The following lockdep splat results from acquiring the init_mutex in acpi_nfit_clear_to_send(): WARNING: possible circular locking dependency detected lt-daxdev-error/7216 is trying to acquire lock: 00000000f694db15 (&acpi_desc->init_mutex){+.+.}, at: acpi_nfit_clear_to_send+0x27/0x80 [nfit] but task is already holding lock: 00000000182298f2 (&nvdimm_bus->reconfig_mutex){+.+.}, at: __nd_ioctl+0x457/0x610 [libnvdimm] which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&nvdimm_bus->reconfig_mutex){+.+.}: nvdimm_badblocks_populate+0x41/0x150 [libnvdimm] nd_region_notify+0x95/0xb0 [libnvdimm] nd_device_notify+0x40/0x50 [libnvdimm] ars_complete+0x7f/0xd0 [nfit] acpi_nfit_scrub+0xbb/0x410 [nfit] process_one_work+0x22b/0x5c0 worker_thread+0x3c/0x390 kthread+0x11e/0x140 ret_from_fork+0x3a/0x50 -> #0 (&acpi_desc->init_mutex){+.+.}: __mutex_lock+0x83/0x980 acpi_nfit_clear_to_send+0x27/0x80 [nfit] __nd_ioctl+0x474/0x610 [libnvdimm] nd_ioctl+0xa4/0xb0 [libnvdimm] do_vfs_ioctl+0xa5/0x6e0 ksys_ioctl+0x70/0x80 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x60/0x210 entry_SYSCALL_64_after_hwframe+0x49/0xbe New infrastructure is needed to be able to perform this check without acquiring the lock. Fixes: 594861215c83 ("acpi, nfit: Further restrict userspace ARS start") Cc: Dave Jiang <dave.jiang(a)intel.com&gt; Signed-off-by: Dan Williams <dan.j.williams(a)intel.com&gt; --- drivers/acpi/nfit/core.c | 15 +++------------ 1 file changed, 3 insertions(+), 12 deletions(-) diff --git a/drivers/acpi/nfit/core.c b/drivers/acpi/nfit/core.c index 5970b8f5f768..14d9f5bea015 100644 --- a/drivers/acpi/nfit/core.c +++ b/drivers/acpi/nfit/core.c @@ -3341,8 +3341,6 @@ static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, unsigned int cmd) { struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc); - struct nfit_spa *nfit_spa; - int rc = 0; if (nvdimm) return 0; @@ -3355,17 +3353,10 @@ static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc, * just needs guarantees that any ARS it initiates are not * interrupted by any intervening start requests from userspace. */ - mutex_lock(&acpi_desc->init_mutex); - list_for_each_entry(nfit_spa, &acpi_desc->spas, list) - if (acpi_desc->scrub_spa - || test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state) - || test_bit(ARS_REQ_LONG, &nfit_spa->ars_state)) { - rc = -EBUSY; - break; - } - mutex_unlock(&acpi_desc->init_mutex); + if (work_busy(&acpi_desc->dwork.work)) + return -EBUSY; - return rc; + return 0; } int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,