libnvdimm/pfn_dev: Don't clear device memmap area during generic namespace probe
Subsystems (4)
PATCH v4 1/2 libnvdimm/pfn_dev: Don't clear device memmap area during generic namespace probe
nvdimm core use nd_pfn_validate when looking for devdax or fsdax namespace. In this case device resources are allocated against nd_namespace_io dev. In-order to allow remap of range in nd_pfn_clear_memmap_error(), move the device memmap area clearing while initializing pfn namespace. With this device resource are allocated against nd_pfn and we can use nd_pfn->dev for remapping.
This also avoids calling nd_pfn_clear_mmap_errors twice. Once while probing the namespace and second while initialzing a pfn namespace.
--- drivers/nvdimm/pfn_devs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/nvdimm/pfn_devs.c +3 -2
@@ -591,7 +591,7 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)return -ENXIO ; } -return nd_pfn_clear_memmap_errors (nd_pfn); +return 0 ; }EXPORT_SYMBOL (nd_pfn_validate);
@@ -730,7 +730,8 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn) rc =nd_pfn_validate (nd_pfn, sig);if (rc != -ENODEV ) -return rc; +/* Clear the memap range of errors */ +return nd_pfn_clear_memmap_errors (nd_pfn);/* no info block, do init */ ;memset (pfn_sb,0 ,sizeof (*pfn_sb));
Signature · 4 lines
2.23.0 _______________________________________________ Linux-nvdimm mailing list -- [email protected] To unsubscribe send an email to [email protected]
PATCH v4 2/2 libnvdimm/namespace: Differentiate between probe mapping and runtime mapping
The nvdimm core currently maps the full namespace to an ioremap range while probing the namespace mode. This can result in probe failures on architectures that have limited ioremap space.
For example, with a large btt namespace that consumes most of I/O remap range, depending on the sequence of namespace initialization, the user can find a pfn namespace initialization failure due to unavailable I/O remap space which nvdimm core uses for temporary mapping.
nvdimm core can avoid this failure by only mapping the reserved info block area to check for pfn superblock type and map the full namespace resource only before using the namespace.
Given that personalities like BTT can be layered on top of any namespace type create a generic form of devm_nsio_enable (devm_namespace_enable) and use it inside the per-personality attach routines. Now devm_namespace_enable() is always paired with disable unless the mapping is going to be used for long term runtime access.
--- drivers/dax/pmem/core.c | 6 +++--- drivers/nvdimm/btt.c | 10 ++++++++-- drivers/nvdimm/claim.c | 14 ++++++-------- drivers/nvdimm/namespace_devs.c | 17 +++++++++++++++++ drivers/nvdimm/nd-core.h | 16 ++++++++++++++++ drivers/nvdimm/nd.h | 22 +++++++++------------- drivers/nvdimm/pfn_devs.c | 18 +++++++++++------- drivers/nvdimm/pmem.c | 17 +++++++++++++---- 8 files changed, 83 insertions(+), 37 deletions(-)
drivers/dax/pmem/core.c +3 -3
@@ -25,20 +25,20 @@ struct dev_dax *__dax_pmem_probe(struct device *dev, enum dev_dax_subsys subsys) ndns =nvdimm_namespace_common_probe (dev);if (IS_ERR (ndns))return ERR_CAST (ndns); - nsio =to_nd_namespace_io (&ndns->dev);/* parse the 'pfn' info block via ->rw_bytes */ - rc =devm_nsio_enable (dev, nsio); + rc =devm_namespace_enable (dev, ndns,nd_info_block_reserve ());if (rc)return ERR_PTR (rc); rc =nvdimm_setup_pfn (nd_pfn, &pgmap);if (rc)return ERR_PTR (rc); -devm_nsio_disable (dev, nsio); +devm_namespace_disable (dev, ndns);/* reserve the metadata area, device-dax will reserve the data */ pfn_sb = nd_pfn->pfn_sb; offset =le64_to_cpu (pfn_sb->dataoff); + nsio =to_nd_namespace_io (&ndns->dev);if (!devm_request_mem_region (dev, nsio->res.start, offset,dev_name (&ndns->dev))) {dev_warn (dev,"could not reserve metadata\n" );
drivers/nvdimm/btt.c +8 -2
@@ -1674,7 +1674,8 @@ int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)struct nd_region *nd_region;struct btt_sb *btt_sb;struct btt *btt; -size_t rawsize; +size_t size, rawsize; +int rc;if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {dev_dbg (&nd_btt->dev,"incomplete btt configuration\n" );
