apci, nfit: DSM improvements
Subsystems (2)
v2 cover apci, nfit: DSM improvements
The first two patches in this series are motivated by feedback from distribution developers, test engineers, and management tool developers. We need the ability to test and support Linux and management tools on systems that support more than one DSM family for NVDIMM-N.
We also need the ability to test and support new functions without requiring users to update their kernel. These changes will also facilate development as we move toward DSM standardization. The ability to restrict DSM functions to the currently documented set is still in place.
The third patch cleans up a cosmetic/modinfo parsing issue with one of the existing module parameters.
Changes in v2:
- Switched from allowing all functions advertised by the firmware through function 0 to an override module parameter.
Linda Knippers (3):
Allow override of built-in bitmasks for NVDIMM DSMs Allow specifying a default DSM family Remove unnecessary newline
drivers/acpi/nfit/core.c | 36 ++++++++++++++++++++++++++++++------ 1 file changed, 30 insertions(+), 6 deletions(-)
Signature · 6 lines
1.8.3.1 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
PATCH v2 1/3 Allow override of built-in bitmasks for NVDIMM DSMs
As it is today, we can't enable or test new NVDIMM management functions provided by new firmware and tools without changing the kernel. We also can't prevent documented DSM functions from being called in the case of buggy firmware. This patch provides a module parameter that overrides the DSM function mask that is built into the kernel.
If the "disable_vendor_specific" module parameter is also used we ignore the new parameter.
--- drivers/acpi/nfit/core.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/nfit/core.c +7 -1
@@ -51,6 +51,10 @@MODULE_PARM_DESC (disable_vendor_specific,"Limit commands to the publicly specified set\n" ); +static unsigned long override_dsm_mask; +module_param (override_dsm_mask, ulong,S_IRUGO ); +MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);
@@ -1402,7 +1406,9 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i; -if (nfit_mem->family ==NVDIMM_FAMILY_INTEL ) { +if (override_dsm_mask && !disable_vendor_specific) + dsm_mask = override_dsm_mask; +else if (nfit_mem->family ==NVDIMM_FAMILY_INTEL ) { dsm_mask =0x3fe ;if (disable_vendor_specific) dsm_mask &= ~(1 <<ND_CMD_VENDOR );
Signature · 6 lines
1.8.3.1 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 1/3] Allow override of built-in bitmasks for NVDIMM DSMs
Quoted 11 lines · Linda Knippers
As it is today, we can't enable or test new NVDIMM management functions provided by new firmware and tools without changing the kernel. We also can't prevent documented DSM functions from being called in the case of buggy firmware. This patch provides a module parameter that overrides the DSM function mask that is built into the kernel.
If the "
disable_vendor_specific" module parameter is also used we ignore the new parameter.Signed-off-by Linda Knippers <[email protected]>---
Looks good to me,
Signature · 10 lines
Johannes Thumshirn Storage [email protected] +49 911 74053 689 SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg GF: Felix Imendörffer, Jane Smithard, Graham Norton HRB 21284 (AG Nürnberg) Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
PATCH v2 2/3 Allow specifying a default DSM family
In reply to Linda KnippersProvide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)
drivers/acpi/nfit/core.c +22 -4
@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);
@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family; +else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;
Signature · 6 lines
1.8.3.1 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;
drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }
I think we can make this simpler and more deterministic with a "force" option? Something like:
@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
* tells us if any commands are reachable through this uuid. */
for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
- if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
- break;
+ if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
+ if (force_dsm_family < 0)
+ break;
+ else if (i == force_dsm_family)
+ break;
+ }
/* limit the supported commands to those that are publicly documented */
nfit_mem->family = i;...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
Quoted 84 lines · Dan Williams
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }I think we can make this simpler and more deterministic with a "force" option? Something like:
drivers/acpi/nfit/core.c +6
-2@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (force_dsm_family <0 ) +break ; +else if (i == force_dsm_family) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;
Quoted 3 lines
...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
It's more complicated when you have a mixture of NVDIMM types, where not all NVDIMMs support the same families. For example, you could have an Intel NVDIMM in the same system as an HPE NVDIMM-N. The reason I called it "default" is because it should be the default for all devices that support that family but I didn't want other types of NVDIMMs to end up with no family at all.
