mm/devm_memremap_pages: fix final page put race

Logan noticed that devm_memremap_pages_release() kills the percpu_ref
drops all the page references that were acquired at init and then
immediately proceeds to unplug, arch_remove_memory(), the backing pages
for the pagemap. If for some reason device shutdown actually collides
with a busy / elevated-ref-count page then arch_remove_memory() should
be deferred until after that reference is dropped.

As it stands the "wait for last page ref drop" happens *after*
devm_memremap_pages_release() returns, which is obviously too late and
can lead to crashes.

Fix this situation by assigning the responsibility to wait for the
percpu_ref to go idle to devm_memremap_pages() with a new ->cleanup()
callback. Implement the new cleanup callback for all
devm_memremap_pages() users: pmem, devdax, hmm, and p2pdma.

Link: http://lkml.kernel.org/r/[e-mail]
Fixes: 41e94a851304 ("add devm_memremap_pages")
Signed-off-by: Dan Williams <[e-mail]>
Reported-by: Logan Gunthorpe <[e-mail]>
Reviewed-by: Ira Weiny <[e-mail]>
Reviewed-by: Logan Gunthorpe <[e-mail]>
Cc: Bjorn Helgaas <[e-mail]>
Cc: "Jérôme Glisse" <[e-mail]>
Cc: Christoph Hellwig <[e-mail]>
Cc: Greg Kroah-Hartman <[e-mail]>
Cc: "Rafael J. Wysocki" <[e-mail]>
Signed-off-by: Andrew Morton <[e-mail]>
Signed-off-by: Linus Torvalds <[e-mail]>

drivers/dax/device.c

···
 	complete(&dev_dax->cmp);
 }
 
-static void dev_dax_percpu_exit(void *data)
+static void dev_dax_percpu_exit(struct percpu_ref *ref)
 {
-	struct percpu_ref *ref = data;
 	struct dev_dax *dev_dax = ref_to_dev_dax(ref);
 
 	dev_dbg(&dev_dax->dev, "%s\n", __func__);
···
 	if (rc)
 		return rc;
 
-	rc = devm_add_action_or_reset(dev, dev_dax_percpu_exit, &dev_dax->ref);
-	if (rc)
-		return rc;
-
 	dev_dax->pgmap.ref = &dev_dax->ref;
 	dev_dax->pgmap.kill = dev_dax_percpu_kill;
+	dev_dax->pgmap.cleanup = dev_dax_percpu_exit;
 	addr = devm_memremap_pages(dev, &dev_dax->pgmap);
-	if (IS_ERR(addr)) {
-		devm_remove_action(dev, dev_dax_percpu_exit, &dev_dax->ref);
-		percpu_ref_exit(&dev_dax->ref);
+	if (IS_ERR(addr))
 		return PTR_ERR(addr);
-	}
 
 	inode = dax_inode(dax_dev);
 	cdev = inode->i_cdev;

drivers/nvdimm/pmem.c

···
 	NULL,
 };
 
-static void pmem_release_queue(void *q)
+static void __pmem_release_queue(struct percpu_ref *ref)
 {
+	struct request_queue *q;
+
+	q = container_of(ref, typeof(*q), q_usage_counter);
 	blk_cleanup_queue(q);
+}
+
+static void pmem_release_queue(void *ref)
+{
+	__pmem_release_queue(ref);
 }
 
 static void pmem_freeze_queue(struct percpu_ref *ref)
···
 	if (!q)
 		return -ENOMEM;
 
