Re: [Linux-nvdimm] [PATCH 1/3] pmem: Initial version of persistent memory driver

[Dan Williams <dan.j.williams@intel.com>]

On Thu, Mar 26, 2015 at 1:32 AM, Christoph Hellwig <hch@lst.de> wrote:
> From: Ross Zwisler <ross.zwisler@linux.intel.com>
>
> 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.
>
> Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> [hch: convert to use a platform_device for discovery, fix partition
>  support]
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Tested-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> ---
>  MAINTAINERS            |   6 +
>  drivers/block/Kconfig  |  13 ++
>  drivers/block/Makefile |   1 +
>  drivers/block/pmem.c   | 373 +++++++++++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 393 insertions(+)
>  create mode 100644 drivers/block/pmem.c
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 358eb01..efacf2b 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -8063,6 +8063,12 @@ 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
> diff --git a/drivers/block/Kconfig b/drivers/block/Kconfig
> index 1b8094d..9284aaf 100644
> --- a/drivers/block/Kconfig
> +++ b/drivers/block/Kconfig
> @@ -404,6 +404,19 @@ 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 block devices.  Memory for PMEM should be
> +         reserved using the "memmap" kernel parameter.
> +
> +         To compile this driver as a module, choose M here: the module will be
> +         called pmem.
> +
> +         Most normal users won't need this functionality, and can thus say N
> +         here.
> +
>  config CDROM_PKTCDVD
>         tristate "Packet writing on CD/DVD media"
>         depends on !UML
> diff --git a/drivers/block/Makefile b/drivers/block/Makefile
> index 02b688d..9cc6c18 100644
> --- a/drivers/block/Makefile
> +++ b/drivers/block/Makefile
> @@ -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
> index 0000000..545b13b
> --- /dev/null
> +++ b/drivers/block/pmem.c
> @@ -0,0 +1,373 @@
> +/*
> + * Persistent Memory Driver
> + * Copyright (c) 2014, Intel Corporation.
> + *
> + * 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.
> + *
> + * This driver is heavily based on drivers/block/brd.c.
> + * Copyright (C) 2007 Nick Piggin
> + * Copyright (C) 2007 Novell Inc.
> + */
> +
> +#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 SECTOR_SHIFT           9
> +#define PAGE_SECTORS_SHIFT     (PAGE_SHIFT - SECTOR_SHIFT)
> +#define PAGE_SECTORS           (1 << PAGE_SECTORS_SHIFT)
> +
> +#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 int pmem_getgeo(struct block_device *bd, struct hd_geometry *geo)
> +{
> +       /* some standard values */
> +       geo->heads = 1 << 6;
> +       geo->sectors = 1 << 5;
> +       geo->cylinders = get_capacity(bd->bd_disk) >> 11;
> +       return 0;
> +}
> +
> +/*
> + * direct translation from (pmem,sector) => void*
> + * We do not require that sector be page aligned.
> + * The return value will point to the beginning of the page containing the
> + * given sector, not to the sector itself.
> + */
> +static void *pmem_lookup_pg_addr(struct pmem_device *pmem, sector_t sector)
> +{
> +       size_t page_offset = sector >> PAGE_SECTORS_SHIFT;
> +       size_t offset = page_offset << PAGE_SHIFT;
> +
> +       BUG_ON(offset >= pmem->size);
> +       return pmem->virt_addr + offset;
> +}
> +
> +/* sector must be page aligned */
> +static unsigned long pmem_lookup_pfn(struct pmem_device *pmem, sector_t sector)
> +{
> +       size_t page_offset = sector >> PAGE_SECTORS_SHIFT;
> +
> +       BUG_ON(sector & (PAGE_SECTORS - 1));
> +       return (pmem->phys_addr >> PAGE_SHIFT) + page_offset;
> +}
> +
> +/*
> + * sector is not required to be page aligned.
> + * n is at most a single page, but could be less.
> + */
> +static void copy_to_pmem(struct pmem_device *pmem, const void *src,
> +                       sector_t sector, size_t n)
> +{
> +       void *dst;
> +       unsigned int offset = (sector & (PAGE_SECTORS - 1)) << SECTOR_SHIFT;
> +       size_t copy;
> +
> +       BUG_ON(n > PAGE_SIZE);
> +
> +       copy = min_t(size_t, n, PAGE_SIZE - offset);
> +       dst = pmem_lookup_pg_addr(pmem, sector);
> +       memcpy(dst + offset, src, copy);
> +
> +       if (copy < n) {
> +               src += copy;
