| Message ID | 20200529220600.225320-3-vaibhav@linux.ibm.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Add support for reporting papr nvdimm health | expand |
On Sat, 2020-05-30 at 03:35 +0530, Vaibhav Jain wrote: > diff --git a/ndctl/lib/libndctl.c b/ndctl/lib/libndctl.c > index d76dbf7e17de..a52c2e208fbe 100644 > --- a/ndctl/lib/libndctl.c > +++ b/ndctl/lib/libndctl.c > @@ -799,6 +799,28 @@ static void parse_nfit_mem_flags(struct ndctl_dimm *dimm, char *flags) > ndctl_dimm_get_devname(dimm), flags); > } > > +static void parse_papr_flags(struct ndctl_dimm *dimm, char *flags) > +{ > + char *start, *end; > + > + start = flags; > + while ((end = strchr(start, ' '))) { > + *end = '\0'; > + if (strcmp(start, "not_armed") == 0) > + dimm->flags.f_arm = 1; > + else if (strcmp(start, "flush_fail") == 0) > + dimm->flags.f_flush = 1; > + else if (strcmp(start, "restore_fail") == 0) > + dimm->flags.f_restore = 1; > + else if (strcmp(start, "smart_notify") == 0) > + dimm->flags.f_smart = 1; > + start = end + 1; > + } > + if (end != start) > + dbg(ndctl_dimm_get_ctx(dimm), "%s: %s\n", > + ndctl_dimm_get_devname(dimm), flags); I think this would be more readable if ctx was obtained and saved in a 'ctx' variable at the start. Also, it might be better if 'flags' was included in the string - something like "%s flags: %s\n" > +} > + > static void parse_dimm_flags(struct ndctl_dimm *dimm, char *flags) > { > char *start, *end; > @@ -856,6 +878,12 @@ static void *add_bus(void *parent, int id, const char *ctl_base) > bus->revision = strtoul(buf, NULL, 0); > } > > + sprintf(path, "%s/device/of_node/compatible", ctl_base); > + if (sysfs_read_attr(ctx, path, buf) < 0) > + bus->has_of_node = 0; > + else > + bus->has_of_node = 1; > + > sprintf(path, "%s/device/nfit/dsm_mask", ctl_base); > if (sysfs_read_attr(ctx, path, buf) < 0) > bus->nfit_dsm_mask = 0; > @@ -964,6 +992,10 @@ NDCTL_EXPORT int ndctl_bus_has_nfit(struct ndctl_bus *bus) > return bus->has_nfit; > } > > +NDCTL_EXPORT int ndctl_bus_has_of_node(struct ndctl_bus *bus) > +{ > + return bus->has_of_node; > +} New libndctl APIs need to update the 'symbol script' - ndctl/lib/libndctl.sym. Create a new 'section' at the bottom, and add all new symbols to that section. The new section would be 'LIBNDCTL_24' (Everything going into a given release gets a new section in that file, so any subsequent patches - throughout the release cycle - that add new APIs can add them to this new section). > /** > * ndctl_bus_get_major - nd bus character device major number > * @bus: ndctl_bus instance returned from ndctl_bus_get_{first|next} > @@ -1441,6 +1473,47 @@ static int ndctl_bind(struct ndctl_ctx *ctx, struct kmod_module *module, > static int ndctl_unbind(struct ndctl_ctx *ctx, const char *devpath); > static struct kmod_module *to_module(struct ndctl_ctx *ctx, const char *alias); > > +static int add_papr_dimm(struct ndctl_dimm *dimm, const char *dimm_base) > +{ > + int rc = -ENODEV; > + char buf[SYSFS_ATTR_SIZE]; > + struct ndctl_ctx *ctx = dimm->bus->ctx; > + char *path = calloc(1, strlen(dimm_base) + 100); > + > + dbg(ctx, "Probing of_pmem dimm %d at %s\n", dimm->id, dimm_base); > + > + if (!path) > + return -ENOMEM; > + > + sprintf(path, "%s/../of_node/compatible", dimm_base); > + if (sysfs_read_attr(ctx, path, buf) < 0) > + goto out; Two things here: 1. Why not use the new ndctl_bus_has_of_node helper here? and 2. This looks redundant. add_papr_dimm() is only called if ndctl_bus_has_of_node() during add_dimm. > + > + Nit: double newline here > + dbg(ctx, "Compatible of_pmem dimm %d at %s\n", dimm->id, buf); > + /* construct path to the papr compatible dimm flags file */ > + sprintf(path, "%s/papr/flags", dimm_base); > + > + if (strcmp(buf, "ibm,pmemory") == 0 && > + sysfs_read_attr(ctx, path, buf) == 0) { > + > + dbg(ctx, "Found papr dimm %d at %s\n", dimm->id, buf); Throughout the series - it would be better to print messages such as: "%s: adding papr dimm with flags: %s\n", devname, buf This would result in the canonical dimm names - nmem1, nmem2 and so on being used instead of "dimm 1" which can be confusing just from a convention standpoint. Similarly for the print a few lines above this: "%s: of_pmem compatible dimm: %s\n", devname, buf (In this case, what does the 'compatible' sysfs attrubute contain? Is it useful to print here? - I havent't looked, just asking). > diff --git a/ndctl/lib/papr.c b/ndctl/lib/papr.c > new file mode 100644 > index 000000000000..5ddf2755a950 > --- /dev/null > +++ b/ndctl/lib/papr.c > @@ -0,0 +1,32 @@ > +/* > + * Copyright (C) 2020 IBM Corporation > + * > + * This program is free software; you can redistribute it and/or modify it > + * under the terms and conditions of the GNU Lesser General Public License, > + * version 2.1, 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 Lesser General Public License for > + * more details. > + */ Prefer SPDX license header instead fo the long form text for new files.
"Verma, Vishal L" <vishal.l.verma@intel.com> writes: > On Sat, 2020-05-30 at 03:35 +0530, Vaibhav Jain wrote: > >> diff --git a/ndctl/lib/libndctl.c b/ndctl/lib/libndctl.c >> index d76dbf7e17de..a52c2e208fbe 100644 >> --- a/ndctl/lib/libndctl.c >> +++ b/ndctl/lib/libndctl.c >> @@ -799,6 +799,28 @@ static void parse_nfit_mem_flags(struct ndctl_dimm *dimm, char *flags) >> ndctl_dimm_get_devname(dimm), flags); >> } >> >> +static void parse_papr_flags(struct ndctl_dimm *dimm, char *flags) >> +{ >> + char *start, *end; >> + >> + start = flags; >> + while ((end = strchr(start, ' '))) { >> + *end = '\0'; >> + if (strcmp(start, "not_armed") == 0) >> + dimm->flags.f_arm = 1; >> + else if (strcmp(start, "flush_fail") == 0) >> + dimm->flags.f_flush = 1; >> + else if (strcmp(start, "restore_fail") == 0) >> + dimm->flags.f_restore = 1; >> + else if (strcmp(start, "smart_notify") == 0) >> + dimm->flags.f_smart = 1; >> + start = end + 1; >> + } >> + if (end != start) >> + dbg(ndctl_dimm_get_ctx(dimm), "%s: %s\n", >> + ndctl_dimm_get_devname(dimm), flags); > > I think this would be more readable if ctx was obtained and saved in a > 'ctx' variable at the start. Also, it might be better if 'flags' was > included in the string - something like "%s flags: %s\n" Sure will get this updated in v6. This function definition was derived from parse_dimm_flags() hence looks this way. > >> +} >> + >> static void parse_dimm_flags(struct ndctl_dimm *dimm, char *flags) >> { >> char *start, *end; >> @@ -856,6 +878,12 @@ static void *add_bus(void *parent, int id, const char *ctl_base) >> bus->revision = strtoul(buf, NULL, 0); >> } >> >> + sprintf(path, "%s/device/of_node/compatible", ctl_base); >> + if (sysfs_read_attr(ctx, path, buf) < 0) >> + bus->has_of_node = 0; >> + else >> + bus->has_of_node = 1; >> + >> sprintf(path, "%s/device/nfit/dsm_mask", ctl_base); >> if (sysfs_read_attr(ctx, path, buf) < 0) >> bus->nfit_dsm_mask = 0; >> @@ -964,6 +992,10 @@ NDCTL_EXPORT int ndctl_bus_has_nfit(struct ndctl_bus *bus) >> return bus->has_nfit; >> } >> >> +NDCTL_EXPORT int ndctl_bus_has_of_node(struct ndctl_bus *bus) >> +{ >> + return bus->has_of_node; >> +} > > New libndctl APIs need to update the 'symbol script' - > ndctl/lib/libndctl.sym. Create a new 'section' at the bottom, and add > all new symbols to that section. The new section would be 'LIBNDCTL_24' > (Everything going into a given release gets a new section in that file, > so any subsequent patches - throughout the release cycle - that add new > APIs can add them to this new section). Sure and thanks for pointing this out. Will add this symbol to the library version script in v6 > >> /** >> * ndctl_bus_get_major - nd bus character device major number >> * @bus: ndctl_bus instance