Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <c20e0d82-0ac8-9e84-9ac5-941ef1893885@loongson.cn>
Date:   Thu, 8 Jun 2023 10:21:24 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.10.0
Subject: Re: [PATCH] LoongArch: let pmd_present return true when splitting pmd
Content-Language: en-US
To:     zhanghongchen@loongson.cn
Cc:     loongarch@lists.linux.dev, linux-kernel@vger.kernel.org,
        Mike Rapoport <rppt@kernel.org>,
        loongson-kernel@lists.loongnix.cn,
        Huacai Chen <chenhuacai@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Anshuman Khandual <anshuman.khandual@arm.com>,
        WANG Xuerui <kernel@xen0n.name>,
        David Hildenbrand <david@redhat.com>,
        Feiyang Chen <chenfeiyang@loongson.cn>,
        Jiaxun Yang <jiaxun.yang@flygoat.com>,
        Qi Zheng <zhengqi.arch@bytedance.com>
References: <20230524074132.10916-1-zhanghongchen@loongson.cn>
 <46ce41a3-16bd-6d92-c8a8-764283859190@loongson.cn>
 <fbe8a1ec-cc81-d4db-8c55-31a60fb9fb58@loongson.cn>
 <c628c83d-424d-5c0e-24ca-702252779940@loongson.cn>
From:   "bibo, mao" <maobibo@loongson.cn>
In-Reply-To: <c628c83d-424d-5c0e-24ca-702252779940@loongson.cn>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk


在 2023/6/8 09:31, Hongchen Zhang 写道:
> Hi Bibo & Anshuman,
> 
> On 2023/6/6 下午9:49, bibo, mao wrote:
>> I do not object to adding sw bit for _PAGE_PRESENT_INVALID,  only that can it use
>> bit12--bit20 since it is for pmd entry only?
> If we change from bit60 to bit12, pmd_pfn will return an incorrect pfn, and it seems that modifying pmd_pfn accordingly is not a good idea.
Good catch. Do you test it or just doubt about it by reviewing the code? bit 12 _PAGE_HGLOBAL_SHIFT is already used by pmd entry. There should be problem also.

if you worry about it, pmd_pfn should be something like x86
static inline pmdval_t pmd_pfn_mask(pmd_t pmd)
{
        if (native_pmd_val(pmd) & _PAGE_PSE)
                return PHYSICAL_PMD_PAGE_MASK;
        else
                return PTE_PFN_MASK;
}

However I think forcefully using bit  60 is not good way. It solves one issue and brings out another problem.

Regards
Bibo, Mao

> 
> Hi Anshuman,
>   What's your opinion about this suggestion? I think arm64 architecture  has similar problems. Why do you choose bit59 instead of bit12--bit20?
>>
>> Regards
>> Bibo, Mao
>>
>> 在 2023/6/6 19:38, Hongchen Zhang 写道:
>>> Hi,
>>>
>>> Gentle ping.
>>>
>>> On 2023/5/24 pm 3:41, Hongchen Zhang wrote:
>>>> when we split a pmd into ptes, pmd_present() and pmd_trans_huge() should
>>>> return true,otherwise it would be treated as a swap pmd.
>>>> As arm64 does in
>>>> commit b65399f6111b ("arm64/mm: Change THP helpers to comply with generic MM semantics")
>>>> we add a _PAGE_PRESENT_INVALID bit for LoongArch.
>>>>
>>>> Signed-off-by: Hongchen Zhang <zhanghongchen@loongson.cn>
>>>> ---
>>>>    arch/loongarch/include/asm/pgtable-bits.h | 2 ++
>>>>    arch/loongarch/include/asm/pgtable.h      | 3 ++-
>>>>    2 files changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/loongarch/include/asm/pgtable-bits.h b/arch/loongarch/include/asm/pgtable-bits.h
>>>> index 8b98d22a145b..a7469d28d9d0 100644
>>>> --- a/arch/loongarch/include/asm/pgtable-bits.h
>>>> +++ b/arch/loongarch/include/asm/pgtable-bits.h
>>>> @@ -22,12 +22,14 @@
>>>>    #define    _PAGE_PFN_SHIFT        12
>>>>    #define    _PAGE_SWP_EXCLUSIVE_SHIFT 23
>>>>    #define    _PAGE_PFN_END_SHIFT    48
>>>> +#define _PAGE_PRESENT_INVALID_SHIFT 60
>>>>    #define    _PAGE_NO_READ_SHIFT    61
>>>>    #define    _PAGE_NO_EXEC_SHIFT    62
>>>>    #define    _PAGE_RPLV_SHIFT    63
>>>>      /* Used by software */
>>>>    #define _PAGE_PRESENT        (_ULCAST_(1) << _PAGE_PRESENT_SHIFT)
>>>> +#define _PAGE_PRESENT_INVALID    (_ULCAST_(1) << _PAGE_PRESENT_INVALID_SHIFT)
>>>>    #define _PAGE_WRITE        (_ULCAST_(1) << _PAGE_WRITE_SHIFT)
>>>>    #define _PAGE_ACCESSED        (_ULCAST_(1) << _PAGE_ACCESSED_SHIFT)
>>>>    #define _PAGE_MODIFIED        (_ULCAST_(1) << _PAGE_MODIFIED_SHIFT)
>>>> diff --git a/arch/loongarch/include/asm/pgtable.h b/arch/loongarch/include/asm/pgtable.h
>>>> index d28fb9dbec59..9a9f9ff9b709 100644
>>>> --- a/arch/loongarch/include/asm/pgtable.h
>>>> +++ b/arch/loongarch/include/asm/pgtable.h
>>>> @@ -213,7 +213,7 @@ static inline int pmd_bad(pmd_t pmd)
>>>>    static inline int pmd_present(pmd_t pmd)
>>>>    {
>>>>        if (unlikely(pmd_val(pmd) & _PAGE_HUGE))
>>>> -        return !!(pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE));
>>>> +        return !!(pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PRESENT_INVALID));
>>>>          return pmd_val(pmd) != (unsigned long)invalid_pte_table;
>>>>    }
>>>> @@ -558,6 +558,7 @@ static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
>>>>      static inline pmd_t pmd_mkinvalid(pmd_t pmd)
>>>>    {
>>>> +    pmd_val(pmd) |= _PAGE_PRESENT_INVALID;
>>>>        pmd_val(pmd) &= ~(_PAGE_PRESENT | _PAGE_VALID | _PAGE_DIRTY | _PAGE_PROTNONE);
>>>>          return pmd;
>>>>
>>>> base-commit: f1fcbaa18b28dec10281551dfe6ed3a3ed80e3d6
>>>>
>>>
>>> Best Regards
>>> Hongchen Zhang
>>>
>>> _______________________________________________
>>> Loongson-kernel mailing list -- loongson-kernel@lists.loongnix.cn
>>> To unsubscribe send an email to loongson-kernel-leave@lists.loongnix.cn
>>
>>
> 
> _______________________________________________
> Loongson-kernel mailing list -- loongson-kernel@lists.loongnix.cn
> To unsubscribe send an email to loongson-kernel-leave@lists.loongnix.cn