Re: Comments on new user events ABI

Re: Comments on new user events ABI

[Alexei Starovoitov]

On Tue, Mar 29, 2022 at 9:17 AM Mathieu Desnoyers
<mathieu.desnoyers@efficios.com> wrote:
>
> >
> >> include/uapi/linux/user_events.h:
> >>
> >> struct user_bpf_iter {
> >>
> >>         /* Offset of the data within the first iovec */
> >>         __u32 iov_offset;
> >>
> >>         /* Number of iovec structures */
> >>         __u32 nr_segs;
> >>
> >>         /* Pointer to iovec structures */
> >>         const struct iovec *iov;
> >>
> >>                            ^ a pointer in a uapi header is usually a no-go. This should be a u64.
> >> };
> >>
> >> include/uapi/linux/user_events.h:
> >>
> >> struct user_bpf_context {
> >>
> >>         /* Data type being passed (see union below) */
> >>         __u32 data_type;
> >>
> >>         /* Length of the data */
> >>         __u32 data_len;
> >>
> >>         /* Pointer to data, varies by data type */
> >>         union {
> >>                 /* Kernel data (data_type == USER_BPF_DATA_KERNEL) */
> >>                 void *kdata;
> >>
> >>                 /* User data (data_type == USER_BPF_DATA_USER) */
> >>                 void *udata;
> >>
> >>                 /* Direct iovec (data_type == USER_BPF_DATA_ITER) */
> >>                 struct user_bpf_iter *iter;
> >>
> >>                                ^ likewise for the 3 pointers above. Should be u64 in uapi headers.
> >>         };
> >> };
> >>
> >
> > The bpf structs are only used within a BPF program. At that point the pointer
> > sizes should all align, right?
>
> I must admit I do not know enough about the eBPF uapi practices to answer this.
> [CCing Alexei on this]

Mathieu,

Thanks for flagging.

Whoever added this user_bpf* stuff please remove it immediately.
It was never reviewed by bpf maintainers.

It's a hard Nack to add a bpf interface to user_events.

Re: Comments on new user events ABI

[Beau Belgrave]

On Tue, Mar 29, 2022 at 09:25:52AM -0700, Alexei Starovoitov wrote:
> On Tue, Mar 29, 2022 at 9:17 AM Mathieu Desnoyers
> <mathieu.desnoyers@efficios.com> wrote:
> >
> > >
> > >> include/uapi/linux/user_events.h:
> > >>
> > >> struct user_bpf_iter {
> > >>
> > >>         /* Offset of the data within the first iovec */
> > >>         __u32 iov_offset;
> > >>
> > >>         /* Number of iovec structures */
> > >>         __u32 nr_segs;
> > >>
> > >>         /* Pointer to iovec structures */
> > >>         const struct iovec *iov;
> > >>
> > >>                            ^ a pointer in a uapi header is usually a no-go. This should be a u64.
> > >> };
> > >>
> > >> include/uapi/linux/user_events.h:
> > >>
> > >> struct user_bpf_context {
> > >>
> > >>         /* Data type being passed (see union below) */
> > >>         __u32 data_type;
> > >>
> > >>         /* Length of the data */
> > >>         __u32 data_len;
> > >>
> > >>         /* Pointer to data, varies by data type */
> > >>         union {
> > >>                 /* Kernel data (data_type == USER_BPF_DATA_KERNEL) */
> > >>                 void *kdata;
> > >>
> > >>                 /* User data (data_type == USER_BPF_DATA_USER) */
> > >>                 void *udata;
> > >>
> > >>                 /* Direct iovec (data_type == USER_BPF_DATA_ITER) */
> > >>                 struct user_bpf_iter *iter;
> > >>
> > >>                                ^ likewise for the 3 pointers above. Should be u64 in uapi headers.
> > >>         };
> > >> };
> > >>
> > >
> > > The bpf structs are only used within a BPF program. At that point the pointer
> > > sizes should all align, right?
> >
> > I must admit I do not know enough about the eBPF uapi practices to answer this.
> > [CCing Alexei on this]
> 
> Mathieu,
> 
> Thanks for flagging.
> 
> Whoever added this user_bpf* stuff please remove it immediately.
> It was never reviewed by bpf maintainers.
> 
> It's a hard Nack to add a bpf interface to user_events.

Sorry about that, I'm sending a patch to remove this.

I'll have another patch to add it back in out to bpf and trace-devel for
review.

Thanks,
-Beau

Re: Comments on new user events ABI

[Steven Rostedt]

On Tue, 29 Mar 2022 09:25:52 -0700
Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:

> Thanks for flagging.
> 
> Whoever added this user_bpf* stuff please remove it immediately.
> It was never reviewed by bpf maintainers.

Heh, now you know how the x86 maintainers feel ;-)

> 
> It's a hard Nack to add a bpf interface to user_events.

Agreed, I'm thinking of marking the entire thing as broken such that it can
be worked on a bit more without a total revert (but still remove the BPF
portion on your request).

Beau, I agree with Mathieu, I don't think it's a good idea to expose the
"ftrace/perf/etc" users. The only thing that the application needs is a bit
to say "call the write now". And let the work be done within the kernel.
I think a single bit may be better, that way you can have many more events
on a page, and since they do not get modified often, it will be in hot
cache and fast.

-- Steve