-	if (devm_add_action_or_reset(dev, pmem_release_queue, q))
-		return -ENOMEM;
-
 	pmem->pfn_flags = PFN_DEV;
 	pmem->pgmap.ref = &q->q_usage_counter;
 	pmem->pgmap.kill = pmem_freeze_queue;
+	pmem->pgmap.cleanup = __pmem_release_queue;
 	if (is_nd_pfn(dev)) {
 		if (setup_pagemap_fsdax(dev, &pmem->pgmap))
 			return -ENOMEM;
···
 		pmem->pfn_flags |= PFN_MAP;
 		memcpy(&bb_res, &pmem->pgmap.res, sizeof(bb_res));
 	} else {
+		if (devm_add_action_or_reset(dev, pmem_release_queue,
+					&q->q_usage_counter))
+			return -ENOMEM;
 		addr = devm_memremap(dev, pmem->phys_addr,
 				pmem->size, ARCH_MEMREMAP_PMEM);
 		memcpy(&bb_res, &nsio->res, sizeof(bb_res));

drivers/pci/p2pdma.c

···
 	percpu_ref_kill(ref);
 }
 
-static void pci_p2pdma_percpu_cleanup(void *ref)
+static void pci_p2pdma_percpu_cleanup(struct percpu_ref *ref)
 {
 	struct p2pdma_pagemap *p2p_pgmap = to_p2p_pgmap(ref);
 
···
 	if (error)
 		goto pgmap_free;
 
-	/*
-	 * FIXME: the percpu_ref_exit needs to be coordinated internal
-	 * to devm_memremap_pages_release(). Duplicate the same ordering
-	 * as other devm_memremap_pages() users for now.
-	 */
-	error = devm_add_action(&pdev->dev, pci_p2pdma_percpu_cleanup,
-			&p2p_pgmap->ref);
-	if (error)
-		goto ref_cleanup;
-
 	pgmap = &p2p_pgmap->pgmap;
 
 	pgmap->res.start = pci_resource_start(pdev, bar) + offset;
···
 	pgmap->pci_p2pdma_bus_offset = pci_bus_address(pdev, bar) -
 		pci_resource_start(pdev, bar);
 	pgmap->kill = pci_p2pdma_percpu_kill;
+	pgmap->cleanup = pci_p2pdma_percpu_cleanup;
 
 	addr = devm_memremap_pages(&pdev->dev, pgmap);
 	if (IS_ERR(addr)) {
 		error = PTR_ERR(addr);
-		goto ref_exit;
+		goto pgmap_free;
 	}
 
 	error = gen_pool_add_owner(pdev->p2pdma->pool, (unsigned long)addr,
···
 
 pages_free:
 	devm_memunmap_pages(&pdev->dev, pgmap);
-ref_cleanup:
-	percpu_ref_exit(&p2p_pgmap->ref);
 pgmap_free:
 	devm_kfree(&pdev->dev, p2p_pgmap);
 	return error;

include/linux/memremap.h

···
  * @res: physical address range covered by @ref
  * @ref: reference count that pins the devm_memremap_pages() mapping
  * @kill: callback to transition @ref to the dead state
+ * @cleanup: callback to wait for @ref to be idle and reap it
  * @dev: host device of the mapping for debug
  * @data: private data pointer for page_free()
  * @type: memory type: see MEMORY_* in memory_hotplug.h
···
 	struct resource res;
 	struct percpu_ref *ref;
 	void (*kill)(struct percpu_ref *ref);
+	void (*cleanup)(struct percpu_ref *ref);
 	struct device *dev;
 	void *data;
 	enum memory_type type;

kernel/memremap.c

···
 	pgmap->kill(pgmap->ref);
 	for_each_device_pfn(pfn, pgmap)
 		put_page(pfn_to_page(pfn));
+	pgmap->cleanup(pgmap->ref);
 
 	/* pages are dead and unused, undo the arch mapping */
 	align_start = res->start & ~(SECTION_SIZE - 1);
···
  * 2/ The altmap field may optionally be initialized, in which case altmap_valid
  *    must be set to true
  *
- * 3/ pgmap->ref must be 'live' on entry and will be killed at
- *    devm_memremap_pages_release() time, or if this routine fails.
+ * 3/ pgmap->ref must be 'live' on entry and will be killed and reaped
+ *    at devm_memremap_pages_release() time, or if this routine fails.
  *
  * 4/ res is expected to be a host memory range that could feasibly be
  *    treated as a "System RAM" range, i.e. not a device mmio range, but
···
 	pgprot_t pgprot = PAGE_KERNEL;
 	int error, nid, is_ram;
 
