| Message ID | 154915640488.3541628.4947340559510114602.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | libnvdimm/dimm: Add a no-BLK quirk based on NVDIMM family | expand |
Hi Dan,
I love your patch! Yet something to improve:
[auto build test ERROR on linux-nvdimm/libnvdimm-for-next]
[also build test ERROR on v5.0-rc4 next-20190201]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Dan-Williams/libnvdimm-dimm-Add-a-no-BLK-quirk-based-on-NVDIMM-family/20190203-213444
base: https://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm.git libnvdimm-for-next
config: x86_64-randconfig-x018-201905 (attached as .config)
compiler: gcc-8 (Debian 8.2.0-14) 8.2.0
reproduce:
# save the attached .config to linux build tree
make ARCH=x86_64
All errors (new ones prefixed by >>):
drivers/acpi//nfit/core.c: In function 'acpi_nfit_register_dimms':
>> drivers/acpi//nfit/core.c:2003:27: error: 'NVDIMM_FAMILY_HYPERV' undeclared (first use in this function); did you mean 'NVDIMM_FAMILY_HPE2'?
if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
^~~~~~~~~~~~~~~~~~~~
NVDIMM_FAMILY_HPE2
drivers/acpi//nfit/core.c:2003:27: note: each undeclared identifier is reported only once for each function it appears in
vim +2003 drivers/acpi//nfit/core.c
1944
1945 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1946 {
1947 struct nfit_mem *nfit_mem;
1948 int dimm_count = 0, rc;
1949 struct nvdimm *nvdimm;
1950
1951 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1952 struct acpi_nfit_flush_address *flush;
1953 unsigned long flags = 0, cmd_mask;
1954 struct nfit_memdev *nfit_memdev;
1955 u32 device_handle;
1956 u16 mem_flags;
1957
1958 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1959 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1960 if (nvdimm) {
1961 dimm_count++;
1962 continue;
1963 }
1964
1965 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1966 set_bit(NDD_ALIASING, &flags);
1967
1968 /* collate flags across all memdevs for this dimm */
1969 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1970 struct acpi_nfit_memory_map *dimm_memdev;
1971
1972 dimm_memdev = __to_nfit_memdev(nfit_mem);
1973 if (dimm_memdev->device_handle
1974 != nfit_memdev->memdev->device_handle)
1975 continue;
1976 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1977 }
1978
1979 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1980 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1981 set_bit(NDD_UNARMED, &flags);
1982
1983 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1984 if (rc)
1985 continue;
1986
1987 /*
1988 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1989 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1990 * userspace interface.
1991 */
1992 cmd_mask = 1UL << ND_CMD_CALL;
1993 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1994 /*
1995 * These commands have a 1:1 correspondence
1996 * between DSM payload and libnvdimm ioctl
1997 * payload format.
1998 */
1999 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2000 }
2001
2002 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
> 2003 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2004 set_bit(NDD_NOBLK, &flags);
2005
2006 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2007 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2008 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2009 }
2010 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2011 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2012
2013 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2014 : NULL;
2015 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2016 acpi_nfit_dimm_attribute_groups,
2017 flags, cmd_mask, flush ? flush->hint_count : 0,
2018 nfit_mem->flush_wpq, &nfit_mem->id[0],
2019 acpi_nfit_get_security_ops(nfit_mem->family));
2020 if (!nvdimm)
2021 return -ENOMEM;
2022
2023 nfit_mem->nvdimm = nvdimm;
2024 dimm_count++;
2025
2026 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2027 continue;
2028
2029 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
2030 nvdimm_name(nvdimm),
2031 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2032 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2033 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2034 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2035 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2036
2037 }
2038
2039 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2040 if (rc)
2041 return rc;
2042
2043 /*
2044 * Now that dimms are successfully registered, and async registration
2045 * is flushed, attempt to enable event notification.