returned from ndctl_bus_get_{first|next} >> @@ -1441,6 +1473,47 @@ static int ndctl_bind(struct ndctl_ctx *ctx, struct kmod_module *module, >> static int ndctl_unbind(struct ndctl_ctx *ctx, const char *devpath); >> static struct kmod_module *to_module(struct ndctl_ctx *ctx, const char *alias); >> >> +static int add_papr_dimm(struct ndctl_dimm *dimm, const char *dimm_base) >> +{ >> + int rc = -ENODEV; >> + char buf[SYSFS_ATTR_SIZE]; >> + struct ndctl_ctx *ctx = dimm->bus->ctx; >> + char *path = calloc(1, strlen(dimm_base) + 100); >> + >> + dbg(ctx, "Probing of_pmem dimm %d at %s\n", dimm->id, dimm_base); >> + >> + if (!path) >> + return -ENOMEM; >> + >> + sprintf(path, "%s/../of_node/compatible", dimm_base); >> + if (sysfs_read_attr(ctx, path, buf) < 0) >> + goto out; > > Two things here: > 1. Why not use the new ndctl_bus_has_of_node helper here? and > 2. This looks redundant. add_papr_dimm() is only called if > ndctl_bus_has_of_node() during add_dimm. Presently we have two different nvdimm implementations: * papr-scm: handled by arch/powerpc/platforms/pseries/papr_scm kernel module. * nvdimm-n: handled by drivers/nvdimm/of_pmem kernel module. Both nvdimms are exposed to the kernel via device tree nodes but different 'compatible' properties. This patchset only adds support for 'papr-scm' compatible nvdimms. 'ndctl_bus_has_of_node()' simply indicates if the nvdimm has an open-firmware compatible devicetree node associated with it and doesnt necessarily indicate if its papr-scm compliant. Hence validating the 'compatible' attribute value is necessary here. Please see a more detailed info below regarding the 'compatible' sysfs attribute. > >> + >> + > > Nit: double newline here Sure. Will fix this in v6 > >> + dbg(ctx, "Compatible of_pmem dimm %d at %s\n", dimm->id, buf); >> + /* construct path to the papr compatible dimm flags file */ >> + sprintf(path, "%s/papr/flags", dimm_base); >> + >> + if (strcmp(buf, "ibm,pmemory") == 0 && >> + sysfs_read_attr(ctx, path, buf) == 0) { >> + >> + dbg(ctx, "Found papr dimm %d at %s\n", dimm->id, buf); > > Throughout the series - it would be better to print messages such as: > > "%s: adding papr dimm with flags: %s\n", devname, buf > > This would result in the canonical dimm names - nmem1, nmem2 and so on > being used instead of "dimm 1" which can be confusing just from a > convention standpoint. > > Similarly for the print a few lines above this: > > "%s: of_pmem compatible dimm: %s\n", devname, buf > Sure will address this in v6. > (In this case, what does the 'compatible' sysfs attrubute contain? Is it > useful to print here? - I havent't looked, just asking). For papr-scm compatible nvdimms: # cat /sys/class/nd/ndctl0/device/of_node/compatible ibm,pmemory This devicetree property is usually in format "<manufacturer>,<model>" that tells the which kernel module to bind to the device. The papr_scm kernel module indicates support for such device value in its device match table ( https://git.io/JfPzF ) so that kernel can call its probe-function. Dumping this value here will be useful in debugging probe failures of nvdimms that are exposed via open-firmware devicetree nodes but are not necessarily papr_scm specification compliant. > >> diff --git a/ndctl/lib/papr.c b/ndctl/lib/papr.c >> new file mode 100644 >> index 000000000000..5ddf2755a950 >> --- /dev/null >> +++ b/ndctl/lib/papr.c >> @@ -0,0 +1,32 @@ >> +/* >> + * Copyright (C) 2020 IBM Corporation >> + * >> + * This program is free software; you can redistribute it and/or modify it >> + * under the terms and conditions of the GNU Lesser General Public License, >> + * version 2.1, 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 Lesser General Public License for >> + * more details. >> + */ > > Prefer SPDX license header instead fo the long form text for new files. > Sure will address this in v6.