-	if (!pgmap->ref || !pgmap->kill)
+	if (!pgmap->ref || !pgmap->kill || !pgmap->cleanup) {
+		WARN(1, "Missing reference count teardown definition\n");
 		return ERR_PTR(-EINVAL);
+	}
 
 	align_start = res->start & ~(SECTION_SIZE - 1);
 	align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE)
···
 	if (conflict_pgmap) {
 		dev_WARN(dev, "Conflicting mapping in same section\n");
 		put_dev_pagemap(conflict_pgmap);
-		return ERR_PTR(-ENOMEM);
+		error = -ENOMEM;
+		goto err_array;
 	}
 
 	conflict_pgmap = get_dev_pagemap(PHYS_PFN(align_end), NULL);
 	if (conflict_pgmap) {
 		dev_WARN(dev, "Conflicting mapping in same section\n");
 		put_dev_pagemap(conflict_pgmap);
-		return ERR_PTR(-ENOMEM);
+		error = -ENOMEM;
+		goto err_array;
 	}
 
 	is_ram = region_intersects(align_start, align_size,
···
 	pgmap_array_delete(res);
  err_array:
 	pgmap->kill(pgmap->ref);
+	pgmap->cleanup(pgmap->ref);
+
 	return ERR_PTR(error);
 }
 EXPORT_SYMBOL_GPL(devm_memremap_pages);

mm/hmm.c

···
 	complete(&devmem->completion);
 }
 
-static void hmm_devmem_ref_exit(void *data)
+static void hmm_devmem_ref_exit(struct percpu_ref *ref)
 {
-	struct percpu_ref *ref = data;
 	struct hmm_devmem *devmem;
 
 	devmem = container_of(ref, struct hmm_devmem, ref);
···
 	if (ret)
 		return ERR_PTR(ret);
 
-	ret = devm_add_action_or_reset(device, hmm_devmem_ref_exit, &devmem->ref);
-	if (ret)
-		return ERR_PTR(ret);
-
 	size = ALIGN(size, PA_SECTION_SIZE);
 	addr = min((unsigned long)iomem_resource.end,
 		   (1UL << MAX_PHYSMEM_BITS) - 1);
···
 	devmem->pagemap.ref = &devmem->ref;
 	devmem->pagemap.data = devmem;
 	devmem->pagemap.kill = hmm_devmem_ref_kill;
+	devmem->pagemap.cleanup = hmm_devmem_ref_exit;
 
 	result = devm_memremap_pages(devmem->device, &devmem->pagemap);
 	if (IS_ERR(result))
···
 	if (ret)
 		return ERR_PTR(ret);
 
-	ret = devm_add_action_or_reset(device, hmm_devmem_ref_exit,
-			&devmem->ref);
-	if (ret)
-		return ERR_PTR(ret);
-
 	devmem->pfn_first = devmem->resource->start >> PAGE_SHIFT;
 	devmem->pfn_last = devmem->pfn_first +
 			   (resource_size(devmem->resource) >> PAGE_SHIFT);
···
 	devmem->pagemap.ref = &devmem->ref;
 	devmem->pagemap.data = devmem;
 	devmem->pagemap.kill = hmm_devmem_ref_kill;
+	devmem->pagemap.cleanup = hmm_devmem_ref_exit;
 
 	result = devm_memremap_pages(devmem->device, &devmem->pagemap);
 	if (IS_ERR(result))

tools/testing/nvdimm/test/iomap.c

···
 {
 	struct dev_pagemap *pgmap = _pgmap;
 
+	WARN_ON(!pgmap || !pgmap->ref || !pgmap->kill || !pgmap->cleanup);
 	pgmap->kill(pgmap->ref);
+	pgmap->cleanup(pgmap->ref);
 }
 
 void *__wrap_devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap)