[v3,0/3] nolibc: add part2 of support for rv32

  Hi, Willy

This is the v3 part2 of support for rv32, differs from the v2 part2 [1],
we only fix up compile issues in this patchset.

With the v3 generic part1 [2] and this patchset, we can compile nolibc
for rv32 now.

This is based on the idea of suggestions from Arnd [3], instead of
'#error' on the unsupported syscall on a target platform, a 'return
-ENOSYS' allow us to compile it at first and then allow we fix up the
test failures reported by nolibc-test one by one.

The first two patches fix up all of the compile failures with '-ENOSYS'
(and '#ifdef' if required):

  tools/nolibc: fix up #error compile failures with -ENOSYS
  tools/nolibc: fix up undeclared syscall macros with #ifdef and -ENOSYS

The last one enables rv32 compile support:
  
  selftests/nolibc: riscv: customize makefile for rv32

The above compile support patch here is only for test currently, as
Thomas suggested, for a full rv32 support, it should wait for the left
parts.

Welcome your feedbacks, will wait for enough discussion on this patchset
and then send the left parts one by one to fix up the test failures
about waitid, llseek and time64 syscalls: ppoll_time64, clock_gettime64,
pselect6_time64.

So, I do recommend to apply this patchset, it allows us to send the left
parts independently, otherwise, all of them should be sent out for
review together. with this patchset, the rv32 users may be able to use
nolibc although some syscalls still missing :-)

Or at least we apply the first two, so, I can manually cherry-pick the
compile support patch to do my local test, and the other platform
developer may also benefit from them.

I'm cleaning up the left parts, but still require some time, I plan to
split them to such parts:

  * part3: waitid, prepared, will send out later
  * part4: llseek, prepared, will send out later
  * part5: time64 syscalls, ppoll_time64 ok, will finish them next week
           (It is a little hard to split them)

Best regards,
Zhangjin
---

[1]: https://lore.kernel.org/linux-riscv/cover.1685387484.git.falcon@tinylab.org/T/#t
[2]: https://lore.kernel.org/linux-riscv/cover.1685777982.git.falcon@tinylab.org/T/#t
[3]: https://lore.kernel.org/linux-riscv/5e7d2adf-e96f-41ca-a4c6-5c87a25d4c9c@app.fastmail.com/

Zhangjin Wu (3):
  tools/nolibc: fix up #error compile failures with -ENOSYS
  tools/nolibc: fix up undeclared syscall macros with #ifdef and -ENOSYS
  selftests/nolibc: riscv: customize makefile for rv32

 tools/include/nolibc/sys.h              | 38 ++++++++++++++++---------
 tools/testing/selftests/nolibc/Makefile | 11 +++++--
 2 files changed, 34 insertions(+), 15 deletions(-)
  

Hi, Arnd

Because this patchset is a 'big' change derived from the idea of suggestion
from you [3], I do very welcome your feedback about this change, just like
Thomas suggested, this requires more discussion before Willy plan to determine
merge it or not.

The first two convert all compile failures to a return of -ENOSYS, if you do
like it, welcome your Reviewed-by. These two are required by the coming new
time64 syscalls for rv32, because they depends on how we cope with the
unsupported syscalls, returning -ENOSYS is really better than simply fail the
compiling.

Hi, Thomas, If you are happy with them, welcome your Reviewed-by too, thanks.
If the first two are ok, then, I can focus on preparing the left time64
patchsets.

The third one is not that urgent, because some important syscalls are
still missing for rv32. It is added here only for compile test.

