Re: [PATCH] crash: Support high memory reservation for range syntax

[Youling Tang <youling.tang@linux.dev>]

On 4/15/26 19:29, Baoquan He wrote:

> On 04/09/26 at 09:55am, Youling Tang wrote:
>> Hi, Baoquan
>>
>> On 4/8/26 21:32, Baoquan He wrote:
>>> On 04/08/26 at 10:01am, Sourabh Jain wrote:
>>>> Hello Youling,
>>>>
>>>> On 04/04/26 13:11, Youling Tang wrote:
>>>>> From: Youling Tang <tangyouling@kylinos.cn>
>>>>>
>>>>> The crashkernel range syntax (range1:size1[,range2:size2,...]) allows
>>>>> automatic size selection based on system RAM, but it always reserves
>>>>> from low memory. When a large crashkernel is selected, this can
>>>>> consume most of the low memory, causing subsequent hardware
>>>>> hotplug or drivers requiring low memory to fail due to allocation
>>>>> failures.
>>>> Support for high crashkernel reservation has been added to
>>>> address the above problem.
>>>>
>>>> However, high crashkernel reservation is not supported with
>>>> range-based crashkernel kernel command-line arguments.
>>>> For example: crashkernel=0M-1G:100M,1G-4G:160M,4G-8G:192M
>>>>
>>>> Many users, including some distributions, use range-based
>>>> crashkernel configuration. So, adding support for high crashkernel
>>>> reservation with range-based configuration would be useful.
>>> Sorry for late response. And I have to say sorry because I have some
>>> negative tendency on this change.
>>>
>>> We use crashkernel=xM|G and crashkernel=range1:size1[,range2:size2,...]
>>> as default setting, so that people only need to set suggested amount
>>> of memory. While crashkernel=,high|low is for advanced user to customize
>>> their crashkernel value. In that case, user knows what's high memory and
>>> low memory, and how much is needed separately to achieve their goal, e.g
>>> saving low memory, taking away more high memory.
>>>
>>> To be honest, above grammers sounds simple, right? I believe both of you
>>> know very well how complicated the current crashkernel code is. I would
>>> suggest not letting them becomre more and more complicated by extending
>>> the grammer further and further. Unless you meet unavoidable issue with
>>> the existing grammer.
>>>
>>> Here comes my question, do you meet unavoidable issue with the existing
>>> grammer when you use crashkernel=range1:size1[,range2:size2,...] and
>>> think it's not satisfactory, and at the same time crashkernel=,high|low
>>> can't meet your demand either?
>> Yes, regular users generally don't know about high memory and low memory,
>> and probably don't know how much crashkernel memory should be reserved
>> either. They mostly just use the default crashkernel parameters configured
>> by the distribution.
>>
>> For advanced users, the current grammar is sufficient, because
>> 'crashkernel=<range1>:<size1>[,<range2>:<size2>,...][@offset],>boundary'
>> can definitely be replaced with 'crashkernel=size,high'.
>>
>> The main purpose of this patch is to provide distributions with a more
>> reasonable default parameter configuration (satisfying most requirements),
>> without having to set different distribution default parameters for
>> different
>> scenarios (physical machines, virtual machines) and different machine
>> models.
> OK, do you have a concrete case? e.g in your distros, what will you set
> with this patchset applied? Let's see if it can cover all cases with one
> simple and satisfying parameter.

For our production deployment across various hardware configurations
(physical servers, VMs with different memory sizes), I'm planning to
use the following crashkernel configuration:
crashkernel=1G-4G:256M,4G-12G:384M,12G-48G:512M,48G-128G:768M,128G-:1024M,>384M

Thanks,
Youling.