1// SPDX-License-Identifier: GPL-2.0
  2
  3#include <linux/bpf.h>
  4#include <bpf/bpf_helpers.h>
  5#include <bpf/bpf_core_read.h>
  6#include "../../../include/linux/filter.h"
  7#include "bpf_misc.h"
  8#include <stdbool.h>
  9#include "bpf_kfuncs.h"
 10
 11SEC("raw_tp")
 12__arch_x86_64
 13__log_level(4) __msg("stack depth 8")
 14__xlated("4: r5 = 5")
 15__xlated("5: w0 = ")
 16__xlated("6: r0 = &(void __percpu *)(r0)")
 17__xlated("7: r0 = *(u32 *)(r0 +0)")
 18__xlated("8: exit")
 19__success
 20__naked void simple(void)
 21{
 22	asm volatile (
 23	"r1 = 1;"
 24	"r2 = 2;"
 25	"r3 = 3;"
 26	"r4 = 4;"
 27	"r5 = 5;"
 28	"*(u64 *)(r10 - 16) = r1;"
 29	"*(u64 *)(r10 - 24) = r2;"
 30	"*(u64 *)(r10 - 32) = r3;"
 31	"*(u64 *)(r10 - 40) = r4;"
 32	"*(u64 *)(r10 - 48) = r5;"
 33	"call %[bpf_get_smp_processor_id];"
 34	"r5 = *(u64 *)(r10 - 48);"
 35	"r4 = *(u64 *)(r10 - 40);"
 36	"r3 = *(u64 *)(r10 - 32);"
 37	"r2 = *(u64 *)(r10 - 24);"
 38	"r1 = *(u64 *)(r10 - 16);"
 39	"exit;"
 40	:
 41	: __imm(bpf_get_smp_processor_id)
 42	: __clobber_all);
 43}
 44
 45/* The logic for detecting and verifying bpf_fastcall pattern is the same for
 46 * any arch, however x86 differs from arm64 or riscv64 in a way
 47 * bpf_get_smp_processor_id is rewritten:
 48 * - on x86 it is done by verifier
 49 * - on arm64 and riscv64 it is done by jit
 50 *
 51 * Which leads to different xlated patterns for different archs:
 52 * - on x86 the call is expanded as 3 instructions
 53 * - on arm64 and riscv64 the call remains as is
 54 *   (but spills/fills are still removed)
 55 *
 56 * It is really desirable to check instruction indexes in the xlated
 57 * patterns, so add this canary test to check that function rewrite by
 58 * jit is correctly processed by bpf_fastcall logic, keep the rest of the
 59 * tests as x86.
 60 */
 61SEC("raw_tp")
 62__arch_arm64
 63__arch_riscv64
 64__xlated("0: r1 = 1")
 65__xlated("1: call bpf_get_smp_processor_id")
 66__xlated("2: exit")
 67__success
 68__naked void canary_arm64_riscv64(void)
 69{
 70	asm volatile (
 71	"r1 = 1;"
 72	"*(u64 *)(r10 - 16) = r1;"
 73	"call %[bpf_get_smp_processor_id];"
 74	"r1 = *(u64 *)(r10 - 16);"
 75	"exit;"
 76	:
 77	: __imm(bpf_get_smp_processor_id)
 78	: __clobber_all);
 79}
 80
 81SEC("raw_tp")
 82__arch_x86_64
 83__xlated("1: r0 = &(void __percpu *)(r0)")
 84__xlated("...")
 85__xlated("3: exit")
 86__success
 87__naked void canary_zero_spills(void)
 88{
 89	asm volatile (
 90	"call %[bpf_get_smp_processor_id];"
 91	"exit;"
 92	:
 93	: __imm(bpf_get_smp_processor_id)
 94	: __clobber_all);
 95}
 96
 97SEC("raw_tp")
 98__arch_x86_64
 99__log_level(4) __msg("stack depth 16")
100__xlated("1: *(u64 *)(r10 -16) = r1")
101__xlated("...")
