514 __msg("mark_precise: frame0: regs=r6 stack= before 7: (27) r6 *= 4")
515 __msg("mark_precise: frame0: regs=r6 stack= before 6: (79) r6 = *(u64 *)(r10 -8)")
516 __msg("mark_precise: frame0: regs= stack=-8 before 5: (85) call pc+6")
517 __msg("mark_precise: frame0: regs= stack=-8 before 4: (b7) r1 = 0")
518 __msg("mark_precise: frame0: regs= stack=-8 before 3: (7b) *(u64 *)(r10 -8) = r6")
519 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 3")
520 __naked int parent_stack_slot_precise_global(void)
521 {
522 	asm volatile (
523 		/* spill reg */
524 		"r6 = 3;"
525 		"*(u64 *)(r10 - 8) = r6;"
526 
527 		/* call subprog and ignore result; we need this call only to
528 		 * complicate jump history
529 		 */
530 		"r1 = 0;"
531 		"call global_identity_subprog;"
532 
533 		/* restore reg from stack; in this case we'll be carrying
534 		 * stack mask when going back into subprog through jump
535 		 * history
536 		 */
537 		"r6 = *(u64 *)(r10 - 8);"
538 
539 		"r6 *= 4;"
540 		"r1 = %[vals];"
541 		/* here r6 is forced to be precise and has to be propagated
542 		 * back to the beginning, handling (and ignoring) subprog call
543 		 */
544 		"r1 += r6;"
545 		"r0 = *(u32 *)(r1 + 0);"
546 		"exit;"
547 		:
548 		: __imm_ptr(vals)
549 		: __clobber_common, "r6"
550 	);
551 }
552 
553 SEC("?raw_tp")
554 __success __log_level(2)
555 /* First simulated path does not include callback body */
556 __msg("14: (0f) r1 += r6")
557 __msg("mark_precise: frame0: last_idx 14 first_idx 10")
558 __msg("mark_precise: frame0: regs=r6 stack= before 13: (bf) r1 = r7")
559 __msg("mark_precise: frame0: regs=r6 stack= before 12: (27) r6 *= 4")
560 __msg("mark_precise: frame0: regs=r6 stack= before 11: (79) r6 = *(u64 *)(r10 -8)")
561 __msg("mark_precise: frame0: regs= stack=-8 before 10: (85) call bpf_loop")
562 __msg("mark_precise: frame0: parent state regs= stack=-8:")
563 __msg("mark_precise: frame0: last_idx 9 first_idx 0 subseq_idx 10")
564 __msg("mark_precise: frame0: regs= stack=-8 before 9: (b7) r4 = 0")
565 __msg("mark_precise: frame0: regs= stack=-8 before 8: (b7) r3 = 0")
566 __msg("mark_precise: frame0: regs= stack=-8 before 7: (bf) r2 = r8")
567 __msg("mark_precise: frame0: regs= stack=-8 before 6: (bf) r1 = r6")
568 __msg("mark_precise: frame0: regs= stack=-8 before 5: (7b) *(u64 *)(r10 -8) = r6")
569 __msg("mark_precise: frame0: regs=r6 stack= before 4: (b7) r6 = 3")
570 /* State entering callback body popped from states stack */
571 __msg("from 10 to 17: frame1:")
572 __msg("17: frame1: R1=scalar() R2=0 R10=fp0 cb")
573 __msg("17: (b7) r0 = 0")
574 __msg("18: (95) exit")
575 __msg("returning from callee:")
576 __msg("to caller at 10:")
577 /* r1, r4 are always precise for bpf_loop(),
578  * fp-8 was marked before backtracking to callback body.
