[PATCH] build_bug.h: Also define static_assert() when __CHECKER__ is defined

[Christophe Leroy]

Le 27/01/2023 à 08:43, Rasmus Villemoes a écrit :
> [Vous ne recevez pas souvent de courriers de rasmus.villemoes at prevas.dk. Découvrez pourquoi ceci est important à https://aka.ms/LearnAboutSenderIdentification ]
> 
> On 26/01/2023 19.17, Christophe Leroy wrote:
>> When doing a build with C=2, the following failure is encountered on
>> several files:
>>
>>          CHECK   arch/powerpc/cpu/mpc8xxx/fsl_lbc.c
>>        arch/powerpc/cpu/mpc8xxx/fsl_lbc.c: note: in included file (through arch/powerpc/include/asm/global_data.h, include/init.h):
>>        include/asm-generic/global_data.h:494:21: error: Expected ) in function declarator
>>        include/asm-generic/global_data.h:494:21: error: got (
>>
>> And because of the error, the interesting part which are the
>> warnings don't appear. This is because static_assert() is defined
>> only when __CHECKER__ is not defined.
>>
>> Add a stub when __CHECKER__ is defined. With that fix, the expected
>> warnings are now seen:
>>
> 
> So sparse has supported _Static_assert since basically forever, and in
> the linux version of build_bug.h, the static_assert definition is not
> inside #ifndef __CHECKER__. So I think a better fix is to synchronize
> build_bug.h with linux again, so that we actually also do those
> static_asserts with sparse (if nothing else then as a sanity check of
> sparse itself, it really should grok ICEs and barf at
> 'static_assert(sizeof(int) == 3)').
> 
> The __CHECKER__ guard around most of the file vanished from upstream in
> 527edbc18a70, just before the static_assert define was added in
> 6bab69c65013, and I didn't notice that discrepancy when porting that to
> U-Boot in ef0f4e834c66.
> 

Ah right, I didn't see that difference with Linux.

By the way, there are other changes in Linux that we are missing in 
Uboot and create problems for me when using sparse:

80591e61a0f7 ("kbuild: tell sparse about the $ARCH")
6c49f359ca14 ("kbuild: disable sparse warnings about unknown attributes")

Also I have a problem with the following u-boot commit : aa9e891c633 
("include/linux/stddef.h: avoid 'warning: preprocessor token offsetof 
redefined'")

On uboot mainline I get:

   CHECK   drivers/pci/pci-uclass.c
drivers/pci/pci-uclass.c:88:9: error: typename in expression
drivers/pci/pci-uclass.c:88:9: error: typename in expression
drivers/pci/pci-uclass.c:88:9: error: undefined identifier 'offsetof'
drivers/pci/pci-uclass.c:88:9: error: cast from unknown type
drivers/pci/pci-uclass.c:88:9: error: undefined identifier 'offsetof'
drivers/pci/pci-uclass.c:88:9: error: cast from unknown type

If I revert that patch I get

   CHECK   drivers/pci/pci-uclass.c
drivers/pci/pci-uclass.c:852:20: warning: function 'board_pci_fixup_dev' 
with external linkage has definition
drivers/pci/pci-uclass.c:594:17: warning: incorrect type in argument 1 
(different address spaces)
drivers/pci/pci-uclass.c:594:17:    expected unsigned char volatile 
[noderef] <asn:2> *addr
drivers/pci/pci-uclass.c:594:17:    got unsigned char volatile [usertype] *
drivers/pci/pci-uclass.c:597:17: warning: incorrect type in argument 1 
(different address spaces)
drivers/pci/pci-uclass.c:597:17:    expected unsigned short volatile 
[noderef] <asn:2> *addr
drivers/pci/pci-uclass.c:597:17:    got unsigned short volatile [usertype] *
drivers/pci/pci-uclass.c:600:17: warning: incorrect type in argument 1 
(different address spaces)
drivers/pci/pci-uclass.c:600:17:    expected unsigned int volatile 
[noderef] <asn:2> *addr
drivers/pci/pci-uclass.c:600:17:    got unsigned int volatile [usertype] *
drivers/pci/pci-uclass.c:625:27: warning: incorrect type in argument 1 
(different address spaces)
drivers/pci/pci-uclass.c:625:27:    expected unsigned char const 
volatile [noderef] <asn:2> *addr
drivers/pci/pci-uclass.c:625:27:    got unsigned char volatile [usertype] *
drivers/pci/pci-uclass.c:628:27: warning: incorrect type in argument 1 
(different address spaces)
drivers/pci/pci-uclass.c:628:27:    expected unsigned short const 
volatile [noderef] <asn:2> *addr
drivers/pci/pci-uclass.c:628:27:    got unsigned short volatile [usertype] *
drivers/pci/pci-uclass.c:631:27: warning: incorrect type in argument 1 
(different address spaces)
drivers/pci/pci-uclass.c:631:27:    expected unsigned int const volatile 
[noderef] <asn:2> *addr
drivers/pci/pci-uclass.c:631:27:    got unsigned int volatile [usertype] *
drivers/pci/pci-uclass.c:1563:16: warning: Using plain integer as NULL 
pointer
drivers/pci/pci-uclass.c:1841:1: warning: symbol 
'_u_boot_list_2_uclass_driver_2_pci' was not declared. Should it be static?
drivers/pci/pci-uclass.c:1863:1: warning: symbol 
'_u_boot_list_2_driver_2_pci_bridge_drv' was not declared. Should it be 
static?
drivers/pci/pci-uclass.c:1870:1: warning: symbol 
'_u_boot_list_2_uclass_driver_2_pci_generic' was not declared. Should it 
be static?
drivers/pci/pci-uclass.c:1880:1: warning: symbol 
'_u_boot_list_2_driver_2_pci_generic_drv' was not declared. Should it be 
static?

But there are other files that then exhibit the original problem:

On mainline:

   CHECK   drivers/rtc/ds1374.c
drivers/rtc/ds1374.c:68:12: warning: symbol 'RtcTodAddr' was not 
declared. Should it be static?

With above commit reverted;

   CHECK   drivers/rtc/ds1374.c
drivers/rtc/ds1374.c: note: in included file (through 
include/compiler.h, include/env.h, include/command.h):
/opt/cldk-1.4.11/lib/gcc/ppc-linux/5.5.0/include/stddef.h:417:9: 
warning: preprocessor token offsetof redefined
drivers/rtc/ds1374.c: note: in included file (through 
include/linux/posix_types.h, include/linux/types.h, include/time.h, 
include/common.h):
include/linux/stddef.h:19:9: this was the original definition
drivers/rtc/ds1374.c:68:12: warning: symbol 'RtcTodAddr' was not 
declared. Should it be static?

What can we do about it ? At least reverting it replaces an error with a 
warning, I think that's better isn't it ?

Thanks
Christophe