102__xlated("3: r0 = &(void __percpu *)(r0)")
103__xlated("...")
104__xlated("5: r2 = *(u64 *)(r10 -16)")
105__success
106__naked void wrong_reg_in_pattern1(void)
107{
108	asm volatile (
109	"r1 = 1;"
110	"*(u64 *)(r10 - 16) = r1;"
111	"call %[bpf_get_smp_processor_id];"
112	"r2 = *(u64 *)(r10 - 16);"
113	"exit;"
114	:
115	: __imm(bpf_get_smp_processor_id)
116	: __clobber_all);
117}
118
119SEC("raw_tp")
120__arch_x86_64
121__xlated("1: *(u64 *)(r10 -16) = r6")
122__xlated("...")
123__xlated("3: r0 = &(void __percpu *)(r0)")
124__xlated("...")
125__xlated("5: r6 = *(u64 *)(r10 -16)")
126__success
127__naked void wrong_reg_in_pattern2(void)
128{
129	asm volatile (
130	"r6 = 1;"
131	"*(u64 *)(r10 - 16) = r6;"
132	"call %[bpf_get_smp_processor_id];"
133	"r6 = *(u64 *)(r10 - 16);"
134	"exit;"
135	:
136	: __imm(bpf_get_smp_processor_id)
137	: __clobber_all);
138}
139
140SEC("raw_tp")
141__arch_x86_64
142__xlated("1: *(u64 *)(r10 -16) = r0")
143__xlated("...")
144__xlated("3: r0 = &(void __percpu *)(r0)")
145__xlated("...")
146__xlated("5: r0 = *(u64 *)(r10 -16)")
147__success
148__naked void wrong_reg_in_pattern3(void)
149{
150	asm volatile (
151	"r0 = 1;"
152	"*(u64 *)(r10 - 16) = r0;"
153	"call %[bpf_get_smp_processor_id];"
154	"r0 = *(u64 *)(r10 - 16);"
155	"exit;"
156	:
157	: __imm(bpf_get_smp_processor_id)
158	: __clobber_all);
159}
160
161SEC("raw_tp")
162__arch_x86_64
163__xlated("2: *(u64 *)(r2 -16) = r1")
164__xlated("...")
165__xlated("4: r0 = &(void __percpu *)(r0)")
166__xlated("...")
167__xlated("6: r1 = *(u64 *)(r10 -16)")
168__success
169__naked void wrong_base_in_pattern(void)
170{
171	asm volatile (
172	"r1 = 1;"
173	"r2 = r10;"
174	"*(u64 *)(r2 - 16) = r1;"
175	"call %[bpf_get_smp_processor_id];"
176	"r1 = *(u64 *)(r10 - 16);"
177	"exit;"
178	:
179	: __imm(bpf_get_smp_processor_id)
180	: __clobber_all);
181}
182
183SEC("raw_tp")
184__arch_x86_64
185__xlated("1: *(u64 *)(r10 -16) = r1")
186__xlated("...")
187__xlated("3: r0 = &(void __percpu *)(r0)")
188__xlated("...")
189__xlated("5: r2 = 1")
190__success
191__naked void wrong_insn_in_pattern(void)
192{
193	asm volatile (
194	"r1 = 1;"
195	"*(u64 *)(r10 - 16) = r1;"
196	"call %[bpf_get_smp_processor_id];"
197	"r2 = 1;"
198	"r1 = *(u64 *)(r10 - 16);"
199	"exit;"
200	:
201	: __imm(bpf_get_smp_processor_id)
202	: __clobber_all);
203}
204
205SEC("raw_tp")
206__arch_x86_64
207__xlated("2: *(u64 *)(r10 -16) = r1")
208__xlated("...")
209__xlated("4: r0 = &(void __percpu *)(r0)")
210__xlated("...")