2046 */
2047 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2048 struct kernfs_node *nfit_kernfs;
2049
2050 nvdimm = nfit_mem->nvdimm;
2051 if (!nvdimm)
2052 continue;
2053
2054 rc = nvdimm_security_setup_events(nvdimm);
2055 if (rc < 0)
2056 dev_warn(acpi_desc->dev,
2057 "security event setup failed: %d\n", rc);
2058
2059 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2060 if (nfit_kernfs)
2061 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2062 "flags");
2063 sysfs_put(nfit_kernfs);
2064 if (!nfit_mem->flags_attr)
2065 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2066 nvdimm_name(nvdimm));
2067 }
2068
2069 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2070 acpi_desc);
2071 }
2072
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
Hi Dan,
I love your patch! Yet something to improve:
[auto build test ERROR on linux-nvdimm/libnvdimm-for-next]
[also build test ERROR on v5.0-rc4 next-20190201]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Dan-Williams/libnvdimm-dimm-Add-a-no-BLK-quirk-based-on-NVDIMM-family/20190203-213444
base: https://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm.git libnvdimm-for-next
config: x86_64-allmodconfig (attached as .config)
compiler: gcc-8 (Debian 8.2.0-14) 8.2.0
reproduce:
# save the attached .config to linux build tree
make ARCH=x86_64
All errors (new ones prefixed by >>):
drivers/acpi/nfit/core.c: In function 'acpi_nfit_register_dimms':
>> drivers/acpi/nfit/core.c:2003:27: error: 'NVDIMM_FAMILY_HYPERV' undeclared (first use in this function); did you mean 'NVDIMM_FAMILY_HPE1'?
if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
^~~~~~~~~~~~~~~~~~~~
NVDIMM_FAMILY_HPE1
drivers/acpi/nfit/core.c:2003:27: note: each undeclared identifier is reported only once for each function it appears in
vim +2003 drivers/acpi/nfit/core.c
1944
1945 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1946 {
1947 struct nfit_mem *nfit_mem;
1948 int dimm_count = 0, rc;
1949 struct nvdimm *nvdimm;
1950
1951 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1952 struct acpi_nfit_flush_address *flush;
1953 unsigned long flags = 0, cmd_mask;
1954 struct nfit_memdev *nfit_memdev;
1955 u32 device_handle;
1956 u16 mem_flags;
1957
1958 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1959 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1960 if (nvdimm) {
1961 dimm_count++;
1962 continue;
1963 }
1964
1965 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1966 set_bit(NDD_ALIASING, &flags);
1967
1968 /* collate flags across all memdevs for this dimm */
1969 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1970 struct acpi_nfit_memory_map *dimm_memdev;
1971
1972 dimm_memdev = __to_nfit_memdev(nfit_mem);
1973 if (dimm_memdev->device_handle
1974 != nfit_memdev->memdev->device_handle)
1975 continue;
1976 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1977 }
1978
1979 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1980 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1981 set_bit(NDD_UNARMED, &flags);
1982
1983 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1984 if (rc)
1985 continue;
1986
1987 /*
1988 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1989 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1990 * userspace interface.
1991 */
1992 cmd_mask = 1UL << ND_CMD_CALL;
1993 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1994 /*
1995 * These commands have a 1:1 correspondence
1996 * between DSM payload and libnvdimm ioctl
1997 * payload format.
1998 */
1999 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2000 }
2001
2002 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
> 2003 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2004 set_bit(NDD_NOBLK, &flags);
2005
2006 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2007 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2008 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2009 }
2010 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2011 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2012
2013 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2014 : NULL;
2015 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2016 acpi_nfit_dimm_attribute_groups,
2017 flags, cmd_mask, flush ? flush->hint_count : 0,
2018 nfit_mem->flush_wpq, &nfit_mem->id[0],
2019 acpi_nfit_get_security_ops(nfit_mem->family));
2020 if (!nvdimm)
2021 return -ENOMEM;
2022
2023 nfit_mem->nvdimm = nvdimm;
2024 dimm_count++;
2025
2026 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2027 continue;
2028
2029 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
2030 nvdimm_name(nvdimm),
2031 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2032 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2033 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2034 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2035 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2036
2037 }
2038
2039 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2040 if (rc)
2041 return rc;
2042
2043 /*
2044 * Now that dimms are successfully registered, and async registration
2045 * is flushed, attempt to enable event notification.
