Re: [PATCH RESEND bpf-next v3 4/6] riscv, bpf: Add necessary Zbb instructions

[Pu Lehui <pulehui@huaweicloud.com>]

On 2024/1/31 1:34, Björn Töpel wrote:
> Pu Lehui <pulehui@huawei.com> writes:
> 
>> On 2024/1/30 14:18, Björn Töpel wrote:
>>> Daniel Borkmann <daniel@iogearbox.net> writes:
>>>
>>>> On 1/29/24 10:13 AM, Pu Lehui wrote:
>>>>> On 2024/1/28 1:16, Björn Töpel wrote:
>>>>>> Pu Lehui <pulehui@huaweicloud.com> writes:
>>>>>>
>>>>>>> From: Pu Lehui <pulehui@huawei.com>
>>>>>>>
>>>>>>> Add necessary Zbb instructions introduced by [0] to reduce code size and
>>>>>>> improve performance of RV64 JIT. Meanwhile, a runtime deteted helper is
>>>>>>> added to check whether the CPU supports Zbb instructions.
>>>>>>>
>>>>>>> Link: https://github.com/riscv/riscv-bitmanip/releases/download/1.0.0/bitmanip-1.0.0-38-g865e7a7.pdf [0]
>>>>>>> Signed-off-by: Pu Lehui <pulehui@huawei.com>
>>>>>>> ---
>>>>>>>     arch/riscv/net/bpf_jit.h | 32 ++++++++++++++++++++++++++++++++
>>>>>>>     1 file changed, 32 insertions(+)
>>>>>>>
>>>>>>> diff --git a/arch/riscv/net/bpf_jit.h b/arch/riscv/net/bpf_jit.h
>>>>>>> index e30501b46f8f..51f6d214086f 100644
>>>>>>> --- a/arch/riscv/net/bpf_jit.h
>>>>>>> +++ b/arch/riscv/net/bpf_jit.h
>>>>>>> @@ -18,6 +18,11 @@ static inline bool rvc_enabled(void)
>>>>>>>         return IS_ENABLED(CONFIG_RISCV_ISA_C);
>>>>>>>     }
>>>>>>> +static inline bool rvzbb_enabled(void)
>>>>>>> +{
>>>>>>> +    return IS_ENABLED(CONFIG_RISCV_ISA_ZBB) && riscv_has_extension_likely(RISCV_ISA_EXT_ZBB);
>>>>>>
>>>>>> Hmm, I'm thinking about the IS_ENABLED(CONFIG_RISCV_ISA_ZBB) semantics
>>>>>> for a kernel JIT compiler.
>>>>>>
>>>>>> IS_ENABLED(CONFIG_RISCV_ISA_ZBB) affects the kernel compiler flags.
>>>>>> Should it be enough to just have the run-time check? Should a kernel
>>>>>> built w/o Zbb be able to emit Zbb from the JIT?
>>>>>
>>>>> Not enough, because riscv_has_extension_likely(RISCV_ISA_EXT_ZBB) is
>>>>> a platform capability check, and the other one is a kernel image
>>>>> capability check. We can pass the check
>>>>> riscv_has_extension_likely(RISCV_ISA_EXT_ZBB) when
>>>>> CONFIG_RISCV_ISA_ZBB=n. And my local test prove it.
>>>
>>> What I'm trying to say (and drew as well in the other reply) is that
>>> "riscv_has_extension_likely(RISCV_ISA_EXT_ZBB) when
>>> CONFIG_RISCV_ISA_ZBB=n" should also make the JIT emit Zbb insns. The
>>> platform check should be sufficient.
>>
>> Ooh, this is really beyond my expectation. The test_progs can pass when
>> with only platform check and it can recognize the zbb instructions. Now
>> I know it. Sorry for misleading.🙁
>>
>> Curious if CONFIG_RISCV_ISA_ZBB is still necessary?
> 
> You don't need IS_ENABLED(CONFIG_RISCV_ISA_ZBB) for the JIT, but the
> kernel needs it.
> 
> Feel free to follow up with a patch to remove it.
> 

Maybe we can implement more extensions about bitmanip.

> 
> Cheers,
> Björn