211__xlated("6: r1 = *(u64 *)(r10 -8)")
212__success
213__naked void wrong_off_in_pattern1(void)
214{
215	asm volatile (
216	"r1 = 1;"
217	"*(u64 *)(r10 - 8) = r1;"
218	"*(u64 *)(r10 - 16) = r1;"
219	"call %[bpf_get_smp_processor_id];"
220	"r1 = *(u64 *)(r10 - 8);"
221	"exit;"
222	:
223	: __imm(bpf_get_smp_processor_id)
224	: __clobber_all);
225}
226
227SEC("raw_tp")
228__arch_x86_64
229__xlated("1: *(u32 *)(r10 -4) = r1")
230__xlated("...")
231__xlated("3: r0 = &(void __percpu *)(r0)")
232__xlated("...")
233__xlated("5: r1 = *(u32 *)(r10 -4)")
234__success
235__naked void wrong_off_in_pattern2(void)
236{
237	asm volatile (
238	"r1 = 1;"
239	"*(u32 *)(r10 - 4) = r1;"
240	"call %[bpf_get_smp_processor_id];"
241	"r1 = *(u32 *)(r10 - 4);"
242	"exit;"
243	:
244	: __imm(bpf_get_smp_processor_id)
245	: __clobber_all);
246}
247
248SEC("raw_tp")
249__arch_x86_64
250__xlated("1: *(u32 *)(r10 -16) = r1")
251__xlated("...")
252__xlated("3: r0 = &(void __percpu *)(r0)")
253__xlated("...")
254__xlated("5: r1 = *(u32 *)(r10 -16)")
255__success
256__naked void wrong_size_in_pattern(void)
257{
258	asm volatile (
259	"r1 = 1;"
260	"*(u32 *)(r10 - 16) = r1;"
261	"call %[bpf_get_smp_processor_id];"
262	"r1 = *(u32 *)(r10 - 16);"
263	"exit;"
264	:
265	: __imm(bpf_get_smp_processor_id)
266	: __clobber_all);
267}
268
269SEC("raw_tp")
270__arch_x86_64
271__xlated("2: *(u32 *)(r10 -8) = r1")
272__xlated("...")
273__xlated("4: r0 = &(void __percpu *)(r0)")
274__xlated("...")
275__xlated("6: r1 = *(u32 *)(r10 -8)")
276__success
277__naked void partial_pattern(void)
278{
279	asm volatile (
280	"r1 = 1;"
281	"r2 = 2;"
282	"*(u32 *)(r10 - 8) = r1;"
283	"*(u64 *)(r10 - 16) = r2;"
284	"call %[bpf_get_smp_processor_id];"
285	"r2 = *(u64 *)(r10 - 16);"
286	"r1 = *(u32 *)(r10 - 8);"
287	"exit;"
288	:
289	: __imm(bpf_get_smp_processor_id)
290	: __clobber_all);
291}
292
293SEC("raw_tp")
294__arch_x86_64
295__xlated("0: r1 = 1")
296__xlated("1: r2 = 2")
297/* not patched, spills for -8, -16 not removed */
298__xlated("2: *(u64 *)(r10 -8) = r1")
299__xlated("3: *(u64 *)(r10 -16) = r2")
300__xlated("...")
301__xlated("5: r0 = &(void __percpu *)(r0)")
302__xlated("...")
303__xlated("7: r2 = *(u64 *)(r10 -16)")
304__xlated("8: r1 = *(u64 *)(r10 -8)")
305/* patched, spills for -24, -32 removed */
306__xlated("...")
307__xlated("10: r0 = &(void __percpu *)(r0)")
308__xlated("...")
