Re: [PATCH -next v2] arm64: fix build warning for ARM64_MEMSTART_SHIFT

[Catalin Marinas <catalin.marinas@arm.com>]

On Tue, Jul 25, 2023 at 04:47:46PM +0800, zhangjianhua (E) wrote:
> 在 2023/7/25 12:22, Anshuman Khandual 写道:
> > On 7/24/23 20:41, Catalin Marinas wrote:
> > > On Mon, Jul 24, 2023 at 05:27:51PM +0000, Zhang Jianhua wrote:
> > > > When building with W=1, the following warning occurs.
> > > > 
> > > > arch/arm64/include/asm/kernel-pgtable.h:129:41: error: "PUD_SHIFT" is not defined, evaluates to 0 [-Werror=undef]
> > > >    129 | #define ARM64_MEMSTART_SHIFT            PUD_SHIFT
> > > >        |                                         ^~~~~~~~~
> > > > arch/arm64/include/asm/kernel-pgtable.h:142:5: note: in expansion of macro ‘ARM64_MEMSTART_SHIFT’
> > > >    142 | #if ARM64_MEMSTART_SHIFT < SECTION_SIZE_BITS
> > > >        |     ^~~~~~~~~~~~~~~~~~~~
> > > 
> > > Another thing that's missing here is that the warning is probably when
> > > this file is included from asm-offests.h or some .S file.
> > > 
> > > > diff --git a/arch/arm64/include/asm/kernel-pgtable.h b/arch/arm64/include/asm/kernel-pgtable.h
> > > > index 577773870b66..51bdce66885d 100644
> > > > --- a/arch/arm64/include/asm/kernel-pgtable.h
> > > > +++ b/arch/arm64/include/asm/kernel-pgtable.h
> > > > @@ -125,12 +125,14 @@
> > > >    * (64k granule), or a multiple that can be mapped using contiguous bits
> > > >    * in the page tables: 32 * PMD_SIZE (16k granule)
> > > >    */
> > > > -#if defined(CONFIG_ARM64_4K_PAGES)
> > > > +#if defined(CONFIG_ARM64_4K_PAGES) && defined(PUD_SHIFT)
> > > >   #define ARM64_MEMSTART_SHIFT		PUD_SHIFT
> > > That's not the correct fix since PUD_SHIFT should always be defined.
> > > When CONFIG_PGTABLE_LEVELS == 3, pgtable-types.h includes
> > > asm-generic/pgtable-nopud.h and this defines PUD_SHIFT. We either got
> > 
> > Right, PUD_SHIFT is always defined irrespective of page table levels.
> > 
> > > ARM64_MEMSTART_SHIFT defined in the wrong file or kernel-pgtable.h does
> > > not pull the relevant headers (either directly or via an included
> > > header). Even if kernel-pgtable.h ends up including the nopud/nopmd
> > > headers, P*D_SHIFT is guarded by an #indef __ASSEMBLY__ in those files.
> > > 
> > > Something like below appears to fix this, though I'm not particularly
> > > fond of guarding the ARM64_MEMSTART_* definitions by #ifndef
> > > __ASSEMBLY__ for no apparent reason (could add a comment though):
> > > 
> > > -----------------------8<---------------------------
> > > diff --git a/arch/arm64/include/asm/kernel-pgtable.h b/arch/arm64/include/asm/kernel-pgtable.h
> > > index 577773870b66..fcea7e87a6ca 100644
> > > --- a/arch/arm64/include/asm/kernel-pgtable.h
> > > +++ b/arch/arm64/include/asm/kernel-pgtable.h
> > > @@ -118,6 +118,8 @@
> > >   #define SWAPPER_RX_MMUFLAGS	(SWAPPER_RW_MMUFLAGS | PTE_RDONLY)
> > >   #endif
> > > +#ifndef __ASSEMBLY__
> > > +
> > >   /*
> > >    * To make optimal use of block mappings when laying out the linear
> > >    * mapping, round down the base of physical memory to a size that can
> > > @@ -145,4 +147,6 @@
> > >   #define ARM64_MEMSTART_ALIGN	(1UL << ARM64_MEMSTART_SHIFT)
> > >   #endif
> > > +#endif /* __ASSEMBLY__ */
> > > +
> > >   #endif	/* __ASM_KERNEL_PGTABLE_H */
> > > diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
> > > index e4944d517c99..22b36f2d5d93 100644
> > > --- a/arch/arm64/include/asm/pgtable-hwdef.h
> > > +++ b/arch/arm64/include/asm/pgtable-hwdef.h
> > > @@ -6,6 +6,7 @@
> > >   #define __ASM_PGTABLE_HWDEF_H
> > >   #include <asm/memory.h>
> > > +#include <asm/pgtable-types.h>
> > >   /*
> > >    * Number of page-table levels required to address 'va_bits' wide
> > > diff --git a/arch/arm64/include/asm/pgtable-types.h b/arch/arm64/include/asm/pgtable-types.h
> > > index b8f158ae2527..ae86e66fdb11 100644
> > > --- a/arch/arm64/include/asm/pgtable-types.h
> > > +++ b/arch/arm64/include/asm/pgtable-types.h
> > > @@ -11,6 +11,8 @@
> > >   #include <asm/types.h>
> > > +#ifndef __ASSEMBLY__
> > > +
> > >   typedef u64 pteval_t;
> > >   typedef u64 pmdval_t;
> > >   typedef u64 pudval_t;
> > > @@ -44,6 +46,8 @@ typedef struct { pteval_t pgprot; } pgprot_t;
> > >   #define pgprot_val(x)	((x).pgprot)
> > >   #define __pgprot(x)	((pgprot_t) { (x) } )
> > > +#endif /* __ASSEMBLY__ */
> > > +
> > >   #if CONFIG_PGTABLE_LEVELS == 2
> > >   #include <asm-generic/pgtable-nopmd.h>
> > >   #elif CONFIG_PGTABLE_LEVELS == 3
> > > -----------------------8<---------------------------
> > > 
> > > To avoid guarding the ARM64_MEMSTART_* definitions, we could instead
> > > move the P*D_SHIFT definitions in asm-generic/pgtable-nop*d.h outside
> > > the #ifndef __ASSEMBLY__ block.
> > 
> > OR could ARM64_MEMSTART_SHIFT and ARM64_MEMSTART_ALIGN computation blocks
> > be moved inside arch/arm64/mm/init.c, where it is used exclusively. Seems
> > to be solving the problem as well.

That's fine by me, better than adding the #ifndef __ASSEMBLY__ around
them.

> This method can avoid the current compilation warning, but does not
> solve the problem that PUD_SHIFT and PMD_SHIFT undefined in fact, and
> it is contrary to XXX_SHIFT should always be defined. Maybe it would
> be more appropriate to solve this issue directly.

For .c files, we can solve this by including asm/pgtable-types.h in
asm/pgtable-hwdef.h. This still leaves P*D_SHIFT undefined for .S files
since the generic nop*d.h headers guard the shifts with !__ASSEMBLY__
but do we really care about this? I haven't seen any other warning of
P*D_SHIFT not being defined. If you do want to solve these, just go and
change the generic headers to take the shift out of the asm guard. I
don't think it's worth it.

-- 
Catalin