Skip to content

Commit cfe30b8

Browse files
djbwdjbw
committed
libnvdimm, pmem: adjust for section collisions with 'System RAM'
On a platform where 'Persistent Memory' and 'System RAM' are mixed within a given sparsemem section, trim the namespace and notify about the sub-optimal alignment. Cc: Toshi Kani <toshi.kani@hpe.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
1 parent d9cbe09 commit cfe30b8

File tree

4 files changed

+111
-36
lines changed

4 files changed

+111
-36
lines changed

‎drivers/nvdimm/namespace_devs.c

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -133,6 +133,7 @@ bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
133133
bool pmem_should_map_pages(struct device *dev)
134134
{
135135
struct nd_region *nd_region = to_nd_region(dev->parent);
136+
struct nd_namespace_io *nsio;
136137

137138
if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
138139
return false;
@@ -143,6 +144,12 @@ bool pmem_should_map_pages(struct device *dev)
143144
if (is_nd_pfn(dev) || is_nd_btt(dev))
144145
return false;
145146

147+
nsio = to_nd_namespace_io(dev);
148+
if (region_intersects(nsio->res.start, resource_size(&nsio->res),
149+
IORESOURCE_SYSTEM_RAM,
150+
IORES_DESC_NONE) == REGION_MIXED)
151+
return false;
152+
146153
#ifdef ARCH_MEMREMAP_PMEM
147154
return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
148155
#else

‎drivers/nvdimm/pfn.h

Lines changed: 8 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -27,10 +27,13 @@ struct nd_pfn_sb {
2727
__le32 flags;
2828
__le16 version_major;
2929
__le16 version_minor;
30-
__le64 dataoff;
30+
__le64 dataoff; /* relative to namespace_base + start_pad */
3131
__le64 npfns;
3232
__le32 mode;
33-
u8 padding[4012];
33+
/* minor-version-1 additions for section alignment */
34+
__le32 start_pad;
35+
__le32 end_trunc;
36+
u8 padding[4004];
3437
__le64 checksum;
3538
};
3639

@@ -45,4 +48,7 @@ struct nd_pfn_sb {
4548
#define PFN_SECTION_ALIGN_DOWN(x) (x)
4649
#define PFN_SECTION_ALIGN_UP(x) (x)
4750
#endif
51+
52+
#define PHYS_SECTION_ALIGN_DOWN(x) PFN_PHYS(PFN_SECTION_ALIGN_DOWN(PHYS_PFN(x)))
53+
#define PHYS_SECTION_ALIGN_UP(x) PFN_PHYS(PFN_SECTION_ALIGN_UP(PHYS_PFN(x)))
4854
#endif /* __NVDIMM_PFN_H */

‎drivers/nvdimm/pfn_devs.c

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -299,6 +299,11 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn)
299299
if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
300300
return -ENODEV;
301301

302+
if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
303+
pfn_sb->start_pad = 0;
304+
pfn_sb->end_trunc = 0;
305+
}
306+
302307
switch (le32_to_cpu(pfn_sb->mode)) {
303308
case PFN_MODE_RAM:
304309
case PFN_MODE_PMEM:

‎drivers/nvdimm/pmem.c

Lines changed: 91 additions & 34 deletions
Original file line numberDiff line numberDiff line change
@@ -43,7 +43,10 @@ struct pmem_device {
4343
phys_addr_t data_offset;
4444
u64 pfn_flags;
4545
void __pmem *virt_addr;
46+
/* immutable base size of the namespace */
4647
size_t size;
48+
/* trim size when namespace capacity has been section aligned */
49+
u32 pfn_pad;
4750
struct badblocks bb;
4851
};
4952

@@ -145,7 +148,7 @@ static long pmem_direct_access(struct block_device *bdev, sector_t sector,
145148
*kaddr = pmem->virt_addr + offset;
146149
*pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
147150

148-
return pmem->size - offset;
151+
return pmem->size - pmem->pfn_pad - offset;
149152
}
150153