> +               sector += copy >> SECTOR_SHIFT;
> +               copy = n - copy;
> +               dst = pmem_lookup_pg_addr(pmem, sector);
> +               memcpy(dst, src, copy);
> +       }
> +}
> +
> +/*
> + * sector is not required to be page aligned.
> + * n is at most a single page, but could be less.
> + */
> +static void copy_from_pmem(void *dst, struct pmem_device *pmem,
> +                         sector_t sector, size_t n)
> +{
> +       void *src;
> +       unsigned int offset = (sector & (PAGE_SECTORS - 1)) << SECTOR_SHIFT;
> +       size_t copy;
> +
> +       BUG_ON(n > PAGE_SIZE);
> +
> +       copy = min_t(size_t, n, PAGE_SIZE - offset);
> +       src = pmem_lookup_pg_addr(pmem, sector);
> +
> +       memcpy(dst, src + offset, copy);
> +
> +       if (copy < n) {
> +               dst += copy;
> +               sector += copy >> SECTOR_SHIFT;
> +               copy = n - copy;
> +               src = pmem_lookup_pg_addr(pmem, sector);
> +               memcpy(dst, src, copy);
> +       }
> +}
> +
> +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);
> +
> +       if (rw == READ) {
> +               copy_from_pmem(mem + off, pmem, sector, len);
> +               flush_dcache_page(page);
> +       } else {
> +               /*
> +                * FIXME: Need more involved flushing to ensure that writes to
> +                * NVDIMMs are actually durable before returning.
> +                */
> +               flush_dcache_page(page);
> +               copy_to_pmem(pmem, mem + off, sector, 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;
> +
> +       sector = bio->bi_iter.bi_sector;
> +       if (bio_end_sector(bio) > get_capacity(bdev->bd_disk)) {
> +               err = -EIO;
> +               goto out;
> +       }
> +
> +       BUG_ON(bio->bi_rw & REQ_DISCARD);
> +
> +       rw = bio_rw(bio);
> +       if (rw == READA)
> +               rw = READ;
> +
> +       bio_for_each_segment(bvec, bio, iter) {
> +               unsigned int len = bvec.bv_len;
> +
> +               BUG_ON(len > PAGE_SIZE);
> +               pmem_do_bvec(pmem, bvec.bv_page, len,
> +                           bvec.bv_offset, rw, sector);
> +               sector += len >> SECTOR_SHIFT;
> +       }
> +
> +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;
> +
> +       if (!pmem)
> +               return -ENODEV;
> +
> +       *kaddr = pmem_lookup_pg_addr(pmem, sector);
> +       *pfn = pmem_lookup_pfn(pmem, sector);
> +
> +       return pmem->size - (sector * 512);
> +}
> +
> +static const struct block_device_operations pmem_fops = {
> +       .owner =                THIS_MODULE,
> +       .rw_page =              pmem_rw_page,
> +       .direct_access =        pmem_direct_access,
> +       .getgeo =               pmem_getgeo,
> +};
> +
> +/* pmem->phys_addr and pmem->size need to be set.
> + * Will then set virt_addr if successful.
> + */
> +static int pmem_mapmem(struct pmem_device *pmem)
> +{
> +       struct resource *res_mem;
> +       int err;
> +
> +       res_mem = request_mem_region_exclusive(pmem->phys_addr, pmem->size,
> +                                              "pmem");

Isn't request_mem_region() enough?  i.e. it seems
request_mem_region_exclusive() assumes no DAX, at least in theory?

> +       if (!res_mem) {
> +               pr_warn("pmem: request_mem_region_exclusive phys=0x%llx size=0x%zx failed\n",
> +                          pmem->phys_addr, pmem->size);
> +               return -EINVAL;
> +       }
> +
> +       /*
> +        * Map the memory as non-cachable, as we can't write back the contents
> +        * of the CPU caches in case of a crash.
> +        */
> +       pmem->virt_addr = ioremap_nocache(pmem->phys_addr, pmem->size);

This is fine for now, but I think we're going to end up with a
continuum of solutions to this problem based on the platform and the
device.  Some ADR platforms have firmware that takes actions like
flushing caches on a "power going away" signal.  Other platforms have
cache management instructions that we can use on either a per-i/o or
per REQ_FUA/FLUSH request.  Hmm, with this being in the memory map by
default I think this poses a challenge for VIVT caches and aliased
accesses?  We can revisit this when arm support shows up.