libnvdimm, pmem, pfn: make pmem_rw_bytes generic and refactor pfn setup
[cascardo/linux.git] / drivers / nvdimm / pmem.c
index 67d48e2..b5f81b0 100644 (file)
@@ -49,19 +49,6 @@ struct pmem_device {
        struct badblocks        bb;
 };
 
-static bool is_bad_pmem(struct badblocks *bb, sector_t sector, unsigned int len)
-{
-       if (bb->count) {
-               sector_t first_bad;
-               int num_bad;
-
-               return !!badblocks_check(bb, sector, len / 512, &first_bad,
-                               &num_bad);
-       }
-
-       return false;
-}
-
 static void pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset,
                unsigned int len)
 {
@@ -209,16 +196,40 @@ void pmem_release_disk(void *disk)
        put_disk(disk);
 }
 
-static struct pmem_device *pmem_alloc(struct device *dev,
-               struct resource *res, int id)
+static struct vmem_altmap *nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
+               struct resource *res, struct vmem_altmap *altmap);
+
+static int pmem_attach_disk(struct device *dev,
+               struct nd_namespace_common *ndns)
 {
+       struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
+       struct vmem_altmap __altmap, *altmap = NULL;
+       struct resource *res = &nsio->res;
+       struct nd_pfn *nd_pfn = NULL;
+       int nid = dev_to_node(dev);
+       struct nd_pfn_sb *pfn_sb;
        struct pmem_device *pmem;
+       struct resource pfn_res;
        struct request_queue *q;
+       struct gendisk *disk;
+       void *addr;
+
+       /* while nsio_rw_bytes is active, parse a pfn info block if present */
+       if (is_nd_pfn(dev)) {
+               nd_pfn = to_nd_pfn(dev);
+               altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap);
+               if (IS_ERR(altmap))
+                       return PTR_ERR(altmap);
+       }
+
+       /* we're attaching a block device, disable raw namespace access */
+       devm_nsio_disable(dev, nsio);
 
        pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
        if (!pmem)
-               return ERR_PTR(-ENOMEM);
+               return -ENOMEM;
 
+       dev_set_drvdata(dev, pmem);
        pmem->phys_addr = res->start;
        pmem->size = resource_size(res);
        if (!arch_has_wmb_pmem())
@@ -227,22 +238,31 @@ static struct pmem_device *pmem_alloc(struct device *dev,
        if (!devm_request_mem_region(dev, res->start, resource_size(res),
                                dev_name(dev))) {
                dev_warn(dev, "could not reserve region %pR\n", res);
-               return ERR_PTR(-EBUSY);
+               return -EBUSY;
        }
 
        q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev));
        if (!q)
-               return ERR_PTR(-ENOMEM);
+               return -ENOMEM;
+       pmem->pmem_queue = q;
 
        pmem->pfn_flags = PFN_DEV;
-       if (pmem_should_map_pages(dev)) {
-               pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res,
+       if (is_nd_pfn(dev)) {
+               addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter,
+                               altmap);
+               pfn_sb = nd_pfn->pfn_sb;
+               pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
+               pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res);
+               pmem->pfn_flags |= PFN_MAP;
+               res = &pfn_res; /* for badblocks populate */
+               res->start += pmem->data_offset;
+       } else if (pmem_should_map_pages(dev)) {
+               addr = devm_memremap_pages(dev, &nsio->res,
                                &q->q_usage_counter, NULL);
                pmem->pfn_flags |= PFN_MAP;
        } else
-               pmem->virt_addr = (void __pmem *) devm_memremap(dev,
-                               pmem->phys_addr, pmem->size,
-                               ARCH_MEMREMAP_PMEM);
+               addr = devm_memremap(dev, pmem->phys_addr,
+                               pmem->size, ARCH_MEMREMAP_PMEM);
 
        /*
         * At release time the queue must be dead before
@@ -250,23 +270,12 @@ static struct pmem_device *pmem_alloc(struct device *dev,
         */
        if (devm_add_action(dev, pmem_release_queue, q)) {
                blk_cleanup_queue(q);
-               return ERR_PTR(-ENOMEM);
+               return -ENOMEM;
        }
 