On Wed, 2020-06-03 at 15:27 +0530, Vaibhav Jain wrote: > > > > Two things here: > > 1. Why not use the new ndctl_bus_has_of_node helper here? and > > 2. This looks redundant. add_papr_dimm() is only called if > > ndctl_bus_has_of_node() during add_dimm. > Presently we have two different nvdimm implementations: > > * papr-scm: handled by arch/powerpc/platforms/pseries/papr_scm kernel module. > * nvdimm-n: handled by drivers/nvdimm/of_pmem kernel module. > > Both nvdimms are exposed to the kernel via device tree nodes but different > 'compatible' properties. This patchset only adds support for 'papr-scm' > compatible nvdimms. > > 'ndctl_bus_has_of_node()' simply indicates if the nvdimm has an > open-firmware compatible devicetree node associated with it and doesnt > necessarily indicate if its papr-scm compliant. > > Hence validating the 'compatible' attribute value is necessary here. > Please see a more detailed info below regarding the 'compatible' sysfs > attribute. > Understood - one more question: Would it be useful to wrap the 'compatible' check into an API similar to _has_of_node - say ndctl_bus_is_papr_compatible()? I'm not too strongly attached this, there is only one user so far after all, but it seemed like an easy thing that might get copy-pasted around in the future.
"Verma, Vishal L" <vishal.l.verma@intel.com> writes: > On Wed, 2020-06-03 at 15:27 +0530, Vaibhav Jain wrote: >> > >> > Two things here: >> > 1. Why not use the new ndctl_bus_has_of_node helper here? and >> > 2. This looks redundant. add_papr_dimm() is only called if >> > ndctl_bus_has_of_node() during add_dimm. >> Presently we have two different nvdimm implementations: >> >> * papr-scm: handled by arch/powerpc/platforms/pseries/papr_scm kernel module. >> * nvdimm-n: handled by drivers/nvdimm/of_pmem kernel module. >> >> Both nvdimms are exposed to the kernel via device tree nodes but different >> 'compatible' properties. This patchset only adds support for 'papr-scm' >> compatible nvdimms. >> >> 'ndctl_bus_has_of_node()' simply indicates if the nvdimm has an >> open-firmware compatible devicetree node associated with it and doesnt >> necessarily indicate if its papr-scm compliant. >> >> Hence validating the 'compatible' attribute value is necessary here. >> Please see a more detailed info below regarding the 'compatible' sysfs >> attribute. >> > Understood - one more question: > > Would it be useful to wrap the 'compatible' check into an API similar to > _has_of_node - say ndctl_bus_is_papr_compatible()? I'm not too strongly > attached this, there is only one user so far after all, but it seemed > like an easy thing that might get copy-pasted around in the future. Yes, sounds good to me. Should simplify the add_papr_dimm() function a bit as I can just call ndctl_bus_is_papr_compatible() and if true then setup the cmd_family. Will roll out this change in v6 iteration.
