On Fri, 14 Aug 2015 22:52:01 +0800
Xiao Guangrong <guangrong.x...@linux.intel.com> wrote:
> The parameter @file is used as backed memory for NVDIMM which is
> divided into two parts if @dataconfig is true:
> - first parts is (0, size - 128K], which is used as PMEM (Persistent
> Memory)
> - 128K at the end of the file, which is used as Config Data Area, it's
> used to store Label namespace data
>
> The @file supports both regular file and block device, of course we
> can assign any these two kinds of files for test and emulation, however,
> in the real word for performance reason, we usually used these files as
> NVDIMM backed file:
> - the regular file in the filesystem with DAX enabled created on NVDIMM
> device on host
> - the raw PMEM device on host, e,g /dev/pmem0
A lot of code in this series could reuse what QEMU already
uses for implementing pc-dimm devices.
here is common concepts that could be reused.
- on physical system both DIMM and NVDIMM devices use
the same slots. We could share QEMU's '-m slots' option between
both devices. An alternative to not sharing would be to introduce
'-machine nvdimm_slots' option.
And yes, we need to know number of NVDIMMs to describe
them all in ACPI table (taking in amount future hotplug
include in this possible NVDIMM devices)
I'd go the same way as on real hardware on make them share the same slots.
- they share the same physical address space and limits
on how much memory system can handle. So I'd suggest sharing existing
'-m maxmem' option and reuse hotplug_memory address space.
Essentially what I'm suggesting is to inherit NVDIMM's implementation
from pc-dimm reusing all of its code/backends and
just override parts that do memory mapping into guest's address space to
accommodate NVDIMM's requirements.
>
> Signed-off-by: Xiao Guangrong <guangrong.x...@linux.intel.com>
> ---
> hw/mem/nvdimm/pc-nvdimm.c | 109
> ++++++++++++++++++++++++++++++++++++++++++++-
> include/hw/mem/pc-nvdimm.h | 7 +++
> 2 files changed, 115 insertions(+), 1 deletion(-)
>
> diff --git a/hw/mem/nvdimm/pc-nvdimm.c b/hw/mem/nvdimm/pc-nvdimm.c
> index 7a270a8..97710d1 100644
> --- a/hw/mem/nvdimm/pc-nvdimm.c
> +++ b/hw/mem/nvdimm/pc-nvdimm.c
> @@ -22,12 +22,20 @@
> * License along with this library; if not, see
> <http://www.gnu.org/licenses/>
> */
>
> +#include <sys/mman.h>
> +#include <sys/ioctl.h>
> +#include <linux/fs.h>
> +
> +#include "exec/address-spaces.h"
> #include "hw/mem/pc-nvdimm.h"
>
> -#define PAGE_SIZE (1UL << 12)
> +#define PAGE_SIZE (1UL << 12)
> +
> +#define MIN_CONFIG_DATA_SIZE (128 << 10)
>
> static struct nvdimms_info {
> ram_addr_t current_addr;
> + int device_index;
> } nvdimms_info;
>
> /* the address range [offset, ~0ULL) is reserved for NVDIMM. */
> @@ -37,6 +45,26 @@ void pc_nvdimm_reserve_range(ram_addr_t offset)
> nvdimms_info.current_addr = offset;
> }
>
> +static ram_addr_t reserved_range_push(uint64_t size)
> +{
> + uint64_t current;
> +
> + current = ROUND_UP(nvdimms_info.current_addr, PAGE_SIZE);
> +
> + /* do not have enough space? */
> + if (current + size < current) {
> + return 0;
> + }
> +
> + nvdimms_info.current_addr = current + size;
> + return current;
> +}
You can't use all memory above hotplug_memory area since
we have to tell guest where 64-bit PCI window starts,
and currently it should start at reserved-memory-end
(but it isn't due to a bug: I've just posted fix to qemu-devel
"[PATCH 0/2] pc: fix 64-bit PCI window clashing with memory hotplug region"
)
> +
> +static uint32_t new_device_index(void)
> +{
> + return nvdimms_info.device_index++;
> +}
> +
> static char *get_file(Object *obj, Error **errp)
> {
> PCNVDIMMDevice *nvdimm = PC_NVDIMM(obj);
> @@ -48,6 +76,11 @@ static void set_file(Object *obj, const char *str, Error
> **errp)
> {
> PCNVDIMMDevice *nvdimm = PC_NVDIMM(obj);
>
> + if (memory_region_size(&nvdimm->mr)) {
> + error_setg(errp, "cannot change property value");
> + return;
> + }
> +
> if (nvdimm->file) {
> g_free(nvdimm->file);
> }
> @@ -76,13 +109,87 @@ static void pc_nvdimm_init(Object *obj)
> set_configdata, NULL);
> }
>
> +static uint64_t get_file_size(int fd)
> +{
> + struct stat stat_buf;
> + uint64_t size;
> +
> + if (fstat(fd, &stat_buf) < 0) {
> + return 0;
> + }
> +
> + if (S_ISREG(stat_buf.st_mode)) {
> + return stat_buf.st_size;
> + }
> +
> + if (S_ISBLK(stat_buf.st_mode) && !ioctl(fd, BLKGETSIZE64, &size)) {
> + return size;
> + }
> +
> + return 0;
> +}
All this file stuff I'd leave to already existing backends like
memory-backend-file or even memory-backend-ram which already do
above and more allowing to configure persistent and volatile
NVDIMMs without changing NVDIMM fronted code.