579  */
580 __msg("frame 0: propagating r1,r4,fp-8")
581 __msg("mark_precise: frame0: last_idx 10 first_idx 10 subseq_idx -1")
582 __msg("mark_precise: frame0: regs=r1,r4 stack=-8 before 18: (95) exit")
583 __msg("mark_precise: frame1: regs= stack= before 17: (b7) r0 = 0")
584 __msg("mark_precise: frame1: regs= stack= before 10: (85) call bpf_loop#181")
585 __msg("mark_precise: frame0: parent state regs= stack=:")
586 __msg("from 18 to 10: safe")
587 __naked int parent_stack_slot_precise_with_callback(void)
588 {
589 	asm volatile (
590 		/* spill reg */
591 		"r6 = 3;"
592 		"*(u64 *)(r10 - 8) = r6;"
593 
594 		/* ensure we have callback frame in jump history */
595 		"r1 = r6;"			/* nr_loops */
596 		"r2 = %[callback_subprog];"	/* callback_fn */
597 		"r3 = 0;"			/* callback_ctx */
598 		"r4 = 0;"			/* flags */
599 		"call %[bpf_loop];"
600 
601 		/* restore reg from stack; in this case we'll be carrying
602 		 * stack mask when going back into subprog through jump
603 		 * history
604 		 */
605 		"r6 = *(u64 *)(r10 - 8);"
606 
607 		"r6 *= 4;"
608 		"r1 = %[vals];"
609 		/* here r6 is forced to be precise and has to be propagated
610 		 * back to the beginning, handling (and ignoring) subprog call
611 		 */
612 		"r1 += r6;"
613 		"r0 = *(u32 *)(r1 + 0);"
614 		"exit;"
615 		:
616 		: __imm_ptr(vals),
617 		  __imm_ptr(callback_subprog),
618 		  __imm(bpf_loop)
619 		: __clobber_common, "r6"
620 	);
621 }
622 
623 __noinline __used
624 static __u64 subprog_with_precise_arg(__u64 x)
625 {
626 	return vals[x]; /* x is forced to be precise */
627 }
628 
629 SEC("?raw_tp")
630 __success __log_level(2)
631 __msg("8: (0f) r2 += r1")
632 __msg("mark_precise: frame1: last_idx 8 first_idx 0")
633 __msg("mark_precise: frame1: regs=r1 stack= before 6: (18) r2 = ")
634 __msg("mark_precise: frame1: regs=r1 stack= before 5: (67) r1 <<= 2")
635 __msg("mark_precise: frame1: regs=r1 stack= before 2: (85) call pc+2")
636 __msg("mark_precise: frame0: regs=r1 stack= before 1: (bf) r1 = r6")
637 __msg("mark_precise: frame0: regs=r6 stack= before 0: (b7) r6 = 3")
638 __naked int subprog_arg_precise(void)
639 {
640 	asm volatile (
641 		"r6 = 3;"
642 		"r1 = r6;"
643 		/* subprog_with_precise_arg expects its argument to be
644 		 * precise, so r1->r6 will be marked precise from inside the
645 		 * subprog
646 		 */
647 		"call subprog_with_precise_arg;"
648 		"r0 += r6;"
649 		"exit;"
650 		:
651 		:
652 		: __clobber_common, "r6"
653 	);
654 }
655 
656 /* r1 is pointer to stack slot;
657  * r2 is a register to spill into that slot
658  * subprog also spills r2 into its own stack slot
659  */
660 __naked __noinline __used
661 static __u64 subprog_spill_reg_precise(void)
662 {
663 	asm volatile (
664 		/* spill to parent stack */
665 		"*(u64 *)(r1 + 0) = r2;"
666 		/* spill to subprog stack (we use -16 offset to avoid
667 		 * accidental confusion with parent's -8 stack slot in
668 		 * verifier log output)
669 		 */
670 		"*(u64 *)(r10 - 16) = r2;"
671 		/* use both spills as return result to propagete precision everywhere */
672 		"r0 = *(u64 *)(r10 - 16);"
673 		"r2 = *(u64 *)(r1 + 0);"
674 		"r0 += r2;"
675 		"exit;"
676 	);
677 }
678 
679 SEC("?raw_tp")
680 __success __log_level(2)
681 __msg("10: (0f) r1 += r7")
682 __msg("mark_precise: frame0: last_idx 10 first_idx 7 subseq_idx -1")
683 __msg("mark_precise: frame0: regs=r7 stack= before 9: (bf) r1 = r8")
684 __msg("mark_precise: frame0: regs=r7 stack= before 8: (27) r7 *= 4")
685 __msg("mark_precise: frame0: regs=r7 stack= before 7: (79) r7 = *(u64 *)(r10 -8)")
686 __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")
687 __msg("mark_precise: frame0: last_idx 18 first_idx 0 subseq_idx 7")
688 __msg("mark_precise: frame0: regs= stack=-8 before 18: (95) exit")
689 __msg("mark_precise: frame1: regs= stack= before 17: (0f) r0 += r2")
690 __msg("mark_precise: frame1: regs= stack= before 16: (79) r2 = *(u64 *)(r1 +0)")
691 __msg("mark_precise: frame1: regs= stack= before 15: (79) r0 = *(u64 *)(r10 -16)")
692 __msg("mark_precise: frame1: regs= stack= before 14: (7b) *(u64 *)(r10 -16) = r2")
693 __msg("mark_precise: frame1: regs= stack= before 13: (7b) *(u64 *)(r1 +0) = r2")
694 __msg("mark_precise: frame1: regs=r2 stack= before 6: (85) call pc+6")
695 __msg("mark_precise: frame0: regs=r2 stack= before 5: (bf) r2 = r6")