2046 */
2047 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2048 struct kernfs_node *nfit_kernfs;
2049
2050 nvdimm = nfit_mem->nvdimm;
2051 if (!nvdimm)
2052 continue;
2053
2054 rc = nvdimm_security_setup_events(nvdimm);
2055 if (rc < 0)
2056 dev_warn(acpi_desc->dev,
2057 "security event setup failed: %d\n", rc);
2058
2059 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2060 if (nfit_kernfs)
2061 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2062 "flags");
2063 sysfs_put(nfit_kernfs);
2064 if (!nfit_mem->flags_attr)
2065 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2066 nvdimm_name(nvdimm));
2067 }
2068
2069 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2070 acpi_desc);
2071 }
2072
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
> From: Dan Williams <dan.j.williams@intel.com> > Sent: Saturday, February 2, 2019 5:13 PM > ... > As Dexuan reports the NVDIMM_FAMILY_HYPERV platform is incompatible with > the existing Linux namespace implementation because it uses > NSLABEL_FLAG_LOCAL for x1-width PMEM interleave sets. Quirk it as an > platform / DIMM that does not provide BLK-aperture access. Allow the > libnvdimm core to assume no potential for aliasing. In case other > implementations make the same mistake, provide a "noblk" module > parameter to force-enable the quirk. Hi Dan, Thanks very much for the patch! With it, "ndctl list" can show the below: root@decui-gen2-u1904:~/ndctl# ndctl list [ { "dev":"namespace1.0", "mode":"raw", "size":137438953472, "uuid":"c258aaab-f72b-e546-bfa5-be5e07761dbc", "blockdev":"pmem1", "name":"Microsoft Hyper-V NVDIMM 1 Label" }, { "dev":"namespace0.0", "mode":"raw", "size":34359738368, "uuid":"9f0497a7-4453-7c40-ad35-21a791e00345", "blockdev":"pmem0", "name":"Microsoft Hyper-V NVDIMM 0 Label" } ] And /dev/pmem0 can appear, but /dev/pmem0p1 can not appear, and the "mode" of "namespace0.0" is not correct. With the Ubuntu 19.04 4.18 kenel, I get the below: root@decui-gen2-u1904:~/ndctl# ndctl list [ { "dev":"namespace1.0", "mode":"raw", "size":137438953472, "blockdev":"pmem1" }, { "dev":"namespace0.0", "mode":"fsdax", "map":"dev", "size":33820770304, "uuid":"ef028c4e-2b1f-4bf8-b92a-1109d7a1c914", "blockdev":"pmem0" } ] and /dev/pmem0p1 can appear. It looks the namespace created by Ubuntu 19.04 (4.18) is incompatible with the libnvdimm-pending branch + this patch. It looks we need to completely disable the label usage for NVDIMM_FAMILY_HYPERV due to compability? Thanks, -- Dexuan
> From: kbuild test robot <lkp@intel.com> > Sent: Sunday, February 3, 2019 6:32 AM > To: Dan Williams <dan.j.williams@intel.com> > Cc: kbuild-all@01.org; linux-nvdimm@lists.01.org; Dexuan Cui > <decui@microsoft.com>; linux-kernel@vger.kernel.org > Subject: Re: [PATCH] libnvdimm/dimm: Add a no-BLK quirk based on NVDIMM > family > > Hi Dan, > > I love your patch! Yet something to improve: > > [auto build test ERROR on linux-nvdimm/libnvdimm-for-next] > [also build test ERROR on v5.0-rc4 next-20190201] > [if your patch is applied to the wrong git tree, please drop us a note to help > improve the system] > ... > All errors (new ones prefixed by >>): > > drivers/acpi//nfit/core.c: In function 'acpi_nfit_register_dimms': > >> drivers/acpi//nfit/core.c:2003:27: error: 'NVDIMM_FAMILY_HYPERV' > undeclared (first use in this function); did you mean 'NVDIMM_FAMILY_HPE2'? > if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) > ^~~~~~~~~~~~~~~~~~~~ > NVDIMM_FAMILY_HPE2 > drivers/acpi//nfit/core.c:2003:27: note: each undeclared identifier is > reported only once for each function it appears in This is a false alarm. The patch is only for https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=libnvdimm-pending where we have the dependant commid 1194c4133195 ("nfit: Add Hyper-V NVDIMM DSM command set to white list") Thanks, -- Dexuan
