| Message ID | 150939642870.4577.1202379190708311186.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Oct 30, 2017 at 01:47:08PM -0700, Dan Williams wrote: > Per v1.6 of the NVDIMM_FAMILY_INTEL command set [1] some of the new > commands require rev-id 2. In addition to enabling ND_CMD_CALL for these > new function numbers, add a lookup table for revision-ids by family > and function number. > > [1]: http://pmem.io/documents/NVDIMM_DSM_Interface-V1.6.pdf > > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > drivers/acpi/nfit/core.c | 45 ++++++++++++++++++++++++++++++++++++++++----- > drivers/acpi/nfit/nfit.h | 26 +++++++++++++++++++++++++- > 2 files changed, 65 insertions(+), 6 deletions(-) > > diff --git a/drivers/acpi/nfit/core.c b/drivers/acpi/nfit/core.c > index 444832b372ec..1b2c613a9d8b 100644 > --- a/drivers/acpi/nfit/core.c > +++ b/drivers/acpi/nfit/core.c > @@ -370,6 +370,32 @@ static union acpi_object *acpi_label_info(acpi_handle handle) > return pkg_to_buf(buf.pointer); > } > > +static u8 nfit_dsm_revid(unsigned family, unsigned func) > +{ > + static const u8 revid_table[NVDIMM_FAMILY_MAX][32] = { > + [NVDIMM_FAMILY_INTEL] = { > + [NVDIMM_INTEL_GET_MODES] = 2, > + [NVDIMM_INTEL_GET_FWINFO] = 2, > + [NVDIMM_INTEL_START_FWUPDATE] = 2, > + [NVDIMM_INTEL_SEND_FWUPDATE] = 2, > + [NVDIMM_INTEL_FINISH_FWUPDATE] = 2, > + [NVDIMM_INTEL_QUERY_FWUPDATE] = 2, > + [NVDIMM_INTEL_SET_THRESHOLD] = 2, > + [NVDIMM_INTEL_INJECT_ERROR] = 2, > + }, > + }; > + u8 id; > + > + if (family > NVDIMM_FAMILY_MAX) > + return 0; > + if (func > 31) > + return 0; > + id = revid_table[family][func]; > + if (id == 0) > + return 1; /* default */ > + return id; > +} > + Elsewhere we find NVDIMM_FAMILY_MAX == NVDIMM_FAMILY_MSFT. Code is off by 1 in the dimension the array revid_table. This approach only ever calls revision ID 1 of a function that was initially defined in revision ID 1 of a spec. Is it required that FW returns same data for a given function if called with different revision IDs? I.e. could firmware implement a function N such that it returns different information if called with revision ID 2 versus being called with revision ID 1? E.g. a field that is reserved in revision 1 could have a definition in revision 2. My reading of the ACPI spec is this is acceptable. Not sure if this makes a difference w/ the Intel FW or not, but some of the structures returns were modified from the earlier spec. I have concerns that this approach can be correctly extended to other vendors. > int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, > unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc) > { > @@ -468,8 +494,15 @@ int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, > > out_obj = acpi_label_write(handle, p->in_offset, p->in_length, > p->in_buf); > - } else > - out_obj = acpi_evaluate_dsm(handle, guid, 1, func, &in_obj); > + } else { > + u8 revid; > + > + if (nfit_mem) > + revid = nfit_dsm_revid(nfit_mem->family, func); > + else > + revid = 1; > + out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj); > + } > > if (!out_obj) { > dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name, > @@ -1640,7 +1673,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, > * different command sets. Note, that checking for function0 (bit0) > * tells us if any commands are reachable through this GUID. > */ > - for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++) > + for (i = 0; i <= NVDIMM_FAMILY_MAX; i++) > if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) > if (family < 0 || i == default_dsm_family) > family = i; > @@ -1650,7 +1683,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, > if (override_dsm_mask && !disable_vendor_specific) > dsm_mask = override_dsm_mask; > else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { > - dsm_mask = 0x3fe; > + dsm_mask = NVDIMM_INTEL_CMDMASK; > if (disable_vendor_specific) > dsm_mask &= ~(1 << ND_CMD_VENDOR); > } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) { > @@ -1670,7 +1703,9 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, > > guid = to_nfit_uuid(nfit_mem->family); > for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) > - if (acpi_check_dsm(adev_dimm->handle, guid, 1, 1ULL << i)) > + if (acpi_check_dsm(adev_dimm->handle, guid, > + nfit_dsm_revid(nfit_mem->family, i), > + 1ULL << i)) > set_bit(i, &nfit_mem->dsm_mask); > > obj = acpi_label_info(adev_dimm->handle); > diff --git a/drivers/acpi/nfit/nfit.h b/drivers/acpi/nfit/nfit.h > index b987196bf132..341be9511d0e 100644 > --- a/drivers/acpi/nfit/nfit.h > +++ b/drivers/acpi/nfit/nfit.h > @@ -24,7 +24,7 @@ > /* ACPI 6.1 */ > #define UUID_NFIT_BUS "2f10e7a4-9e91-11e4-89d3-123b93f75cba" > > -/* http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf */ > +/* http://pmem.io/documents/NVDIMM_DSM_Interface-V1.6.pdf */ > #define UUID_NFIT_DIMM "4309ac30-0d11-11e4-9191-0800200c9a66" > > /* https://github.com/HewlettPackard/hpe-nvm/blob/master/Documentation/ */ > @@ -38,12 +38,36 @@ > | ACPI_NFIT_MEM_RESTORE_FAILED | ACPI_NFIT_MEM_FLUSH_FAILED \ > | ACPI_NFIT_MEM_NOT_ARMED | ACPI_NFIT_MEM_MAP_FAILED) > > +#define NVDIMM_FAMILY_MAX NVDIMM_FAMILY_MSFT Suggest this be defined immediately after where NVDIMM_FAMILY_MSFT is defined. That way if a new family is ever defined, its more obvious that NVDIMM_FAMILY_MAX needs to be redefined as well. > + > #define NVDIMM_STANDARD_CMDMASK \ > (1 << ND_CMD_SMART | 1 << ND_CMD_SMART_THRESHOLD | 1 << ND_CMD_DIMM_FLAGS \ > | 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA \ > | 1 << ND_CMD_SET_CONFIG_DATA | 1 << ND_CMD_VENDOR_EFFECT_LOG_SIZE \ > | 1 << ND_CMD_VENDOR_EFFECT_LOG | 1 << ND_CMD_VENDOR) > > +/* > + * Command numbers that the kernel needs to know about to handle > + * non-default DSM revision ids > + */ > +enum nvdimm_family_cmds { > + NVDIMM_INTEL_GET_MODES = 11, > + NVDIMM_INTEL_GET_FWINFO = 12, > + NVDIMM_INTEL_START_FWUPDATE = 13, > + NVDIMM_INTEL_SEND_FWUPDATE = 14, > + NVDIMM_INTEL_FINISH_FWUPDATE = 15, > + NVDIMM_INTEL_QUERY_FWUPDATE = 16, > + NVDIMM_INTEL_SET_THRESHOLD = 17, > + NVDIMM_INTEL_INJECT_ERROR = 18, > +}; > + > +#define NVDIMM_INTEL_CMDMASK \ > +(NVDIMM_STANDARD_CMDMASK | 1 << NVDIMM_INTEL_GET_MODES \ > + | 1 << NVDIMM_INTEL_GET_FWINFO | 1 << NVDIMM_INTEL_START_FWUPDATE \ > + | 1 << NVDIMM_INTEL_SEND_FWUPDATE | 1 << NVDIMM_INTEL_FINISH_FWUPDATE \ > + | 1 << NVDIMM_INTEL_QUERY_FWUPDATE | 1 << NVDIMM_INTEL_SET_THRESHOLD \ > + | 1 << NVDIMM_INTEL_INJECT_ERROR) > + > enum nfit_uuids { > /* for simplicity alias the uuid index with the family id */ > NFIT_DEV_DIMM = NVDIMM_FAMILY_INTEL, > > _______________________________________________ > Linux-nvdimm mailing list > Linux-nvdimm@lists.01.org > https://lists.01.org/mailman/listinfo/linux-nvdimm
On Tue, Oct 31, 2017 at 2:27 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote: > On Mon, Oct 30, 2017 at 01:47:08PM -0700, Dan Williams wrote: >> Per v1.6 of the NVDIMM_FAMILY_INTEL command set [1] some of the new >> commands require rev-id 2. In addition to enabling ND_CMD_CALL for these >> new function numbers, add a lookup table for revision-ids by family >> and function number. >> >> [1]: http://pmem.io/documents/NVDIMM_DSM_Interface-V1.6.pdf >> >> Signed-off-by: Dan Williams <dan.j.williams@intel.com> >> --- >> drivers/acpi/nfit/core.c | 45 ++++++++++++++++++++++++++++++++++++++++----- >> drivers/acpi/nfit/nfit.h | 26 +++++++++++++++++++++++++- >> 2 files changed, 65 insertions(+), 6 deletions(-) >> >> diff --git a/drivers/acpi/nfit/core.c b/drivers/acpi/nfit/core.c >> index 444832b372ec..1b2c613a9d8b 100644 >> --- a/drivers/acpi/nfit/core.c >> +++ b/drivers/acpi/nfit/core.c >> @@ -370,6 +370,32 @@ static union acpi_object *acpi_label_info(acpi_handle handle) >> return pkg_to_buf(buf.pointer); >> } >> >> +static u8 nfit_dsm_revid(unsigned family, unsigned func) >> +{ >> + static const u8 revid_table[NVDIMM_FAMILY_MAX][32] = { >> + [NVDIMM_FAMILY_INTEL] = { >> + [NVDIMM_INTEL_GET_MODES] = 2, >> + [NVDIMM_INTEL_GET_FWINFO] = 2, >> + [NVDIMM_INTEL_START_FWUPDATE] = 2, >> + [NVDIMM_INTEL_SEND_FWUPDATE] = 2, >> + [NVDIMM_INTEL_FINISH_FWUPDATE] = 2, >> + [NVDIMM_INTEL_QUERY_FWUPDATE] = 2, >> + [NVDIMM_INTEL_SET_THRESHOLD] = 2, >> + [NVDIMM_INTEL_INJECT_ERROR] = 2, >> + }, >> + }; >> + u8 id; >> + >> + if (family > NVDIMM_FAMILY_MAX) >> + return 0; >> + if (func > 31) >> + return 0; >> + id = revid_table[family][func]; >> + if (id == 0) >> + return 1; /* default */ >> + return id; >> +} >> + > > > Elsewhere we find NVDIMM_FAMILY_MAX == NVDIMM_FAMILY_MSFT. > Code is off by 1 in the dimension the array revid_table. Nice catch, will fix. At least the compiler will complain if we use c99 style initialization for that NVDIMM_FAMILY_MAX index (error: array index in initializer exceeds array bounds). > This approach only ever calls revision ID 1 of a function that > was initially defined in revision ID 1 of a spec. > > Is it required that FW returns same data for a given function if > called with different revision IDs? I.e. could firmware implement > a function N such that it returns different information if called > with revision ID 2 versus being called with revision ID 1? > E.g. a field that is reserved in revision 1 could have a definition > in revision 2. My reading of the ACPI spec is this is acceptable. Yes, this can happen. However, for the kernel implementation it can decide to move to rev-id2 at its leisure since ACPI mandates that rev-id1 implementations stick around, and if the function number was reserved in rev-id1 the kernel will already not support it. > > Not sure if this makes a difference w/ the Intel FW or not, but > some of the structures returns were modified from the earlier > spec. I have concerns that this approach can be correctly > extended to other vendors. > [..] >> +#define NVDIMM_FAMILY_MAX NVDIMM_FAMILY_MSFT > > > Suggest this be defined immediately after where NVDIMM_FAMILY_MSFT > is defined. That way if a new family is ever defined, its more obvious > that NVDIMM_FAMILY_MAX needs to be redefined as well. These are currently defined in the userspace api header and that value is not relevant to userspace since the kernel is free to support more families than NVDIMM_FAMILY_MAX would imply. Also, if we as an industry continue to add NVDIMM families something is broken with the standardization process.
On Tue, Oct 31, 2017 at 02:42:01PM -0700, Dan Williams wrote: > On Tue, Oct 31, 2017 at 2:27 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote: > > On Mon, Oct 30, 2017 at 01:47:08PM -0700, Dan Williams wrote: > >> Per v1.6 of the NVDIMM_FAMILY_INTEL command set [1] some of the new > >> commands require rev-id 2. In addition to enabling ND_CMD_CALL for these > >> new function numbers, add a lookup table for revision-ids by family > >> and function number. > >> > >> [1]: http://pmem.io/documents/NVDIMM_DSM_Interface-V1.6.pdf > >> > >> Signed-off-by: Dan Williams <dan.j.williams@intel.com> > >> --- > >> drivers/acpi/nfit/core.c | 45 ++++++++++++++++++++++++++++++++++++++++----- > >> drivers/acpi/nfit/nfit.h | 26 +++++++++++++++++++++++++- > >> 2 files changed, 65 insertions(+), 6 deletions(-) > >> > >> diff --git a/drivers/acpi/nfit/core.c b/drivers/acpi/nfit/core.c > >> index 444832b372ec..1b2c613a9d8b 100644 > >> --- a/drivers/acpi/nfit/core.c > >> +++ b/drivers/acpi/nfit/core.c > >> @@ -370,6 +370,32 @@ static union acpi_object *acpi_label_info(acpi_handle handle) > >> return pkg_to_buf(buf.pointer); > >> } > >> > >> +static u8 nfit_dsm_revid(unsigned family, unsigned func) > >> +{ > >> + static const u8 revid_table[NVDIMM_FAMILY_MAX][32] = { > >> + [NVDIMM_FAMILY_INTEL] = { > >> + [NVDIMM_INTEL_GET_MODES] = 2, > >> + [NVDIMM_INTEL_GET_FWINFO] = 2, > >> + [NVDIMM_INTEL_START_FWUPDATE] = 2, > >> + [NVDIMM_INTEL_SEND_FWUPDATE] = 2, > >> + [NVDIMM_INTEL_FINISH_FWUPDATE] = 2, > >> + [NVDIMM_INTEL_QUERY_FWUPDATE] = 2, > >> + [NVDIMM_INTEL_SET_THRESHOLD] = 2, > >> + [NVDIMM_INTEL_INJECT_ERROR] = 2, > >> + }, > >> + }; > >> + u8 id; > >> + > >> + if (family > NVDIMM_FAMILY_MAX) > >> + return 0; > >> + if (func > 31) > >> + return 0; > >> + id = revid_table[family][func]; > >> + if (id == 0) > >> + return 1; /* default */ > >> + return id; > >> +} > >> + > > > > > > Elsewhere we find NVDIMM_FAMILY_MAX == NVDIMM_FAMILY_MSFT. > > Code is off by 1 in the dimension the array revid_table. > > Nice catch, will fix. > > At least the compiler will complain if we use c99 style initialization > for that NVDIMM_FAMILY_MAX index (error: array index in initializer > exceeds array bounds). > > > This approach only ever calls revision ID 1 of a function that > > was initially defined in revision ID 1 of a spec. > > > > Is it required that FW returns same data for a given function if > > called with different revision IDs? I.e. could firmware implement > > a function N such that it returns different information if called > > with revision ID 2 versus being called with revision ID 1? > > E.g. a field that is reserved in revision 1 could have a definition > > in revision 2. My reading of the ACPI spec is this is acceptable. > > Yes, this can happen. However, for the kernel implementation it can > decide to move to rev-id2 at its leisure since ACPI mandates that > rev-id1 implementations stick around, and if the function number was > reserved in rev-id1 the kernel will already not support it. Unfortunately, we can't do that. The kernel will still need to support firmware that only knows about Rev ID 1. Example, HPE currently supports customers w/ NVDIMMS with DSM that only responds to rev id 1. Say in the future we ship a new generation of NVDIMMs that extends the DSM health function. We could make that additional info available only for rev id 2. If the table above were extended to call health function with rev id 2 so that we get the additional info for the new nvdimm, that call will stop working on what will then be legacy systems that only know rev id 1. > > > > > Not sure if this makes a difference w/ the Intel FW or not, but > > some of the structures returns were modified from the earlier > > spec. I have concerns that this approach can be correctly > > extended to other vendors. > > > [..] > >> +#define NVDIMM_FAMILY_MAX NVDIMM_FAMILY_MSFT > > > > > > Suggest this be defined immediately after where NVDIMM_FAMILY_MSFT > > is defined. That way if a new family is ever defined, its more obvious > > that NVDIMM_FAMILY_MAX needs to be redefined as well. > > These are currently defined in the userspace api header and that value > is not relevant to userspace since the kernel is free to support more > families than NVDIMM_FAMILY_MAX would imply. Also, if we as an > industry continue to add NVDIMM families something is broken with the > standardization process. > -- > To unsubscribe from this list: send the line "unsubscribe linux-acpi" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Oct 31, 2017 at 3:05 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote: > On Tue, Oct 31, 2017 at 02:42:01PM -0700, Dan Williams wrote: [..] >> Yes, this can