151154
static const struct block_device_operations pmem_fops = {
@@ -236,7 +239,8 @@ static int pmem_attach_disk(struct device *dev,
236239
disk->flags = GENHD_FL_EXT_DEVT;
237240
nvdimm_namespace_disk_name(ndns, disk->disk_name);
238241
disk->driverfs_dev = dev;
239-
set_capacity(disk, (pmem->size - pmem->data_offset) / 512);
242+
set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
243+
/ 512);
240244
pmem->pmem_disk = disk;
241245
devm_exit_badblocks(dev, &pmem->bb);
242246
if (devm_init_badblocks(dev, &pmem->bb))
@@ -279,6 +283,9 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
279283
struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL);
280284
struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev);
281285
struct nd_namespace_common *ndns = nd_pfn->ndns;
286+
u32 start_pad = 0, end_trunc = 0;
287+
resource_size_t start, size;
288+
struct nd_namespace_io *nsio;
282289
struct nd_region *nd_region;
283290
unsigned long npfns;
284291
phys_addr_t offset;
@@ -304,28 +311,66 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
304311
}
305312

306313
memset(pfn_sb, 0, sizeof(*pfn_sb));
307-
npfns = (pmem->size - SZ_8K) / SZ_4K;
314+
315+
/*
316+
* Check if pmem collides with 'System RAM' when section aligned and
317+
* trim it accordingly
318+
*/
319+
nsio = to_nd_namespace_io(&ndns->dev);
320+
start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
321+
size = resource_size(&nsio->res);
322+
if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
323+
IORES_DESC_NONE) == REGION_MIXED) {
324+
325+
start = nsio->res.start;
326+
start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
327+
}
328+
329+
start = nsio->res.start;
330+
size = PHYS_SECTION_ALIGN_UP(start + size) - start;
331+
if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
332+
IORES_DESC_NONE) == REGION_MIXED) {
333+
size = resource_size(&nsio->res);
334+
end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size);
335+
}
336+
337+
if (start_pad + end_trunc)
338+
dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n",
339+
dev_name(&ndns->dev), start_pad + end_trunc);
340+
308341
/*
309342
* Note, we use 64 here for the standard size of struct page,
310343
* debugging options may cause it to be larger in which case the
311344
* implementation will limit the pfns advertised through
312345
* ->direct_access() to those that are included in the memmap.
313346
*/
347+
start += start_pad;
348+
npfns = (pmem->size - start_pad - end_trunc - SZ_8K) / SZ_4K;
314349
if (nd_pfn->mode == PFN_MODE_PMEM)
315-
offset = ALIGN(SZ_8K + 64 * npfns, nd_pfn->align);
350+
offset = ALIGN(start + SZ_8K + 64 * npfns, nd_pfn->align)
351+
- start;
316352
else if (nd_pfn->mode == PFN_MODE_RAM)
317-
offset = ALIGN(SZ_8K, nd_pfn->align);
353+
offset = ALIGN(start + SZ_8K, nd_pfn->align) - start;
318354
else
319355
goto err;
320356

321-
npfns = (pmem->size - offset) / SZ_4K;
357+
if (offset + start_pad + end_trunc >= pmem->size) {
358+
dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
359+
dev_name(&ndns->dev));
360+
goto err;
361+
}
362+
363+
npfns = (pmem->size - offset - start_pad - end_trunc) / SZ_4K;
322364
pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
323365
pfn_sb->dataoff = cpu_to_le64(offset);
324366
pfn_sb->npfns = cpu_to_le64(npfns);
325367
memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN);
326368
memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
327369
memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
328370
pfn_sb->version_major = cpu_to_le16(1);
371+
pfn_sb->version_minor = cpu_to_le16(1);
372+
pfn_sb->start_pad = cpu_to_le32(start_pad);
373+
pfn_sb->end_trunc = cpu_to_le32(end_trunc);
329374
checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
330375
pfn_sb->checksum = cpu_to_le64(checksum);
331376

@@ -376,60 +421,57 @@ static unsigned long init_altmap_reserve(resource_size_t base)
376421
return reserve;
377422
}
378423