-- ljk
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
Quoted 85 lines · Linda Knippers
Quoted 84 lines · Dan Williams
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }I think we can make this simpler and more deterministic with a "force" option? Something like:
drivers/acpi/nfit/core.c +6
-2@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (force_dsm_family <0 ) +break ; +else if (i == force_dsm_family) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;
Quoted 11 lines
Quoted 3 lines
...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
It's more complicated when you have a mixture of NVDIMM types, where not all NVDIMMs support the same families. For example, you could have an Intel NVDIMM in the same system as an HPE NVDIMM-N. The reason I called it "default" is because it should be the default for all devices that support that family but I didn't want other types of NVDIMMs to end up with no family at all.
Ok, but I still think we can make the logic a bit simpler. I'm reacting to the new "if (i == -1)" branch and 2 locations where we call "acpi_check_dsm()". How about this to centralize everything?
@@ -1397,11 +1397,15 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
* tells us if any commands are reachable through this uuid. */
for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
- if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
- break;
+ if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
+ if (family < 0 || i == default_dsm_family) {
+ family = i;
+ break;
+ }
+ }
/* limit the supported commands to those that are publicly documented */
- nfit_mem->family = i;
+ nfit_mem->family = family;
if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
dsm_mask = 0x3fe;
if (disable_vendor_specific)_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
Quoted 121 lines · Dan Williams
Quoted 85 lines · Linda Knippers
Quoted 84 lines · Dan Williams
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }I think we can make this simpler and more deterministic with a "force" option? Something like:
drivers/acpi/nfit/core.c +6
-2@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (force_dsm_family <0 ) +break ; +else if (i == force_dsm_family) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;Quoted 11 lines
Quoted 3 lines
...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
It's more complicated when you have a mixture of NVDIMM types, where not all NVDIMMs support the same families. For example, you could have an Intel NVDIMM in the same system as an HPE NVDIMM-N. The reason I called it "default" is because it should be the default for all devices that support that family but I didn't want other types of NVDIMMs to end up with no family at all.
Ok, but I still think we can make the logic a bit simpler. I'm reacting to the new "if (i == -1)" branch and 2 locations where we call "
acpi_check_dsm()". How about this to centralize everything?(quoted diff) +7
-3@@ -1397,11 +1397,15 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++) - if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) - break; + if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) { + if (family < 0 || i == default_dsm_family) { + family = i; + break; + } + } /* limit the supported commands to those that are publicly documented */ - nfit_mem->family = i; + nfit_mem->family = family; if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { dsm_mask = 0x3fe;
Quoted 2 lines
if (
disable_vendor_specific)
I think that will work. I'll test it in the morning.
Thanks for the feedback. I wasn't crazy about the code myself.
-- ljk
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
In reply to Dan WilliamsQuoted 103 lines · Dan Williams
Quoted 85 lines · Linda Knippers
Quoted 84 lines · Dan Williams
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }I think we can make this simpler and more deterministic with a "force" option? Something like:
drivers/acpi/nfit/core.c +6
-2@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (force_dsm_family <0 ) +break ; +else if (i == force_dsm_family) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;Quoted 11 lines
Quoted 3 lines
...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
It's more complicated when you have a mixture of NVDIMM types, where not all NVDIMMs support the same families. For example, you could have an Intel NVDIMM in the same system as an HPE NVDIMM-N. The reason I called it "default" is because it should be the default for all devices that support that family but I didn't want other types of NVDIMMs to end up with no family at all.
Ok, but I still think we can make the logic a bit simpler. I'm reacting to the new "if (i == -1)" branch and 2 locations where we call "
acpi_check_dsm()". How about this to centralize everything?
drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct · line 1404
Quoted 10 lines
drivers/acpi/nfit/core.c +4
-2@@ -1397,11 +1397,15 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (family <0 || i == default_dsm_family) { + family = i; +break ;
This actually doesn't work with the break since the for() will break on the first match, just like the current code. It works if I remove the break, but then we cycle through all the families even if the default_dsm_family parameter isn't used. Do you care about that?
