Re: [PATCH v3 04/13] mm/execmem, arch: convert remaining overrides of module_alloc to execmem

[Mike Rapoport <rppt@kernel.org>]

Hi Will,

On Mon, Oct 23, 2023 at 06:14:20PM +0100, Will Deacon wrote:
> Hi Mike,
> 
> On Mon, Sep 18, 2023 at 10:29:46AM +0300, Mike Rapoport wrote:
> > From: "Mike Rapoport (IBM)" <rppt@kernel.org>
> > 
> > Extend execmem parameters to accommodate more complex overrides of
> > module_alloc() by architectures.
> > 
> > This includes specification of a fallback range required by arm, arm64
> > and powerpc and support for allocation of KASAN shadow required by
> > arm64, s390 and x86.
> > 
> > The core implementation of execmem_alloc() takes care of suppressing
> > warnings when the initial allocation fails but there is a fallback range
> > defined.
> > 
> > Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org>
> > ---
> >  arch/arm/kernel/module.c     | 38 ++++++++++++---------
> >  arch/arm64/kernel/module.c   | 57 ++++++++++++++------------------
> >  arch/powerpc/kernel/module.c | 52 ++++++++++++++---------------
> >  arch/s390/kernel/module.c    | 52 +++++++++++------------------
> >  arch/x86/kernel/module.c     | 64 +++++++++++-------------------------
> >  include/linux/execmem.h      | 14 ++++++++
> >  mm/execmem.c                 | 43 ++++++++++++++++++++++--
> >  7 files changed, 167 insertions(+), 153 deletions(-)
> 
> [...]
> 
> > diff --git a/arch/arm64/kernel/module.c b/arch/arm64/kernel/module.c
> > index dd851297596e..cd6320de1c54 100644
> > --- a/arch/arm64/kernel/module.c
> > +++ b/arch/arm64/kernel/module.c
> > @@ -20,6 +20,7 @@
> >  #include <linux/random.h>
> >  #include <linux/scs.h>
> >  #include <linux/vmalloc.h>
> > +#include <linux/execmem.h>
> >  
> >  #include <asm/alternative.h>
> >  #include <asm/insn.h>
> > @@ -108,46 +109,38 @@ static int __init module_init_limits(void)
> >  
> >  	return 0;
> >  }
> > -subsys_initcall(module_init_limits);
> >  
> > -void *module_alloc(unsigned long size)
> > +static struct execmem_params execmem_params __ro_after_init = {
> > +	.ranges = {
> > +		[EXECMEM_DEFAULT] = {
> > +			.flags = EXECMEM_KASAN_SHADOW,
> > +			.alignment = MODULE_ALIGN,
> > +		},
> > +	},
> > +};
> > +
> > +struct execmem_params __init *execmem_arch_params(void)
> >  {
> > -	void *p = NULL;
> > +	struct execmem_range *r = &execmem_params.ranges[EXECMEM_DEFAULT];
> >  
> > -	/*
> > -	 * Where possible, prefer to allocate within direct branch range of the
> > -	 * kernel such that no PLTs are necessary.
> > -	 */
> 
> Why are you removing this comment? I think you could just move it next
> to the part where we set a 128MiB range.
 
Oops, my bad. Will add it back.

> > -	if (module_direct_base) {
> > -		p = __vmalloc_node_range(size, MODULE_ALIGN,
> > -					 module_direct_base,
> > -					 module_direct_base + SZ_128M,
> > -					 GFP_KERNEL | __GFP_NOWARN,
> > -					 PAGE_KERNEL, 0, NUMA_NO_NODE,
> > -					 __builtin_return_address(0));
> > -	}
> > +	module_init_limits();
> 
> Hmm, this used to be run from subsys_initcall(), but now you're running
> it _really_ early, before random_init(), so randomization of the module
> space is no longer going to be very random if we don't have early entropy
> from the firmware or the CPU, which is likely to be the case on most SoCs.

Well, it will be as random as KASLR. Won't that be enough?
 
> > diff --git a/mm/execmem.c b/mm/execmem.c
> > index f25a5e064886..a8c2f44d0133 100644
> > --- a/mm/execmem.c
> > +++ b/mm/execmem.c
> > @@ -11,12 +11,46 @@ static void *execmem_alloc(size_t size, struct execmem_range *range)
> >  {
> >  	unsigned long start = range->start;
> >  	unsigned long end = range->end;
> > +	unsigned long fallback_start = range->fallback_start;
> > +	unsigned long fallback_end = range->fallback_end;
> >  	unsigned int align = range->alignment;
> >  	pgprot_t pgprot = range->pgprot;
> > +	bool kasan = range->flags & EXECMEM_KASAN_SHADOW;
> > +	unsigned long vm_flags  = VM_FLUSH_RESET_PERMS;
> > +	bool fallback  = !!fallback_start;
> > +	gfp_t gfp_flags = GFP_KERNEL;
> > +	void *p;
> >  
> > -	return __vmalloc_node_range(size, align, start, end,
> > -				   GFP_KERNEL, pgprot, VM_FLUSH_RESET_PERMS,
> > -				   NUMA_NO_NODE, __builtin_return_address(0));
> > +	if (PAGE_ALIGN(size) > (end - start))
> > +		return NULL;
> > +
> > +	if (kasan)
> > +		vm_flags |= VM_DEFER_KMEMLEAK;
> 
> Hmm, I don't think we passed this before on arm64, should we have done?

It was there on arm64 before commit 8339f7d8e178 ("arm64: module: remove
old !KASAN_VMALLOC logic").
There's no need to pass VM_DEFER_KMEMLEAK when KASAN_VMALLOC is enabled and
arm64 always selects KASAN_VMALLOC with KASAN.

And for the generic case, I should have made the condition to check for
KASAN_VMALLOC as well.
 
> Will

-- 
Sincerely yours,
Mike.