379-
static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
424+
static int __nvdimm_namespace_attach_pfn(struct nd_pfn *nd_pfn)
380425
{
381-
struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
382-
struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
383-
struct device *dev = &nd_pfn->dev;
384-
struct nd_region *nd_region;
385-
struct vmem_altmap *altmap;
386-
struct nd_pfn_sb *pfn_sb;
387-
struct pmem_device *pmem;
388-
struct request_queue *q;
389-
phys_addr_t offset;
390426
int rc;
427+
struct resource res;
428+
struct request_queue *q;
429+
struct pmem_device *pmem;
430+
struct vmem_altmap *altmap;
431+
struct device *dev = &nd_pfn->dev;
432+
struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
433+
struct nd_namespace_common *ndns = nd_pfn->ndns;
434+
u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
435+
u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
436+
struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
437+
resource_size_t base = nsio->res.start + start_pad;
391438
struct vmem_altmap __altmap = {
392-
.base_pfn = init_altmap_base(nsio->res.start),
393-
.reserve = init_altmap_reserve(nsio->res.start),
439+
.base_pfn = init_altmap_base(base),
440+
.reserve = init_altmap_reserve(base),
394441
};
395442

396-
if (!nd_pfn->uuid || !nd_pfn->ndns)
397-
return -ENODEV;
398-
399-
nd_region = to_nd_region(dev->parent);
400-
rc = nd_pfn_init(nd_pfn);
401-
if (rc)
402-
return rc;
403-
404-
pfn_sb = nd_pfn->pfn_sb;
405-
offset = le64_to_cpu(pfn_sb->dataoff);
443+
pmem = dev_get_drvdata(dev);
444+
pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
445+
pmem->pfn_pad = start_pad + end_trunc;
406446
nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode);
407447
if (nd_pfn->mode == PFN_MODE_RAM) {
408-
if (offset < SZ_8K)
448+
if (pmem->data_offset < SZ_8K)
409449
return -EINVAL;
410450
nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
411451
altmap = NULL;
412452
} else if (nd_pfn->mode == PFN_MODE_PMEM) {
413-
nd_pfn->npfns = (resource_size(&nsio->res) - offset)
453+
nd_pfn->npfns = (pmem->size - pmem->pfn_pad - pmem->data_offset)
414454
/ PAGE_SIZE;
415455
if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
416456
dev_info(&nd_pfn->dev,
417457
"number of pfns truncated from %lld to %ld\n",
418458
le64_to_cpu(nd_pfn->pfn_sb->npfns),
419459
nd_pfn->npfns);
420460
altmap = & __altmap;
421-
altmap->free = __phys_to_pfn(offset - SZ_8K);
461+
altmap->free = __phys_to_pfn(pmem->data_offset - SZ_8K);
422462
altmap->alloc = 0;
423463
} else {
424464
rc = -ENXIO;
425465
goto err;
426466
}
427467

428468
/* establish pfn range for lookup, and switch to direct map */
429-
pmem = dev_get_drvdata(dev);
430469
q = pmem->pmem_queue;
470+
memcpy(&res, &nsio->res, sizeof(res));
471+
res.start += start_pad;
472+
res.end -= end_trunc;
431473
devm_memunmap(dev, (void __force *) pmem->virt_addr);
432-
pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &nsio->res,
474+
pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &res,
433475
&q->q_usage_counter, altmap);
434476
pmem->pfn_flags |= PFN_MAP;
435477
if (IS_ERR(pmem->virt_addr)) {
@@ -438,7 +480,6 @@ static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
438480
}
439481

440482
/* attach pmem disk in "pfn-mode" */
441-
pmem->data_offset = offset;
442483
rc = pmem_attach_disk(dev, ndns, pmem);
443484
if (rc)
444485
goto err;
@@ -447,6 +488,22 @@ static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
447488
err:
448489
nvdimm_namespace_detach_pfn(ndns);
449490
return rc;
491+
492+
}
493+
494+
static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
495+
{
496+
struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
497+
int rc;
498+
499+
if (!nd_pfn->uuid || !nd_pfn->ndns)
500+
return -ENODEV;
501+
502+
rc = nd_pfn_init(nd_pfn);
503+
if (rc)
504+
return rc;
505+
/* we need a valid pfn_sb before we can init a vmem_altmap */
506+
return __nvdimm_namespace_attach_pfn(nd_pfn);
450507
}
451508

452509
static int nd_pmem_probe(struct device *dev)

0 commit comments

Comments
 (0)