scx_lavd: fix/work around a verifier error

[multics69]

Without this, the BPF verifier spits the following errors with some version of vmlinux.h. So added +1 to work around the problem.

; bpf_for(j, 0, 64) { @ main.bpf.c:1926
509: (bf) r1 = r8                     ; R1_w=fp-32 R8_w=fp-32 refs=66,2035
510: (b4) w2 = 0                      ; R2_w=0 refs=66,2035
511: (b4) w3 = 64                     ; R3_w=64 refs=66,2035
512: (85) call bpf_iter_num_new#104189        ; R0=scalar() fp-32=iter_num(ref_id=2048,state=active,depth=0) refs=66,2035,2048
513: (bf) r1 = r8                     ; R1=fp-32 R8=fp-32 refs=66,2035,2048
514: (85) call bpf_iter_num_next#104191 515: R0_w=rdonly_mem(id=2049,ref_obj_id=2048,sz=4) R6=scalar(id=2047,smin=smin32=0,smax=umax=smax32=umax32=7,var_off=(0x0; 0x7)) R7=scalar() R8=fp-32 R9=map_value(map=bpf_bpf.bss,ks=4,vs=4584,off=384,smin=smin32=0,smax=umax=smax32=umax32=3968,var_off=(0x0; 0xf80)) R10=fp0 fp-16=iter_num(ref_id=66,state=active,depth=1) fp-24=iter_num(ref_id=2035,state=active,depth=1) fp-32=iter_num(ref_id=2048,state=active,depth=1) fp-80=scalar(id=1) fp-88=map_value(map=.data.LAVD,ks=4,vs=1320,off=40,smin=smin32=0,smax=umax=smax32=umax32=1240,var_off=(0x0; 0x7f8)) fp-96=????0 fp-112=rcu_ptr_bpf_cpumask() fp-120=rcu_ptr_bpf_cpumask() fp-128=rcu_ptr_bpf_cpumask() fp-136=rcu_ptr_bpf_cpumask() refs=66,2035,2048
; bpf_for(j, 0, 64) { @ main.bpf.c:1926
515: (15) if r0 == 0x0 goto pc+49     ; R0_w=rdonly_mem(id=2049,ref_obj_id=2048,sz=4) refs=66,2035,2048
516: (64) w6 <<= 6                    ; R6=scalar(smin=smin32=0,smax=umax=smax32=umax32=448,var_off=(0x0; 0x1c0)) refs=66,2035,2048
517: (61) r8 = *(u32 *)(r0 +0)        ; R0=rdonly_mem(id=2049,ref_obj_id=2048,sz=4) R8_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) refs=66,2035,2048
518: (26) if w8 > 0x3f goto pc+46     ; R8_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=63,var_off=(0x0; 0x3f)) refs=66,2035,2048
; if (cpumask & 0x1LLU << j) { @ main.bpf.c:1927
519: (bf) r1 = r7                     ; R1_w=scalar(id=2053) R7=scalar(id=2053) refs=66,2035,2048
520: (7f) r1 >>= r8                   ; R1_w=scalar() R8_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=63,var_off=(0x0; 0x3f)) refs=66,2035,2048
521: (57) r1 &= 1                     ; R1_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=1,var_off=(0x0; 0x1)) refs=66,2035,2048
522: (15) if r1 == 0x0 goto pc+38     ; R1_w=1 refs=66,2035,2048
; cpu = (i * 64) + j; @ main.bpf.c:1928
523: (4c) w8 |= w6                    ; R6=scalar(smin=smin32=0,smax=umax=smax32=umax32=448,var_off=(0x0; 0x1c0)) R8_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=511,var_off=(0x0; 0x1ff)) refs=66,2035,2048
; bpf_cpumask_set_cpu(cpu, cd_cpumask); @ main.bpf.c:1929
524: (bc) w1 = w8                     ; R1_w=scalar(id=2054,smin=smin32=0,smax=umax=smax32=umax32=511,var_off=(0x0; 0x1ff)) R8_w=scalar(id=2054,smin=smin32=0,smax=umax=smax32=umax32=511,var_off=(0x0; 0x1ff)) refs=66,2035,2048
525: (79) r2 = *(u64 *)(r10 -88)      ; R2_w=map_value(map=.data.LAVD,ks=4,vs=1320,off=40,smin=smin32=0,smax=umax=smax32=umax32=1240,var_off=(0x0; 0x7f8)) R10=fp0 fp-88=map_value(map=.data.LAVD,ks=4,vs=1320,off=40,smin=smin32=0,smax=umax=smax32=umax32=1240,var_off=(0x0; 0x7f8)) refs=66,2035,2048
526: (85) call bpf_cpumask_set_cpu#93595
invalid access to map value, value_size=1320 off=1280 size=48
R2 max value is outside of the allowed memory range
processed 24200 insns (limit 1000000) max_states_per_insn 19 total_states 961 peak_states 789 mark_read 44

[hodgesds]

I guess alternatively you could add a bounds check here on cpuc_prev->cpdom_id and that would maybe be the same.

[multics69]

Thanks @hodgesds for the suggestion. Actually this is where the verifier complains. I failed to find where to place more bound checks.

Weirdly, everything is okay with the vmlinux header for 6.12-rc0.

[hodgesds]

Thanks @hodgesds for the suggestion. Actually this is where the verifier complains. I failed to find where to place more bound checks.

Good catch, it might be easier to do the bounds checks in the long run than having maybe off by one errors get introduced into the code.

[hodgesds]

Yeah, looking at this again I see the verifier issue.