Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH 2/6] arm64/mm: Enable memory hot remove
To:     Steven Price <steven.price@arm.com>,
        Robin Murphy <robin.murphy@arm.com>,
        linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-mm@kvack.org, akpm@linux-foundation.org, will.deacon@arm.com,
        catalin.marinas@arm.com
Cc:     mark.rutland@arm.com, mhocko@suse.com, david@redhat.com,
        logang@deltatee.com, cai@lca.pw, pasha.tatashin@oracle.com,
        james.morse@arm.com, cpandya@codeaurora.org, arunks@codeaurora.org,
        dan.j.williams@intel.com, mgorman@techsingularity.net,
        osalvador@suse.de
References: <1554265806-11501-1-git-send-email-anshuman.khandual@arm.com>
 <1554265806-11501-3-git-send-email-anshuman.khandual@arm.com>
 <ed4ceac4-b92c-47f4-33b0-ed1d0833b40d@arm.com>
 <e191ddcb-271c-57f3-091f-eacaac2e86e0@arm.com>
From:   Anshuman Khandual <anshuman.khandual@arm.com>
Message-ID: <c52f0adf-5764-19b8-235e-4c37148388aa@arm.com>
Date:   Thu, 4 Apr 2019 12:21:28 +0530
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <e191ddcb-271c-57f3-091f-eacaac2e86e0@arm.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 04/03/2019 06:45 PM, Steven Price wrote:
> On 03/04/2019 13:37, Robin Murphy wrote:
>> [ +Steve ]
>>
>> Hi Anshuman,

Hi Steve,

>>
>> On 03/04/2019 05:30, Anshuman Khandual wrote:
> 
> <snip>
> 
>>> diff --git a/arch/arm64/include/asm/pgtable.h
>>> b/arch/arm64/include/asm/pgtable.h
>>> index de70c1e..858098e 100644
>>> --- a/arch/arm64/include/asm/pgtable.h
>>> +++ b/arch/arm64/include/asm/pgtable.h
>>> @@ -355,6 +355,18 @@ static inline int pmd_protnone(pmd_t pmd)
>>>   }
>>>   #endif
>>>   +#if (CONFIG_PGTABLE_LEVELS > 2)
>>> +#define pmd_large(pmd)    (pmd_val(pmd) && !(pmd_val(pmd) &
>>> PMD_TABLE_BIT))
>>> +#else
>>> +#define pmd_large(pmd) 0
>>> +#endif
>>> +
>>> +#if (CONFIG_PGTABLE_LEVELS > 3)
>>> +#define pud_large(pud)    (pud_val(pud) && !(pud_val(pud) &
>>> PUD_TABLE_BIT))
>>> +#else
>>> +#define pud_large(pmd) 0
>>> +#endif
>>
>> These seem rather different from the versions that Steve is proposing in
>> the generic pagewalk series - can you reach an agreement on which
>> implementation is preferred?
> 
> Indeed this doesn't match the version in my series although is quite
> similar.
> 
> My desire is that p?d_large represents the hardware architectural
> definition of large page/huge page/section (pick your naming). Although
> now I look more closely this is actually broken in my series (I'll fix
> that up and send a new version shortly) - p?d_sect() is similarly
> conditional.
> 
> Is there a good reason not to use the existing p?d_sect() macros
> available on arm64?

Nothing specific. Now I just tried using pud|pmd_sect() which looks good on
multiple configs for 4K/16K/64K. Will migrate pmd|pud_large() to more arch
specific pmd|pud_sect() which would also help in staying clear from your
series.

> 
> I'm also surprised by the CONFIG_PGTABLE_LEVEL conditions as they don't
> match the existing conditions for p?d_sect(). Might be worth double
> checking it actually does what you expect.

Right they are bit different. Surely will check. But if pmd|pud_sect() works
out okay will probably go with it as its been there for sometime.