dax: enable dax fault handler to report VM_FAULT_HWPOISON
Subsystems (10)
v5 cover dax: enable dax fault handler to report VM_FAULT_HWPOISON
Change from v4:
Add comments describing when and why dax_mem2blk_err() is used. Suggested by Dan.
Change from v3:
Prevent leaking EHWPOISON to user level block IO calls such as zero_range_range, and truncate. Suggested by Dan.
Change from v2:
Convert EHWPOISON to EIO to prevent EHWPOISON errno from leaking out to block read(2). Suggested by Matthew.
Jane Chu (1):
dax: enable dax fault handler to report VM_FAULT_HWPOISON
drivers/dax/super.c | 5 ++++- drivers/nvdimm/pmem.c | 2 +- drivers/s390/block/dcssblk.c | 3 ++- fs/dax.c | 11 ++++++----- fs/fuse/virtio_fs.c | 3 ++- include/linux/dax.h | 13 +++++++++++++ include/linux/mm.h | 2 ++ 7 files changed, 30 insertions(+), 9 deletions(-)
Signature · 1 line
2.18.4
PATCH v5 1/1 dax: enable dax fault handler to report VM_FAULT_HWPOISON
When multiple processes mmap() a dax file, then at some point, a process issues a 'load' and consumes a hwpoison, the process receives a SIGBUS with si_code = BUS_MCEERR_AR and with si_lsb set for the poison scope. Soon after, any other process issues a 'load' to the poisoned page (that is unmapped from the kernel side by memory_failure), it receives a SIGBUS with si_code = BUS_ADRERR and without valid si_lsb.
This is confusing to user, and is different from page fault due to poison in RAM memory, also some helpful information is lost.
Channel dax backend driver's poison detection to the filesystem such that instead of reporting VM_FAULT_SIGBUS, it could report VM_FAULT_HWPOISON.
If user level block IO syscalls fail due to poison, the errno will be converted to EIO to maintain block API consistency.
--- drivers/dax/super.c | 5 ++++- drivers/nvdimm/pmem.c | 2 +- drivers/s390/block/dcssblk.c | 3 ++- fs/dax.c | 11 ++++++----- fs/fuse/virtio_fs.c | 3 ++- include/linux/dax.h | 13 +++++++++++++ include/linux/mm.h | 2 ++ 7 files changed, 30 insertions(+), 9 deletions(-)
drivers/dax/super.c +4 -1
@@ -203,6 +203,8 @@ size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,int dax_zero_page_range (struct dax_device *dax_dev,pgoff_t pgoff,size_t nr_pages) { +int ret; +if (!dax_alive (dax_dev))return -ENXIO ;/*
@@ -213,7 +215,8 @@ int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,if (nr_pages !=1 )return -EIO ; -return dax_dev->ops->zero_page_range (dax_dev, pgoff, nr_pages); + ret = dax_dev->ops->zero_page_range (dax_dev, pgoff, nr_pages); +return dax_mem2blk_err (ret); }EXPORT_SYMBOL_GPL (dax_zero_page_range);
drivers/nvdimm/pmem.c +1 -1
@@ -260,7 +260,7 @@ __weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,long actual_nr;if (mode !=DAX_RECOVERY_WRITE ) -return -EIO ; +return -EHWPOISON ;/* * Set the recovery stride is set to kernel page size because
drivers/s390/block/dcssblk.c +2 -1
@@ -54,7 +54,8 @@ static int dcssblk_dax_zero_page_range(struct dax_device *dax_dev, rc =dax_direct_access (dax_dev, pgoff, nr_pages,DAX_ACCESS , &kaddr,NULL );if (rc <0 ) -return rc; +return dax_mem2blk_err (rc); +memset (kaddr,0 , nr_pages <<PAGE_SHIFT );dax_flush (dax_dev, kaddr, nr_pages <<PAGE_SHIFT );return 0 ;
fs/dax.c +6 -5
@@ -1148,7 +1148,7 @@ static int dax_iomap_copy_around(loff_t pos, uint64_t length, size_t align_size,if (!zero_edge) { ret =dax_iomap_direct_access (srcmap, pos, size, &saddr,NULL );if (ret) -return ret; +return dax_mem2blk_err (ret); }if (copy_all) {
@@ -1310,7 +1310,7 @@ static s64 dax_unshare_iter(struct iomap_iter *iter) out_unlock:dax_read_unlock (id); -return ret; +return dax_mem2blk_err (ret); } int dax_file_unshare(struct inode *inode, loff_t pos, loff_t len,
@@ -1342,7 +1342,8 @@ static int dax_memzero(struct iomap_iter *iter, loff_t pos, size_t size) ret =dax_direct_access (iomap->dax_dev, pgoff,1 ,DAX_ACCESS , &kaddr,NULL );if (ret <0 ) -return ret; +return dax_mem2blk_err (ret); +memset (kaddr + offset,0 , size);if (iomap->flags &IOMAP_F_SHARED ) ret = dax_iomap_copy_around(pos, size, PAGE_SIZE, srcmap,
@@ -1498,7 +1499,7 @@ static loff_t dax_iomap_iter(const struct iomap_iter *iomi, map_len =dax_direct_access (dax_dev, pgoff,PHYS_PFN (size),DAX_ACCESS , &kaddr,NULL ); -if (map_len == -EIO &&iov_iter_rw (iter) ==WRITE ) { +if (map_len == -EHWPOISON &&iov_iter_rw (iter) ==WRITE ) { map_len =dax_direct_access (dax_dev, pgoff,PHYS_PFN (size),DAX_RECOVERY_WRITE , &kaddr,NULL );
@@ -1506,7 +1507,7 @@ static loff_t dax_iomap_iter(const struct iomap_iter *iomi, recovery = true; }if (map_len <0 ) { - ret = map_len; + ret =dax_mem2blk_err (map_len);break ; }
fs/fuse/virtio_fs.c +2 -1
@@ -775,7 +775,8 @@ static int virtio_fs_zero_page_range(struct dax_device *dax_dev, rc =dax_direct_access (dax_dev, pgoff, nr_pages,DAX_ACCESS , &kaddr,NULL );if (rc <0 ) -return rc; +return dax_mem2blk_err (rc); +memset (kaddr,0 , nr_pages <<PAGE_SHIFT );dax_flush (dax_dev, kaddr, nr_pages <<PAGE_SHIFT );return 0 ;