696 __msg("mark_precise: frame0: regs=r6 stack= before 4: (07) r1 += -8")
697 __msg("mark_precise: frame0: regs=r6 stack= before 3: (bf) r1 = r10")
698 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 1")
699 __naked int subprog_spill_into_parent_stack_slot_precise(void)
700 {
701 	asm volatile (
702 		"r6 = 1;"
703 
704 		/* pass pointer to stack slot and r6 to subprog;
705 		 * r6 will be marked precise and spilled into fp-8 slot, which
706 		 * also should be marked precise
707 		 */
708 		"r1 = r10;"
709 		"r1 += -8;"
710 		"r2 = r6;"
711 		"call subprog_spill_reg_precise;"
712 
713 		/* restore reg from stack; in this case we'll be carrying
714 		 * stack mask when going back into subprog through jump
715 		 * history
716 		 */
717 		"r7 = *(u64 *)(r10 - 8);"
718 
719 		"r7 *= 4;"
720 		"r1 = %[vals];"
721 		/* here r7 is forced to be precise and has to be propagated
722 		 * back to the beginning, handling subprog call and logic
723 		 */
724 		"r1 += r7;"
725 		"r0 = *(u32 *)(r1 + 0);"
726 		"exit;"
727 		:
728 		: __imm_ptr(vals)
729 		: __clobber_common, "r6", "r7"
730 	);
731 }
732 
733 SEC("?raw_tp")
734 __success __log_level(2)
735 __msg("17: (0f) r1 += r0")
736 __msg("mark_precise: frame0: last_idx 17 first_idx 0 subseq_idx -1")
737 __msg("mark_precise: frame0: regs=r0 stack= before 16: (bf) r1 = r7")
738 __msg("mark_precise: frame0: regs=r0 stack= before 15: (27) r0 *= 4")
739 __msg("mark_precise: frame0: regs=r0 stack= before 14: (79) r0 = *(u64 *)(r10 -16)")
740 __msg("mark_precise: frame0: regs= stack=-16 before 13: (7b) *(u64 *)(r7 -8) = r0")
741 __msg("mark_precise: frame0: regs=r0 stack= before 12: (79) r0 = *(u64 *)(r8 +16)")
742 __msg("mark_precise: frame0: regs= stack=-16 before 11: (7b) *(u64 *)(r8 +16) = r0")
743 __msg("mark_precise: frame0: regs=r0 stack= before 10: (79) r0 = *(u64 *)(r7 -8)")
744 __msg("mark_precise: frame0: regs= stack=-16 before 9: (7b) *(u64 *)(r10 -16) = r0")
745 __msg("mark_precise: frame0: regs=r0 stack= before 8: (07) r8 += -32")
746 __msg("mark_precise: frame0: regs=r0 stack= before 7: (bf) r8 = r10")
747 __msg("mark_precise: frame0: regs=r0 stack= before 6: (07) r7 += -8")
748 __msg("mark_precise: frame0: regs=r0 stack= before 5: (bf) r7 = r10")
749 __msg("mark_precise: frame0: regs=r0 stack= before 21: (95) exit")
750 __msg("mark_precise: frame1: regs=r0 stack= before 20: (bf) r0 = r1")
751 __msg("mark_precise: frame1: regs=r1 stack= before 4: (85) call pc+15")
752 __msg("mark_precise: frame0: regs=r1 stack= before 3: (bf) r1 = r6")
753 __msg("mark_precise: frame0: regs=r6 stack= before 2: (b7) r6 = 1")
754 __naked int stack_slot_aliases_precision(void)
755 {
756 	asm volatile (
757 		"r6 = 1;"
758 		/* pass r6 through r1 into subprog to get it back as r0;
759 		 * this whole chain will have to be marked as precise later
760 		 */
761 		"r1 = r6;"
762 		"call identity_subprog;"
763 		/* let's setup two registers that are aliased to r10 */
764 		"r7 = r10;"
765 		"r7 += -8;"			/* r7 = r10 - 8 */
766 		"r8 = r10;"
767 		"r8 += -32;"			/* r8 = r10 - 32 */
768 		/* now spill subprog's return value (a r6 -> r1 -> r0 chain)
769 		 * a few times through different stack pointer regs, making
770 		 * sure to use r10, r7, and r8 both in LDX and STX insns, and
771 		 * *importantly* also using a combination of const var_off and
772 		 * insn->off to validate that we record final stack slot
773 		 * correctly, instead of relying on just insn->off derivation,
774 		 * which is only valid for r10-based stack offset
775 		 */
776 		"*(u64 *)(r10 - 16) = r0;"
777 		"r0 = *(u64 *)(r7 - 8);"	/* r7 - 8 == r10 - 16 */
778 		"*(u64 *)(r8 + 16) = r0;"	/* r8 + 16 = r10 - 16 */
779 		"r0 = *(u64 *)(r8 + 16);"
780 		"*(u64 *)(r7 - 8) = r0;"
781 		"r0 = *(u64 *)(r10 - 16);"
782 		/* get ready to use r0 as an index into array to force precision */
783 		"r0 *= 4;"
784 		"r1 = %[vals];"
785 		/* here r0->r1->r6 chain is forced to be precise and has to be
786 		 * propagated back to the beginning, including through the
787 		 * subprog call and all the stack spills and loads
788 		 */
789 		"r1 += r0;"
790 		"r0 = *(u32 *)(r1 + 0);"
791 		"exit;"
792 		:
793 		: __imm_ptr(vals)
794 		: __clobber_common, "r6"
795 	);
796 }
797 
798 char _license[] SEC("license") = "GPL";
799