Re: [PATCH] meson: Allow BPF code to be built with GCC

[Sean Young <sean@mess.org>]

On Tue, Aug 05, 2025 at 11:32:54PM +0100, Sam James wrote:
> Sam James <sam@gentoo.org> writes:
> 
> > Sean Young <sean@mess.org> writes:
> >
> >> On Mon, Aug 04, 2025 at 10:13:18AM +0100, Sean Young wrote:
> >>> On Sun, Aug 03, 2025 at 09:17:59AM +0100, James Le Cuirot wrote:
> >>> > Disclaimer: I haven't actually tested the result of the GCC build.
> >>> 
> >>> Let me test this.
> >>
> >> It doesn't work:
> >>
> >> # ir-keytable -p ./imon_rsc.o 
> >> Protocols changed to 
> >> symbol  has unknown section 6
> >> bpf_load_program() err=Permission denied
> >> 0: R1=ctx() R10=fp0
> >> 0: (bf) r6 = r1                       ; R1=ctx() R6_w=ctx()
> >> 1: (62) *(u32 *)(r10 -4) = 0          ; R10=fp0 fp-8=0000????
> >> 2: (bf) r2 = r10                      ; R2_w=fp0 R10=fp0
> >> 3: (18) r1 = 0xffff89ed1318f800       ; R1_w=map_ptr(ks=4,vs=16)
> >> 5: (07) r2 += -4                      ; R2_w=fp-4
> >> 6: (85) call bpf_map_lookup_elem#1    ; R0_w=map_value(ks=4,vs=16)
> >> 7: (bf) r4 = r0                       ; R0_w=map_value(ks=4,vs=16) R4_w=map_value(ks=4,vs=16)
> >> 8: (15) if r0 == 0x0 goto pc+7        ; R0_w=map_value(ks=4,vs=16)
> >> 9: (61) r0 = *(u32 *)(r6 +0)          ; R0_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R6_w=ctx()
> >> 10: (bf) r1 = r0                      ; R0_w=scalar(id=1,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R1_w=scalar(id=1,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
> >> 11: (bf) r2 = r0                      ; R0_w=scalar(id=1,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R2_w=scalar(id=1,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
> >> 12: (54) w1 &= -16777216              ; R1_w=scalar(smin=0,smax=umax=umax32=0xff000000,smax32=0x7f000000,var_off=(0x0; 0xff000000))
> >> 13: (54) w2 &= -33554432              ; R2=scalar(smin=0,smax=umax=umax32=0xfe000000,smax32=0x7e000000,var_off=(0x0; 0xfe000000))
> >> 14: (16) if w2 == 0x0 goto pc+3       ; R2=scalar(smin=umin=umin32=1,smax=umax=umax32=0xfe000000,smax32=0x7e000000,var_off=(0x0; 0xfe000000))
> >> 15: (16) if w1 == 0x3000000 goto pc+31 47:
> >> R0=scalar(id=1,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
> >> R1=0x3000000
> >> R2=scalar(smin=umin=umin32=1,smax=umax=umax32=0xfe000000,smax32=0x7e000000,var_off=(0x0;
> >> 0xfe000000)) R4=map_value(ks=4,vs=16) R6=ctx() R10=fp0 fp-8=0000????
> >> 47: (61) r5 = *(u32 *)(r4 +8)         ; R4=map_value(ks=4,vs=16) R5_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
> >> 48: (16) if w5 == 0x1 goto pc+7 56:
> >> R0=scalar(id=1,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
> >> R1=0x3000000
> >> R2=scalar(smin=umin=umin32=1,smax=umax=umax32=0xfe000000,smax32=0x7e000000,var_off=(0x0;
> >> 0xfe000000)) R4=map_value(ks=4,vs=16) R5_w=1 R6=ctx() R10=fp0
> >> fp-8=0000????
> >> 56: (57) r0 &= 16777215               ; R0_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=0xffffff,var_off=(0x0; 0xffffff))
> >> 57: (b6) if w0 <= 0x960 goto pc+22    ; R0_w=scalar(smin=umin=smin32=umin32=2401,smax=umax=smax32=umax32=0xffffff,var_off=(0x0; 0xffffff))
> >> 58: (61) r1 = *(u32 *)(r4 +12)        ; R1_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R4=map_value(ks=4,vs=16)
> >> 59: (04) w1 += -4                     ; R1=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
> >> 60: (26) if w1 > 0x1 goto pc-11       ; R1=scalar(smin=smin32=0,smax=umax=smax32=umax32=1,var_off=(0x0; 0x1))
> >> 61: (79) r3 = *(u64 *)(r4 +0)         ; R3_w=scalar() R4=map_value(ks=4,vs=16)
> >> 62: (57) r3 &= 15                     ; R3_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=15,var_off=(0x0; 0xf))
> >> 63: (07) r3 += -1                     ; R3_w=scalar(smin=smin32=-1,smax=smax32=14)
> >> 64: (25) if r3 > 0xe goto pc+46       ; R3_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=14,var_off=(0x0; 0xf))
> >> 65: (67) r3 <<= 2                     ; R3_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=56,var_off=(0x0; 0x3c))
> >> 66: (18) r2 = 0x3c                    ; R2_w=60
> >> 68: (18) r9 = 0x0                     ; R9_w=0
> >> 70: (0f) r2 += r3 ;
> >> R2_w=scalar(smin=umin=smin32=umin32=60,smax=umax=smax32=umax32=116,var_off=(0x0;
> >> 0x7c))
> >> R3_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=56,var_off=(0x0;
> >> 0x3c))
> >> 71: (0f) r9 += r3 ;
> >> R3_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=56,var_off=(0x0;
> >> 0x3c))
> >> R9_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=56,var_off=(0x0;
> >> 0x3c))
> >> 72: (81) r2 = *(s32 *)(r2 +0)
> >> R2 invalid mem access 'scalar'
> >> processed 40 insns (limit 1000000) max_states_per_insn 0 total_states 3 peak_states 3 mark_read 1
> >
> > I'll have a look. Thanks.
> 
> OK, I can't reproduce it yet because I don't have the hardware (using
> loopback doesn't seem to be sufficient, even with Clang).
> 
> Could you get the full bpf-* gcc invocation with `ninja --verbose`, and
> then append -save-temps to it, and attach the .i it makes?

Works fine with rc-loopback.

The problem is that gcc compiles this down to a lookup table which is then
stored in the .rodata section.

https://git.linuxtv.org/v4l-utils.git/tree/utils/keytable/bpf_protocols/imon_rsc.c#n97

Then ir-keytable fails to load and relocate the rodata section. Note the 
error:

symbol  has unknown section 6

section 6 is the rodata section.

ir-keytable simply has no handling for .rodata right now. I am not sure how
bpf handles .rodata (if at all), needs more investigation.
 

Sean