-       if (IS_ERR(pmem->virt_addr))
-               return (void __force *) pmem->virt_addr;
-
-       pmem->pmem_queue = q;
-       return pmem;
-}
-
-static int pmem_attach_disk(struct device *dev,
-               struct nd_namespace_common *ndns, struct pmem_device *pmem)
-{
-       struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
-       int nid = dev_to_node(dev);
-       struct resource bb_res;
-       struct gendisk *disk;
+       if (IS_ERR(addr))
+               return PTR_ERR(addr);
+       pmem->virt_addr = (void __pmem *) addr;
 
        blk_queue_make_request(pmem->pmem_queue, pmem_make_request);
        blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE);
@@ -291,20 +300,9 @@ static int pmem_attach_disk(struct device *dev,
        set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
                        / 512);
        pmem->pmem_disk = disk;
-       devm_exit_badblocks(dev, &pmem->bb);
        if (devm_init_badblocks(dev, &pmem->bb))
                return -ENOMEM;
-       bb_res.start = nsio->res.start + pmem->data_offset;
-       bb_res.end = nsio->res.end;
-       if (is_nd_pfn(dev)) {
-               struct nd_pfn *nd_pfn = to_nd_pfn(dev);
-               struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
-
-               bb_res.start += __le32_to_cpu(pfn_sb->start_pad);
-               bb_res.end -= __le32_to_cpu(pfn_sb->end_trunc);
-       }
-       nvdimm_badblocks_populate(to_nd_region(dev->parent), &pmem->bb,
-                       &bb_res);
+       nvdimm_badblocks_populate(to_nd_region(dev->parent), &pmem->bb, res);
        disk->bb = &pmem->bb;
        add_disk(disk);
        revalidate_disk(disk);
@@ -312,33 +310,8 @@ static int pmem_attach_disk(struct device *dev,
        return 0;
 }
 
-static int pmem_rw_bytes(struct nd_namespace_common *ndns,
-               resource_size_t offset, void *buf, size_t size, int rw)
-{
-       struct pmem_device *pmem = dev_get_drvdata(ndns->claim);
-
-       if (unlikely(offset + size > pmem->size)) {
-               dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
-               return -EFAULT;
-       }
-
-       if (rw == READ) {
-               unsigned int sz_align = ALIGN(size + (offset & (512 - 1)), 512);
-
-               if (unlikely(is_bad_pmem(&pmem->bb, offset / 512, sz_align)))
-                       return -EIO;
-               return memcpy_from_pmem(buf, pmem->virt_addr + offset, size);
-       } else {
-               memcpy_to_pmem(pmem->virt_addr + offset, buf, size);
-               wmb_pmem();
-       }
-
-       return 0;
-}
-
 static int nd_pfn_init(struct nd_pfn *nd_pfn)
 {
-       struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev);
        struct nd_namespace_common *ndns = nd_pfn->ndns;
        u32 start_pad = 0, end_trunc = 0;
        resource_size_t start, size;
@@ -404,7 +377,8 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
         * ->direct_access() to those that are included in the memmap.
         */
        start += start_pad;
-       npfns = (pmem->size - start_pad - end_trunc - SZ_8K) / SZ_4K;
+       size = resource_size(&nsio->res);
+       npfns = (size - start_pad - end_trunc - SZ_8K) / SZ_4K;
        if (nd_pfn->mode == PFN_MODE_PMEM)
                offset = ALIGN(start + SZ_8K + 64 * npfns, nd_pfn->align)
                        - start;
@@ -413,13 +387,13 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
        else
                return -ENXIO;
 
-       if (offset + start_pad + end_trunc >= pmem->size) {
+       if (offset + start_pad + end_trunc >= size) {
                dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
                                dev_name(&ndns->dev));
                return -ENXIO;
        }
 
-       npfns = (pmem->size - offset - start_pad - end_trunc) / SZ_4K;
+       npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
        pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
        pfn_sb->dataoff = cpu_to_le64(offset);
        pfn_sb->npfns = cpu_to_le64(npfns);
@@ -456,17 +430,14 @@ static unsigned long init_altmap_reserve(resource_size_t base)
        return reserve;
 }
 
-static int __nvdimm_namespace_attach_pfn(struct nd_pfn *nd_pfn)
+static struct vmem_altmap *__nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
+               struct resource *res, struct vmem_altmap *altmap)
 {
-       struct resource res;
-       struct request_queue *q;
-       struct pmem_device *pmem;
-       struct vmem_altmap *altmap;
-       struct device *dev = &nd_pfn->dev;
        struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
-       struct nd_namespace_common *ndns = nd_pfn->ndns;
+       u64 offset = le64_to_cpu(pfn_sb->dataoff);
        u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
        u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
+       struct nd_namespace_common *ndns = nd_pfn->ndns;
        struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
        resource_size_t base = nsio->res.start + start_pad;
        struct vmem_altmap __altmap = {
@@ -474,112 +445,75 @@ static int __nvdimm_namespace_attach_pfn(struct nd_pfn *nd_pfn)
                .reserve = init_altmap_reserve(base),
        };
 
