The recent addition of CXL Region setup to the 6.0-rc1 Linux kernel has allowed us to test a few paths that weren't exercised fully until now. That threw up a mixture of bugs in QEMU emulation and the kernel (kernel fixes already posted).
The first patch is down to a wrong assumption about RO MemoryRegions and was hit due to a kernel bug overruning the mailbox and into another register region. Kernel bug fix: https://lore.kernel.org/linux-cxl/20220815154044.24733-1-jonathan.came...@huawei.com The later 2 have shown up because the Get / Set LSA commands have started getting proper user in the kernel. With these fixes in place I've tested an 8 way interleave on a host bridge to directly connected CXL devices and zeroed LSA regions. On that we can bring up a CXL region on which we can create an fsdax namespace and make a filesystem. For now, the cookie used to identify device for re establishing that region/namespace will be 0. A new feature patch will follow that will add the ability to provide the necessary serial numbers for correct namespace cookie creation allowing the namespace/filesystem to be remounted after a full restart. Jonathan Cameron (3): hw/cxl: Add stub write function for RO MemoryRegionOps entries. hw/cxl: Fix Get LSA input payload size which should be 8 bytes. hw/cxl: Correctly handle variable sized mailbox input payloads. hw/cxl/cxl-device-utils.c | 12 +++++++++--- hw/cxl/cxl-mailbox-utils.c | 4 ++-- 2 files changed, 11 insertions(+), 5 deletions(-) -- 2.32.0
