ACK/Cmnt: [PATCH 0/1][B][D][E][Unstable] intel-lpss driver conflicts with write-combining MTRR region

Andrea Righi andrea.righi at
Fri Sep 27 14:36:21 UTC 2019

On Fri, Sep 27, 2019 at 04:07:44PM +0800, AceLan Kao wrote:
> BugLink:
> [Impact]
> The memory region intel-lpss-pci uses has been declared as
> write-combining
> [ 0.001728] 5 base 4000000000 mask 6000000000 write-combining
> This leads to the system hangs up during booting up.
> This is a BIOS issue, but there are some platforms on the market and
> users are struggling on booting up the machines. So, we may have to
> fix this in the kernel.
> [Fix]
> Tuowen Zhao(ztuowen at[1] provides a diff patch for intel-lpss
> driver to claim to use un-cacheable memory while calling
> __devm_ioremap(), and it works well. But it haven't been accepted by
> maintainer yet.

Minor syntax nitpick:


> To avoid the potential impact on other machines, I add a quirk to list
> the machines which has the write-combining area in MTRR which overlaps
> with the address that intel-lpss uses, only the machines in the list
> pass DEVM_IOREMAP_UC to __devm_ioremap().
> [Test]
> Verified on Dell XPS 13 7390 2-in-1
> [Regression Potential]
> Low, without this patch, the machine even can't boot. And the quirk only
> for specific machines, so the modification won't affect other machines.
> 1.

It seems reasonable to me considering that this change affects only the
specific hardware (and without it we can't even boot).

IIUC with a fixed BIOS this change is not required. I'm wondering if
this change may have a performance impact for those people that upgrade
the BIOS and may not require this quirk enabled. However, considering
that the scope of the change (low power subsystem devices), performance
shouldn't be extremely critical.

So, under all the assumptions above:

Acked-by: Andrea Righi <andrea.righi at>

More information about the kernel-team mailing list