> From: kbuild test robot <lkp@intel.com> > Sent: Sunday, February 3, 2019 6:38 AM > ... > Hi Dan, > > I love your patch! Yet something to improve: > > [auto build test ERROR on linux-nvdimm/libnvdimm-for-next] > [also build test ERROR on v5.0-rc4 next-20190201] > [if your patch is applied to the wrong git tree, please drop us a note to help > improve the system] > ... > config: x86_64-allmodconfig (attached as .config) > compiler: gcc-8 (Debian 8.2.0-14) 8.2.0 > reproduce: > # save the attached .config to linux build tree > make ARCH=x86_64 > > All errors (new ones prefixed by >>): > > drivers/acpi/nfit/core.c: In function 'acpi_nfit_register_dimms': > >> drivers/acpi/nfit/core.c:2003:27: error: 'NVDIMM_FAMILY_HYPERV' > undeclared (first use in this function); did you mean 'NVDIMM_FAMILY_HPE1'? > if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) > ^~~~~~~~~~~~~~~~~~~~ > NVDIMM_FAMILY_HPE1 > drivers/acpi/nfit/core.c:2003:27: note: each undeclared identifier is > reported only once for each function it appears in This is a false alarm, too. The patch is only for https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=libnvdimm-pending where we have the dependant commit: 1194c4133195 ("nfit: Add Hyper-V NVDIMM DSM command set to white list") Thanks, -- Dexuan
On Sun, Feb 3, 2019 at 9:22 AM Dexuan Cui <decui@microsoft.com> wrote: > > > From: Dan Williams <dan.j.williams@intel.com> > > Sent: Saturday, February 2, 2019 5:13 PM > > ... > > As Dexuan reports the NVDIMM_FAMILY_HYPERV platform is incompatible with > > the existing Linux namespace implementation because it uses > > NSLABEL_FLAG_LOCAL for x1-width PMEM interleave sets. Quirk it as an > > platform / DIMM that does not provide BLK-aperture access. Allow the > > libnvdimm core to assume no potential for aliasing. In case other > > implementations make the same mistake, provide a "noblk" module > > parameter to force-enable the quirk. > > Hi Dan, > Thanks very much for the patch! With it, "ndctl list" can show the below: > > root@decui-gen2-u1904:~/ndctl# ndctl list > [ > { > "dev":"namespace1.0", > "mode":"raw", > "size":137438953472, > "uuid":"c258aaab-f72b-e546-bfa5-be5e07761dbc", > "blockdev":"pmem1", > "name":"Microsoft Hyper-V NVDIMM 1 Label" > }, > { > "dev":"namespace0.0", > "mode":"raw", > "size":34359738368, > "uuid":"9f0497a7-4453-7c40-ad35-21a791e00345", > "blockdev":"pmem0", > "name":"Microsoft Hyper-V NVDIMM 0 Label" > } > ] > > And /dev/pmem0 can appear, but /dev/pmem0p1 can not appear, and the "mode" of > "namespace0.0" is not correct. With the Ubuntu 19.04 4.18 kenel, I get the below: > > root@decui-gen2-u1904:~/ndctl# ndctl list > [ > { > "dev":"namespace1.0", > "mode":"raw", > "size":137438953472, > "blockdev":"pmem1" > }, > { > "dev":"namespace0.0", > "mode":"fsdax", > "map":"dev", > "size":33820770304, > "uuid":"ef028c4e-2b1f-4bf8-b92a-1109d7a1c914", > "blockdev":"pmem0" > } > ] > and /dev/pmem0p1 can appear. > > It looks the namespace created by Ubuntu 19.04 (4.18) is incompatible with > the libnvdimm-pending branch + this patch. This is correct, the configuration switched from label-less by default to labeled. > It looks we need to completely disable the label usage for NVDIMM_FAMILY_HYPERV > due to compability? It's either going to be a quirk in Linux or a quirk in the Hyper-V configuration. I think it's more manageable for Hyper-V to ship a "label disable" switch than to try to ship a workaround in Linux. The "noblk" quirk in Linux works around the LOCAL bit in the labels. However, the namespace mode regression can only be resolved by hiding the label capability until the administrator manually switches from label-less to labeled operation.
> From: Dan Williams <dan.j.williams@intel.com> > Sent: Sunday, February 3, 2019 9:49 AM > > ... > > It looks the namespace created by Ubuntu 19.04 (4.18) is incompatible with > > the libnvdimm-pending branch + this patch. > > This is correct, the configuration switched from label-less by default > to labeled. Thanks for the confirmation! > > It looks we need to completely disable the label usage for > NVDIMM_FAMILY_HYPERV > > due to compability? > > It's either going to be a quirk in Linux or a quirk in the Hyper-V > configuration. I think it's more manageable for Hyper-V to ship a > "label disable" switch than to try to ship a workaround in Linux. The > "noblk" quirk in Linux works around the LOCAL bit in the labels. > However, the namespace mode regression can only be resolved by hiding > the label capability until the administrator manually switches from > label-less to labeled operation. As I understand, the essence of the