happen. However, for the kernel implementation it can >> decide to move to rev-id2 at its leisure since ACPI mandates that >> rev-id1 implementations stick around, and if the function number was >> reserved in rev-id1 the kernel will already not support it. > > Unfortunately, we can't do that. The kernel will still need to support > firmware that only knows about Rev ID 1. > > Example, HPE currently supports customers w/ NVDIMMS with DSM that only > responds to rev id 1. > > Say in the future we ship a new generation of NVDIMMs > that extends the DSM health function. We could make that > additional info available only for rev id 2. If the table > above were extended to call health function with rev id 2 > so that we get the additional info for the new nvdimm, that > call will stop working on what will then be legacy systems > that only know rev id 1. Right, I'm explicitly not supporting that degree of compatibility freedom. If a platform wants to change the payload of an existing function it had better expand the payload with new information that can optionally be ignored by older kernels rather than spin the revid. The kernel can't break legacy and neither should the platform, either the kernel can call revid1 forever, the function is only available in revid2 or later, or that function number is burned and we need to live with the breakage that the platform shipped. We are already in this position with the "smart threshold" payload that has been broken in recent revisions of the Intel spec. In this specific case we have ndctl binaries in the wild that are broken, but are somewhat saved by the fact that the interface for *setting* alarms and thresholds is brand new in v1.6. So my plan is to document that smart threshold is broken and require new ndctl binaries with v1.6 support to implement the new format via the ND_CMD_CALL interface. The fact that this compatibility management is on a case by case basis and somewhat painful is a feature. Platform implementations must hide DSM update thrash from the kernel and shipping binaries as much as possible.
On Tue, Oct 31, 2017 at 04:19:32PM -0700, Dan Williams wrote: > On Tue, Oct 31, 2017 at 3:05 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote: > > On Tue, Oct 31, 2017 at 02:42:01PM -0700, Dan Williams wrote: > [..] > >> Yes, this can happen. However, for the kernel implementation it can > >> decide to move to rev-id2 at its leisure since ACPI mandates that > >> rev-id1 implementations stick around, and if the function number was > >> reserved in rev-id1 the kernel will already not support it. > > > > Unfortunately, we can't do that. The kernel will still need to support > > firmware that only knows about Rev ID 1. > > > > Example, HPE currently supports customers w/ NVDIMMS with DSM that only > > responds to rev id 1. > > > > Say in the future we ship a new generation of NVDIMMs > > that extends the DSM health function. We could make that > > additional info available only for rev id 2. If the table > > above were extended to call health function with rev id 2 > > so that we get the additional info for the new nvdimm, that > > call will stop working on what will then be legacy systems > > that only know rev id 1. > > Right, I'm explicitly not supporting that degree of compatibility > freedom. If a platform wants to change the payload of an existing > function it had better expand the payload with new information that > can optionally be ignored by older kernels rather than spin the revid. > The kernel can't break legacy and neither should the platform, either > the kernel can call revid1 forever, the function is only available in > revid2 or later, or that function number is burned and we need to live > with the breakage that the platform shipped. Old