Re: [PATCH v5.10.y] xen: replace xen_remap() with memremap()

["Teddy Astie" <teddy.astie@xxxxxxxxxx>]

Le 02/09/2025 à 13:18, Greg Kroah-Hartman a écrit :
> On Tue, Sep 02, 2025 at 09:28:32AM +0000, Teddy Astie wrote:
>> From: Juergen Gross <jgross@xxxxxxxx>
>>
>> From: Juergen Gross <jgross@xxxxxxxx>
>>
>> [ upstream commit 41925b105e345ebc84cedb64f59d20cb14a62613 ]
>>
>> xen_remap() is used to establish mappings for frames not under direct
>> control of the kernel: for Xenstore and console ring pages, and for
>> grant pages of non-PV guests.
>>
>> Today xen_remap() is defined to use ioremap() on x86 (doing uncached
>> mappings), and ioremap_cache() on Arm (doing cached mappings).
>>
>> Uncached mappings for those use cases are bad for performance, so they
>> should be avoided if possible. As all use cases of xen_remap() don't
>> require uncached mappings (the mapped area is always physical RAM),
>> a mapping using the standard WB cache mode is fine.
>>
>> As sparse is flagging some of the xen_remap() use cases to be not
>> appropriate for iomem(), as the result is not annotated with the
>> __iomem modifier, eliminate xen_remap() completely and replace all
>> use cases with memremap() specifying the MEMREMAP_WB caching mode.
>>
>> xen_unmap() can be replaced with memunmap().
>>
>> Reported-by: kernel test robot <lkp@xxxxxxxxx>
>> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
>> Reviewed-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
>> Acked-by: Stefano Stabellini <sstabellini@xxxxxxxxxx>
>> Link: https://lore.kernel.org/r/20220530082634.6339-1-jgross@xxxxxxxx
>> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
>> Signed-off-by: Teddy Astie <teddy.astie@xxxxxxxxxx> [backport to 5.10.y]
>> ---
>
> Why is this needed for 5.10.y at all?  What bug does it fix?  And why
> are you still using Xen on a 5.10.y kernel?  What prevents you from
> moving to a newer one?
>

This patch is only useful for virtual machines (DomU) that runs this
Linux version (a notable Linux distribution with this kernel branch is
Debian 11); it's not useful for Dom0 kernels.

On AMD platforms (and future Intel ones with TME); this patch along with
[1] makes the caching attribute for access as WB instead of falling back
to UC due to ioremap (within xen_remap) being used improving the
performance as explained in the commit.

[1] x86/hvmloader: select xen platform pci MMIO BAR UC or WB MTRR cache
attribute
https://xenbits.xen.org/gitweb/?p=xen.git;a=commit;h=22650d6054625be10172fe0c78b9cadd1a39bd63

> thanks,
>
> greg k-h
>

Teddy


--
Teddy Astie | Vates XCP-ng Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech