Message-ID: <ZGIETQZWp3scdS/m@MiWiFi-R3L-srv>
Date:   Mon, 15 May 2023 18:07:09 +0800
From:   Baoquan He <bhe@...hat.com>
To:     linux-kernel@...r.kernel.org
Cc:     catalin.marinas@....com, will@...nel.org, horms@...nel.org,
        thunder.leizhen@...wei.com, John.p.donnelly@...cle.com,
        kexec@...ts.infradead.org, linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH v6 1/2] arm64: kdump: simplify the reservation behaviour
 of crashkernel=,high

On 05/15/23 at 02:02pm, Baoquan He wrote:
> On arm64, reservation for 'crashkernel=xM,high' is taken by searching for
> suitable memory region top down. If the 'xM' of crashkernel high memory
> is reserved from high memory successfully, it will try to reserve
> crashkernel low memory later accoringly. Otherwise, it will try to search
> low memory area for the 'xM' suitable region. Please see the details in
> Documentation/admin-guide/kernel-parameters.txt.
> 
> While we observed an unexpected case where a reserved region crosses the
> high and low meomry boundary. E.g on a system with 4G as low memory end,
> user added the kernel parameters like: 'crashkernel=512M,high', it could
> finally have [4G-126M, 4G+386M], [1G, 1G+128M] regions in running kernel.
> The crashkernel high region crossing low and high memory boudary will bring
> issues:
> 
> 1) For crashkernel=x,high, if getting crashkernel high region across
> low and high memory boundary, then user will see two memory regions in
> low memory, and one memory region in high memory. The two crashkernel
> low memory regions are confusing as shown in above example.
> 
> 2) If people explicityly specify "crashkernel=x,high crashkernel=y,low"
> and y <= 128M, when crashkernel high region crosses low and high memory
> boundary and the part of crashkernel high reservation below boundary is
> bigger than y, the expected crahskernel low reservation will be skipped.
> But the expected crashkernel high reservation is shrank and could not
> satisfy user space requirement.
> 
> 3) The crossing boundary behaviour of crahskernel high reservation is
> different than x86 arch. On x86_64, the low memory end is 4G fixedly,
> and the memory near 4G is reserved by system, e.g for mapping firmware,
> pci mapping, so the crashkernel reservation crossing boundary never happens.
> From distros point of view, this brings inconsistency and confusion. Users
> need to dig into x86 and arm64 system details to find out why.
> 
> For kernel itself, the impact of issue 3) could be slight. While issue
> 1) and 2) cause actual impact because it brings obscure semantics and
> behaviour to crashkernel=,high reservation.
> 
> Here, for crashkernel=xM,high, search the high memory for the suitable
> region only in high memory. If failed, try reserving the suitable
> region only in low memory. Like this, the crashkernel high region will
> only exist in high memory, and crashkernel low region only exists in low
> memory. The reservation behaviour for crashkernel=,high is clearer and
> simpler.
> 
> Note: RPi4 has different zone ranges than normal memory. Its DMA zone is
> 0~1G, and DMA32 zone is 1G~4G if CONFIG_ZONE_DMA|DMA32 are enabled by
> default. The low memory end is 1G in order to validate all devices, high
> memory starts at 1G memory. However, for being consistent with normla
> arm64 system, its low memory end is still 1G, while reserving crashkernel
> high memory from 4G if crashkernel=size,high specified. This will remove
> confusion.
> 
> With above change applied, summary of arm64 crashkernel reservation range:
> 1)
> RPi4(zone DMA:0~1G; DMA32:1G~4G):
>  crashkernel=size
>   0~1G: low memory | 1G~top: high memory
> 
>  crashkernel=size,high
>   0~1G: low memory | 4G~top: high memory
> 
> 2)
> Other normal system:
>  crashkernel=size
>  crashkernel=size,high
>   0~4G: low memory | 4G~top: high memory
> 
> 3)
> Systems w/o zone DMA|DMA32
>  crashkernel=size
>  crashkernel=size,high
>   0~top: low memory
> 
> Signed-off-by: Baoquan He <bhe@...hat.com>
> 
> arm64: kdump: fix warning reported by static checker
> Signed-off-by: Baoquan He <bhe@...hat.com>

Sorry, forgot cleaning up this relic of local patch merging, have resent
one to remove it, and add Catalin's Reviewed-by tag.

Thanks
Baoquan

Powered by blists - more mailing lists