Re: [BUG soft lockup] Re: [PATCH bpf-next v3] bpf: Propagate stack bounds to registers in atomics w/ BPF_FETCH

[Jiri Olsa]

On Tue, Jun 29, 2021 at 06:41:24PM +0200, Jiri Olsa wrote:
> On Tue, Jun 29, 2021 at 06:25:33PM +0200, Brendan Jackman wrote:
> > On Tue, 29 Jun 2021 at 18:04, Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Jun 29, 2021 at 04:10:12PM +0200, Jiri Olsa wrote:
> > > > On Mon, Jun 28, 2021 at 11:21:42AM +0200, Brendan Jackman wrote:
> > > > > On Sun, 27 Jun 2021 at 17:34, Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Tue, Feb 02, 2021 at 01:50:02PM +0000, Brendan Jackman wrote:
> > [snip]
> > > > > Hmm, is the test prog from atomic_bounds.c getting JITed there (my
> > > > > dumb guess at what '0xc0000000119efb30 (unreliable)' means)? That
> > > > > shouldn't happen - should get 'eBPF filter atomic op code %02x (@%d)
> > > > > unsupported\n' in dmesg instead. I wonder if I missed something in
> > > > > commit 91c960b0056 (bpf: Rename BPF_XADD and prepare to encode other
> > >
> > > I see that for all the other atomics tests:
> > >
> > > [root@ibm-p9z-07-lp1 bpf]# ./test_verifier 21
> > > #21/p BPF_ATOMIC_AND without fetch FAIL
> > > Failed to load prog 'Unknown error 524'!
> > > verification time 32 usec
> > > stack depth 8
> > > processed 10 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 1
> > > Summary: 0 PASSED, 0 SKIPPED, 2 FAILED
> > 
> > Hm that's also not good - failure to JIT shouldn't mean failure to
> > load. Are there other test_verifier failures or is it just the atomics
> > ones?
> 
> I have CONFIG_BPF_JIT_ALWAYS_ON=y so I think that's fine
> 
> > 
> > > console:
> > >
> > >         [   51.850952] eBPF filter atomic op code db (@2) unsupported
> > >         [   51.851134] eBPF filter atomic op code db (@2) unsupported
> > >
> > >
> > > [root@ibm-p9z-07-lp1 bpf]# ./test_verifier 22
> > > #22/u BPF_ATOMIC_AND with fetch FAIL
> > > Failed to load prog 'Unknown error 524'!
> > > verification time 38 usec
> > > stack depth 8
> > > processed 14 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 1
> > > #22/p BPF_ATOMIC_AND with fetch FAIL
> > > Failed to load prog 'Unknown error 524'!
> > > verification time 26 usec
> > > stack depth 8
> > > processed 14 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 1
> > >
> > > console:
> > >         [  223.231420] eBPF filter atomic op code db (@3) unsupported
> > >         [  223.231596] eBPF filter atomic op code db (@3) unsupported
> > >
> > > ...
> > >
> > >
> > > but no such console output for:
> > >
> > > [root@ibm-p9z-07-lp1 bpf]# ./test_verifier 24
> > > #24/u BPF_ATOMIC bounds propagation, mem->reg OK
> > >
> > >
> > > > > atomics in .imm). Any idea if this test was ever passing on PowerPC?
> > > > >
> > > >
> > > > hum, I guess not.. will check
> > >
> > > nope, it locks up the same:
> > 
> > Do you mean it locks up at commit 91c960b0056 too?
> > 
> 
> I tried this one:
>   37086bfdc737 bpf: Propagate stack bounds to registers in atomics w/ BPF_FETCH
> 
> I will check also 91c960b0056, but I think it's the new test issue

for i91c960b0056 in HEAD I'm getting just 2 fails:

	#1097/p xadd/w check whether src/dst got mangled, 1 FAIL
	Failed to load prog 'Unknown error 524'!
	verification time 25 usec
	stack depth 8
	processed 12 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 0
	#1098/p xadd/w check whether src/dst got mangled, 2 FAIL
	Failed to load prog 'Unknown error 524'!
	verification time 30 usec
	stack depth 8
	processed 12 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 0

with console output:

	[  289.499341] eBPF filter atomic op code db (@4) unsupported
	[  289.499510] eBPF filter atomic op code c3 (@4) unsupported

no lock up

jirka