linux-kernel - Re: [PATCH 03/11] MIPS: support RAM beyond 32-bit

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <878r89b4jh.fsf@BL-laptop>
Date:   Wed, 11 Oct 2023 16:46:26 +0200
From:   Gregory CLEMENT <gregory.clement@...tlin.com>
To:     Jiaxun Yang <jiaxun.yang@...goat.com>,
        "paulburton@...nel.org" <paulburton@...nel.org>,
        Thomas Bogendoerfer <tsbogend@...ha.franken.de>,
        "linux-mips@...r.kernel.org" <linux-mips@...r.kernel.org>,
        Rob Herring <robh+dt@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org
Cc:     Vladimir Kondratiev <vladimir.kondratiev@...el.com>,
        Tawfik Bayouk <tawfik.bayouk@...ileye.com>,
        Alexandre Belloni <alexandre.belloni@...tlin.com>,
        Théo Lebrun <theo.lebrun@...tlin.com>,
        Thomas Petazzoni <thomas.petazzoni@...tlin.com>
Subject: Re: [PATCH 03/11] MIPS: support RAM beyond 32-bit

Hello Jiaxun,

> 在2023年10月9日十月 下午4:59，Gregory CLEMENT写道：
>> Hello Jiaxun,
>>
>>> 在2023年10月4日十月 下午5:10，Gregory CLEMENT写道：
>>>> From: Vladimir Kondratiev <vladimir.kondratiev@...el.com>
>>>>
>>>> Support platforms where RAM is mapped beyond 32-bit.
>>>>
>>>> The kernel parameter ddr32_alias allows to setup the alias to point
>>>> outside the first 4 GB of memory.
>>>
>>> Are you trying to fix the problem that if kernel text is loaded in
>>> XKPHYS there is no way to to set EBASE to that region?
>>
>> Yes that exactly we try to fix.
>>
>>>
>>> The common practice for other 64bit MIPS system is to load kernel
>>> in KSEG0 and add low 4G mirror with rest of the high memory to buddy
>>> system. By doing this Kernel still have access to all memory beyond
>>> 32 bit, the only draw back is Kernel's text and data can't be relocted
>>> beyond 32-bit.
>>>
>>> Loading kernel into KSEG0 (i.e. with KBUILD_SYM32) have significant benefit
>>> on performance, so I think you shouldn't try to load kernel into XKPHYS
>>> without a good reason, but it might be helpful to add a BUG_ON at
>>> CPS driver to handle such situation.
>>
>> I guess that being in KSEG0 allows to use shorter pointer.  But in our
>> case the RAM is physically connected beyond 32bits, so it is not
>> accessible in KSEG0.
>
> For most system there should be a mirror of part of DDR which is accessible
> at KSEG0 and kernel runs from here. As per my interpretion of your code EyeQ5
> is also doing this? If not could you please briefly describe the memory map?
>
> For Kernel in KSEG0 the pointer is still 64bit but we can use fewer inst
> to load ABS pointer into register, see [1].
>

There is a kind of mirror but its physical address start at 0x8000000
so beyond the first 512MBytes that are used for KSEG0.

In short the 32bits mapping is the following:

 - the controllers registers of the SoC are located  until 0x8000000,
 - then from 0x8000000 to 0x10000000 there is the alias to low addresses
   of the DDR
 - then the SPIflash is mapped to from 0x10000000 to 0x20000000
 - after the PCIe Memory 32-bit addr space is from 0x20000000 to
   0x40000000

Gregory

> [1]: https://elinux.org/images/1/1f/New-tricks-mips-linux.pdf

-- 
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com