Received-SPF: pass (google.com: domain of linux-kernel+bounces-9861-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Feedback-ID: ifd894703:Fastmail
Message-ID: <a25d3736-f6ca-478d-8e58-69a7c0e529f7@flygoat.com>
Date: Fri, 22 Dec 2023 16:01:43 +0000
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH 1/1] riscv: set ARCH_DMA_DEFAULT_COHERENT if
 RISCV_DMA_NONCOHERENT is not set
Content-Language: en-US
To: Maxim Kochetkov <fido_max@inbox.ru>, Conor Dooley <conor@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>, linux-riscv@lists.infradead.org,
 linux-kernel@vger.kernel.org, robh@kernel.org, mpe@ellerman.id.au,
 aou@eecs.berkeley.edu, palmer@dabbelt.com, paul.walmsley@sifive.com
References: <20231221185152.327231-1-fido_max@inbox.ru>
 <20231221-discount-decade-e306e5878c46@spud>
 <f31d929c-fa0a-4046-be05-38e92afa5d92@flygoat.com>
 <20231222041428.GA2803@lst.de>
 <c53c22dd-c482-4808-bdd7-e81c01c04f9e@inbox.ru>
 <20231222-outburst-spoiling-75082a7826dd@spud>
 <be10eb21-a0ef-416e-9c0c-d53326996fba@inbox.ru>
 <7bc138f7-d5e9-471a-ae82-d6c490645623@flygoat.com>
 <6fe34090-656d-491f-8173-217d20144166@inbox.ru>
From: Jiaxun Yang <jiaxun.yang@flygoat.com>
In-Reply-To: <6fe34090-656d-491f-8173-217d20144166@inbox.ru>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit


在 2023/12/22 15:53, Maxim Kochetkov 写道:
> 
> 
> On 22.12.2023 18:45, Jiaxun Yang wrote:
>>
>>
>> 在 2023/12/22 15:38, Maxim Kochetkov 写道:
>>>
>>>
>>> On 22.12.2023 17:54, Conor Dooley wrote:
>>>
>>>>> etc..) do not have this feature. These devices will use value from
>>>>> device_initialize(). And we have no possibility to change
>>>>> dma_default_coherent value by disabling ARCH_DMA_DEFAULT_COHERENT.
>>>>> Moreover, changing dma_default_coherent from false to true may cause
>>>>> regression for other devices.
>>>>
>>>> How can there be a regression when dma has been coherent by default for
>>>> the RISC-V kernel from day 1?
>>>
>>> Before ARCH_DMA_DEFAULT_COHERENT patch dma_default_coherent was used 
>>> unassigned as "false" in device_initialize():
>>> ..........
>>> #if defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE) || \
>>>     defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU) || \
>>>     defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU_ALL)
>>>     dev->dma_coherent = dma_default_coherent;
>>> #endif
>>> ..........
>>> And now it becomes "true". It may change behavior of other non-DT 
>>> drivers.
>> I don't see any problem here, default is default.
>> Actually leaving those device with  dev->dma_coherent = false is 
>> risky, because
>> we can't guarantee underlying cache flush functions are here.
>>
>> If a non-dt device do need to override it, it should be done in 
>> arch_setup_dma_ops.
> 
> But arch_setup_dma_ops() is called only from of_dma_configure_id() and 
> acpi_dma_configure_id(). So it works only for DT and ACPI devices. What 
> about platform_device?

Ah I see, that's the problem, in MIPS's use case all DMA capable devices
are following platform's default coherency. For RISC-V we assume all 
device are enabled by ACPI or DT.

Perhaps you can override it in driver, but that will make drivers 
platform dependent.

I'll leave this question to Christoph.

Thanks
- Jiaxun