libbpf-tools/softirqs: Fix logarithmic calculation issue

[kknjh]

Should use log2l.
When there is a function call, R1 unbounded memory access problem occurs, so inline is used.

[kknjh]

I encountered R1 unbounded memory access when changing log2 to log2l. The log is in softirq.log.
And there are other modifications that do not affect the code logic to avoid this problem, such as the following modifications

 if (slot >= MAX_SLOTS) {
            slot = MAX_SLOTS - 1;
            __sync_fetch_and_add(&hist->slots[slot], 1);
    }
    else
            __sync_fetch_and_add(&hist->slots[slot], 1);

The logs of both are as follows:

success:
    3: (67) r6 <<= 32                     ; R6_w=scalar(smax=0x7fffffff00000000,umax=0xffffffff00000000,smin32=0,smax32=umax32=0,var_off=(0x0; 0xffffffff00000000))
    4: (77) r6 >>= 32                     ; R6_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
    ; if (vec_nr >= NR_SOFTIRQS)
    5: (25) if r6 > 0x9 goto pc+60        ; R6_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=9,var_off=(0x0; 0xf))
    ......
    13: (18) r1 = 0xffffb2af418320a0      ; R1_w=map_value(map=softirqs.bss,ks=4,vs=960,off=160)
    15: (0f) r1 += r6                     ; R1_w=map_value(map=softirqs.bss,ks=4,vs=960,off=160,smin=smin32=0,smax=umax=smax32=umax32=720,var_off=(0x0; 0x3f0)) R6_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=720,var_off=(0x0; 0x3f0))
    
Failed:
    7: (bf) r1 = r6                       ; frame1: R1_w=scalar(id=1) R6_w=scalar(id=1)
    8: (67) r1 <<= 32                     ; frame1: R1_w=scalar(smax=0x7fffffff00000000,umax=0xffffffff00000000,smin32=0,smax32=umax32=0,var_off=(0x0; 0xffffffff00000000))
    9: (77) r1 >>= 32                     ; frame1: R1_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
    ; if (vec_nr >= NR_SOFTIRQS)
    10: (25) if r1 > 0x9 goto pc+89       ; frame1: R1_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=9,var_off=(0x0; 0xf))
    ......
    46: (18) r1 = 0xffffba0f8095a0a0      ; frame1: R1_w=map_value(map=softirqs.bss,ks=4,vs=960,off=160)
    80: (0f) r1 += r6                     ; frame1: R1_w=map_value(map=softirqs.bss,ks=4,vs=960,off=160,smin=0,smax=umax=0x4fffffffb0,smax32=0x7ffffff0,umax32=0xfffffff0,var_off=(0x0; 0x7ffffffff0)) R6=scalar(smin=0,smax=umax=0x4fffffffb0,smax32=0x7ffffff0,umax32=0xfffffff0,var_off=(0x0; 0x7ffffffff0))

r1: hists
r6: vec_nr
r1+r6: hists[vec_nr]
Normally, after r6 determines the maximum value smax=umax=smax32=umax32=9, r1+r6 also limits r1.
The reason for the failure is that r6 was passed to r1, causing r6 to not determine the maximum value.
Subsequently, r1+r6 was used, resulting in R1 unbounded memory access
From the logs, it appears that this issue occurred during the function call process, so I used inline instead

However, may I ask why this situation has occurred?
softirq.log

[yonghong-song]

The failed code is due to compiler optimizations.
For the failed code:

7: (bf) r1 = r6                       ; frame1: R1_w=scalar(id=1) R6_w=scalar(id=1)
    8: (67) r1 <<= 32                     ; frame1: R1_w=scalar(smax=0x7fffffff00000000,umax=0xffffffff00000000,smin32=0,smax32=umax32=0,var_off=(0x0; 0xffffffff00000000))
    9: (77) r1 >>= 32                     ; frame1: R1_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
    ; if (vec_nr >= NR_SOFTIRQS)
    10: (25) if r1 > 0x9 goto pc+89       ; frame1: R1_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=9,var_off=(0x0; 0xf))
    ......
    46: (18) r1 = 0xffffba0f8095a0a0      ; frame1: R1_w=map_value(map=softirqs.bss,ks=4,vs=960,off=160)
    80: (0f) r1 += r6                     ; frame1: R1_w=map_value(map=softirqs.bss,ks=4,vs=960,off=160,smin=0,smax=umax=0x4fffffffb0,smax32=0x7ffffff0,umax32=0xfffffff0,var_off=(0x0; 0x7ffffffff0)) R6=scalar(smin=0,smax=umax=0x4fffffffb0,smax32=0x7ffffff0,umax32=0xfffffff0,var_off=(0x0; 0x7ffffffff0))

R1 did some range refinement. But unfortunately, register R6 didn't inherit refined R1 so later it caused verification.

In Makefile, we have

  BPFCFLAGS_softirqs := $(BPFCFLAGS) -mcpu=v3

why you didn't use -mcpu=v3? Which compiler you are using?

[kknjh]

I'm so sorry. I happened to not update the patch 'libbpf tools/softirqs: Add - c/-- cpu option to filter CPU'. It's normal now. I removed the inline of this patch