Re: [PATCH 10/17] arm64: generate 64-bit syscall.tbl

[Arnd Bergmann]

On Tue, Jul 9, 2024, at 19:23, Mark Brown wrote:
> On Thu, Jul 04, 2024 at 04:36:04PM +0200, Arnd Bergmann wrote:
>
>>  #define __ARCH_WANT_SYS_CLONE
>> +#define __ARCH_WANT_NEW_STAT
>>  
>> -#ifndef __COMPAT_SYSCALL_NR
>> -#include <uapi/asm/unistd.h>
>> -#endif
>> +#include <asm/unistd_64.h>
>
> It looks like this is causing widespread build breakage in kselftest in
> -next for arm64, there are *many* errors in the form:
>
> In file included from test_signals_utils.c:14:
> /build/stage/build-work/usr/include/asm/unistd.h:2:10: fatal error: 
> unistd_64.h: No such file or directory
>     2 | #include <unistd_64.h>
>       |          ^~~~~~~~~~~~~
>
> which obviously looks like it's tied to the above but I've not fully
> understood the patch/series yet.  Build log at:
>

Thanks for the report! I just panicked a bit and thought I had
done something entirely wrong here, but after having a closer
look it turned out to be a silly typo:

diff --git a/arch/arm64/include/uapi/asm/unistd.h b/arch/arm64/include/uapi/asm/unistd.h
index 038dddf8f554..df36f23876e8 100644
--- a/arch/arm64/include/uapi/asm/unistd.h
+++ b/arch/arm64/include/uapi/asm/unistd.h
@@ -1,2 +1,2 @@
 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
-#include <unistd_64.h>
+#include <asm/unistd_64.h>

I'm folding the fix into the tree now, in addition to the Acks
I received and another small fixup.

I checked that arm64 is the only architecture that has this
particular bug, and I tried building kselftest now, which seems
to work better. There are still a few warnings and errors
I get doing that, but I suspect those are all preexisting
issues.

     Arnd