cxl/pmem: Fix module reload vs workqueue state

A test of the form:

    while true; do modprobe -r cxl_pmem; modprobe cxl_pmem; done

May lead to a crash signature of the form:

    BUG: unable to handle page fault for address: ffffffffc0660030
    #PF: supervisor instruction fetch in kernel mode
    #PF: error_code(0x0010) - not-present page
    [..]
    Workqueue: cxl_pmem 0xffffffffc0660030
    RIP: 0010:0xffffffffc0660030
    Code: Unable to access opcode bytes at RIP 0xffffffffc0660006.
    [..]
    Call Trace:
     ? process_one_work+0x4ec/0x9c0
     ? pwq_dec_nr_in_flight+0x100/0x100
     ? rwlock_bug.part.0+0x60/0x60
     ? worker_thread+0x2eb/0x700

In that report the 0xffffffffc0660030 address corresponds to the former
function address of cxl_nvb_update_state() from a previous load of the
module, not the current address. Fix that by arranging for ->state_work
in the 'struct cxl_nvdimm_bridge' object to be reinitialized on cxl_pmem
module reload.

Details:

Recall that CXL subsystem wants to link a CXL memory expander device to
an NVDIMM sub-hierarchy when both a persistent memory range has been
registered by the CXL platform driver (cxl_acpi) *and* when that CXL
memory expander has published persistent memory capacity (Get Partition
Info). To this end the cxl_nvdimm_bridge driver arranges to rescan the
CXL bus when either of those conditions change. The helper
bus_rescan_devices() can not be called underneath the device_lock() for
any device on that bus, so the cxl_nvdimm_bridge driver uses a workqueue
for the rescan.

Typically a driver allocates driver data to hold a 'struct work_struct'
for a driven device, but for a workqueue that may run after ->remove()
returns, driver data will have been freed. The 'struct
cxl_nvdimm_bridge' object holds the state and work_struct directly.
Unfortunately it was only arranging for that infrastructure to be
initialized once per device creation rather than the necessary once per
workqueue (cxl_pmem_wq) creation.

Introduce is_cxl_nvdimm_bridge() and cxl_nvdimm_bridge_reset() in
support of invalidating stale references to a recently destroyed
cxl_pmem_wq.

Cc: <stable@vger.kernel.org>
Fixes: 8fdcb1704f61 ("cxl/pmem: Add initial infrastructure for pmem support")
Reported-by: Vishal Verma <vishal.l.verma@intel.com>
Tested-by: Vishal Verma <vishal.l.verma@intel.com>
Link: https://lore.kernel.org/r/163665474585.3505991.8397182770066720755.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

1 files changed, 7 insertions, 1 deletions

diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
index 2e402f0b2a10..b5fca97b0a07 100644
--- a/drivers/cxl/core/pmem.c
+++ b/drivers/cxl/core/pmem.c
@@ -51,10 +51,16 @@ struct cxl_nvdimm_bridge *to_cxl_nvdimm_bridge(struct device *dev)
 }
 EXPORT_SYMBOL_NS_GPL(to_cxl_nvdimm_bridge, CXL);
 
-__mock int match_nvdimm_bridge(struct device *dev, const void *data)
+bool is_cxl_nvdimm_bridge(struct device *dev)
 {
 	return dev->type == &cxl_nvdimm_bridge_type;
 }
+EXPORT_SYMBOL_NS_GPL(is_cxl_nvdimm_bridge, CXL);
+
+__mock int match_nvdimm_bridge(struct device *dev, const void *data)
+{
+	return is_cxl_nvdimm_bridge(dev);
+}
 
 struct cxl_nvdimm_bridge *cxl_find_nvdimm_bridge(struct cxl_nvdimm *cxl_nvd)
 {