issue is: Hyper-V emulates the label mechanism (i.e. it supports _LSI and LSR), but doesn't do it right (i.e. it doesn't support _LSW). To manage the namespaces, Linux can choose to use label or not. If _LSI/_LSR are supported, Linux assumes _LSW is supported as well and chooses to use label (i.e. the label mode), but since Hyper-V doesn't support _LSW, Linux fails to change the namespace configuration. In Ubuntu 19.04 (4.18), the kernel is not aware of Hyper-V _LSI/_LSR, so the created namespace is in "label-less" mode, and hence can't be used with the libnvdimm-pending branch + this patch, unless we add a quirk in Linux to explicitly not use the label. I agree ideally Hyper-V should hide the label capability, and I'll request Hyper-V team to do that. I hope Hyper-V guys are still able to do that in time so we won't need a quirk in Linux kernel. BTW, I suppose Windows VM should be in "label-less" mode. Thanks for the help, Dan! Thanks, -- Dexuan
On Sun, Feb 3, 2019 at 10:13 AM Dexuan Cui <decui@microsoft.com> wrote: > > > From: Dan Williams <dan.j.williams@intel.com> > > Sent: Sunday, February 3, 2019 9:49 AM > > > ... > > > It looks the namespace created by Ubuntu 19.04 (4.18) is incompatible with > > > the libnvdimm-pending branch + this patch. > > > > This is correct, the configuration switched from label-less by default > > to labeled. > Thanks for the confirmation! > > > > It looks we need to completely disable the label usage for > > NVDIMM_FAMILY_HYPERV > > > due to compability? > > > > It's either going to be a quirk in Linux or a quirk in the Hyper-V > > configuration. I think it's more manageable for Hyper-V to ship a > > "label disable" switch than to try to ship a workaround in Linux. The > > "noblk" quirk in Linux works around the LOCAL bit in the labels. > > However, the namespace mode regression can only be resolved by hiding > > the label capability until the administrator manually switches from > > label-less to labeled operation. > > As I understand, the essence of the issue is: Hyper-V emulates the > label mechanism (i.e. it supports _LSI and LSR), but doesn't do it > right (i.e. it doesn't support _LSW). > > To manage the namespaces, Linux can choose to use label or not. > > If _LSI/_LSR are supported, Linux assumes _LSW is supported as well > and chooses to use label (i.e. the label mode), but since Hyper-V doesn't > support _LSW, Linux fails to change the namespace configuration. No, that's not quite right. The reason Linux does not see the fsdax mode configuration is due to the missing "address abstraction GUID" in the label produced the default Hyper-V configuration. In label-less mode there is no "address abstraction GUID" to validate so it falls back to just using the info-block directly. The _LSW issue is if you wanted to reconfigure a raw namespace to fsdax mode. The current flow tries to delete the label, but that's only for reconfiguration, not the initial boot-up case that is currently failing. The deletion would fail on finding no label-write capability, but to be clear the boot-up case does not perform any writes. > In > Ubuntu 19.04 (4.18), the kernel is not aware of Hyper-V _LSI/_LSR, so > the created namespace is in "label-less" mode, and hence can't be used > with the libnvdimm-pending branch + this patch, unless we add a quirk > in Linux to explicitly not use the label. > > I agree ideally Hyper-V should hide the label capability, and I'll request > Hyper-V team to do that. I hope Hyper-V guys are still able to do that > in time so we won't need a quirk in Linux kernel. After some more thought, the "no regressions" rule means that Linux should ship a quirk for this by default. I think a good heuristic is to disable label support by default if no _LSW method is detected. An opt-in can be specified to accept "read-only" configurations, but that's an exceptional case. I'll send a patch for this. > BTW, I suppose Windows VM should be in "label-less" mode. I expect Windows mandates labeled operation. This label-less concept was something Linux shipped in advance of a specification being ratified and to support early NVDIMMs that don't advertise label space. > Thanks for the help, Dan! Thanks for the quick feedback and testing.