diff --git a/ndctl/lib/Makefile.am b/ndctl/lib/Makefile.am index d6be5c3acd26..e15bb2288575 100644 --- a/ndctl/lib/Makefile.am +++ b/ndctl/lib/Makefile.am @@ -23,6 +23,7 @@ libndctl_la_SOURCES =\ hpe1.c \ msft.c \ hyperv.c \ + papr.c \ ars.c \ firmware.c \ libndctl.c \ diff --git a/ndctl/lib/libndctl.c b/ndctl/lib/libndctl.c index d76dbf7e17de..a52c2e208fbe 100644 --- a/ndctl/lib/libndctl.c +++ b/ndctl/lib/libndctl.c @@ -799,6 +799,28 @@ static void parse_nfit_mem_flags(struct ndctl_dimm *dimm, char *flags) ndctl_dimm_get_devname(dimm), flags); } +static void parse_papr_flags(struct ndctl_dimm *dimm, char *flags) +{ + char *start, *end; + + start = flags; + while ((end = strchr(start, ' '))) { + *end = '\0'; + if (strcmp(start, "not_armed") == 0) + dimm->flags.f_arm = 1; + else if (strcmp(start, "flush_fail") == 0) + dimm->flags.f_flush = 1; + else if (strcmp(start, "restore_fail") == 0) + dimm->flags.f_restore = 1; + else if (strcmp(start, "smart_notify") == 0) + dimm->flags.f_smart = 1; + start = end + 1; + } + if (end != start) + dbg(ndctl_dimm_get_ctx(dimm), "%s: %s\n", + ndctl_dimm_get_devname(dimm), flags); +} + static void parse_dimm_flags(struct ndctl_dimm *dimm, char *flags) { char *start, *end; @@ -856,6 +878,12 @@ static void *add_bus(void *parent, int id, const char *ctl_base) bus->revision = strtoul(buf, NULL, 0); } + sprintf(path, "%s/device/of_node/compatible", ctl_base); + if (sysfs_read_attr(ctx, path, buf) < 0) + bus->has_of_node = 0; + else + bus->has_of_node = 1; + sprintf(path, "%s/device/nfit/dsm_mask", ctl_base); if (sysfs_read_attr(ctx, path, buf) < 0) bus->nfit_dsm_mask = 0; @@ -964,6 +992,10 @@ NDCTL_EXPORT int ndctl_bus_has_nfit(struct ndctl_bus *bus) return bus->has_nfit; } +NDCTL_EXPORT int ndctl_bus_has_of_node(struct ndctl_bus *bus) +{ + return bus->has_of_node; +} /** * ndctl_bus_get_major - nd bus character device major number * @bus: ndctl_bus instance returned from ndctl_bus_get_{first|next} @@ -1441,6 +1473,47 @@ static int ndctl_bind(struct ndctl_ctx *ctx, struct kmod_module *module, static int ndctl_unbind(struct ndctl_ctx *ctx, const char *devpath); static struct kmod_module *to_module(struct ndctl_ctx *ctx, const char *alias); +static int add_papr_dimm(struct ndctl_dimm *dimm, const char *dimm_base) +{ + int rc = -ENODEV; + char buf[SYSFS_ATTR_SIZE]; + struct ndctl_ctx *ctx = dimm->bus->ctx; + char *path = calloc(1, strlen(dimm_base) + 100); + + dbg(ctx, "Probing of_pmem dimm %d at %s\n", dimm->id, dimm_base); + + if (!path) + return -ENOMEM; + + sprintf(path, "%s/../of_node/compatible", dimm_base); + if (sysfs_read_attr(ctx, path, buf) < 0) + goto out; + + + dbg(ctx, "Compatible of_pmem dimm %d at %s\n", dimm->id, buf); + /* construct path to the papr compatible dimm flags file */ + sprintf(path, "%s/papr/flags", dimm_base); + + if (strcmp(buf, "ibm,pmemory") == 0 && + sysfs_read_attr(ctx, path, buf) == 0) { + + dbg(ctx, "Found papr dimm %d at %s\n", dimm->id, buf); + dimm->cmd_family = NVDIMM_FAMILY_PAPR; + + /* Parse dimm flags */ + parse_papr_flags(dimm, buf); + + /* Allocate monitor mode fd */ + dimm->health_eventfd = open(path, O_RDONLY|O_CLOEXEC); + rc = 0; + } + /* Ignore dimm if unknown type */ + + out: + free(path); + return rc; +} + static int add_nfit_dimm(struct ndctl_dimm *dimm, const char *dimm_base) { int i, rc = -1; @@ -1619,6 +1692,8 @@ static void *add_dimm(void *parent, int id, const char *dimm_base) if (ndctl_bus_has_nfit(bus)) { dimm->formats = formats; rc = add_nfit_dimm(dimm, dimm_base); + } else if (ndctl_bus_has_of_node(bus)) { + rc = add_papr_dimm(dimm, dimm_base); } if (rc == -ENODEV) { @@ -1636,6 +1711,9 @@ static void *add_dimm(void *parent, int id, const char *dimm_base) dimm->ops = msft_dimm_ops; if (dimm->cmd_family == NVDIMM_FAMILY_HYPERV) dimm->ops = hyperv_dimm_ops; + if (dimm->cmd_family == NVDIMM_FAMILY_PAPR) + dimm->ops = papr_dimm_ops; + out: if (rc) { err(ctx, "Unable to probe dimm:%d. Err:%d\n", id, rc); diff --git a/ndctl/lib/papr.c b/ndctl/lib/papr.c new file mode 100644 index 000000000000..5ddf2755a950 --- /dev/null +++ b/ndctl/lib/papr.c @@ -0,0 +1,32 @@ +/* + * Copyright (C) 2020 IBM Corporation + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU Lesser General Public License, + * version 2.1, 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 Lesser General Public License for + * more details. + */ +#include <stdint.h> +#include <stdlib.h> +#include <limits.h> +#include <util/log.h> +#include <ndctl.h> +#include <ndctl/libndctl.h> +#include <lib/private.h> + +static bool papr_cmd_is_supported(struct ndctl_dimm *dimm, int cmd) +{ + /* Handle this separately to support monitor mode */ + if (cmd == ND_CMD_SMART) + return true; + + return !!(dimm->cmd_mask & (1ULL << cmd)); +} + +struct ndctl_dimm_ops * const papr_dimm_ops = &(struct ndctl_dimm_ops) { + .cmd_is_supported = papr_cmd_is_supported, +}; diff --git a/ndctl/lib/private.h b/ndctl/lib/private.h index 2e537f0a8649..d90236b1f98b 100644 --- a/ndctl/lib/private.h +++ b/ndctl/lib/private.h @@ -167,6 +167,7 @@ struct ndctl_bus { int dimms_init; int regions_init; int has_nfit; + int has_of_node; char *bus_path; char *bus_buf; size_t buf_len; @@ -352,6 +353,7 @@ extern struct ndctl_dimm_ops * const intel_dimm_ops; extern struct ndctl_dimm_ops * const hpe1_dimm_ops; extern struct ndctl_dimm_ops * const msft_dimm_ops; extern struct ndctl_dimm_ops * const hyperv_dimm_ops; +extern struct ndctl_dimm_ops * const papr_dimm_ops; static inline struct ndctl_bus *cmd_to_bus(struct ndctl_cmd *cmd) { diff --git a/ndctl/libndctl.h b/ndctl/libndctl.h index 2580f433ade8..daf11b8ce4ea 100644 --- a/ndctl/libndctl.h +++ b/ndctl/libndctl.h @@ -119,6 +119,7 @@ struct ndctl_bus *ndctl_bus_get_next(struct ndctl_bus *bus); bus = ndctl_bus_get_next(bus)) struct ndctl_ctx *ndctl_bus_get_ctx(struct ndctl_bus *bus); int ndctl_bus_has_nfit(struct ndctl_bus *bus); +int ndctl_bus_has_of_node(struct ndctl_bus *bus); unsigned int ndctl_bus_get_major(struct ndctl_bus *bus); unsigned int ndctl_bus_get_minor(struct ndctl_bus *bus); const char *ndctl_bus_get_devname(struct ndctl_bus *bus); diff --git a/ndctl/ndctl.h b/ndctl/ndctl.h index 008f81cdeb9f..3b64f66d58cc 100644 --- a/ndctl/ndctl.h +++ b/ndctl/ndctl.h @@ -263,6 +263,7 @@ struct nd_cmd_pkg { #define NVDIMM_FAMILY_HPE2 2 #define NVDIMM_FAMILY_MSFT 3 #define NVDIMM_FAMILY_HYPERV 4 +#define NVDIMM_FAMILY_PAPR 5 #define ND_IOCTL_CALL _IOWR(ND_IOCTL, ND_CMD_CALL,\ struct nd_cmd_pkg)
