[PATCH v2 3/7] perf syscalltbl: Remove struct syscalltbl

[Ian Rogers]

On Mon, Feb 10, 2025 at 11:48 PM Arnd Bergmann <arnd at arndb.de> wrote:
>
> On Mon, Feb 10, 2025, at 17:51, Ian Rogers wrote:
> > The syscalltbl held entries of system call name and number pairs,
> > generated from a native syscalltbl at start up. As there are gaps in
> > the system call number there is a notion of index into the
> > table. Going forward we want the system call table to be identifiable
> > by a machine type, for example, i386 vs x86-64. Change the interface
> > to the syscalltbl so (1) a (currently unused machine type of EM_HOST)
> > is passed (2) the index to syscall number and system call name mapping
> > is computed at build time.
> >
> > Two tables are used for this, an array of system call number to name,
> > an array of system call numbers sorted by the system call name. The
> > sorted array doesn't store strings in part to save memory and
> > relocations. The index notion is carried forward and is an index into
> > the sorted array of system call numbers, the data structures are
> > opaque (held only in syscalltbl.c), and so the number of indices for a
> > machine type is exposed as a new API.
> >
> > The arrays are computed in the syscalltbl.sh script and so no start-up
> > time computation and storage is necessary.
> >
> > Signed-off-by: Ian Rogers <irogers at google.com>
> > Reviewed-by: Howard Chu <howardchu95 at gmail.com>
>
> Your changes look fine to me, but I noticed one part that may
> be wrong before and after your patch:
>
> >
> > -const int syscalltbl_native_max_id = SYSCALLTBL_MAX_ID;
> > -static const char *const *syscalltbl_native = syscalltbl;
> > +const char *syscalltbl__name(int e_machine __maybe_unused, int id)
> > +{
> > +     if (id >= 0 && id <= (int)ARRAY_SIZE(syscall_num_to_name))
> > +             return syscall_num_to_name[id];
> > +     return NULL;
> > +}
>
> The syscall numbers on mips (and previously on ia64) are offset by
> a large number depending on the ABI (o32/n32/n64). I assume what
> we want here is to have the small numbers without the offset in
> syscall_num_to_name[], but that requires adding the offset during
> the lookup. Can you check if this is handled correctly?

Thanks Arnd! I agree the tables are large and can be sparse, they'll
also be full of relocations. MIPS doesn't look like an outlier to me
here:
```
#if defined(ALL_SYSCALLTBL) || defined(__mips__)
static const char *const syscall_num_to_name_EM_MIPS[] = {
       [0] = "read",
       [1] = "write",
       [2] = "open",
...
        [465] = "listxattrat",
        [466] = "removexattrat",
};
```
For contrast x86:
```
#if defined(ALL_SYSCALLTBL) || defined(__i386__) || defined(__x86_64__)
static const char *const syscall_num_to_name_EM_386[] = {
        [0] = "restart_syscall",
        [1] = "exit",
        [2] = "fork",
...
        [464] = "getxattrat",
        [465] = "listxattrat",
        [466] = "removexattrat",
};
```
Looking through the tables I see alpha having the highest number
syscall with 572 being mseal.

I don't think this is great but in the current code (on x86-64) we
have in arch/x86/include/generated/asm/syscalls_64.h:
```
static const char *const syscalltbl[] = {
       [0] = "read",
       [1] = "write",
       [2] = "open",
...
        [465] = "listxattrat",
        [466] = "removexattrat",
};
#define SYSCALLTBL_MAX_ID 466
```
So the change is carrying forward a bad behavior, the table is still
only around 4kb. We could be more aggressive in compressing the
strings and pointers, for example how we compress the perf events and
metrics. I think it is getting out-of-scope here as that logic is
written in python, with the aid of lots of dictionaries, whilst this
code is currently a shell script. It becomes more of an issue if we
enable all of the tables in the build at once.

Thanks,
Ian