> From: Dan Williams <dan.j.williams@intel.com> > Sent: Sunday, February 3, 2019 11:14 AM > > ... > > As I understand, the essence of the issue is: Hyper-V emulates the > > label mechanism (i.e. it supports _LSI and LSR), but doesn't do it > > right (i.e. it doesn't support _LSW). > > > > To manage the namespaces, Linux can choose to use label or not. > > > > If _LSI/_LSR are supported, Linux assumes _LSW is supported as well > > and chooses to use label (i.e. the label mode), but since Hyper-V doesn't > > support _LSW, Linux fails to change the namespace configuration. > > No, that's not quite right. The reason Linux does not see the fsdax > mode configuration is due to the missing "address abstraction GUID" in > the label produced the default Hyper-V configuration. Hi Dan, Do you mean NVDIMM_DAX_GUID? > In label-less mode there is no "address abstraction GUID" to validate so > it falls back to just using the info-block directly. In the case of not using label storage area, as I understand the info-block resides in regular data storage area. Can you please tell me where exactly the info-block is and how its location is decided? And I suppose the info-block contains the NVDIMM_DAX_GUID? I'm asking because I found I lose ~500MBytes of the 32GBytes Hyper-V NVDIMM device, when the namespace is in fsdax mode. When it's in raw mode, I'm able to use all of the 32GB space. > The _LSW issue is if you wanted to reconfigure a raw namespace to > fsdax mode. The current flow tries to delete the label, but that's > only for reconfiguration, not the initial boot-up case that is > currently failing. The deletion would fail on finding no label-write > capability, but to be clear the boot-up case does not perform any > writes. Thanks for the explanation! > > In Ubuntu 19.04 (4.18), the kernel is not aware of Hyper-V _LSI/_LSR, so > > the created namespace is in "label-less" mode, and hence can't be used > > with the libnvdimm-pending branch + this patch, unless we add a quirk > > in Linux to explicitly not use the label. > > > > I agree ideally Hyper-V should hide the label capability, and I'll request > > Hyper-V team to do that. I hope Hyper-V guys are still able to do that > > in time so we won't need a quirk in Linux kernel. > > After some more thought, the "no regressions" rule means that Linux > should ship a quirk for this by default. I think a good heuristic is > to disable label support by default if no _LSW method is detected. An > opt-in can be specified to accept "read-only" configurations, but > that's an exceptional case. I'll send a patch for this. > > > BTW, I suppose Windows VM should be in "label-less" mode. > > I expect Windows mandates labeled operation. This label-less concept > was something Linux shipped in advance of a specification being > ratified and to support early NVDIMMs that don't advertise label > space. Since Hyper-V NVDIMM doesn't support _LSW, IMO Windows VM can't update the LSI area either, so I guess Windows VM can't use label either, just like Linux VM? I guess Windows VM also stores an "info block" when the namespace is in dax mode, just like Linux VM? BTW, it looks the dax mode configuration in Linux VM is incompatible with that in Windows VM(?) When I create a DAX-enabled NTFS partition in Windows VM in the same Hyper-V NVDIMM, Windows VM is able to use all of the 32GB space (As I mentioned above, in Linux I lose ~500MB). And the "dax" mode namespace created in Windows VM is detected as "raw", and vice verse (I think). I'm trying to understand why I lose ~500MB in Linux dax mode. Your insights are appreciated! Thanks, --Dexuan
On Tue, Feb 5, 2019 at 8:53 AM Dexuan Cui <decui@microsoft.com> wrote: > > > From: Dan Williams <dan.j.williams@intel.com> > > Sent: Sunday, February 3, 2019 11:14 AM > > > ... > > > As I understand, the essence of the issue is: Hyper-V emulates the > > > label mechanism (i.e. it supports _LSI and LSR), but doesn't do it > > > right (i.e. it doesn't support _LSW). > > > > > > To manage the namespaces, Linux can choose to use label or not. > > > > > > If _LSI/_LSR are supported, Linux assumes _LSW is supported as well > > > and chooses to use label (i.e. the label mode), but since Hyper-V doesn't > > > support _LSW, Linux fails to change the namespace configuration. > > > > No, that's not quite right. The reason Linux does not see the fsdax > > mode configuration is due to the missing "address abstraction GUID" in > > the label produced the default Hyper-V configuration. > Hi Dan, > Do you mean NVDIMM_DAX_GUID? Correct. > > > In label-less mode there is no "address abstraction GUID" to validate so > > it falls back to just using the info-block directly. > In the