On Mon, Oct 19, 2020 at 8:43 PM sheng chen <eason.sheng.chen@xxxxxxxxx> wrote: > > Hi Andrii, > [...] > > I'm using vendor qcom's kernel 4.19, and want to use bpf_probe_read_user(available on linux 5.5) on linux 4.19, so I need to apply the bpf_probe_read_user > function patches from upstream linux 5.5 to my local kernel 4.19, is that a suitable solution? On older kernels that don't yet support bpf_probe_read_user() you can just use bpf_probe_read() instead and get the same results. Again, I don't know bpftrace specifics and whether you can do that easily, but you don't really have to backport any kernel changes just to read user-space memory from BPF program. > > Thanks > Eason > > > > > Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> 于2020年10月20日周二 上午9:58写道： >> >> This is the third identical email you've sent, please don't spam the >> mailing list. Sometimes it takes a bit of time on the mailing list to >> get an answer. Re-sending your email will just annoy people and won't >> help you get an answer. >> >> As an advice for the future, please try to formulate your problem >> clearly, before asking a seemingly-random set of questions. See below, >> I tried to answer your questions as best as I could. >> >> >> On Mon, Oct 19, 2020 at 6:45 PM sheng chen <eason.sheng.chen@xxxxxxxxx> wrote: >> > >> > Hi Andrii, >> > >> > I'm developing bpftrace tools for Android aarch64 devices to analyze performance, mostly using uprobe/uretprobe and kprobe/kretprobe. >> > >> > I'm using the project https://github.com/facebookexperimental/ExtendedAndroidTools for build bpftrace cmd tool. libbpf still not included. >> > >> > >> > First question: >> > >> > Currently there is an issue(https://github.com/iovisor/bpftrace/issues/1503) block me to correctly access the pointer address of the uprobe parameters. >> > >> > Seems like this require bpf_probe_read_user on linux 5.5(as mension in https://github.com/iovisor/bcc/blob/master/docs/kernel-versions.md), if I use the older kernel, I need to apply the patch about the function bpf_probe_read_user, is there any reference code I need to apply as well? >> >> I don't understand what reference code you mean. And given this is a >> bpftrace question, it's probably best to route it to bpftrace Github >> repo? This mailing list is discussing kernel BPF subsystem and libbpf, >> for the most part. >> >> > >> > like the following parts: >> > >> > linux/include/linux/bpf.h >> > linux/include/uapi/bpf.h >> > linux/include/linux/filter.h >> > linux/include/uapi/filter.h >> > linux/kernel/bpf/ >> > linux/net/core/filter.c >> > linux/kernel/trace/bpf_trace.c >> > linux/tools/bpf/ >> > linux/tools/lib/bpf/ >> > >> > >> > Second question: >> > >> > Does the trace program like using uprobe/uretprobe and kprobe/kretprobe need libbpf built-in? >> > >> >> I don't think bpftrace relies on libbpf, so I suppose no? >> >> > For a specific kernel(like 4.19), how to choose the right version of libbpf for build as the dependency for bcc? >> >> You should build with whatever version of libbpf BCC depends on. But >> then I'm even more confused between you talking about bpftrace, BCC, >> and libbpf. All three are quite independent projects, with libbpf used >> by BCC for some functionality. >> >> > >> > Third question: >> > >> > Does my kernel need support BTF? Since I need to access the struct members from kernel and userspace. >> >> I don't know, because I don't know which kernel is *your* kernel. >> bpftrace has --btf parameter with which it can use kernel BTF, so I'm >> guessing that's what you are asking about? If yes, kernels starting >> from 5.2 version support emitting kernel BTF, but you need to enable >> it through CONFIG_DEBUG_INFO_BTF=y config value. >> >> > >> > >> > Thanks >> > >> > Eason
