Re: ndctl cxl test suite fails in arm64 QEMU

[Itaru Kitayama <itaru.kitayama@linux.dev>]

On Tue, Feb 25, 2025 at 02:40:25PM -0800, Alison Schofield wrote:
> On Tue, Feb 25, 2025 at 01:09:57PM +0900, Itaru Kitayama wrote:
> > Hi,
> > 
> > Has anyone noticed the ndctl cxl test suite failures I reported below on arm64, QEMU emulation?
> > 
> > https://github.com/pmem/ndctl/issues/278
> > 
> > I’m using Jonathan’s latest CXL capable QEMU [1], and the latest CXL kernel [2].
> 
> Hi Itaru,
> 
> Looking at region.c:size_store() it seems that alloc_hpa() is failing.
> (kstrtou64 is arch independent and it works for me with same script &
> value)
> 
> Check the dmesg log at the time of the failure. alloc_hpa() may be
> emitting a message.  (In the run_qemu.sh cmdline add '--cxl-debug')

dynamic debug reported this (kernel is without kaslr, today's cxl/next):

[  193.267649] device: 'region4': device_add
[  193.267931] bus: 'cxl': add device region4
[  193.268284] cxl region4: bus: 'cxl': __driver_probe_device: matched device with driver cxl_region
[  193.268333] cxl region4: bus: 'cxl': really_probe: probing driver cxl_region with device
[  193.268401] cxl_region region4: no default pinctrl state
[  193.268482] cxl_region region4: probe with driver cxl_region rejects match -6

does the above show the region4 was not added properly? The sysfs
entries are there after the out of range failure.

Itaru.

> 
> Sounds like Marc is on to something and maybe the KASLR is known
> to cause this problem.
> 
> Thanks for posting to this list and send any further questions or
> comments. Happy to have you using run_qemu.sh and the cxl tests!
> 
> Alison
> 
> 
> > 
> > [1] https://gitlab.com/jic23/qemu/ cxl-2025-02-20
> > [2] https://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl.git/
> > 
> > Thanks,
> > Itaru.