case of not using label storage area, as I understand the info-block > resides in regular data storage area. Can you please tell me where exactly > the info-block is and how its location is decided? > And I suppose the info-block contains the NVDIMM_DAX_GUID? The info-block always lives in the data-storage area, even with labels. It's placed at namespace base address + 4K. When labels are present the label gives the namespace a uuid and the info-block "parent uuid" field must match that value. Without labels the "parent uuid" field is unused / filled with zero's. Also with v1.2 labels the address abstraction uuid must match the info-block type. > I'm asking because I found I lose ~500MBytes of the 32GBytes Hyper-V > NVDIMM device, when the namespace is in fsdax mode. When it's in > raw mode, I'm able to use all of the 32GB space. Yes. A portion of the capacity is reserved for page structures. https://lwn.net/Articles/656197/
> From: Dan Williams <dan.j.williams@intel.com> > Sent: Tuesday, February 5, 2019 9:12 AM > On Tue, Feb 5, 2019 at 8:53 AM Dexuan Cui <decui@microsoft.com> wrote: > > > > > From: Dan Williams <dan.j.williams@intel.com> > > > Sent: Sunday, February 3, 2019 11:14 AM > > > > ... > > > > As I understand, the essence of the issue is: Hyper-V emulates the > > > > label mechanism (i.e. it supports _LSI and LSR), but doesn't do it > > > > right (i.e. it doesn't support _LSW). > > > > > > > > To manage the namespaces, Linux can choose to use label or not. > > > > > > > > If _LSI/_LSR are supported, Linux assumes _LSW is supported as well > > > > and chooses to use label (i.e. the label mode), but since Hyper-V doesn't > > > > support _LSW, Linux fails to change the namespace configuration. > > > > > > No, that's not quite right. The reason Linux does not see the fsdax > > > mode configuration is due to the missing "address abstraction GUID" in > > > the label produced the default Hyper-V configuration. > > Hi Dan, > > Do you mean NVDIMM_DAX_GUID? > > Correct. FYI: in create_namespace_pmem(), label0->abstraction_guid is guid_null in the case of Hyper-V NVDIMM. It looks Hyper-V does't use the guid. > > > In label-less mode there is no "address abstraction GUID" to validate so > > > it falls back to just using the info-block directly. > > In the case of not using label storage area, as I understand the info-block > > resides in regular data storage area. Can you please tell me where exactly > > the info-block is and how its location is decided? > > And I suppose the info-block contains the NVDIMM_DAX_GUID? > > The info-block always lives in the data-storage area, even with > labels. It's placed at namespace base address + 4K. > > When labels are present the label gives the namespace a uuid and the > info-block "parent uuid" field must match that value. Without labels > the "parent uuid" field is unused / filled with zero's. Also with v1.2 > labels the address abstraction uuid must match the info-block type. Thanks for elaborating on this! It looks the label mechanism is for advanced usages, e.g., a PMEM namespace can consist of multiple NVDIMMs, or a namespace only uses part of the capacity of a NVDIMM, or a NVDIMM contains both PMEM and Block-Mode namespaces. For a simple usage, e.g. if a NVDIMM only contains one PMEM namespace which consumes all the storage space of the NVDIMM, it looks the namespace can be normally used with the help of the info-block, and we don't really need the label. IMO this is the case of Hyper-V, i.e. we don't use the label at all, since _LSW is not implemented. > > I'm asking because I found I lose ~500MBytes of the 32GBytes Hyper-V > > NVDIMM device, when the namespace is in fsdax mode. When it's in > > raw mode, I'm able to use all of the 32GB space. > > Yes. A portion of the capacity is reserved for page structures. Got it. It looks the size of the info-block (and the related padding?) is 2MB, and "the overhead is 64-bytes per 4K (16GB per 1TB) on x86", so the total overhead in my 32GB case is: 2MB + 32GB/(4096/64) = 514MB. Thanks for sharing the link! Now I realized we can use the -M parameter to not store the page metadata in the NVDIMM: -M; --map= A pmem namespace in “fsdax” or “devdax” mode requires allocation of per-page metadata. The allocation can be drawn from either: “mem”: typical system memory “dev”: persistent memory reserved from the namespace Thanks, --Dexuan