Thanks,
Zhangjin

> Hi, Willy
> 
> This is the v3 part2 of support for rv32, differs from the v2 part2 [1],
> we only fix up compile issues in this patchset.
> 
> With the v3 generic part1 [2] and this patchset, we can compile nolibc
> for rv32 now.
> 
> This is based on the idea of suggestions from Arnd [3], instead of
> '#error' on the unsupported syscall on a target platform, a 'return
> -ENOSYS' allow us to compile it at first and then allow we fix up the
> test failures reported by nolibc-test one by one.
> 
> The first two patches fix up all of the compile failures with '-ENOSYS'
> (and '#ifdef' if required):
> 
>   tools/nolibc: fix up #error compile failures with -ENOSYS
>   tools/nolibc: fix up undeclared syscall macros with #ifdef and -ENOSYS
> 
> The last one enables rv32 compile support:
>   
>   selftests/nolibc: riscv: customize makefile for rv32
> 
> The above compile support patch here is only for test currently, as
> Thomas suggested, for a full rv32 support, it should wait for the left
> parts.
> 
> Welcome your feedbacks, will wait for enough discussion on this patchset
> and then send the left parts one by one to fix up the test failures
> about waitid, llseek and time64 syscalls: ppoll_time64, clock_gettime64,
> pselect6_time64.
> 
> So, I do recommend to apply this patchset, it allows us to send the left
> parts independently, otherwise, all of them should be sent out for
> review together. with this patchset, the rv32 users may be able to use
> nolibc although some syscalls still missing :-)
> 
> Or at least we apply the first two, so, I can manually cherry-pick the
> compile support patch to do my local test, and the other platform
> developer may also benefit from them.
> 
> I'm cleaning up the left parts, but still require some time, I plan to
> split them to such parts:
> 
>   * part3: waitid, prepared, will send out later
>   * part4: llseek, prepared, will send out later
>   * part5: time64 syscalls, ppoll_time64 ok, will finish them next week
>            (It is a little hard to split them)
> 
> Best regards,
> Zhangjin
> ---
> 
> [1]: https://lore.kernel.org/linux-riscv/cover.1685387484.git.falcon@tinylab.org/T/#t
> [2]: https://lore.kernel.org/linux-riscv/cover.1685777982.git.falcon@tinylab.org/T/#t
> [3]: https://lore.kernel.org/linux-riscv/5e7d2adf-e96f-41ca-a4c6-5c87a25d4c9c@app.fastmail.com/
> 
> Zhangjin Wu (3):
>   tools/nolibc: fix up #error compile failures with -ENOSYS
>   tools/nolibc: fix up undeclared syscall macros with #ifdef and -ENOSYS
>   selftests/nolibc: riscv: customize makefile for rv32
> 
>  tools/include/nolibc/sys.h              | 38 ++++++++++++++++---------
>  tools/testing/selftests/nolibc/Makefile | 11 +++++--
>  2 files changed, 34 insertions(+), 15 deletions(-)
  

Hi Zhangjin,

On Tue, Jun 06, 2023 at 12:25:35PM +0800, Zhangjin Wu wrote:
> The first two convert all compile failures to a return of -ENOSYS, if you do
> like it, welcome your Reviewed-by. These two are required by the coming new
> time64 syscalls for rv32, because they depends on how we cope with the
> unsupported syscalls, returning -ENOSYS is really better than simply fail the
> compiling.

I had a look now and I can sya that I like this. Initially the supported
syscalls were so restricted that it was not even imaginable to accept to
build without any of them, but now that we're completing the list, some
of them are less critical and I don't see why we'd fail to build just
because one is missing. So yeah, a big +1 for -ENOSYS.

> The third one is not that urgent, because some important syscalls are
> still missing for rv32. It is added here only for compile test.

I personally have no opinion on this one. I can't judge whether it will
make things easier or more complicated at this point. It seems to me
that for now it's just avoiding one extra line at the expense of some
$(if) on several lines. Maybe it could help add more such archs, or
maybe it can make them more complicated to debug, I don't know. I'm
interested in others' opinions as well.

Thanks,
Willy
  

Willy, Thomas, Arnd

> Hi Zhangjin,
> 
> On Tue, Jun 06, 2023 at 12:25:35PM +0800, Zhangjin Wu wrote:
> > The first two convert all compile failures to a return of -ENOSYS, if you do
> > like it, welcome your Reviewed-by. These two are required by the coming new
> > time64 syscalls for rv32, because they depends on how we cope with the
> > unsupported syscalls, returning -ENOSYS is really better than simply fail the
> > compiling.
> 
> I had a look now and I can sya that I like this. Initially the supported
> syscalls were so restricted that it was not even imaginable to accept to
> build without any of them, but now that we're completing the list, some
> of them are less critical and I don't see why we'd fail to build just
> because one is missing. So yeah, a big +1 for -ENOSYS.
>

Cool, I will prepare the new patchsets on them, welcome your new branch
with both of them, of course, still weclome Reviewed-by from Arnd and Thomas.

