drivers/block/pmem: Add a driver for persistent memory
authorRoss Zwisler <ross.zwisler@linux.intel.com>
Wed, 1 Apr 2015 07:12:19 +0000 (09:12 +0200)
committerIngo Molnar <mingo@kernel.org>
Wed, 1 Apr 2015 15:03:56 +0000 (17:03 +0200)
PMEM is a new driver that presents a reserved range of memory as
a block device.  This is useful for developing with NV-DIMMs,
and can be used with volatile memory as a development platform.

This patch contains the initial driver from Ross Zwisler, with
various changes: converted it to use a platform_device for
discovery, fixed partition support and merged various patches
from Boaz Harrosh.

Tested-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Boaz Harrosh <boaz@plexistor.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matthew Wilcox <willy@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-nvdimm@ml01.01.org
Link: http://lkml.kernel.org/r/1427872339-6688-3-git-send-email-hch@lst.de
[ Minor cleanups. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
MAINTAINERS
drivers/block/Kconfig
drivers/block/Makefile
drivers/block/pmem.c [new file with mode: 0644]

index 1de6afa..4517613 100644 (file)
@@ -8071,6 +8071,12 @@ S:       Maintained
 F:     Documentation/blockdev/ramdisk.txt
 F:     drivers/block/brd.c
 
+PERSISTENT MEMORY DRIVER
+M:     Ross Zwisler <ross.zwisler@linux.intel.com>
+L:     linux-nvdimm@lists.01.org
+S:     Supported
+F:     drivers/block/pmem.c
+
 RANDOM NUMBER DRIVER
 M:     "Theodore Ts'o" <tytso@mit.edu>
 S:     Maintained
index 1b8094d..eb1fed5 100644 (file)
@@ -404,6 +404,17 @@ config BLK_DEV_RAM_DAX
          and will prevent RAM block device backing store memory from being
          allocated from highmem (only a problem for highmem systems).
 