software would still be calling firmware with revision id 1 and firmware would still respond with the revision id 1 version of the call. So there is no problem there. This would be the case for legacy firmware that only understand revision id 1, or new firmware that understands both revision id 1 and revision id 2. Calling a function with revision id 1 should get the revision id 1 version of the call. If not, then bug in firmware. The idea behind revision ids here is to allow both software and firmware to progress but still work together. Firmware can extend functionality by increasing revision ID, but will still answer to old software that calls with older revision IDS. Likewise, OS can add support for new revision IDs/functions but still needs to be aware that firmware at customer sites may be legacy firmware that doesn't understand the newer firmware revision. So software needs to call firmware with revision IDs that the firmware understands. Specifically, I am pushing back on the idea that the OS could just start calling a function with revision ID 2. That doesn't work on firmware that only understands revision ID 1. The call will fail. > > We are already in this position with the "smart threshold" payload > that has been broken in recent revisions of the Intel spec. In this > specific case we have ndctl binaries in the wild that are broken, but > are somewhat saved by the fact that the interface for *setting* alarms > and thresholds is brand new in v1.6. So my plan is to document that > smart threshold is broken and require new ndctl binaries with v1.6 > support to implement the new format via the ND_CMD_CALL interface. The > fact that this compatibility management is on a case by case basis and > somewhat painful is a feature. Platform implementations must hide DSM > update thrash from the kernel and shipping binaries as much as > possible. I haven't thought through the changes in v1.6. The threshold one doesn't seem backwards compatible and it would have been better for this to have been caught during design review. Oh, well..... Fortunately, this isn't a customer/legacy bios issue yet, so solutions are a bit more flexible. I am concerned how the ND_CMD_CALL interface can handle revision ID changes in general. The interface doesn't specify a revision ID so it can't control the revision of data being returned to the level it can handle. Likewise, there is no mechanism to communicate back to the application the revision of the call being returned. This make me think that any application that uses ND_CMD_CALL will be rev locked to specific kernel revisions. I need to think about this more.
On Wed, Nov 1, 2017 at 10:35 AM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote: > On Tue, Oct 31, 2017 at 04:19:32PM -0700, Dan Williams wrote: >> On Tue, Oct 31, 2017 at 3:05 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote: >> > On Tue, Oct 31, 2017 at 02:42:01PM -0700, Dan Williams wrote: >> [..] >> >> Yes, this can happen. However, for the kernel implementation it can >> >> decide to move to rev-id2 at its leisure since ACPI mandates that >> >> rev-id1 implementations stick around, and if the function number was >> >> reserved in rev-id1 the kernel will already not support it. >> > >> > Unfortunately, we can't do that. The kernel will still need to support >> > firmware that only knows about Rev ID 1. >> > >> > Example, HPE currently supports customers w/ NVDIMMS with DSM that only >> > responds to rev id 1. >> > >> > Say in the future we ship a new generation of NVDIMMs >> > that extends the DSM health function. We could make that >> > additional info available only for rev id 2. If the table >> > above were extended to call health function with rev id 2 >> > so that we get the