> +
> static void pc_nvdimm_realize(DeviceState *dev, Error **errp)
> {
> PCNVDIMMDevice *nvdimm = PC_NVDIMM(dev);
> + char name[512];
> + void *buf;
> + ram_addr_t addr;
> + uint64_t size, nvdimm_size, config_size = MIN_CONFIG_DATA_SIZE;
> + int fd;
>
> if (!nvdimm->file) {
> error_setg(errp, "file property is not set");
> }
> +
> + fd = open(nvdimm->file, O_RDWR);
> + if (fd < 0) {
> + error_setg(errp, "can not open %s", nvdimm->file);
> + return;
> + }
> +
> + size = get_file_size(fd);
> + buf = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
> + if (buf == MAP_FAILED) {
> + error_setg(errp, "can not do mmap on %s", nvdimm->file);
> + goto do_close;
> + }
> +
> + nvdimm->config_data_size = config_size;
> + if (nvdimm->configdata) {
> + /* reserve MIN_CONFIGDATA_AREA_SIZE for configue data. */
> + nvdimm_size = size - config_size;
> + nvdimm->config_data_addr = buf + nvdimm_size;
> + } else {
> + nvdimm_size = size;
> + nvdimm->config_data_addr = NULL;
> + }
> +
> + if ((int64_t)nvdimm_size <= 0) {
> + error_setg(errp, "file size is too small to store NVDIMM"
> + " configure data");
> + goto do_unmap;
> + }
> +
> + addr = reserved_range_push(nvdimm_size);
> + if (!addr) {
> + error_setg(errp, "do not have enough space for size %#lx.\n", size);
> + goto do_unmap;
> + }
> +
> + nvdimm->device_index = new_device_index();
> + sprintf(name, "NVDIMM-%d", nvdimm->device_index);
> + memory_region_init_ram_ptr(&nvdimm->mr, OBJECT(dev), name, nvdimm_size,
> + buf);
> + vmstate_register_ram(&nvdimm->mr, DEVICE(dev));
> + memory_region_add_subregion(get_system_memory(), addr, &nvdimm->mr);
> +
> + return;
> +
> +do_unmap:
> + munmap(buf, size);
> +do_close:
> + close(fd);
> }
>
> static void pc_nvdimm_class_init(ObjectClass *oc, void *data)
> diff --git a/include/hw/mem/pc-nvdimm.h b/include/hw/mem/pc-nvdimm.h
> index 8601e9b..f617fd2 100644
> --- a/include/hw/mem/pc-nvdimm.h
> +++ b/include/hw/mem/pc-nvdimm.h
> @@ -21,6 +21,13 @@ typedef struct PCNVDIMMDevice {
>
> char *file;
> bool configdata;
> +
> + int device_index;
> +
> + uint64_t config_data_size;
> + void *config_data_addr;
> +
> + MemoryRegion mr;
> } PCNVDIMMDevice;
>
> #define TYPE_PC_NVDIMM "pc-nvdimm"