I think it can be simple or efficient but not both.
-- ljk
Quoted 10 lines
drivers/acpi/nfit/core.c +3
-1+ } + }/* limit the supported commands to those that are publicly documented */ - nfit_mem->family = i; + nfit_mem->family = family;if (nfit_mem->family ==NVDIMM_FAMILY_INTEL ) { dsm_mask =0x3fe ;if (
disable_vendor_specific)
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
Quoted 120 lines · Linda Knippers
Quoted 103 lines · Dan Williams
Quoted 85 lines · Linda Knippers
Quoted 84 lines · Dan Williams
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }I think we can make this simpler and more deterministic with a "force" option? Something like:
drivers/acpi/nfit/core.c +6
-2@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (force_dsm_family <0 ) +break ; +else if (i == force_dsm_family) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;Quoted 11 lines
Quoted 3 lines
...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
It's more complicated when you have a mixture of NVDIMM types, where not all NVDIMMs support the same families. For example, you could have an Intel NVDIMM in the same system as an HPE NVDIMM-N. The reason I called it "default" is because it should be the default for all devices that support that family but I didn't want other types of NVDIMMs to end up with no family at all.
Ok, but I still think we can make the logic a bit simpler. I'm reacting to the new "if (i == -1)" branch and 2 locations where we call "
acpi_check_dsm()". How about this to centralize everything?drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct · line 1404
Quoted 10 lines
drivers/acpi/nfit/core.c +4
-2@@ -1397,11 +1397,15 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (family <0 || i == default_dsm_family) { + family = i; +break ;This actually doesn't work with the break since the
for()will break on the first match, just like the current code. It works if I remove the break, but then we cycle through all the families even if thedefault_dsm_familyparameter isn't used. Do you care about that?
Quoted 1 line
I think it can be simple or efficient but not both.
Something like this does work.
for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
if (default_dsm_family != -1) {
if (i == default_dsm_family) {
family = i;
break;
}
if (family < 0)
family = i;
}
else {
family = i;
break;
}
}Quoted 13 lines
-- ljk
Quoted 10 lines
drivers/acpi/nfit/core.c +3
-1+ } + }/* limit the supported commands to those that are publicly documented */ - nfit_mem->family = i; + nfit_mem->family = family;if (nfit_mem->family ==NVDIMM_FAMILY_INTEL ) { dsm_mask =0x3fe ;if (
disable_vendor_specific)
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 2/3] Allow specifying a default DSM family
In reply to Linda KnippersQuoted 120 lines · Linda Knippers
Quoted 103 lines · Dan Williams
Quoted 85 lines · Linda Knippers
Quoted 84 lines · Dan Williams
Quoted 52 lines · Linda Knippers
Provide the ability to request a default DSM family. If it is not supported, then fall back to the normal discovery order.
This is helpful for testing platforms that support multiple DSM families. It will also allow administrators to request the DSM family that their management tools support, which may not be the first one found using the current discovery order.