@@ -1685,6 +1686,11 @@ int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)if (!btt_sb)return -ENOMEM ; + size =nvdimm_namespace_capacity (ndns); + rc =devm_namespace_enable (&nd_btt->dev, ndns, size); +if (rc) +return rc; +/* * If this returns < 0, that is ok as it just means there wasn't * an existing BTT, and we're creating a new one. We still need to
@@ -1693,7 +1699,7 @@ int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)*/ nd_btt_version (nd_btt, ndns, btt_sb); - rawsize =nvdimm_namespace_capacity (ndns) - nd_btt->initial_offset; + rawsize = size - nd_btt->initial_offset;if (rawsize <ARENA_MIN_SIZE ) { dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n", dev_name(&ndns->dev),
drivers/nvdimm/claim.c +6 -8
@@ -300,13 +300,14 @@ static int nsio_rw_bytes(struct nd_namespace_common *ndns,return rc; } -int devm_nsio_enable (struct device *dev,struct nd_namespace_io *nsio) +int devm_nsio_enable (struct device *dev,struct nd_namespace_io *nsio, +resource_size_t size) {struct resource *res = &nsio->res;struct nd_namespace_common *ndns = &nsio->common; - nsio->size =resource_size (res); -if (!devm_request_mem_region (dev, res->start,resource_size (res), + nsio->size = size; +if (!devm_request_mem_region (dev, res->start, size,dev_name (&ndns->dev))) {dev_warn (dev,"could not reserve region %pR\n" , res);return -EBUSY ;
@@ -318,12 +319,10 @@ int devm_nsio_enable(struct device *dev, struct nd_namespace_io *nsio)nvdimm_badblocks_populate (to_nd_region (ndns->dev.parent), &nsio->bb, &nsio->res); - nsio->addr =devm_memremap (dev, res->start,resource_size (res), -ARCH_MEMREMAP_PMEM ); + nsio->addr =devm_memremap (dev, res->start, size,ARCH_MEMREMAP_PMEM );return PTR_ERR_OR_ZERO (nsio->addr); } -EXPORT_SYMBOL_GPL (devm_nsio_enable);void devm_nsio_disable (struct device *dev,struct nd_namespace_io *nsio) {
@@ -331,6 +330,5 @@ void devm_nsio_disable(struct device *dev, struct nd_namespace_io *nsio)devm_memunmap (dev, nsio->addr);devm_exit_badblocks (dev, &nsio->bb); -devm_release_mem_region (dev, res->start,resource_size (res)); +devm_release_mem_region (dev, res->start, nsio->size); } -EXPORT_SYMBOL_GPL (devm_nsio_disable);
drivers/nvdimm/namespace_devs.c +17 -0
@@ -1759,6 +1759,23 @@ struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev) }EXPORT_SYMBOL (nvdimm_namespace_common_probe); +int devm_namespace_enable (struct device *dev,struct nd_namespace_common *ndns, +resource_size_t size) +{ +if (is_namespace_blk (&ndns->dev)) +return 0 ; +return devm_nsio_enable (dev,to_nd_namespace_io (&ndns->dev), size); +} +EXPORT_SYMBOL_GPL (devm_namespace_enable); + +void devm_namespace_disable (struct device *dev,struct nd_namespace_common *ndns) +{ +if (is_namespace_blk (&ndns->dev)) +return ; +devm_nsio_disable (dev,to_nd_namespace_io (&ndns->dev)); +} +EXPORT_SYMBOL_GPL (devm_namespace_disable); +static struct device **create_namespace_io (struct nd_region *nd_region) {struct nd_namespace_io *nsio;
drivers/nvdimm/nd-core.h +16 -0
@@ -171,6 +171,22 @@ ssize_t nd_namespace_store(struct device *dev,struct nd_pfn *to_nd_pfn_safe (struct device *dev);bool is_nvdimm_bus (struct device *dev); +#if IS_ENABLED (CONFIG_ND_CLAIM ) +int devm_nsio_enable (struct device *dev,struct nd_namespace_io *nsio, +resource_size_t size); +void devm_nsio_disable (struct device *dev,struct nd_namespace_io *nsio); +#else +static inline int devm_nsio_enable (struct device *dev, +struct nd_namespace_io *nsio,resource_size_t size) +{ +return -ENXIO ; +} + +void devm_nsio_disable (struct device *dev,struct nd_namespace_io *nsio) +{ +} +#endif +#ifdef CONFIG_PROVE_LOCKING extern struct class *nd_class;
drivers/nvdimm/nd.h +9 -13
@@ -212,6 +212,11 @@ struct nd_dax {struct nd_pfn nd_pfn; }; +static inline u32 nd_info_block_reserve (void ) +{ +return ALIGN (SZ_8K ,PAGE_SIZE ); +} +enum nd_async_mode {ND_SYNC ,ND_ASYNC ,