additional info for the new nvdimm, that >> > call will stop working on what will then be legacy systems >> > that only know rev id 1. >> >> Right, I'm explicitly not supporting that degree of compatibility >> freedom. If a platform wants to change the payload of an existing >> function it had better expand the payload with new information that >> can optionally be ignored by older kernels rather than spin the revid. >> The kernel can't break legacy and neither should the platform, either >> the kernel can call revid1 forever, the function is only available in >> revid2 or later, or that function number is burned and we need to live >> with the breakage that the platform shipped. > > > Old software would still be calling firmware with revision id 1 and > firmware would still respond with the revision id 1 version of the > call. So there is no problem there. This would be the case for > legacy firmware that only understand revision id 1, or new firmware > that understands both revision id 1 and revision id 2. Calling a > function with revision id 1 should get the revision id 1 version > of the call. If not, then bug in firmware. > > The idea behind revision ids here is to allow both software and firmware > to progress but still work together. Firmware can extend functionality > by increasing revision ID, but will still answer to old software that > calls with older revision IDS. > > Likewise, OS can add support for new revision IDs/functions but still > needs to be aware that firmware at customer sites may be legacy firmware that > doesn't understand the newer firmware revision. So software needs > to call firmware with revision IDs that the firmware understands. > > Specifically, I am pushing back on the idea that the OS could > just start calling a function with revision ID 2. That doesn't > work on firmware that only understands revision ID 1. The call > will fail. Right, the OS can only switch to rev-id2 of the same function after a long deprecation period. The timeframe to deprecate support for rev-id1 and switch the kernel to start using rev-id2 of the same function is on the order of several years until all old hardware with only rev-id1 is end-of-life. The way to upgrade functionality in a manner that does not break the world is to extend payloads with optional functionality. This is why DSM is a dangerous interface, it gives entirely too much power to the platform to do something incompatible, so we should mitigate that by communicating to platform implementers that the kernel will continue to use rev-id1 indefinitely for a given function.
diff --git a/drivers/acpi/nfit/core.c b/drivers/acpi/nfit/core.c index 444832b372ec..1b2c613a9d8b 100644 --- a/drivers/acpi/nfit/core.c +++ b/drivers/acpi/nfit/core.c @@ -370,6 +370,32 @@ static union acpi_object *acpi_label_info(acpi_handle handle) return pkg_to_buf(buf.pointer); } +static u8 nfit_dsm_revid(unsigned family, unsigned func) +{ + static const u8 revid_table[NVDIMM_FAMILY_MAX][32] = { + [NVDIMM_FAMILY_INTEL] = { + [NVDIMM_INTEL_GET_MODES] = 2, + [NVDIMM_INTEL_GET_FWINFO] = 2, + [NVDIMM_INTEL_START_FWUPDATE] = 2, + [NVDIMM_INTEL_SEND_FWUPDATE] = 2, + [NVDIMM_INTEL_FINISH_FWUPDATE] = 2, + [NVDIMM_INTEL_QUERY_FWUPDATE] = 2, + [NVDIMM_INTEL_SET_THRESHOLD] = 2, + [NVDIMM_INTEL_INJECT_ERROR] = 2, + }, + }; + u8 id; + + if (family > NVDIMM_FAMILY_MAX) + return 0; + if (func > 31) + return 0; + id = revid_table[family][func]; + if (id == 0) + return 1; /* default */ + return id; +} + int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc) { @@ -468,8 +494,15 @@ int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, out_obj = acpi_label_write(handle, p->in_offset, p->in_length, p->in_buf); - } else - out_obj = acpi_evaluate_dsm(handle, guid, 