Add necessary scaffolding in libndctl for dimms that support papr_scm specification[1]. Since there can be platforms that support Open-Firmware[2] but not the papr_scm specification, hence the changes proposed first add support for probing if the dimm bus supports Open-Firmware. This is done via querying for sysfs attribute 'of_node' in dimm device sysfs directory. If available newly introduced member 'struct ndctl_bus.has_of_node' is set. During the probe of the dimm and execution of add_dimm(), the newly introduced add_papr_dimm() is called if dimm bus reports supports Open-Firmware. Function add_papr_dimm() queries the 'compatible' device tree attribute and based on its value assign NVDIMM_FAMILY_PAPR to the dimm command family. In future, based on the contents of 'compatible' attribute more of_pmem dimm families can be queried. We also add support for parsing the dimm flags for NVDIMM_FAMILY_PAPR supporting nvdimms as described at [3]. A newly introduced function parse_papr_flags() reads the contents of this flag file and sets appropriate flag bits in 'struct ndctl_dimm.flags'. Also we advertise support for monitor mode by allocating a file descriptor to the dimm 'flags' file and assigning it to 'struct ndctl_dimm.health_event_fd'. The dimm-ops implementation for NVDIMM_FAMILY_PAPR is available in global variable 'papr_dimm_ops' which points to skeleton implementation in newly introduced file 'lib/papr.c'. References: [1] Documentation/powerpc/papr_hcalls.rst https://lore.kernel.org/linux-nvdimm/20200529214719.223344-2-vaibhav@linux.ibm.com [2] https://en.wikipedia.org/wiki/Open_Firmware [3] Documentation/ABI/testing/sysfs-bus-papr-pmem https://lore.kernel.org/linux-nvdimm/20200529214719.223344-4-vaibhav@linux.ibm.com Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com> --- Changelog: v4..v5: * Renamed file 'papr_scm.c' to 'papr.c' * s/NVDIMM_FAMILY_PAPR_SCM/NVDIMM_FAMILY_PAPR/g * s/papr_scm_dimm_ops/papr_dimm_ops/g * s/parse_papr_scm_flags()/parse_papr_flags()/g * Updated patch description & title to reflect new command family name and file-names. v3..v4: * None v2..v3: * Renamed add_of_pmem() to add_papr_dimm() [ Aneesh ] v1..v2: * Squashed the patch to parse dimm flags * Updated the dimm flags parsing to add case for 'restore_fail' and 'flush_fail'. * Renamed parse_of_pmem_flags() to parse_papr_scm_flags(). * Updated the path to dimm flags file to 'papr/flags'. * Introduced 'papr_scm.c' file in this patch rather than later in the patch series. * Update add_of_pmem_dimm() to parse dimm flags and enable monitoring only if 'ibm,pmemory' compatible nvdimm is found. --- ndctl/lib/Makefile.am | 1 + ndctl/lib/libndctl.c | 78 +++++++++++++++++++++++++++++++++++++++++++ ndctl/lib/papr.c | 32 ++++++++++++++++++ ndctl/lib/private.h | 2 ++ ndctl/libndctl.h | 1 + ndctl/ndctl.h | 1 + 6 files changed, 115 insertions(+) create mode 100644 ndctl/lib/papr.c