diff --git a/drivers/acpi/nfit/core.c b/drivers/acpi/nfit/core.c index 4a7e8b1fa43b..811c399a3a76 100644 --- a/drivers/acpi/nfit/core.c +++ b/drivers/acpi/nfit/core.c @@ -2016,6 +2016,10 @@ static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc) cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK; } + /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */ + if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) + set_bit(NDD_NOBLK, &flags); + if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) { set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask); set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask); diff --git a/drivers/nvdimm/dimm_devs.c b/drivers/nvdimm/dimm_devs.c index 4890310df874..186d63f16434 100644 --- a/drivers/nvdimm/dimm_devs.c +++ b/drivers/nvdimm/dimm_devs.c @@ -11,6 +11,7 @@ * General Public License for more details. */ #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt +#include <linux/moduleparam.h> #include <linux/vmalloc.h> #include <linux/device.h> #include <linux/ndctl.h> @@ -25,6 +26,10 @@ static DEFINE_IDA(dimm_ida); +static bool noblk; +module_param(noblk, bool, 0444); +MODULE_PARM_DESC(noblk, "force disable BLK / local alias support"); + /* * Retrieve bus and dimm handle and return if this bus supports * get_config_data commands @@ -551,7 +556,7 @@ struct nvdimm *__nvdimm_create(struct nvdimm_bus *nvdimm_bus, nvdimm->dimm_id = dimm_id; nvdimm->provider_data = provider_data; - nvdimm->flags = flags; + nvdimm->flags = flags | noblk ? (1 << NDD_NOBLK) : 0; nvdimm->cmd_mask = cmd_mask; nvdimm->num_flush = num_flush; nvdimm->flush_wpq = flush_wpq; diff --git a/drivers/nvdimm/label.c b/drivers/nvdimm/label.c index 6d6e9a12150b..f3d753d3169c 100644 --- a/drivers/nvdimm/label.c +++ b/drivers/nvdimm/label.c @@ -392,6 +392,7 @@ int nd_label_reserve_dpa(struct nvdimm_drvdata *ndd) return 0; /* no label, nothing to reserve */ for_each_clear_bit_le(slot, free, nslot) { + struct nvdimm *nvdimm = to_nvdimm(ndd->dev); struct nd_namespace_label *nd_label; struct nd_region *nd_region = NULL; u8 label_uuid[NSLABEL_UUID_LEN]; @@ -406,6 +407,8 @@ int nd_label_reserve_dpa(struct nvdimm_drvdata *ndd) memcpy(label_uuid, nd_label->uuid, NSLABEL_UUID_LEN); flags = __le32_to_cpu(nd_label->flags); + if (test_bit(NDD_NOBLK, &nvdimm->flags)) + flags &= ~NSLABEL_FLAG_LOCAL; nd_label_gen_id(&label_id, label_uuid, flags); res = nvdimm_allocate_dpa(ndd, &label_id, __le64_to_cpu(nd_label->dpa), diff --git a/drivers/nvdimm/namespace_devs.c b/drivers/nvdimm/namespace_devs.c index 4b077555ac70..3677b0c4a33d 100644 --- a/drivers/nvdimm/namespace_devs.c +++ b/drivers/nvdimm/namespace_devs.c @@ -2492,6 +2492,12 @@ static int init_active_labels(struct nd_region *nd_region) if (!label_ent) break; label = nd_label_active(ndd, j); + if (test_bit(NDD_NOBLK, &nvdimm->flags)) { + u32 flags = __le32_to_cpu(label->flags); + + flags &= ~NSLABEL_FLAG_LOCAL; + label->flags = __cpu_to_le32(flags); + } label_ent->label = label; mutex_lock(&nd_mapping->lock); diff --git a/drivers/nvdimm/region_devs.c b/drivers/nvdimm/region_devs.c index e2818f94f292..3b58baa44b5c 100644 --- a/drivers/nvdimm/region_devs.c +++ b/drivers/nvdimm/region_devs.c @@ -1003,6 +1003,13 @@ static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus, if (test_bit(NDD_UNARMED, &nvdimm->flags)) ro = 1; + + if (test_bit(NDD_NOBLK, &nvdimm->flags) + && dev_type == &nd_blk_device_type) { + dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n", + caller, dev_name(&nvdimm->dev), i); + return NULL; + } } if (dev_type == &nd_blk_device_type) { diff --git a/include/linux/libnvdimm.h b/include/linux/libnvdimm.h index 5440f11b0907..7da406ae3a2b 100644 --- a/include/linux/libnvdimm.h +++ b/include/linux/libnvdimm.h @@ -42,6 +42,8 @@ enum { NDD_SECURITY_OVERWRITE = 3, /* tracking whether or not there is a pending device reference */ NDD_WORK_PENDING = 4, + /* ignore / filter NSLABEL_FLAG_LOCAL for this DIMM, i.e. no aliasing */ + NDD_NOBLK = 5, /* need to set a limit somewhere, but yes, this is likely overkill */ ND_IOCTL_MAX_BUFLEN = SZ_4M,
As Dexuan reports the NVDIMM_FAMILY_HYPERV platform is incompatible with the existing Linux namespace implementation because it uses NSLABEL_FLAG_LOCAL for x1-width PMEM interleave sets. Quirk it as an platform / DIMM that does not provide BLK-aperture access. Allow the libnvdimm core to assume no potential for aliasing. In case other implementations make the same mistake, provide a "noblk" module parameter to force-enable the quirk. Link: https://lkml.kernel.org/r/PU1P153MB0169977604493B82B662A01CBF920@PU1P153MB0169.APCP153.PROD.OUTLOOK.COM Reported-by: Dexuan Cui <decui@microsoft.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- drivers/acpi/nfit/core.c | 4 ++++ drivers/nvdimm/dimm_devs.c | 7 ++++++- drivers/nvdimm/label.c | 3 +++ drivers/nvdimm/namespace_devs.c | 6 ++++++ drivers/nvdimm/region_devs.c | 7 +++++++ include/linux/libnvdimm.h | 2 ++ 6 files changed, 28 insertions(+), 1 deletion(-)