On 08/11/2017 11:57 AM, Wangnan (F) wrote:
>
>
> On 2017/8/11 17:17, Thomas-Mich Richter wrote:
>> On 08/11/2017 07:19 AM, Wangnan (F) wrote:
[....]
> Your analysis is correct.
>
>>
>> Maybe the solution is to add an endianness convertion into the gen_read_mem()
>> function.
>
> That means alert the prologue. We can utilize the size field in BPF_LDX_MEM.
>
> Could you please try this patch:
>
> diff --git a/tools/perf/util/bpf-prologue.c b/tools/perf/util/bpf-prologue.c
> index 6cdbee1..aadbda4 100644
> --- a/tools/perf/util/bpf-prologue.c
> +++ b/tools/perf/util/bpf-prologue.c
> @@ -258,7 +258,7 @@ gen_prologue_slowpath(struct bpf_insn_pos *pos,
>
> /* Final pass: read to registers */
> for (i = 0; i < nargs; i++)
> - ins(BPF_LDX_MEM(BPF_DW, BPF_PROLOGUE_START_ARG_REG + i,
> + ins(BPF_LDX_MEM(BPF_W, BPF_PROLOGUE_START_ARG_REG + i,
> BPF_REG_FP, -BPF_REG_SIZE * (i + 1)), pos);
>
> ins(BPF_JMP_IMM(BPF_JA, BPF_REG_0, 0, JMP_TO_SUCCESS_CODE), pos);
>
>
> On your platform and check if the value of f_mode is correct?
>
> NOTE: 64 bit load becomes incorrect after applying this patch. I want to make
> sure
> the behavior of 32 bit BPF load instruction works correct on big endian
> machine.
>
> Thank you.
>
This patch solves the issue with the first parameter. Now I see the correct
value of f_mode member of struct-file.
As you said the other loads are wrong. Here is the output of the
trace:
perf-13438 [000] d..2 2102.392312: : f_mode 6001f offset:0 orig:0
perf-13438 [000] d..2 2102.392414: : f_mode 2001d offset:0 orig:0
perf-13438 [000] d..2 2102.392440: : f_mode 2001d offset:0 orig:0
perf-13438 [000] d..2 2102.392453: : f_mode 6001f offset:0 orig:0
PS: When I ran this test I undid the changes to
tests/bpf-test-script-prologue.c.
It was in its original state when this test was executed.
--
Thomas Richter, Dept 3303, IBM LTC Boeblingen Germany
--
Vorsitzende des Aufsichtsrats: Martina Koederitz
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB
243294