1, func, &in_obj); + } else { + u8 revid; + + if (nfit_mem) + revid = nfit_dsm_revid(nfit_mem->family, func); + else + revid = 1; + out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj); + } if (!out_obj) { dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name, @@ -1640,7 +1673,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this GUID. */ - for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++) + for (i = 0; i <= NVDIMM_FAMILY_MAX; i++) if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) if (family < 0 || i == default_dsm_family) family = i; @@ -1650,7 +1683,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, if (override_dsm_mask && !disable_vendor_specific) dsm_mask = override_dsm_mask; else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { - dsm_mask = 0x3fe; + dsm_mask = NVDIMM_INTEL_CMDMASK; if (disable_vendor_specific) dsm_mask &= ~(1 << ND_CMD_VENDOR); } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) { @@ -1670,7 +1703,9 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, guid = to_nfit_uuid(nfit_mem->family); for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) - if (acpi_check_dsm(adev_dimm->handle, guid, 1, 1ULL << i)) + if (acpi_check_dsm(adev_dimm->handle, guid, + nfit_dsm_revid(nfit_mem->family, i), + 1ULL << i)) set_bit(i, &nfit_mem->dsm_mask); obj = acpi_label_info(adev_dimm->handle); diff --git a/drivers/acpi/nfit/nfit.h b/drivers/acpi/nfit/nfit.h index b987196bf132..341be9511d0e 100644 --- a/drivers/acpi/nfit/nfit.h +++ b/drivers/acpi/nfit/nfit.h @@ -24,7 +24,7 @@ /* ACPI 6.1 */ #define UUID_NFIT_BUS "2f10e7a4-9e91-11e4-89d3-123b93f75cba" -/* http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf */ +/* http://pmem.io/documents/NVDIMM_DSM_Interface-V1.6.pdf */ #define UUID_NFIT_DIMM "4309ac30-0d11-11e4-9191-0800200c9a66" /* https://github.com/HewlettPackard/hpe-nvm/blob/master/Documentation/ */ @@ -38,12 +38,36 @@ | ACPI_NFIT_MEM_RESTORE_FAILED | ACPI_NFIT_MEM_FLUSH_FAILED \ | ACPI_NFIT_MEM_NOT_ARMED | ACPI_NFIT_MEM_MAP_FAILED) +#define NVDIMM_FAMILY_MAX NVDIMM_FAMILY_MSFT + #define NVDIMM_STANDARD_CMDMASK \ (1 << ND_CMD_SMART | 1 << ND_CMD_SMART_THRESHOLD | 1 << ND_CMD_DIMM_FLAGS \ | 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA \ | 1 << ND_CMD_SET_CONFIG_DATA | 1 << ND_CMD_VENDOR_EFFECT_LOG_SIZE \ | 1 << ND_CMD_VENDOR_EFFECT_LOG | 1 << ND_CMD_VENDOR) +/* + * Command numbers that the kernel needs to know about to handle + * non-default DSM revision ids + */ +enum nvdimm_family_cmds { + NVDIMM_INTEL_GET_MODES = 11, + NVDIMM_INTEL_GET_FWINFO = 12, + NVDIMM_INTEL_START_FWUPDATE = 13, + NVDIMM_INTEL_SEND_FWUPDATE = 14, + NVDIMM_INTEL_FINISH_FWUPDATE = 15, + NVDIMM_INTEL_QUERY_FWUPDATE = 16, + NVDIMM_INTEL_SET_THRESHOLD = 17, + NVDIMM_INTEL_INJECT_ERROR = 18, +}; + +#define NVDIMM_INTEL_CMDMASK \ +(NVDIMM_STANDARD_CMDMASK | 1 << NVDIMM_INTEL_GET_MODES \ + | 1 << NVDIMM_INTEL_GET_FWINFO | 1 << NVDIMM_INTEL_START_FWUPDATE \ + | 1 << NVDIMM_INTEL_SEND_FWUPDATE | 1 << NVDIMM_INTEL_FINISH_FWUPDATE \ + | 1 << NVDIMM_INTEL_QUERY_FWUPDATE | 1 << NVDIMM_INTEL_SET_THRESHOLD \ + | 1 << NVDIMM_INTEL_INJECT_ERROR) + enum nfit_uuids { /* for simplicity alias the uuid index with the family id */ NFIT_DEV_DIMM = NVDIMM_FAMILY_INTEL,
Per v1.6 of the NVDIMM_FAMILY_INTEL command set [1] some of the new commands require rev-id 2. In addition to enabling ND_CMD_CALL for these new function numbers, add a lookup table for revision-ids by family and function number. [1]: http://pmem.io/documents/NVDIMM_DSM_Interface-V1.6.pdf Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- drivers/acpi/nfit/core.c | 45 ++++++++++++++++++++++++++++++++++++++++----- drivers/acpi/nfit/nfit.h | 26 +++++++++++++++++++++++++- 2 files changed, 65 insertions(+), 6 deletions(-)