Re: git: 3473f2832284 - main - Switch the arm64 VM_MEMATTR_DEVICE to nGnRE

From: Mark Millard <marklmi_at_yahoo.com>
Date: Thu, 16 Mar 2023 18:53:29 UTC 
On Mar 16, 2023, at 09:53, Andrew Turner <andrew@fubar.geek.nz> wrote:

> I’ve pushed a change to relax the default device memory type on arm64. This is now equivalent to the device memory type used on arm.
> 
> Because of this I expect most devices to just work, however if you are seeing strange issues you can set kern.force_nonposted=1 in loader.conf to revert to the previous memory type globally, or hint.<device>.<unit>.force_nonposted=1 to disable it for a specific device. For example to disable it for xhci0 you would use hint.xhci.0.force_nonposted=1.
> 
> If you do find this fixes an issue please report it so we can fix the driver.

A probably unfortunate issue relative to some types of
testing is that drm-515-kmod and drm-510-kmod fail to
build for aarch64 (arm64). (drm-510-kmod might not be
reasonable for aarch64 for alll I know.)

The current problem for drm-515-kmod is lack of
in_interrupt via lack of sufficient
#include <linux/preempt.h> use (that provides the
macro definition involved):

error: call to undeclared function 'in_interrupt'

(I'm guessing that amd64 works only via some amd64
specific header polution that aarch64 does not get.)

My vague understanding of issues that exist in
Linux land may suggest that Cortex-A72's may have
additional issues shown via video card use. An
example is the video-related activity for the
HoneyComb to avoid the issues.

> Andrew
> 
>> Begin forwarded message:
>> 
>> From: Andrew Turner <andrew@FreeBSD.org>
>> Subject: git: 3473f2832284 - main - Switch the arm64 VM_MEMATTR_DEVICE to nGnRE
>> Date: 16 March 2023 at 16:46:11 GMT
>> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org
>> 
>> The branch main has been updated by andrew:
>> 
>> URL: https://cgit.FreeBSD.org/src/commit/?id=3473f283228401ba585044ac6dea8c8eb27237a5
>> 
>> commit 3473f283228401ba585044ac6dea8c8eb27237a5
>> Author:     Andrew Turner <andrew@FreeBSD.org>
>> AuthorDate: 2023-03-16 15:36:06 +0000
>> Commit:     Andrew Turner <andrew@FreeBSD.org>
>> CommitDate: 2023-03-16 16:45:42 +0000
>> 
>>    Switch the arm64 VM_MEMATTR_DEVICE to nGnRE
>> 
>>    Move device memory to a weaker type. The new device memory type allows
>>    the system to acknowledge a write to a device before the write has
>>    completed. This is inline with VM_MEMATTR_DEVICE on armv6/armv7.
>> 
>>    Sponsored by:   Arm Ltd
>>    Differential Revision:  https://reviews.freebsd.org/D38945
>> ---
>> sys/arm64/include/vm.h | 6 +-----
>> 1 file changed, 1 insertion(+), 5 deletions(-)
>> 
>> diff --git a/sys/arm64/include/vm.h b/sys/arm64/include/vm.h
>> index e479aab52e26..a8c1ea97b513 100644
>> --- a/sys/arm64/include/vm.h
>> +++ b/sys/arm64/include/vm.h
>> @@ -36,11 +36,7 @@
>> #define VM_MEMATTR_WRITE_THROUGH 3
>> #define VM_MEMATTR_DEVICE_nGnRE 4
>> 
>> -/*
>> - * VM_MEMATTR_DEVICE can be changed to VM_MEMATTR_DEVICE_nGnRE when
>> - * the PCI drivers use VM_MEMATTR_DEVICE_NP for their config space.
>> - */
>> -#define VM_MEMATTR_DEVICE VM_MEMATTR_DEVICE_nGnRnE
>> +#define VM_MEMATTR_DEVICE VM_MEMATTR_DEVICE_nGnRE
>> #define VM_MEMATTR_DEVICE_NP VM_MEMATTR_DEVICE_nGnRnE
>> 
>> #ifdef _KERNEL
>> 
> 

===
Mark Millard
marklmi at yahoo.com