309__xlated("12: exit")
310__success
311__naked void min_stack_offset(void)
312{
313	asm volatile (
314	"r1 = 1;"
315	"r2 = 2;"
316	/* this call won't be patched */
317	"*(u64 *)(r10 - 8) = r1;"
318	"*(u64 *)(r10 - 16) = r2;"
319	"call %[bpf_get_smp_processor_id];"
320	"r2 = *(u64 *)(r10 - 16);"
321	"r1 = *(u64 *)(r10 - 8);"
322	/* this call would be patched */
323	"*(u64 *)(r10 - 24) = r1;"
324	"*(u64 *)(r10 - 32) = r2;"
325	"call %[bpf_get_smp_processor_id];"
326	"r2 = *(u64 *)(r10 - 32);"
327	"r1 = *(u64 *)(r10 - 24);"
328	"exit;"
329	:
330	: __imm(bpf_get_smp_processor_id)
331	: __clobber_all);
332}
333
334SEC("raw_tp")
335__arch_x86_64
336__xlated("1: *(u64 *)(r10 -8) = r1")
337__xlated("...")
338__xlated("3: r0 = &(void __percpu *)(r0)")
339__xlated("...")
340__xlated("5: r1 = *(u64 *)(r10 -8)")
341__success
342__naked void bad_fixed_read(void)
343{
344	asm volatile (
345	"r1 = 1;"
346	"*(u64 *)(r10 - 8) = r1;"
347	"call %[bpf_get_smp_processor_id];"
348	"r1 = *(u64 *)(r10 - 8);"
349	"r1 = r10;"
350	"r1 += -8;"
351	"r1 = *(u64 *)(r1 - 0);"
352	"exit;"
353	:
354	: __imm(bpf_get_smp_processor_id)
355	: __clobber_all);
356}
357
358SEC("raw_tp")
359__arch_x86_64
360__xlated("1: *(u64 *)(r10 -8) = r1")
361__xlated("...")
362__xlated("3: r0 = &(void __percpu *)(r0)")
363__xlated("...")
364__xlated("5: r1 = *(u64 *)(r10 -8)")
365__success
366__naked void bad_fixed_write(void)
367{
368	asm volatile (
369	"r1 = 1;"
370	"*(u64 *)(r10 - 8) = r1;"
371	"call %[bpf_get_smp_processor_id];"
372	"r1 = *(u64 *)(r10 - 8);"
373	"r1 = r10;"
374	"r1 += -8;"
375	"*(u64 *)(r1 - 0) = r1;"
376	"exit;"
377	:
378	: __imm(bpf_get_smp_processor_id)
379	: __clobber_all);
380}
381
382SEC("raw_tp")
383__arch_x86_64
384__xlated("6: *(u64 *)(r10 -16) = r1")
385__xlated("...")
386__xlated("8: r0 = &(void __percpu *)(r0)")
387__xlated("...")
388__xlated("10: r1 = *(u64 *)(r10 -16)")
389__success
390__naked void bad_varying_read(void)
391{
392	asm volatile (
393	"r6 = *(u64 *)(r1 + 0);" /* random scalar value */
394	"r6 &= 0x7;"		 /* r6 range [0..7] */
395	"r6 += 0x2;"		 /* r6 range [2..9] */
396	"r7 = 0;"
397	"r7 -= r6;"		 /* r7 range [-9..-2] */
398	"r1 = 1;"
399	"*(u64 *)(r10 - 16) = r1;"
400	"call %[bpf_get_smp_processor_id];"
401	"r1 = *(u64 *)(r10 - 16);"
402	"r1 = r10;"
403	"r1 += r7;"
404	"r1 = *(u8 *)(r1 - 0);" /* touches slot [-16..-9] where spills are stored */
405	"exit;"
406	:
407	: __imm(bpf_get_smp_processor_id)
408	: __clobber_all);
409}
410
411SEC("raw_tp")
412__arch_x86_64
413__xlated("6: *(u64 *)(r10 -16) = r1")
414__xlated("...")
415__xlated("8: r0 = &(void __percpu *)(r0)")
416__xlated("...")