@@ -370,29 +375,20 @@ const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,unsigned int pmem_sector_size (struct nd_namespace_common *ndns);void nvdimm_badblocks_populate (struct nd_region *nd_region,struct badblocks *bb,const struct resource *res); +int devm_namespace_enable (struct device *dev,struct nd_namespace_common *ndns, +resource_size_t size); +void devm_namespace_disable (struct device *dev, +struct nd_namespace_common *ndns);#if IS_ENABLED (CONFIG_ND_CLAIM ) -/* max struct page size independent of kernel config */ #define MAX_STRUCT_PAGE_SIZE 64 -int nvdimm_setup_pfn (struct nd_pfn *nd_pfn,struct dev_pagemap *pgmap); -int devm_nsio_enable (struct device *dev,struct nd_namespace_io *nsio); -void devm_nsio_disable (struct device *dev,struct nd_namespace_io *nsio);#else static inline int nvdimm_setup_pfn (struct nd_pfn *nd_pfn,struct dev_pagemap *pgmap) {return -ENXIO ; } -static inline int devm_nsio_enable (struct device *dev, -struct nd_namespace_io *nsio) -{ -return -ENXIO ; -} -static inline void devm_nsio_disable (struct device *dev, -struct nd_namespace_io *nsio) -{ -}#endif int nd_blk_region_init (struct nd_region *nd_region);int nd_region_activate (struct nd_region *nd_region);
drivers/nvdimm/pfn_devs.c +11 -7
@@ -382,6 +382,15 @@ static int nd_pfn_clear_memmap_errors(struct nd_pfn *nd_pfn) meta_start = (SZ_4K +sizeof (*pfn_sb)) >>9 ; meta_num = (le64_to_cpu (pfn_sb->dataoff) >>9 ) - meta_start; +/* +* re-enable the namespace with correct size so that we can access +* the device memmap area. +*/ +devm_namespace_disable (&nd_pfn->dev, ndns); + rc =devm_namespace_enable (&nd_pfn->dev, ndns,le64_to_cpu (pfn_sb->dataoff)); +if (rc) +return rc; + do { unsigned long zero_len; u64 nsoff;
@@ -635,11 +644,6 @@ int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns) }EXPORT_SYMBOL (nd_pfn_probe); -static u32 info_block_reserve (void ) -{ -return ALIGN (SZ_8K ,PAGE_SIZE ); -} -/* * We hotplug memory at sub-section granularity, pad the reserved area * from the previous section base to the namespace base address.
@@ -653,7 +657,7 @@ static unsigned long init_altmap_base(resource_size_t base)static unsigned long init_altmap_reserve (resource_size_t base) { -unsigned long reserve =info_block_reserve () >>PAGE_SHIFT ; +unsigned long reserve =nd_info_block_reserve () >>PAGE_SHIFT ;unsigned long base_pfn =PHYS_PFN (base); reserve += base_pfn -SUBSECTION_ALIGN_DOWN (base_pfn);
@@ -668,7 +672,7 @@ static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)u64 offset =le64_to_cpu (pfn_sb->dataoff);u32 start_pad =__le32_to_cpu (pfn_sb->start_pad);u32 end_trunc =__le32_to_cpu (pfn_sb->end_trunc); -u32 reserve =info_block_reserve (); +u32 reserve =nd_info_block_reserve ();struct nd_namespace_common *ndns = nd_pfn->ndns;struct nd_namespace_io *nsio =to_nd_namespace_io (&ndns->dev);resource_size_t base = nsio->res.start + start_pad;
drivers/nvdimm/pmem.c +13 -4
@@ -372,6 +372,10 @@ static int pmem_attach_disk(struct device *dev,if (!pmem)return -ENOMEM ; + rc =devm_namespace_enable (dev, ndns,nd_info_block_reserve ()); +if (rc) +return rc; +/* while nsio_rw_bytes is active, parse a pfn info block if present */ if (is_nd_pfn (dev)) { nd_pfn =to_nd_pfn (dev);
@@ -381,7 +385,7 @@ static int pmem_attach_disk(struct device *dev, }/* we're attaching a block device, disable raw namespace access */ -devm_nsio_disable (dev, nsio); +devm_namespace_disable (dev, ndns);dev_set_drvdata (dev, pmem); pmem->phys_addr = res->start;
@@ -497,15 +501,16 @@ static int nd_pmem_probe(struct device *dev)if (IS_ERR (ndns))return PTR_ERR (ndns); -if (devm_nsio_enable (dev,to_nd_namespace_io (&ndns->dev))) -return -ENXIO ; -if (is_nd_btt (dev))return nvdimm_namespace_attach_btt (ndns);if (is_nd_pfn (dev))return pmem_attach_disk (dev, ndns); + ret =devm_namespace_enable (dev, ndns,nd_info_block_reserve ()); +if (ret) +return ret; + ret =nd_btt_probe (dev, ndns);if (ret ==0 )return -ENXIO ;
@@ -532,6 +537,10 @@ static int nd_pmem_probe(struct device *dev)return -ENXIO ; else if (ret == -EOPNOTSUPP) return ret; + +/* probe complete, attach handles namespace enabling */ +devm_namespace_disable (dev ,ndns ); +return pmem_attach_disk (dev, ndns); }
Signature · 4 lines
2.23.0 _______________________________________________ Linux-nvdimm mailing list -- [email protected] To unsubscribe send an email to [email protected]
Re: [PATCH v4 2/2] libnvdimm/namespace: Differentiate between probe mapping and runtime mapping
Quoted 14 lines · Aneesh Kumar K.V
The nvdimm core currently maps the full namespace to an ioremap range while probing the namespace mode. This can result in probe failures on architectures that have limited ioremap space.
