On Tue, Jun 22, 2021 at 07:50:38PM +0800, Qu Wenruo wrote:

On 2021/6/22 下午7:14, David Sterba wrote:

quoted

On Fri, Jun 18, 2021 at 06:46:51AM +0800, Qu Wenruo wrote:

Although the biggest challenge for testing is the hardware.

I guess even with subpage merged into upstream in v5.14, it won't really
get many tests from real world, unlike x86_64 that every guys in the
mail list is testing on.

Although we have cheap ARM SoC boards in recent days, there aren't
really much good candidates for us to utilize.

Either they don't have fast enough CPUs (2x 2GHz+ A72 or more) or don't
have even a single PCIe lane for NVME, or don't have good enough
upstream kernel support.

So far RPI CM4 would be a pretty good candidate, and I'm already using
it, but without overclocking and good heatsink, its CPU can be a little
bit slow, and the PCIe2.0 x1 lane is far from enough.

But I totally understand how difficult it could be for even kernel
developers to setup all the small things up.

Like TTY, libvirt, edk2 firmware for VM, RPI specific boot sequence etc.

Thus even subpage get merged, I still think we need way more rounds of
upstream release to really get some feedback.

Back in January I got a hold of a decent arm64 machine and tested the 64K
build in a VM, the hardware should be good enough for running at least
3 in parallel and I think there were more identical physical machines
available so we can use that for testing once all the prep work is done.
Right now all people involved know that the subpage support may be buggy
or incomplete so we can rely on external testing for some time.