> > The third one is not that urgent, because some important syscalls are
> > still missing for rv32. It is added here only for compile test.
> 
> I personally have no opinion on this one. I can't judge whether it will
> make things easier or more complicated at this point. It seems to me
> that for now it's just avoiding one extra line at the expense of some
> $(if) on several lines. Maybe it could help add more such archs, or
> maybe it can make them more complicated to debug, I don't know. I'm
> interested in others' opinions as well.

As I explained why we did it in current way in this reply [1], Thomas had no
more questions on it, so I think Thomas was happy with it now and I got his
only left suggestion is that may be this patch should be applied after the
missing 64bit syscalls being added for there are several important test
failures currently, for me, it is ok before or after.

Thomas, welcome your Reviewed-by on the makefile patch itself If you are really
happy with it now, thanks very much ;-)

Willy, I will send the v2 syscalls helpers (also required by the coming 64bit
syscalls) and some other patches (mainly for test with faster kernel build)
about selftests/nolibc later, because we have not enough time for v6.5 test,
so, I suggest we can create new branch for v6.6 and my new patchsets will be
for v6.6.

Best regards,
Zhangjin

---

[1]: https://lore.kernel.org/linux-riscv/20230526092029.149351-1-falcon@tinylab.org/

> 
> Thanks,
> Willy
  

On Tue, Jun 06, 2023 at 02:34:21PM +0800, Zhangjin Wu wrote:
> Willy, Thomas, Arnd
> 
> > Hi Zhangjin,
> > 
> > On Tue, Jun 06, 2023 at 12:25:35PM +0800, Zhangjin Wu wrote:
> > > The first two convert all compile failures to a return of -ENOSYS, if you do
> > > like it, welcome your Reviewed-by. These two are required by the coming new
> > > time64 syscalls for rv32, because they depends on how we cope with the
> > > unsupported syscalls, returning -ENOSYS is really better than simply fail the
> > > compiling.
> > 
> > I had a look now and I can sya that I like this. Initially the supported
> > syscalls were so restricted that it was not even imaginable to accept to
> > build without any of them, but now that we're completing the list, some
> > of them are less critical and I don't see why we'd fail to build just
> > because one is missing. So yeah, a big +1 for -ENOSYS.
> >
> 
> Cool, I will prepare the new patchsets on them, welcome your new branch
> with both of them, of course, still weclome Reviewed-by from Arnd and Thomas.

I don't even think a new branch is needed, I can take them as-is it seems.

> > > The third one is not that urgent, because some important syscalls are
> > > still missing for rv32. It is added here only for compile test.
> > 
> > I personally have no opinion on this one. I can't judge whether it will
> > make things easier or more complicated at this point. It seems to me
> > that for now it's just avoiding one extra line at the expense of some
> > $(if) on several lines. Maybe it could help add more such archs, or
> > maybe it can make them more complicated to debug, I don't know. I'm
> > interested in others' opinions as well.
> 
> As I explained why we did it in current way in this reply [1], Thomas had no
> more questions on it, so I think Thomas was happy with it now and I got his
> only left suggestion is that may be this patch should be applied after the
> missing 64bit syscalls being added for there are several important test
> failures currently, for me, it is ok before or after.
> 
> Thomas, welcome your Reviewed-by on the makefile patch itself If you are really
> happy with it now, thanks very much ;-)
> 
> Willy, I will send the v2 syscalls helpers (also required by the coming 64bit
> syscalls) and some other patches (mainly for test with faster kernel build)
> about selftests/nolibc later, because we have not enough time for v6.5 test,
> so, I suggest we can create new branch for v6.6 and my new patchsets will be
> for v6.6.

Agreed, thank you!
Willy