-       pmem = dev_get_drvdata(dev);
-       pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
-       pmem->pfn_pad = start_pad + end_trunc;
+       memcpy(res, &nsio->res, sizeof(*res));
+       res->start += start_pad;
+       res->end -= end_trunc;
+
        nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode);
        if (nd_pfn->mode == PFN_MODE_RAM) {
-               if (pmem->data_offset < SZ_8K)
-                       return -EINVAL;
+               if (offset < SZ_8K)
+                       return ERR_PTR(-EINVAL);
                nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
                altmap = NULL;
        } else if (nd_pfn->mode == PFN_MODE_PMEM) {
-               nd_pfn->npfns = (pmem->size - pmem->pfn_pad - pmem->data_offset)
-                       / PAGE_SIZE;
+               nd_pfn->npfns = (resource_size(res) - offset) / PAGE_SIZE;
                if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
                        dev_info(&nd_pfn->dev,
                                        "number of pfns truncated from %lld to %ld\n",
                                        le64_to_cpu(nd_pfn->pfn_sb->npfns),
                                        nd_pfn->npfns);
-               altmap = & __altmap;
-               altmap->free = PHYS_PFN(pmem->data_offset - SZ_8K);
+               memcpy(altmap, &__altmap, sizeof(*altmap));
+               altmap->free = PHYS_PFN(offset - SZ_8K);
                altmap->alloc = 0;
        } else
-               return -ENXIO;
+               return ERR_PTR(-ENXIO);
 
-       /* establish pfn range for lookup, and switch to direct map */
-       q = pmem->pmem_queue;
-       memcpy(&res, &nsio->res, sizeof(res));
-       res.start += start_pad;
-       res.end -= end_trunc;
-       devm_remove_action(dev, pmem_release_queue, q);
-       devm_memunmap(dev, (void __force *) pmem->virt_addr);
-       pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &res,
-                       &q->q_usage_counter, altmap);
-       pmem->pfn_flags |= PFN_MAP;
-
-       /*
-        * At release time the queue must be dead before
-        * devm_memremap_pages is unwound
-        */
-       if (devm_add_action(dev, pmem_release_queue, q)) {
-               blk_cleanup_queue(q);
-               return -ENOMEM;
-       }
-       if (IS_ERR(pmem->virt_addr))
-               return PTR_ERR(pmem->virt_addr);
-
-       /* attach pmem disk in "pfn-mode" */
-       return pmem_attach_disk(dev, ndns, pmem);
+       return altmap;
 }
 
-static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
+/*
+ * Determine the effective resource range and vmem_altmap from an nd_pfn
+ * instance.
+ */
+static struct vmem_altmap *nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
+               struct resource *res, struct vmem_altmap *altmap)
 {
-       struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
        int rc;
 
        if (!nd_pfn->uuid || !nd_pfn->ndns)
-               return -ENODEV;
+               return ERR_PTR(-ENODEV);
 
        rc = nd_pfn_init(nd_pfn);
        if (rc)
-               return rc;
+               return ERR_PTR(rc);
+
        /* we need a valid pfn_sb before we can init a vmem_altmap */
-       return __nvdimm_namespace_attach_pfn(nd_pfn);
+       return __nvdimm_setup_pfn(nd_pfn, res, altmap);
 }
 
 static int nd_pmem_probe(struct device *dev)
 {
-       struct nd_region *nd_region = to_nd_region(dev->parent);
        struct nd_namespace_common *ndns;
-       struct nd_namespace_io *nsio;
-       struct pmem_device *pmem;
 
        ndns = nvdimm_namespace_common_probe(dev);
        if (IS_ERR(ndns))
                return PTR_ERR(ndns);
 
-       nsio = to_nd_namespace_io(&ndns->dev);
-       pmem = pmem_alloc(dev, &nsio->res, nd_region->id);
-       if (IS_ERR(pmem))
-               return PTR_ERR(pmem);
-
-       dev_set_drvdata(dev, pmem);
-       ndns->rw_bytes = pmem_rw_bytes;
-       if (devm_init_badblocks(dev, &pmem->bb))
-               return -ENOMEM;
-       nvdimm_badblocks_populate(nd_region, &pmem->bb, &nsio->res);
+       if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev)))
+               return -ENXIO;
 
-       if (is_nd_btt(dev)) {
-               /* btt allocates its own request_queue */
-               devm_remove_action(dev, pmem_release_queue, pmem->pmem_queue);
-               blk_cleanup_queue(pmem->pmem_queue);
+       if (is_nd_btt(dev))
                return nvdimm_namespace_attach_btt(ndns);
-       }
 
        if (is_nd_pfn(dev))
-               return nvdimm_namespace_attach_pfn(ndns);
+               return pmem_attach_disk(dev, ndns);
 
-       if (nd_btt_probe(dev, ndns, pmem) == 0
-                       || nd_pfn_probe(dev, ndns, pmem) == 0) {
-               /*
-                * We'll come back as either btt-pmem, or pfn-pmem, so
-                * drop the queue allocation for now.
-                */
+       /* if we find a valid info-block we'll come back as that personality */
+       if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0)
                return -ENXIO;
-       }
 
-       return pmem_attach_disk(dev, ndns, pmem);
+       /* ...otherwise we're just a raw pmem device */
+       return pmem_attach_disk(dev, ndns);
 }
 
 static int nd_pmem_remove(struct device *dev)