417__xlated("10: r1 = *(u64 *)(r10 -16)")
418__success
419__naked void bad_varying_write(void)
420{
421	asm volatile (
422	"r6 = *(u64 *)(r1 + 0);" /* random scalar value */
423	"r6 &= 0x7;"		 /* r6 range [0..7] */
424	"r6 += 0x2;"		 /* r6 range [2..9] */
425	"r7 = 0;"
426	"r7 -= r6;"		 /* r7 range [-9..-2] */
427	"r1 = 1;"
428	"*(u64 *)(r10 - 16) = r1;"
429	"call %[bpf_get_smp_processor_id];"
430	"r1 = *(u64 *)(r10 - 16);"
431	"r1 = r10;"
432	"r1 += r7;"
433	"*(u8 *)(r1 - 0) = r7;" /* touches slot [-16..-9] where spills are stored */
434	"exit;"
435	:
436	: __imm(bpf_get_smp_processor_id)
437	: __clobber_all);
438}
439
440SEC("raw_tp")
441__arch_x86_64
442__xlated("1: *(u64 *)(r10 -8) = r1")
443__xlated("...")
444__xlated("3: r0 = &(void __percpu *)(r0)")
445__xlated("...")
446__xlated("5: r1 = *(u64 *)(r10 -8)")
447__success
448__naked void bad_write_in_subprog(void)
449{
450	asm volatile (
451	"r1 = 1;"
452	"*(u64 *)(r10 - 8) = r1;"
453	"call %[bpf_get_smp_processor_id];"
454	"r1 = *(u64 *)(r10 - 8);"
455	"r1 = r10;"
456	"r1 += -8;"
457	"call bad_write_in_subprog_aux;"
458	"exit;"
459	:
460	: __imm(bpf_get_smp_processor_id)
461	: __clobber_all);
462}
463
464__used
465__naked static void bad_write_in_subprog_aux(void)
466{
467	asm volatile (
468	"r0 = 1;"
469	"*(u64 *)(r1 - 0) = r0;"	/* invalidates bpf_fastcall contract for caller: */
470	"exit;"				/* caller stack at -8 used outside of the pattern */
471	::: __clobber_all);
472}
473
474SEC("raw_tp")
475__arch_x86_64
476__xlated("1: *(u64 *)(r10 -8) = r1")
477__xlated("...")
478__xlated("3: r0 = &(void __percpu *)(r0)")
479__xlated("...")
480__xlated("5: r1 = *(u64 *)(r10 -8)")
481__success
482__naked void bad_helper_write(void)
483{
484	asm volatile (
485	"r1 = 1;"
486	/* bpf_fastcall pattern with stack offset -8 */
487	"*(u64 *)(r10 - 8) = r1;"
488	"call %[bpf_get_smp_processor_id];"
489	"r1 = *(u64 *)(r10 - 8);"
490	"r1 = r10;"
491	"r1 += -8;"
492	"r2 = 1;"
493	"r3 = 42;"
494	/* read dst is fp[-8], thus bpf_fastcall rewrite not applied */
495	"call %[bpf_probe_read_kernel];"
496	"exit;"
497	:
498	: __imm(bpf_get_smp_processor_id),
499	  __imm(bpf_probe_read_kernel)
500	: __clobber_all);
501}
502
503SEC("raw_tp")
504__arch_x86_64
505/* main, not patched */
506__xlated("1: *(u64 *)(r10 -8) = r1")
507__xlated("...")
508__xlated("3: r0 = &(void __percpu *)(r0)")
509__xlated("...")
510__xlated("5: r1 = *(u64 *)(r10 -8)")
511__xlated("...")
512__xlated("9: call pc+1")
513__xlated("...")
514__xlated("10: exit")
515/* subprogram, patched */
516__xlated("11: r1 = 1")
517__xlated("...")
518__xlated("13: r0 = &(void __percpu *)(r0)")
519__xlated("...")
520__xlated("15: exit")
521__success
522__naked void invalidate_one_subprog(void)
523{
524	asm volatile (
525	"r1 = 1;"
526	"*(u64 *)(r10 - 8) = r1;"
527	"call %[bpf_get_smp_processor_id];"
528	"r1 = *(u64 *)(r10 - 8);"
529	"r1 = r10;"
530	"r1 += -8;"
531	"r1 = *(u64 *)(r1 - 0);"
532	"call invalidate_one_subprog_aux;"
533	"exit;"
534	:
535	: __imm(bpf_get_smp_processor_id)
536	: __clobber_all);
537}
538
539__used
540__naked static void invalidate_one_subprog_aux(void)
541{
542	asm volatile (
543	"r1 = 1;"
544	"*(u64 *)(r10 - 8) = r1;"
545	"call %[bpf_get_smp_processor_id];"
546	"r1 = *(u64 *)(r10 - 8);"
547	"exit;"
548	:
549	: __imm(bpf_get_smp_processor_id)
550	: __clobber_all);
551}
552
553SEC("raw_tp")
554__arch_x86_64
555/* main */
556__xlated("0: r1 = 1")
557__xlated("...")
558__xlated("2: r0 = &(void __percpu *)(r0)")
559__xlated("...")
560__xlated("4: call pc+1")
561__xlated("5: exit")
562/* subprogram */
563__xlated("6: r1 = 1")
564__xlated("...")
565__xlated("8: r0 = &(void __percpu *)(r0)")
566__xlated("...")
567__xlated("10: *(u64 *)(r10 -16) = r1")
568__xlated("11: exit")
569__success
570__naked void subprogs_use_independent_offsets(void)
571{
572	asm volatile (
573	"r1 = 1;"
574	"*(u64 *)(r10 - 16) = r1;"
575	"call %[bpf_get_smp_processor_id];"
576	"r1 = *(u64 *)(r10 - 16);"
577	"call subprogs_use_independent_offsets_aux;"
578	"exit;"
579	:
580	: __imm(bpf_get_smp_processor_id)
581	: __clobber_all);
582}
583
584__used
585__naked static void subprogs_use_independent_offsets_aux(void)
586{
587	asm volatile (
588	"r1 = 1;"
589	"*(u64 *)(r10 - 24) = r1;"
590	"call %[bpf_get_smp_processor_id];"
591	"r1 = *(u64 *)(r10 - 24);"
592	"*(u64 *)(r10 - 16) = r1;"
593	"exit;"
594	:
595	: __imm(bpf_get_smp_processor_id)
596	: __clobber_all);
597}
598
599SEC("raw_tp")
600__arch_x86_64
601__log_level(4) __msg("stack depth 8")
602__xlated("2: r0 = &(void __percpu *)(r0)")
603__success
604__naked void helper_call_does_not_prevent_bpf_fastcall(void)
605{
606	asm volatile (
607	"r1 = 1;"
608	"*(u64 *)(r10 - 8) = r1;"
609	"call %[bpf_get_smp_processor_id];"
610	"r1 = *(u64 *)(r10 - 8);"
611	"*(u64 *)(r10 - 8) = r1;"
612	"call %[bpf_get_prandom_u32];"
613	"r1 = *(u64 *)(r10 - 8);"
614	"exit;"
615	:
616	: __imm(bpf_get_smp_processor_id),
617	  __imm(bpf_get_prandom_u32)
618	: __clobber_all);
619}
620
621SEC("raw_tp")
622__arch_x86_64
623__log_level(4) __msg("stack depth 16")
624/* may_goto counter at -16 */
625__xlated("0: *(u64 *)(r10 -16) =")
626__xlated("1: r1 = 1")
627__xlated("...")
628__xlated("3: r0 = &(void __percpu *)(r0)")
629__xlated("...")
630/* may_goto expansion starts */
631__xlated("5: r11 = *(u64 *)(r10 -16)")
632__xlated("6: if r11 == 0x0 goto pc+3")
633__xlated("7: r11 -= 1")
634__xlated("8: *(u64 *)(r10 -16) = r11")
635/* may_goto expansion ends */
636__xlated("9: *(u64 *)(r10 -8) = r1")
637__xlated("10: exit")
638__success
639__naked void may_goto_interaction(void)
640{
641	asm volatile (
642	"r1 = 1;"
643	"*(u64 *)(r10 - 16) = r1;"
644	"call %[bpf_get_smp_processor_id];"
645	"r1 = *(u64 *)(r10 - 16);"
646	".8byte %[may_goto];"
647	/* just touch some stack at -8 */
648	"*(u64 *)(r10 - 8) = r1;"
649	"exit;"
650	:
651	: __imm(bpf_get_smp_processor_id),
652	  __imm_insn(may_goto, BPF_RAW_INSN(BPF_JMP | BPF_JCOND, 0, 0, +1 /* offset */, 0))
653	: __clobber_all);
654}
655
656__used
657__naked static void dummy_loop_callback(void)
658{
659	asm volatile (
660	"r0 = 0;"
661	"exit;"
662	::: __clobber_all);
663}
664
665SEC("raw_tp")
666__arch_x86_64
667__log_level(4) __msg("stack depth 32+0")
668__xlated("2: r1 = 1")
669__xlated("3: w0 =")
670__xlated("4: r0 = &(void __percpu *)(r0)")
671__xlated("5: r0 = *(u32 *)(r0 +0)")
672/* bpf_loop params setup */
673__xlated("6: r2 =")
674__xlated("7: r3 = 0")
675__xlated("8: r4 = 0")
676__xlated("...")
677/* ... part of the inlined bpf_loop */
678__xlated("12: *(u64 *)(r10 -32) = r6")
679__xlated("13: *(u64 *)(r10 -24) = r7")
680__xlated("14: *(u64 *)(r10 -16) = r8")
681__xlated("...")
682__xlated("21: call pc+8") /* dummy_loop_callback */
683/* ... last insns of the bpf_loop_interaction1 */
684__xlated("...")
685__xlated("28: r0 = 0")
686__xlated("29: exit")
687/* dummy_loop_callback */
688__xlated("30: r0 = 0")
689__xlated("31: exit")
690__success
691__naked int bpf_loop_interaction1(void)
692{
693	asm volatile (
694	"r1 = 1;"
695	/* bpf_fastcall stack region at -16, but could be removed */
696	"*(u64 *)(r10 - 16) = r1;"
697	"call %[bpf_get_smp_processor_id];"
698	"r1 = *(u64 *)(r10 - 16);"
699	"r2 = %[dummy_loop_callback];"
700	"r3 = 0;"
701	"r4 = 0;"
702	"call %[bpf_loop];"
703	"r0 = 0;"
704	"exit;"
705	:
706	: __imm_ptr(dummy_loop_callback),
707	  __imm(bpf_get_smp_processor_id),
708	  __imm(bpf_loop)
709	: __clobber_common
710	);
711}
712
713SEC("raw_tp")
714__arch_x86_64
715__log_level(4) __msg("stack depth 40+0")
716/* call bpf_get_smp_processor_id */
717__xlated("2: r1 = 42")
718__xlated("3: w0 =")
719__xlated("4: r0 = &(void __percpu *)(r0)")
720__xlated("5: r0 = *(u32 *)(r0 +0)")
721/* call bpf_get_prandom_u32 */
722__xlated("6: *(u64 *)(r10 -16) = r1")
723__xlated("7: call")
724__xlated("8: r1 = *(u64 *)(r10 -16)")
725__xlated("...")
