Re: [PATCH v3 10/13] arch: make execmem setup available regardless of CONFIG_MODULES

[Mike Rapoport]

 Hi Arnd,

On Tue, Sep 26, 2023 at 09:33:48AM +0200, Arnd Bergmann wrote:
> On Mon, Sep 18, 2023, at 09:29, Mike Rapoport wrote:
> > index a42e4cd11db2..c0b536e398b4 100644
> > --- a/arch/arm/mm/init.c
> > +++ b/arch/arm/mm/init.c
> > +#ifdef CONFIG_XIP_KERNEL
> > +/*
> > + * The XIP kernel text is mapped in the module area for modules and
> > + * some other stuff to work without any indirect relocations.
> > + * MODULES_VADDR is redefined here and not in asm/memory.h to avoid
> > + * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned 
> > on/off.
> > + */
> > +#undef MODULES_VADDR
> > +#define MODULES_VADDR	(((unsigned long)_exiprom + ~PMD_MASK) & 
> > PMD_MASK)
> > +#endif
> > +
> > +#if defined(CONFIG_MMU) && defined(CONFIG_EXECMEM)
> > +static struct execmem_params execmem_params __ro_after_init = {
> > +	.ranges = {
> > +		[EXECMEM_DEFAULT] = {
> > +			.start = MODULES_VADDR,
> > +			.end = MODULES_END,
> > +			.alignment = 1,
> > +		},
> 
> This causes a randconfig build failure for me on linux-next now:
> 
> arch/arm/mm/init.c:499:25: error: initializer element is not constant
>   499 | #define MODULES_VADDR   (((unsigned long)_exiprom + ~PMD_MASK) & PMD_MASK)
>       |                         ^
> arch/arm/mm/init.c:506:34: note: in expansion of macro 'MODULES_VADDR'
>   506 |                         .start = MODULES_VADDR,
>       |                                  ^~~~~~~~~~~~~
> arch/arm/mm/init.c:499:25: note: (near initialization for 'execmem_params.ranges[0].start')
>   499 | #define MODULES_VADDR   (((unsigned long)_exiprom + ~PMD_MASK) & PMD_MASK)
>       |                         ^
> arch/arm/mm/init.c:506:34: note: in expansion of macro 'MODULES_VADDR'
>   506 |                         .start = MODULES_VADDR,
>       |                                  ^~~~~~~~~~~~~
>
> I have not done any analysis on the issue so far, I hope
> you can see the problem directly. See
> https://pastebin.com/raw/xVqAyakH for a .config that runs into
> this problem with gcc-13.2.0.

The first patch that breaks XIP build is rather patch 04/13, currently
commit 52a34d45419f ("mm/execmem, arch: convert remaining overrides of
module_alloc to execmem") in mm.git/mm-unstable.

The hunk below is a fix for that and the attached patch is the updated
version of 835bc9685f45 ("arch: make execmem setup available regardless of
CONFIG_MODULES")

Andrew, please let me know if you'd like to me to resend these differently.

diff --git a/arch/arm/kernel/module.c b/arch/arm/kernel/module.c
index 2c7651a2d84c..096cc1ead635 100644
--- a/arch/arm/kernel/module.c
+++ b/arch/arm/kernel/module.c
@@ -38,8 +38,6 @@
 static struct execmem_params execmem_params __ro_after_init = {
 	.ranges = {
 		[EXECMEM_DEFAULT] = {
-			.start = MODULES_VADDR,
-			.end = MODULES_END,
 			.alignment = 1,
 		},
 	},
@@ -49,6 +47,8 @@ struct execmem_params __init *execmem_arch_params(void)
 {
 	struct execmem_range *r = &execmem_params.ranges[EXECMEM_DEFAULT];
 
+	r->start = MODULES_VADDR;
+	r->end = MODULES_END;
 	r->pgprot = PAGE_KERNEL_EXEC;
 
 	if (IS_ENABLED(CONFIG_ARM_MODULE_PLTS)) {

 
> 
>       Arnd

-- 
Sincerely yours,
Mike.