[PATCH] riscv: mm: Pre-allocate module page tables for execmem

[PATCH] riscv: mm: Pre-allocate module page tables for execmem

[Karl Mehltretter]

RV64 pre-allocates page-table levels for late kernel mappings. The module
range is guarded by CONFIG_MODULES, although execmem can use it without
modules.

With CONFIG_BPF_JIT=y and CONFIG_MODULES=n, execmem adds a mapping to
init_mm while the current process retains an empty PGD entry. Initializing
a BPF program pack then faults:

  Unable to handle kernel paging request ...
  [ffffffff78000000] pgd=0000000000000000
  ...
  bpf_jit_fill_hole_with_zero
  bpf_prog_pack_alloc

Use CONFIG_EXECMEM for the module-range pre-allocation, as powerpc does
for its execmem page-table setup. Both CONFIG_MODULES and CONFIG_BPF_JIT
select it.

Fixes: 2c9e5d4a0082 ("bpf: remove CONFIG_BPF_JIT dependency on CONFIG_MODULES of")
Cc: stable@vger.kernel.org
Assisted-by: LLM
Signed-off-by: Karl Mehltretter <kmehltretter@gmail.com>
---

Notes:
    Tested on a BeagleV Ahead with matched LLVM 18.1.3 builds based on
    v7.3-rc7:
    
    - unpatched: a native XDP attach triggered the Oops above
    - patched: native XDP attach and detach completed without an Oops

 arch/riscv/mm/init.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index fb37b0b67efe..d92342c162e2 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -1437,7 +1437,7 @@ static void __init preallocate_pgd_pages_range(unsigned long start, unsigned lon
 void __init pgtable_cache_init(void)
 {
 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
-	if (IS_ENABLED(CONFIG_MODULES))
+	if (IS_ENABLED(CONFIG_EXECMEM))
 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
 	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
 		preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap");
-- 
2.43.0

Re: [PATCH] riscv: mm: Pre-allocate module page tables for execmem

[Mike Rapoport]

On Thu, Oct 15, 2026 at 05:36:43AM +0200, Karl Mehltretter wrote:
> RV64 pre-allocates page-table levels for late kernel mappings. The module
> range is guarded by CONFIG_MODULES, although execmem can use it without
> modules.
> 
> With CONFIG_BPF_JIT=y and CONFIG_MODULES=n, execmem adds a mapping to
> init_mm while the current process retains an empty PGD entry. Initializing
> a BPF program pack then faults:
> 
>   Unable to handle kernel paging request ...
>   [ffffffff78000000] pgd=0000000000000000
>   ...
>   bpf_jit_fill_hole_with_zero
>   bpf_prog_pack_alloc
> 
> Use CONFIG_EXECMEM for the module-range pre-allocation, as powerpc does
> for its execmem page-table setup. Both CONFIG_MODULES and CONFIG_BPF_JIT
> select it.

I think the commit message should mention that BPF_JIT_REGION is inside
MODULES region.
 
> Fixes: 2c9e5d4a0082 ("bpf: remove CONFIG_BPF_JIT dependency on CONFIG_MODULES of")
> Cc: stable@vger.kernel.org
> Assisted-by: LLM
> Signed-off-by: Karl Mehltretter <kmehltretter@gmail.com>

Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>

> ---
> 
> Notes:
>     Tested on a BeagleV Ahead with matched LLVM 18.1.3 builds based on
>     v7.3-rc7:
>     
>     - unpatched: a native XDP attach triggered the Oops above
>     - patched: native XDP attach and detach completed without an Oops
>
>  arch/riscv/mm/init.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
> index fb37b0b67efe..d92342c162e2 100644
> --- a/arch/riscv/mm/init.c
> +++ b/arch/riscv/mm/init.c
> @@ -1437,7 +1437,7 @@ static void __init preallocate_pgd_pages_range(unsigned long start, unsigned lon
>  void __init pgtable_cache_init(void)
>  {
>  	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
> -	if (IS_ENABLED(CONFIG_MODULES))
> +	if (IS_ENABLED(CONFIG_EXECMEM))
>  		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
>  	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
>  		preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap");
> -- 
> 2.43.0
> 

-- 
Sincerely yours,
Mike.