726/* ... part of the inlined bpf_loop */
727__xlated("15: *(u64 *)(r10 -40) = r6")
728__xlated("16: *(u64 *)(r10 -32) = r7")
729__xlated("17: *(u64 *)(r10 -24) = r8")
730__success
731__naked int bpf_loop_interaction2(void)
732{
733	asm volatile (
734	"r1 = 42;"
735	/* bpf_fastcall stack region at -16, cannot be removed */
736	"*(u64 *)(r10 - 16) = r1;"
737	"call %[bpf_get_smp_processor_id];"
738	"r1 = *(u64 *)(r10 - 16);"
739	"*(u64 *)(r10 - 16) = r1;"
740	"call %[bpf_get_prandom_u32];"
741	"r1 = *(u64 *)(r10 - 16);"
742	"r2 = %[dummy_loop_callback];"
743	"r3 = 0;"
744	"r4 = 0;"
745	"call %[bpf_loop];"
746	"r0 = 0;"
747	"exit;"
748	:
749	: __imm_ptr(dummy_loop_callback),
750	  __imm(bpf_get_smp_processor_id),
751	  __imm(bpf_get_prandom_u32),
752	  __imm(bpf_loop)
753	: __clobber_common
754	);
755}
756
757SEC("raw_tp")
758__arch_x86_64
759__log_level(4)
760__msg("stack depth 512+0")
761/* just to print xlated version when debugging */
762__xlated("r0 = &(void __percpu *)(r0)")
763__success
764/* cumulative_stack_depth() stack usage is MAX_BPF_STACK,
765 * called subprogram uses an additional slot for bpf_fastcall spill/fill,
766 * since bpf_fastcall spill/fill could be removed the program still fits
767 * in MAX_BPF_STACK and should be accepted.
768 */
769__naked int cumulative_stack_depth(void)
770{
771	asm volatile(
772	"r1 = 42;"
773	"*(u64 *)(r10 - %[max_bpf_stack]) = r1;"
774	"call cumulative_stack_depth_subprog;"
775	"exit;"
776	:
777	: __imm_const(max_bpf_stack, MAX_BPF_STACK)
778	: __clobber_all
779	);
780}
781
782__used
783__naked static void cumulative_stack_depth_subprog(void)
784{
785	asm volatile (
786	"*(u64 *)(r10 - 8) = r1;"
787	"call %[bpf_get_smp_processor_id];"
788	"r1 = *(u64 *)(r10 - 8);"
789	"exit;"
790	:: __imm(bpf_get_smp_processor_id) : __clobber_all);
791}
792
793SEC("cgroup/getsockname_unix")
794__xlated("0: r2 = 1")
795/* bpf_cast_to_kern_ctx is replaced by a single assignment */
796__xlated("1: r0 = r1")
797__xlated("2: r0 = r2")
798__xlated("3: exit")
799__success
800__naked void kfunc_bpf_cast_to_kern_ctx(void)
801{
802	asm volatile (
803	"r2 = 1;"
804	"*(u64 *)(r10 - 32) = r2;"
805	"call %[bpf_cast_to_kern_ctx];"
806	"r2 = *(u64 *)(r10 - 32);"
807	"r0 = r2;"
808	"exit;"
809	:
810	: __imm(bpf_cast_to_kern_ctx)
811	: __clobber_all);
812}
813
814SEC("raw_tp")
815__xlated("3: r3 = 1")
816/* bpf_rdonly_cast is replaced by a single assignment */
817__xlated("4: r0 = r1")
818__xlated("5: r0 = r3")
819void kfunc_bpf_rdonly_cast(void)
820{
821	asm volatile (
822	"r2 = %[btf_id];"
823	"r3 = 1;"
824	"*(u64 *)(r10 - 32) = r3;"
825	"call %[bpf_rdonly_cast];"
826	"r3 = *(u64 *)(r10 - 32);"
827	"r0 = r3;"
828	:
829	: __imm(bpf_rdonly_cast),
830	 [btf_id]"r"(bpf_core_type_id_kernel(union bpf_attr))
831	: __clobber_common);
832}
833
834/* BTF FUNC records are not generated for kfuncs referenced
835 * from inline assembly. These records are necessary for
836 * libbpf to link the program. The function below is a hack
837 * to ensure that BTF FUNC records are generated.
838 */
839void kfunc_root(void)
840{
841	bpf_cast_to_kern_ctx(0);
842	bpf_rdonly_cast(0, 0);
843}
844
845char _license[] SEC("license") = "GPL";