+config BLK_DEV_PMEM
+       tristate "Persistent memory block device support"
+       help
+         Saying Y here will allow you to use a contiguous range of reserved
+         memory as one or more persistent block devices.
+
+         To compile this driver as a module, choose M here: the module will be
+         called 'pmem'.
+
+         If unsure, say N.
+
 config CDROM_PKTCDVD
        tristate "Packet writing on CD/DVD media"
        depends on !UML
index 02b688d..9cc6c18 100644 (file)
@@ -14,6 +14,7 @@ obj-$(CONFIG_PS3_VRAM)                += ps3vram.o
 obj-$(CONFIG_ATARI_FLOPPY)     += ataflop.o
 obj-$(CONFIG_AMIGA_Z2RAM)      += z2ram.o
 obj-$(CONFIG_BLK_DEV_RAM)      += brd.o
+obj-$(CONFIG_BLK_DEV_PMEM)     += pmem.o
 obj-$(CONFIG_BLK_DEV_LOOP)     += loop.o
 obj-$(CONFIG_BLK_CPQ_DA)       += cpqarray.o
 obj-$(CONFIG_BLK_CPQ_CISS_DA)  += cciss.o
diff --git a/drivers/block/pmem.c b/drivers/block/pmem.c
new file mode 100644 (file)
index 0000000..988f384
--- /dev/null
@@ -0,0 +1,263 @@
+/*
+ * Persistent Memory Driver
+ *
+ * Copyright (c) 2014, Intel Corporation.
+ * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>.
+ * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ */
+
+#include <asm/cacheflush.h>
+#include <linux/blkdev.h>
+#include <linux/hdreg.h>
+#include <linux/init.h>
+#include <linux/platform_device.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/slab.h>
+
+#define PMEM_MINORS            16
+
+struct pmem_device {
+       struct request_queue    *pmem_queue;
+       struct gendisk          *pmem_disk;
+
+       /* One contiguous memory region per device */
+       phys_addr_t             phys_addr;
+       void                    *virt_addr;
+       size_t                  size;
+};
+
+static int pmem_major;
+static atomic_t pmem_index;
+
+static void pmem_do_bvec(struct pmem_device *pmem, struct page *page,
+                       unsigned int len, unsigned int off, int rw,
+                       sector_t sector)
+{
+       void *mem = kmap_atomic(page);
+       size_t pmem_off = sector << 9;
+
+       if (rw == READ) {
+               memcpy(mem + off, pmem->virt_addr + pmem_off, len);
+               flush_dcache_page(page);
+       } else {
+               flush_dcache_page(page);
+               memcpy(pmem->virt_addr + pmem_off, mem + off, len);
+       }
+
+       kunmap_atomic(mem);
+}
+
+static void pmem_make_request(struct request_queue *q, struct bio *bio)
+{
+       struct block_device *bdev = bio->bi_bdev;
+       struct pmem_device *pmem = bdev->bd_disk->private_data;
+       int rw;
+       struct bio_vec bvec;
+       sector_t sector;
+       struct bvec_iter iter;
+       int err = 0;
+
+       if (bio_end_sector(bio) > get_capacity(bdev->bd_disk)) {
+               err = -EIO;
+               goto out;
+       }
+
+       BUG_ON(bio->bi_rw & REQ_DISCARD);
+
+       rw = bio_data_dir(bio);
+       sector = bio->bi_iter.bi_sector;
+       bio_for_each_segment(bvec, bio, iter) {
+               pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, bvec.bv_offset,
+                            rw, sector);
+               sector += bvec.bv_len >> 9;
+       }
+
+out:
+       bio_endio(bio, err);
+}
+
+static int pmem_rw_page(struct block_device *bdev, sector_t sector,
+                      struct page *page, int rw)
+{
+       struct pmem_device *pmem = bdev->bd_disk->private_data;
+
+       pmem_do_bvec(pmem, page, PAGE_CACHE_SIZE, 0, rw, sector);
+       page_endio(page, rw & WRITE, 0);
+
+       return 0;
+}
+
+static long pmem_direct_access(struct block_device *bdev, sector_t sector,
+                             void **kaddr, unsigned long *pfn, long size)
+{
+       struct pmem_device *pmem = bdev->bd_disk->private_data;
+       size_t offset = sector << 9;
+
+       if (!pmem)
+               return -ENODEV;
+
+       *kaddr = pmem->virt_addr + offset;
+       *pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT;
+
+       return pmem->size - offset;
+}
+
+static const struct block_device_operations pmem_fops = {
+       .owner =                THIS_MODULE,
+       .rw_page =              pmem_rw_page,
+       .direct_access =        pmem_direct_access,
+};
+
+static struct pmem_device *pmem_alloc(struct device *dev, struct resource *res)
+{
+       struct pmem_device *pmem;
+       struct gendisk *disk;
+       int idx, err;
+
+       err = -ENOMEM;
+       pmem = kzalloc(sizeof(*pmem), GFP_KERNEL);
+       if (!pmem)
+               goto out;
+
+       pmem->phys_addr = res->start;
+       pmem->size = resource_size(res);
+
+       err = -EINVAL;
+       if (!request_mem_region(pmem->phys_addr, pmem->size, "pmem")) {
+               dev_warn(dev, "could not reserve region [0x%llx:0x%zx]\n",
+                          pmem->phys_addr, pmem->size);
+               goto out_free_dev;
+       }
+
+       /*
+        * Map the memory as non-cachable, as we can't write back the contents
+        * of the CPU caches in case of a crash.
+        */
+       err = -ENOMEM;
+       pmem->virt_addr = ioremap_nocache(pmem->phys_addr, pmem->size);
+       if (!pmem->virt_addr)
+               goto out_release_region;
+
+       pmem->pmem_queue = blk_alloc_queue(GFP_KERNEL);
+       if (!pmem->pmem_queue)
+               goto out_unmap;
+
+       blk_queue_make_request(pmem->pmem_queue, pmem_make_request);
+       blk_queue_max_hw_sectors(pmem->pmem_queue, 1024);
+       blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY);
+
+       disk = alloc_disk(PMEM_MINORS);
+       if (!disk)
+               goto out_free_queue;
+
+       idx = atomic_inc_return(&pmem_index) - 1;
+
+       disk->major             = pmem_major;
+       disk->first_minor       = PMEM_MINORS * idx;
+       disk->fops              = &pmem_fops;
+       disk->private_data      = pmem;
+       disk->queue             = pmem->pmem_queue;
+       disk->flags             = GENHD_FL_EXT_DEVT;
+       sprintf(disk->disk_name, "pmem%d", idx);
+       disk->driverfs_dev = dev;
+       set_capacity(disk, pmem->size >> 9);
+       pmem->pmem_disk = disk;
+
+       add_disk(disk);
+
+       return pmem;
+
+out_free_queue:
+       blk_cleanup_queue(pmem->pmem_queue);
+out_unmap:
+       iounmap(pmem->virt_addr);
+out_release_region:
+       release_mem_region(pmem->phys_addr, pmem->size);
+out_free_dev:
+       kfree(pmem);
+out:
+       return ERR_PTR(err);
+}
+
+static void pmem_free(struct pmem_device *pmem)
+{
+       del_gendisk(pmem->pmem_disk);
+       put_disk(pmem->pmem_disk);
+       blk_cleanup_queue(pmem->pmem_queue);
+       iounmap(pmem->virt_addr);
+       release_mem_region(pmem->phys_addr, pmem->size);
+       kfree(pmem);
+}
+
+static int pmem_probe(struct platform_device *pdev)
+{
+       struct pmem_device *pmem;
+       struct resource *res;
+
+       if (WARN_ON(pdev->num_resources > 1))
+               return -ENXIO;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res)
+               return -ENXIO;
+
+       pmem = pmem_alloc(&pdev->dev, res);
+       if (IS_ERR(pmem))
+               return PTR_ERR(pmem);
+
+       platform_set_drvdata(pdev, pmem);
+
+       return 0;
+}
+
+static int pmem_remove(struct platform_device *pdev)
+{
+       struct pmem_device *pmem = platform_get_drvdata(pdev);
+
+       pmem_free(pmem);
+       return 0;
+}
+
+static struct platform_driver pmem_driver = {
+       .probe          = pmem_probe,
+       .remove         = pmem_remove,
+       .driver         = {
+               .owner  = THIS_MODULE,
+               .name   = "pmem",
+       },
+};
+
+static int __init pmem_init(void)
+{
+       int error;
+
+       pmem_major = register_blkdev(0, "pmem");
+       if (pmem_major < 0)
+               return pmem_major;
+
+       error = platform_driver_register(&pmem_driver);
+       if (error)
+               unregister_blkdev(pmem_major, "pmem");
+       return error;
+}
+module_init(pmem_init);
+
+static void pmem_exit(void)
+{
+       platform_driver_unregister(&pmem_driver);
+       unregister_blkdev(pmem_major, "pmem");
+}
+module_exit(pmem_exit);
+
+MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
+MODULE_LICENSE("GPL v2");