For example, with a large btt namespace that consumes most of I/O remap range, depending on the sequence of namespace initialization, the user can find a pfn namespace initialization failure due to unavailable I/O remap space which nvdimm core uses for temporary mapping.
nvdimm core can avoid this failure by only mapping the reserved info block area to check for pfn superblock type and map the full namespace resource only before using the namespace.
Quoted 5 lines
Given that personalities like BTT can be layered on top of any namespace type create a generic form of
devm_nsio_enable(devm_namespace_enable) and use it inside the per-personality attach routines. Nowdevm_namespace_enable()is always paired with disable unless the mapping is going to be used for long term runtime access.
Looks good to me, applied.
_______________________________________________
Linux-nvdimm mailing list -- [email protected] To unsubscribe send an email to [email protected]
Quoted 9 lines · Aneesh Kumar K.V
nvdimm core use
nd_pfn_validatewhen looking for devdax or fsdax namespace. In this case device resources are allocated againstnd_namespace_iodev. In-order to allow remap of range innd_pfn_clear_memmap_error(), move the device memmap area clearing while initializing pfn namespace. With this device resource are allocated againstnd_pfnand we can usend_pfn->devfor remapping.This also avoids calling
nd_pfn_clear_mmap_errorstwice. Once while probing the namespace and second while initialzing a pfn namespace.
Nice find!
For the resend: s/initialzing/initializing/
Quoted 20 lines
Signed-off-by Aneesh Kumar K.V <[email protected]>--- drivers/nvdimm/pfn_devs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/nvdimm/pfn_devs.c +1
-1@@ -591,7 +591,7 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)return -ENXIO ; } -return nd_pfn_clear_memmap_errors (nd_pfn); +return 0 ; }EXPORT_SYMBOL (nd_pfn_validate);
drivers/nvdimm/pfn_devs.c static int nd_pfn_init(struct nd_pfn *nd_pfn) · line 733
Quoted 5 lines
drivers/nvdimm/pfn_devs.c +1
-1@@ -730,7 +730,8 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn) rc =nd_pfn_validate (nd_pfn, sig);if (rc != -ENODEV ) -return rc; +/* Clear the memap range of errors */
No need for this comment the following function name is descriptive enough.
drivers/nvdimm/pfn_devs.c static int nd_pfn_init(struct nd_pfn *nd_pfn) · line 734
Quoted 1 line
drivers/nvdimm/pfn_devs.c +1
-0+return nd_pfn_clear_memmap_errors (nd_pfn);
While this does allow for the errors to be cleared once on each activation, it blocks validation errors from being reported up the stack. It also does not address errors being cleared on create, but that was also a problem shared with the current implementation.
I think you want something like this? This passes my testing, but please review / test to make sure I'm not overlooking something, and update the changelog to reflect that errors are now also cleared on create.
drivers/nvdimm/pfn_devs.c +7 -3
@@ -725,9 +725,10 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn) sig =PFN_SIG ; rc =nd_pfn_validate (nd_pfn, sig); -if (rc != -ENODEV ) -/* Clear the memap range of errors */ +if (rc ==0 )return nd_pfn_clear_memmap_errors (nd_pfn); +if (rc != -ENODEV ) +return rc;/* no info block, do init */ ;memset (pfn_sb,0 ,sizeof (*pfn_sb));
@@ -793,7 +794,10 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn) checksum =nd_sb_checksum ((struct nd_gen_sb *) pfn_sb); pfn_sb->checksum =cpu_to_le64 (checksum); -return nvdimm_write_bytes (ndns,SZ_4K , pfn_sb,sizeof (*pfn_sb),0 ); + rc =nvdimm_write_bytes (ndns,SZ_4K , pfn_sb,sizeof (*pfn_sb),0 ); +if (rc) +return rc; +return nd_pfn_clear_memmap_errors (nd_pfn); }/*
_______________________________________________
Linux-nvdimm mailing list -- [email protected] To unsubscribe send an email to [email protected]