Re: [PATCH 11/13] perf bpf: Add helper header files

[Arnaldo Carvalho de Melo]

Em Mon, Mar 12, 2018 at 08:20:20PM +0100, Jiri Olsa escreveu:
> On Mon, Mar 12, 2018 at 11:44:57AM -0700, Alexei Starovoitov wrote:
> > On Mon, Mar 12, 2018 at 10:43:11AM +0100, Jiri Olsa wrote:
> > > diff --git a/tools/perf/util/bpf-userapi.h b/tools/perf/util/bpf-userapi.h
> > > new file mode 100644
> > > index 000000000000..63f2b4c13a5c
> > > --- /dev/null
> > > +++ b/tools/perf/util/bpf-userapi.h
> > > @@ -0,0 +1,11 @@
> > > +#ifndef __BPF_USERAPI_H
> > > +#define __BPF_USERAPI_H
> > > +
> > > +enum {
> > > +	BPF_FUNC_USER_START			= 10000,
> > > +	BPF_FUNC_USER_print			= BPF_FUNC_USER_START,
> > > +	BPF_FUNC_USER_bpf_map_get_next_key,
> > > +	BPF_FUNC_USER_bpf_map_lookup_elem,
> > > +};
> > > +
> > > +#endif /* __BPF_USERFUNCS_H */
> > > diff --git a/tools/perf/util/bpf-userfuncs.h b/tools/perf/util/bpf-userfuncs.h
> > > new file mode 100644
> > > index 000000000000..e920420237ee
> > > --- /dev/null
> > > +++ b/tools/perf/util/bpf-userfuncs.h
> > > @@ -0,0 +1,19 @@
> > > +#ifndef __BPF_USERFUNCS_H
> > > +#define __BPF_USERFUNCS_H
> > > +
> > > +#include <bpf-userapi.h>
> > > +
> > > +static int (*bpfu_print)(const char *fmt, ...) =
> > > +	(void *) BPF_FUNC_USER_print;
> > > +static int (*bpfu_map_get_next_key)(void *map, void *key, void *value) =
> > > +	(void *) BPF_FUNC_USER_bpf_map_get_next_key;
> > > +static int (*bpfu_map_lookup_elem)(void *map, void *key, void *value) =
> > > +	(void *) BPF_FUNC_USER_bpf_map_lookup_elem;
> > > +
> > > +#define print(fmt, ...)                                  \
> > > +({                                                       \
> > > +	char ____fmt[] = fmt;                            \
> > > +	bpfu_print(____fmt, ##__VA_ARGS__);              \
> > > +})
> > 
> > since there is no verifier for this user space bpf interpreter
> > there is no need to restrict what BEGIN/END progs can do and can call.
> > llvm will happily compile any C code into bpf instructions.
> > With little bit of elf magic it's possible to let it call any libc
> > function instead of only 3 above.
> > perf loader will see bpf_call into normal printf, memcpy, memcmp, etc
> > then during the loading need to associate symbol with actual address
> > of these functions inside perf binary and let call_cb do the call.
> 
> right, I kept this way, because it was already there for kernel,
> so it was fast to write ;-)
> 
> but it's true we could leave it symbol based and have some sort of
> dynamic loader behaviour.. but we'd need to sort out passing the
> arguments in some generic form.. I'll check on that

Does this answer it or am I missing something?

Documentation/bpf/bpf_design_QA.txt

Q: Is BPF a generic virtual machine ?
A: NO.

BPF is generic instruction set _with_ C calling convention.

Q: Why C calling convention was chosen?
A: Because BPF programs are designed to run in the linux kernel
   which is written in C, hence BPF defines instruction set compatible
   with two most used architectures x64 and arm64 (and takes into
   consideration important quirks of other architectures) and
   defines calling convention that is compatible with C calling
   convention of the linux kernel on those architectures.

- Arnaldo