include/linux/dax.h +13 -0
@@ -261,6 +261,19 @@ static inline bool dax_mapping(struct address_space *mapping)return mapping->host &&IS_DAX (mapping->host); } +/* +* Due to dax's memory and block duo personalities, hwpoison reporting +* takes into consideration which personality is presently visible. +* When dax acts like a block device, such as in block IO, an encounter of +* dax hwpoison is reported as -EIO. +* When dax acts like memory, such as in page fault, a detection of hwpoison +* is reported as -EHWPOISON which leads to VM_FAULT_HWPOISON. +*/ +static inline int dax_mem2blk_err (int err) +{ +return (err == -EHWPOISON ) ? -EIO : err; +} +#ifdef CONFIG_DEV_DAX_HMEM_DEVICES void hmem_register_resource (int target_nid,struct resource *r);#else
include/linux/mm.h +2 -0
@@ -3342,6 +3342,8 @@ static inline vm_fault_t vmf_error(int err) {if (err == -ENOMEM )return VM_FAULT_OOM ; +else if (err == -EHWPOISON ) +return VM_FAULT_HWPOISON ;return VM_FAULT_SIGBUS ; }
Signature · 1 line
2.18.4
Quoted 11 lines · Jane Chu
When multiple processes
mmap()a dax file, then at some point, a process issues a 'load' and consumes a hwpoison, the process receives a SIGBUS withsi_code=BUS_MCEERR_ARand withsi_lsbset for the poison scope. Soon after, any other process issues a 'load' to the poisoned page (that is unmapped from the kernel side bymemory_failure), it receives a SIGBUS withsi_code=BUS_ADRERRand without validsi_lsb.This is confusing to user, and is different from page fault due to poison in RAM memory, also some helpful information is lost.
Quoted 8 lines
Channel dax backend driver's poison detection to the filesystem such that instead of reporting
VM_FAULT_SIGBUS, it could reportVM_FAULT_HWPOISON.If user level block IO syscalls fail due to poison, the errno will be converted to EIO to maintain block API consistency.
Signed-off-by Jane Chu <[email protected]>
Quoted 1 line
Change from v4:
…
I suggest to omit the cover letter for a single patch.
Will any patch series evolve for your proposed changes?
Regards,
Markus
Quoted 7 lines · Markus Elfring
Quoted 1 line
Change from v4:
…
I suggest to omit the cover letter for a single patch.
Will any patch series evolve for your proposed changes?
No. The thought was to put descriptions unsuitable for commit header in the cover letter.
thanks,
jane
Quoted 2 lines
Regards,
Markus
Quoted 1 line
The thought was to put descriptions unsuitable for commit header in the cover letter.
How do you think about to put additional information below triple dashes (or even into improved change descriptions)?
Regards,
Markus
Matthew Wilcox
reviewerQuoted 7 lines · Markus Elfring
Quoted 1 line
The thought was to put descriptions unsuitable for commit header in the cover letter.
How do you think about to put additional information below triple dashes (or even into improved change descriptions)?
Markus,
Please go away. Your feedback is not helpful. Thank you.
Quoted 9 lines
Quoted 5 lines
How do you think about to put additional information below triple dashes (or even into improved change descriptions)?
Markus,
Please go away. Your feedback is not helpful.
Would you insist on the usage of cover letters also for single patches?
Regards,
Markus
Matthew Wilcox
reviewerQuoted 11 lines · Markus Elfring
Quoted 9 lines
Quoted 5 lines
How do you think about to put additional information below triple dashes (or even into improved change descriptions)?
Markus,
Please go away. Your feedback is not helpful.
Would you insist on the usage of cover letters also for single patches?
I would neither insist on it, nor prohibit it. It simply does not make enough difference.
Quoted 3 lines
Quoted 1 line
Would you insist on the usage of cover letters also for single patches?
I would neither insist on it, nor prohibit it.
It seems that you can tolerate an extra message here.
Quoted 1 line
It simply does not make enough difference.
Can it occasionally be a bit nicer to receive all relevant information within a single patch instead of a combination of two messages?
Regards,
Markus
Quoted 11 lines · Markus Elfring
Quoted 3 lines
Quoted 1 line
Would you insist on the usage of cover letters also for single patches?
I would neither insist on it, nor prohibit it.
It seems that you can tolerate an extra message here.
Quoted 1 line
It simply does not make enough difference.
Can it occasionally be a bit nicer to receive all relevant information within a single patch instead of a combination of two messages?
No, I am the maintainer of this code and everything I needed to judge this patch was provided. This cover letter only included inter-version details and anything relevant for the kernel history is included in the patch itself.
For any code I maintain inter-version details below the --- line or in a 0/1 cover letter are perfectly acceptable.
Please, if the review feedback is arbitrary, as it is in this case, strongly consider not offering it.