Received-SPF: pass (google.com: domain of linux-kernel+bounces-171653-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Message-ID: <e60295b6-caa1-40a3-bc37-ecae00902565@arm.com>
Date: Tue, 7 May 2024 16:06:27 +0100
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v4 1/4] arm64/mm: generalize PMD_PRESENT_INVALID for all
 levels
Content-Language: en-GB
To: Catalin Marinas <catalin.marinas@arm.com>
Cc: David Hildenbrand <david@redhat.com>, Will Deacon <will@kernel.org>,
 Joey Gouly <joey.gouly@arm.com>, Ard Biesheuvel <ardb@kernel.org>,
 Mark Rutland <mark.rutland@arm.com>,
 Anshuman Khandual <anshuman.khandual@arm.com>, Peter Xu <peterx@redhat.com>,
 Mike Rapoport <rppt@linux.ibm.com>, Shivansh Vij <shivanshvij@outlook.com>,
 linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org
References: <20240503144604.151095-1-ryan.roberts@arm.com>
 <20240503144604.151095-2-ryan.roberts@arm.com>
 <d8373b6d-ec81-4010-8e01-cfc7d7e219a7@redhat.com>
 <4b63f70c-8a48-4290-90c7-25627de0d52d@arm.com> <Zjo114ph-vAN-yPY@arm.com>
From: Ryan Roberts <ryan.roberts@arm.com>
In-Reply-To: <Zjo114ph-vAN-yPY@arm.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

On 07/05/2024 15:08, Catalin Marinas wrote:
> On Tue, May 07, 2024 at 01:34:36PM +0100, Ryan Roberts wrote:
>> On 07/05/2024 12:38, David Hildenbrand wrote:
>>> On 03.05.24 16:45, Ryan Roberts wrote:
>>>> --- a/arch/arm64/include/asm/pgtable-prot.h
>>>> +++ b/arch/arm64/include/asm/pgtable-prot.h
>>>> @@ -21,11 +21,11 @@
>>>> ï¿½ #define PTE_PROT_NONEï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ (_AT(pteval_t, 1) << 58) /* only when !PTE_VALID */
>>>> ï¿½ ï¿½ /*
>>>> - * This bit indicates that the entry is present i.e. pmd_page()
>>>> - * still points to a valid huge page in memory even if the pmd
>>>> - * has been invalidated.
>>>> + * PTE_PRESENT_INVALID=1 & PTE_VALID=0 indicates that the pte's fields should be
>>>> + * interpreted according to the HW layout by SW but any attempted HW access to
>>>> + * the address will result in a fault. pte_present() returns true.
>>>> ï¿½ï¿½ */
>>>> -#define PMD_PRESENT_INVALIDï¿½ï¿½ï¿½ (_AT(pteval_t, 1) << 59) /* only when !PMD_SECT_VALID */
>>>> +#define PTE_PRESENT_INVALIDï¿½ï¿½ï¿½ (_AT(pteval_t, 1) << 59) /* only when !PTE_VALID */
>>>
>>> Ah, so PTE_VALID == PMD_SECT_VALID. Would that also be a reasonable
>>> generalization independent of this? (or do we keep it as is because it's a HW def?)
>>
>> To be honest, I'm not sure of the history, but some things are implemented as
>> wrappers around pte functions and others are implemented specifically for
>> pmd/pud/etc.
> 
> There's also a bit of historical arm32 code moved over to arm64 when we
> did the port. On classic arm32 page tables, the ptes and pmds were
> pretty different.>
>> On arm64, block mappings (all levels except last level) have the same HW format
>> as page mappings (last level) except that bit 1 must be 0 for block and 1 for
>> page. And with this series, SW/non-present bits are all matching too. So my vote
>> would be to harmonise toward a single implementation in future (modulus the bit
>> 1 problem), which would include getting rid of things like PMD_SECT_VALID.
> 
> For PMD_SECT_VALID vs PTE_VALID, it's fine to only use the latter. The
> PMD_TABLE_BIT, however, only makes sense for p*d levels. I think we can
> get rid of all the PMD_SECT_* macros, just keeping PMD_TYPE_* and the
> table bit.

OK, perhaps I'll get around to sending a patch at some point.

> 
> For a bit of architecture history, the reason the pmd block entries have
> bit 1 clear while the ptes have it set is to allow recursive mappings
> where an entry in the pgd points to the pgd itself. The hardware page
> table walk would end on the pmd entry when accessed at the specific VA,
> giving quick access to the pte. The downside is wasting a bit of the VA
> space.

Ahh ok. I had to think about that for a while, but makes sense. Thanks!