| Message ID | 20170526022438.17478-1-haozhong.zhang@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
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.zhang@intel.com> > Message-id: CAPcyv4hV2-ZW8SMCRtD0P_86KgR3DHOvNe+6T5SY2u7wXg3gEg@mail.gmail.com > --- > Cc: "Michael S. Tsirkin" <mst@redhat.com> > Cc: Igor Mammedov <imammedo@redhat.com> > Cc: Xiao Guangrong <guangrong.xiao@linux.intel.com> > Cc: Stefan Hajnoczi <stefanha@gmail.com> > Cc: Dan Williams <dan.j.williams@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". > + 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. > + > + 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/ > + "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 >
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.zhang@intel.com> > > Message-id: CAPcyv4hV2-ZW8SMCRtD0P_86KgR3DHOvNe+6T5SY2u7wXg3gEg@mail.gmail.com > > --- > > Cc: "Michael S. Tsirkin" <mst@redhat.com> > > Cc: Igor Mammedov <imammedo@redhat.com> > > Cc: Xiao Guangrong <guangrong.xiao@linux.intel.com> > > Cc: Stefan Hajnoczi <stefanha@gmail.com> > > Cc: Dan Williams <dan.j.williams@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 > >
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")) { + 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; + + out: + if (!is_dax) { + error_report("warning: nvdimm backend %s is not DAX device, " + "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;
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.zhang@intel.com> Message-id: CAPcyv4hV2-ZW8SMCRtD0P_86KgR3DHOvNe+6T5SY2u7wXg3gEg@mail.gmail.com --- Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Xiao Guangrong <guangrong.xiao@linux.intel.com> Cc: Stefan Hajnoczi <stefanha@gmail.com> Cc: Dan Williams <dan.j.williams@intel.com> --- hw/mem/nvdimm.c | 37 +++++++++++++++++++++++++++++++++++++ 1 file changed, 37 insertions(+)