On 05/30/17 10:15 +0100, Stefan Hajnoczi wrote:
> On Fri, May 26, 2017 at 10:24:37AM +0800, Haozhong Zhang wrote:
> > Applications in Linux guest that use device-dax never trigger flush
> > that can be trapped by KVM/QEMU. Meanwhile, if the host backend is not
> > device-dax, QEMU cannot guarantee the persistence of guest writes.
> > Before solving this flushing problem, QEMU should warn users if the
> > host backend is not device-dax.
> >
> > Signed-off-by: Haozhong Zhang <haozhong.zh...@intel.com>
> > Message-id:
> > capcyv4hv2-zw8smcrtd0p_86kgr3dhovne+6t5sy2u7wxg3...@mail.gmail.com
> > ---
> > Cc: "Michael S. Tsirkin" <m...@redhat.com>
> > Cc: Igor Mammedov <imamm...@redhat.com>
> > Cc: Xiao Guangrong <guangrong.x...@linux.intel.com>
> > Cc: Stefan Hajnoczi <stefa...@gmail.com>
> > Cc: Dan Williams <dan.j.willi...@intel.com>
> > ---
> > hw/mem/nvdimm.c | 37 +++++++++++++++++++++++++++++++++++++
> > 1 file changed, 37 insertions(+)
> >
> > diff --git a/hw/mem/nvdimm.c b/hw/mem/nvdimm.c
> > index db896b0bb6..c7bb407f33 100644
> > --- a/hw/mem/nvdimm.c
> > +++ b/hw/mem/nvdimm.c
> > @@ -26,6 +26,7 @@
> > #include "qapi/error.h"
> > #include "qapi/visitor.h"
> > #include "hw/mem/nvdimm.h"
> > +#include "qemu/error-report.h"
> >
> > static void nvdimm_get_label_size(Object *obj, Visitor *v, const char
> > *name,
> > void *opaque, Error **errp)
> > @@ -78,12 +79,48 @@ static MemoryRegion
> > *nvdimm_get_memory_region(PCDIMMDevice *dimm)
> > return &nvdimm->nvdimm_mr;
> > }
> >
> > +static void nvdimm_check_dax(HostMemoryBackend *hostmem)
> > +{
> > + char *mem_path =
> > + object_property_get_str(OBJECT(hostmem), "mem-path", NULL);
> > + char *dev_name = NULL, *sysfs_path = NULL;
> > + bool is_dax = false;
> > +
> > + if (!mem_path) {
> > + goto out;
> > + }
> > +
> > + if (!g_str_has_prefix(mem_path, "/dev/dax")) {
>
> Does not work with relative paths, symlinks, or device nodes in
> non-standard locations.
>
> A more accurate check is to fstat the file descriptor for the char
> device major/minor number and verify that
> /sys/dev/char/MAJ:MIN/device/devtype is "nd_dax".
>
Yes, Dan also suggested to use this approach. I'll turn to it in v2.
> > + goto out;
> > + }
> > +
> > + dev_name = mem_path + strlen("/dev/");
> > + sysfs_path = g_strdup_printf("/sys/class/dax/%s", dev_name);
> > + if (access(sysfs_path, F_OK)) {
> > + goto out;
> > + }
> > +
> > + is_dax = true;
>
> This check is Linux-specific, please use #ifdef __linux__. It's okay to
> always print an error on other operating systems.
>
will change
> > +
> > + out:
> > + if (!is_dax) {
> > + error_report("warning: nvdimm backend %s is not DAX device, "
>
> The check is not 100% accurate so:
>
> s/is not/does not look like a/
ditto
Thanks,
Haozhong
>
> > + "unable to guarantee persistence of guest writes",
> > + mem_path ?: "RAM");
> > + }
> > +
> > + g_free(sysfs_path);
> > + g_free(mem_path);
> > +}
> > +
> > static void nvdimm_realize(PCDIMMDevice *dimm, Error **errp)
> > {
> > MemoryRegion *mr = host_memory_backend_get_memory(dimm->hostmem, errp);
> > NVDIMMDevice *nvdimm = NVDIMM(dimm);
> > uint64_t align, pmem_size, size = memory_region_size(mr);
> >
> > + nvdimm_check_dax(dimm->hostmem);
> > +
> > align = memory_region_get_alignment(mr);
> >
> > pmem_size = size - nvdimm->label_size;
> > --
> > 2.11.0
> >
