CVE-2023-54323

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> cxl/pmem: Fix nvdimm registration races<br /> <br /> A loop of the form:<br /> <br /> while true; do modprobe cxl_pci; modprobe -r cxl_pci; done<br /> <br /> ...fails with the following crash signature:<br /> <br /> BUG: kernel NULL pointer dereference, address: 0000000000000040<br /> [..]<br /> RIP: 0010:cxl_internal_send_cmd+0x5/0xb0 [cxl_core]<br /> [..]<br /> Call Trace:<br /> <br /> cxl_pmem_ctl+0x121/0x240 [cxl_pmem]<br /> nvdimm_get_config_data+0xd6/0x1a0 [libnvdimm]<br /> nd_label_data_init+0x135/0x7e0 [libnvdimm]<br /> nvdimm_probe+0xd6/0x1c0 [libnvdimm]<br /> nvdimm_bus_probe+0x7a/0x1e0 [libnvdimm]<br /> really_probe+0xde/0x380<br /> __driver_probe_device+0x78/0x170<br /> driver_probe_device+0x1f/0x90<br /> __device_attach_driver+0x85/0x110<br /> bus_for_each_drv+0x7d/0xc0<br /> __device_attach+0xb4/0x1e0<br /> bus_probe_device+0x9f/0xc0<br /> device_add+0x445/0x9c0<br /> nd_async_device_register+0xe/0x40 [libnvdimm]<br /> async_run_entry_fn+0x30/0x130<br /> <br /> ...namely that the bottom half of async nvdimm device registration runs<br /> after the CXL has already torn down the context that cxl_pmem_ctl()<br /> needs. Unlike the ACPI NFIT case that benefits from launching multiple<br /> nvdimm device registrations in parallel from those listed in the table,<br /> CXL is already marked PROBE_PREFER_ASYNCHRONOUS. So provide for a<br /> synchronous registration path to preclude this scenario.