Re: [v2 PATCH] arm64: kprobes: call set_memory_rox() for kprobe page

[Yang Shi <yang@os.amperecomputing.com>]

On 9/21/25 12:36 AM, Mike Rapoport wrote:
> On Thu, Sep 18, 2025 at 10:33:26AM -0700, Yang Shi wrote:
>>
>> On 9/18/25 10:26 AM, Will Deacon wrote:
>>> On Thu, Sep 18, 2025 at 09:23:49AM -0700, Yang Shi wrote:
>>>> The kprobe page is allocated by execmem allocator with ROX permission.
>>>> It needs to call set_memory_rox() to set proper permission for the
>>>> direct map too. It was missed.
>>>>
>>>> Fixes: 10d5e97c1bf8 ("arm64: use PAGE_KERNEL_ROX directly in alloc_insn_page")
>>>> Cc: <stable@vger.kernel.org>
>>>> Signed-off-by: Yang Shi <yang@os.amperecomputing.com>
>>>> Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
>>>> ---
>>>> v2: Separated the patch from BBML2 series since it is an orthogonal bug
>>>>       fix per Ryan.
>>>>       Fixed the variable name nit per Catalin.
>>>>       Collected R-bs from Catalin.
>>>>
>>>>    arch/arm64/kernel/probes/kprobes.c | 12 ++++++++++++
>>>>    1 file changed, 12 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/kernel/probes/kprobes.c b/arch/arm64/kernel/probes/kprobes.c
>>>> index 0c5d408afd95..8ab6104a4883 100644
>>>> --- a/arch/arm64/kernel/probes/kprobes.c
>>>> +++ b/arch/arm64/kernel/probes/kprobes.c
>>>> @@ -10,6 +10,7 @@
>>>>    #define pr_fmt(fmt) "kprobes: " fmt
>>>> +#include <linux/execmem.h>
>>>>    #include <linux/extable.h>
>>>>    #include <linux/kasan.h>
>>>>    #include <linux/kernel.h>
>>>> @@ -41,6 +42,17 @@ DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
>>>>    static void __kprobes
>>>>    post_kprobe_handler(struct kprobe *, struct kprobe_ctlblk *, struct pt_regs *);
>>>> +void *alloc_insn_page(void)
>>>> +{
>>>> +	void *addr;
>>>> +
>>>> +	addr = execmem_alloc(EXECMEM_KPROBES, PAGE_SIZE);
>>>> +	if (!addr)
>>>> +		return NULL;
>>>> +	set_memory_rox((unsigned long)addr, 1);
>>>> +	return addr;
>>>> +}
>>> Why isn't execmem taking care of this? It looks to me like the
>>> execmem_cache_alloc() path calls set_memory_rox() but the
>>> execmem_vmalloc() path doesn't?
> execmem_alloc() -> execmem_vmalloc() consolidated __vmalloc_node_range()
> for executable allocations. Those also didn't update the linear map alias.
>
> It could be added to execmem_vmalloc(), but as of now we don't have a way
> for generic code to tell which set_memory method to call based on pgprot,
> so making execmem_vmalloc() to deal with direct map alias is quite
> involved.
>
> It would be easier to just remove the direct map alias. It works on x86 so
> I don't see what can possibly go wrong :)
>   
>> execmem_cache_alloc() is just called if execmem ROX cache is enabled, but it
>> currently just supported by x86. Included Mike to this thread who is the
>> author of execmem ROX cache.
>>
>>> It feels a bit bizarre to me that we have to provide our own wrapper
>>> (which is identical to what s390 does). Also, how does alloc_insn_page()
>>> handle the direct map alias on x86?
> s390 had its version of alloc_insn_page() long before execmem so there I
> just replaced module_alloc() with exemem_alloc().
>
> arm64 version of alloc_insn_page() didn't update the direct map before
> execmem, so I overlooked this issue when I was converting arm64 to execmem.

Thanks, Mike. It is not your fault. The set_memory_rox() was called in 
arm64's alloc_insn_page() before, but it was mistakenly removed by 
commit 10d5e97c1bf8 ("arm64: use PAGE_KERNEL_ROX directly in 
alloc_insn_page") before execmem.

Yang

>   
>> x86 handles it via execmem ROX cache.
>>
>> Thanks,
>> Yang
>>
>>> Will