Re: [RFC PATCH] mm: Enable CONFIG_PT_RECLAIM on all architectures

[Qi Zheng <zhengqi.arch@bytedance.com>]

On 11/4/25 12:02 PM, Dev Jain wrote:
> 
> On 03/11/25 2:37 pm, Qi Zheng wrote:
>> Hi Dev,
>>
>> On 11/3/25 4:43 PM, Dev Jain wrote:
>>>
>>> On 03/11/25 12:33 pm, Qi Zheng wrote:
>>>> Hi Dev,
>>>>
>>>> On 11/3/25 2:37 PM, Dev Jain wrote:
>>>>> The implementation of CONFIG_PT_RECLAIM is completely contained in 
>>>>> generic
>>>>> mm code. It depends on the RCU callback which will reclaim the 
>>>>> pagetables -
>>>>> there is nothing arch-specific about that. So, enable this config for
>>>>> all architectures.
>>>>
>>>> Thanks for doing this!
>>>>
>>>> But unfortunately, not all architectures call tlb_remove_ptdesc() in
>>>> __pte_free_tlb(). Some architectures directly call pte_free() to
>>>> free PTE pages (without RCU).
>>>
>>> Thanks! This was not obvious to figure out.
>>>
>>> Is there an arch bottleneck because of which they do this? I mean to 
>>> say,
>>>
>>> is something stopping us from simply redirecting __pte_free_tlb to 
>>> tlb_remove_ptdesc
>>
>> Some architectures have special handling in __pte_free_tlb(), and cannot
>> simple redirect __pte_free_tlb() to tlb_remove_ptdesc(), such as m68k,
>> powerpc, etc.
>>
>> For those architectures that call pte_free() in __pte_free_tlb(), it
>> should be easy to modify them.
>>
>> If you're not in a rush, I can take the time to finish the above tasks.
> 
> Right then, I'll leave that up to you!

OK, I will do it ASAP.

> 
> 
>>
>>>
>>> or pte_free_defer?
>>>
>>>
>>> I am looking to enable this config at least on arm64 by default, I 
>>> believe it will be legal
>>>
>>> to do this at least here.
>>
>> IIRC, arm64 can directly enable CONFIG_PT_RECLAIM, as it is supported
>> at the architecture level.
>>
>> Thanks,
>> Qi
>>
>>>
>>>
>>>>
>>>> We need to modify these architectures first, otherwise it will
>>>> lead to UAF. This approach is feasible because Hugh provides similar
>>>> support in pte_free_defer().
>>>>
>>>> Enabling PT_RECLAIM on all architecture has always been on my
>>>> TODO list, but it's been blocked by other things. :(
>>>>
>>>> Thanks,
>>>> Qi
>>>>
>>>>>
>>>>> Signed-off-by: Dev Jain <dev.jain@arm.com>
>>>>> ---
>>>>>   arch/x86/Kconfig | 1 -
>>>>>   mm/Kconfig       | 5 +----
>>>>>   mm/pt_reclaim.c  | 2 +-
>>>>>   3 files changed, 2 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>>>>> index fa3b616af03a..5681308a5650 100644
>>>>> --- a/arch/x86/Kconfig
>>>>> +++ b/arch/x86/Kconfig
>>>>> @@ -327,7 +327,6 @@ config X86
>>>>>       select FUNCTION_ALIGNMENT_4B
>>>>>       imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
>>>>>       select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
>>>>> -    select ARCH_SUPPORTS_PT_RECLAIM        if X86_64
>>>>>       select ARCH_SUPPORTS_SCHED_SMT        if SMP
>>>>>       select SCHED_SMT            if SMP
>>>>>       select ARCH_SUPPORTS_SCHED_CLUSTER    if SMP
>>>>> diff --git a/mm/Kconfig b/mm/Kconfig
>>>>> index 0e26f4fc8717..903c37d02555 100644
>>>>> --- a/mm/Kconfig
>>>>> +++ b/mm/Kconfig
>>>>> @@ -1355,13 +1355,10 @@ config ARCH_HAS_USER_SHADOW_STACK
>>>>>         The architecture has hardware support for userspace shadow 
>>>>> call
>>>>>             stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
>>>>>   -config ARCH_SUPPORTS_PT_RECLAIM
>>>>> -    def_bool n
>>>>> -
>>>>>   config PT_RECLAIM
>>>>>       bool "reclaim empty user page table pages"
>>>>>       default y
>>>>> -    depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
>>>>> +    depends on MMU && SMP
>>>>>       select MMU_GATHER_RCU_TABLE_FREE
>>>>>       help
>>>>>         Try to reclaim empty user page table pages in paths other 
>>>>> than munmap
>>>>> diff --git a/mm/pt_reclaim.c b/mm/pt_reclaim.c
>>>>> index 7e9455a18aae..049e17f08c6a 100644
>>>>> --- a/mm/pt_reclaim.c
>>>>> +++ b/mm/pt_reclaim.c
>>>>> @@ -1,6 +1,6 @@
>>>>>   // SPDX-License-Identifier: GPL-2.0
>>>>>   #include <linux/hugetlb.h>
>>>>> -#include <asm-generic/tlb.h>
>>>>> +#include <asm/tlb.h>
>>>>>   #include <asm/pgalloc.h>
>>>>>     #include "internal.h"
>>>>
>>