Signed-off-by Linda Knippers <[email protected]>--- drivers/acpi/nfit/core.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-)drivers/acpi/nfit/core.c +12
-4@@ -55,6 +55,11 @@module_param (override_dsm_mask, ulong,S_IRUGO );MODULE_PARM_DESC (override_dsm_mask,"Bitmask of allowed NVDIMM DSM functions" ); +static int default_dsm_family =-1 ; +module_param (default_dsm_family, int,S_IRUGO ); +MODULE_PARM_DESC (default_dsm_family, +"Try this DSM type first when identifying NVDIMM family" ); +LIST_HEAD (acpi_descs);DEFINE_MUTEX (acpi_desc_lock);@@ -1371,7 +1376,7 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,struct device *dev = acpi_desc->dev;unsigned long dsm_mask;const u8 *uuid; -int i; +int i =-1 ;/* nfit test assumes 1:1 relationship between commands and dsms */ nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;@@ -1399,10 +1404,23 @@ static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,* Until standardization materializes we need to consider 4 * different command sets. Note, that checking for function0 (bit0) * tells us if any commands are reachable through this uuid. +* First check for a requested default. */ -for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (default_dsm_family >=NVDIMM_FAMILY_INTEL && + default_dsm_family <=NVDIMM_FAMILY_MSFT ) { +if (acpi_check_dsm (adev_dimm->handle, +to_nfit_uuid (default_dsm_family),1 ,1 )) + i = default_dsm_family;drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, · line 1423
Quoted 10 lines
drivers/acpi/nfit/core.c +10
-0+else +dev_dbg (dev,"default_dsm_family %d not supported\n" , + default_dsm_family); + } +if (i ==-1 ) { +for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i), +1 ,1 )) +break ; + }I think we can make this simpler and more deterministic with a "force" option? Something like:
drivers/acpi/nfit/core.c +6
-2@@ -1397,8 +1397,12 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (force_dsm_family <0 ) +break ; +else if (i == force_dsm_family) +break ; + }/* limit the supported commands to those that are publicly documented */ nfit_mem->family = i;Quoted 11 lines
Quoted 3 lines
...because the user specifying the override should know ahead of time if that family is available, and we should fail the detection otherwise.
It's more complicated when you have a mixture of NVDIMM types, where not all NVDIMMs support the same families. For example, you could have an Intel NVDIMM in the same system as an HPE NVDIMM-N. The reason I called it "default" is because it should be the default for all devices that support that family but I didn't want other types of NVDIMMs to end up with no family at all.
Ok, but I still think we can make the logic a bit simpler. I'm reacting to the new "if (i == -1)" branch and 2 locations where we call "
acpi_check_dsm()". How about this to centralize everything?drivers/acpi/nfit/core.c static int acpi_nfit_add_dimm(struct · line 1404
Quoted 10 lines
drivers/acpi/nfit/core.c +4
-2@@ -1397,11 +1397,15 @@ static int acpi_nfit_add_dimm(structacpi_nfit_desc *acpi_desc, * tells us if any commands are reachable through this uuid. */ for (i =NVDIMM_FAMILY_INTEL ; i <=NVDIMM_FAMILY_MSFT ; i++) -if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) -break ; +if (acpi_check_dsm (adev_dimm->handle,to_nfit_uuid (i),1 ,1 )) { +if (family <0 || i == default_dsm_family) { + family = i; +break ;This actually doesn't work with the break since the
for()will break on the first match, just like the current code. It works if I remove the break, but then we cycle through all the families even if thedefault_dsm_familyparameter isn't used. Do you care about that?
Quoted 1 line
I think it can be simple or efficient but not both.
It doesn't need to be efficient, this is a slow path.
_______________________________________________
Linux-nvdimm mailing list [email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
PATCH v2 3/3 Remove unnecessary newline
In reply to Linda KnippersThe newline in MODULE_PARM_DESC causes modinfo to print the parameter data type on a separate line, which is different from all the other module parameters and could potentially cause a problem for someone parsing the output of modinfo.
--- drivers/acpi/nfit/core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/acpi/nfit/core.c +1 -1
@@ -49,7 +49,7 @@static bool disable_vendor_specific;module_param (disable_vendor_specific, bool,S_IRUGO );MODULE_PARM_DESC (disable_vendor_specific, -"Limit commands to the publicly specified set\n" ); +"Limit commands to the publicly specified set" );static unsigned long override_dsm_mask;module_param (override_dsm_mask, ulong,S_IRUGO );
Signature · 6 lines
1.8.3.1 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
Re: [PATCH 3/3] Remove unnecessary newline
Quoted 7 lines · Linda Knippers
The newline in
MODULE_PARM_DESCcauses modinfo to print the parameter data type on a separate line, which is different from all the other module parameters and could potentially cause a problem for someone parsing the output of modinfo.Signed-off-by Linda Knippers <[email protected]>---
Thanks,
Signature · 10 lines
Johannes Thumshirn Storage [email protected] +49 911 74053 689 SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg GF: Felix Imendörffer, Jane Smithard, Graham Norton HRB 21284 (AG Nürnberg) Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm