- 14 Oct, 2020 40 commits
-
-
Yu Zhao authored
To activate a page, mark_page_accessed() always holds a reference on it. It either gets a new reference when adding a page to lru_pvecs.activate_page or reuses an existing one it previously got when it added a page to lru_pvecs.lru_add. So it doesn't call SetPageActive() on a page that doesn't have any reference left. Therefore, the race is impossible these days (I didn't brother to dig into its history). For other paths, namely reclaim and migration, a reference count is always held while calling SetPageActive() on a page. SetPageSlabPfmemalloc() also uses SetPageActive(), but it's irrelevant to LRU pages. Signed-off-by:
Yu Zhao <yuzhao@google.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Reviewed-by:
Yang Shi <shy828301@gmail.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Qian Cai <cai@lca.pw> Link: http://lkml.kernel.org/r/20200818184704.3625199-2-yuzhao@google.comSigned-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Yu Zhao authored
We don't initially add anon pages to active lruvec after commit b518154e ("mm/vmscan: protect the workingset on anonymous LRU"). Remove activate_page() from unuse_pte(), which seems to be missed by the commit. And make the function static while we are at it. Before the commit, we called lru_cache_add_active_or_unevictable() to add new ksm pages to active lruvec. Therefore, activate_page() wasn't necessary for them in the first place. Signed-off-by:
-
Gao Xiang authored
SWP_FS is used to make swap_{read,write}page() go through the filesystem, and it's only used for swap files over NFS for now. Otherwise it will directly submit IO to blockdev according to swapfile extents reported by filesystems in advance. As Matthew pointed out [1], SWP_FS naming is somewhat confusing, so let's rename to SWP_FS_OPS. [1] https://lore.kernel.org/r/20200820113448.GM17456@casper.infradead.orgSuggested-by:Matthew Wilcox <willy@infradead.org> Signed-off-by:
Gao Xiang <hsiangkao@redhat.com> Signed-off-by:
-
John Hubbard authored
As suggested by Dan Carpenter, fortify unpin_user_pages() just a bit, against a typical caller mistake: check if the npages arg is really a -ERRNO value, which would blow up the unpinning loop: WARN and return. If this new WARN_ON() fires, then the system *might* be leaking pages (by leaving them pinned), but probably not. More likely, gup/pup returned a hard -ERRNO error to the caller, who erroneously passed it here. Signed-off-by:
John Hubbard <jhubbard@nvidia.com> Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
Barry Song authored
gup prohibits users from calling get_user_pages() with FOLL_PIN. But it allows users to call get_user_pages() with FOLL_LONGTERM only. It seems insensible. Since FOLL_LONGTERM is a stricter case of FOLL_PIN, we should prohibit users from calling get_user_pages() with FOLL_LONGTERM while not with FOLL_PIN. mm/gup_benchmark.c used to be the only user who did this improperly. But it has been fixed by moving to use pin_user_pages(). [akpm@linux-foundation.org: fix CONFIG_MMU=n build] Link: https://lkml.kernel.org/r/CA+G9fYuNS3k0DVT62twfV746pfNhCSrk5sVMcOcQ1PGGnEseyw@mail.gmail.comSigned-off-by:
Barry Song <song.bao.hua@hisilicon.com> Signed-off-by:
Ira Weiny <ira.weiny@intel.com> Cc: John Hubbard <jhubbard@nvidia.com> Cc: Jan Kara <jack@suse.cz> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Naresh Kamboju <naresh.kamboju@linaro.org> Link: http://lkml.kernel.org/r/20200819110100.23504-1-song.bao.hua@hisilicon.comSigned-off-by:
-
Barry Song authored
According to Documentation/core-api/pin_user_pages.rst, FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying that is, FOLL_LONGTERM is a specific case, more restrictive case of FOLL_PIN. Almost all kernel modules are using pin_user_pages() with FOLL_LONGTERM, mm/gup_benchmark.c seems to the only exception in which FOLL_PIN is not a prerequisite to FOLL_LONGTERM. Signed-off-by:
-
Barry Song authored
In the beginning, mm/gup_benchmark.c supported get_user_pages_fast() only, but right now, it supports the benchmarking of a couple of get_user_pages() related calls like: * get_user_pages_fast() * get_user_pages() * pin_user_pages_fast() * pin_user_pages() The documentation is confusing and needs update. Signed-off-by:
-
Yafang Shao authored
Our users reported that there're some random latency spikes when their RT process is running. Finally we found that latency spike is caused by FADV_DONTNEED. Which may call lru_add_drain_all() to drain LRU cache on remote CPUs, and then waits the per-cpu work to complete. The wait time is uncertain, which may be tens millisecond. That behavior is unreasonable, because this process is bound to a specific CPU and the file is only accessed by itself, IOW, there should be no pagecache pages on a per-cpu pagevec of a remote CPU. That unreasonable behavior is partially caused by the wrong comparation of the number of invalidated pages and the number of the target. For example, if (count < (end_index - start_index + 1)) The count above is how many pages were invalidated in the local CPU, and (end_index - start_index + 1) is how many pages should be invalidated. The usage of (end_index - start_index + 1) is incorrect, because they are virtual addresses, which may not mapped to pages. Besides that, there may be holes between start and end. So we'd better check whether there are still pages on per-cpu pagevec after drain the local cpu, and then decide whether or not to call lru_add_drain_all(). After I applied it with a hotfix to our production environment, most of the lru_add_drain_all() can be avoided. Suggested-by:Mel Gorman <mgorman@suse.de> Signed-off-by:
Yafang Shao <laoar.shao@gmail.com> Signed-off-by:
-
Matthew Wilcox (Oracle) authored
We dereference page->mapping and page->index directly after calling find_subpage() and these fields are not valid for tail pages. While commit 4101196b ("mm: page cache: store only head pages in i_pages") introduced the call to find_subpage(), the problem existed prior to this; I'm going to suggest all the way back to when THPs first existed. The user-visible effects of this are almost negligible. To hit it, you have to mmap a tmpfs file at an unaligned address and then it's only a disabled optimisation causing page faults to happen more frequently than they otherwise would. Fix this by keeping both head and page pointers and checking the appropriate one. We could use page_mapping() and page_to_index(), but that's higher overhead. Signed-off-by:
Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: William Kucharski <william.kucharski@oracle.com> Link: https://lkml.kernel.org/r/20200911012532.24761-1-willy@infradead.orgSigned-off-by:
-
Matthew Wilcox (Oracle) authored
Add a new FGP_HEAD flag which avoids calling find_subpage() and add a convenience wrapper for it. Signed-off-by:
-
Matthew Wilcox (Oracle) authored
Convert shmem_getpage_gfp() (the only remaining caller of find_lock_entry()) to cope with a head page being returned instead of the subpage for the index. [willy@infradead.org: fix BUG()s] Link https://lore.kernel.org/linux-mm/20200912032042.GA6583@casper.infradead.org/Signed-off-by:
-
Matthew Wilcox (Oracle) authored
There are only four callers remaining of find_get_entry(). get_shadow_from_swap_cache() only wants to see shadow entries and doesn't care about which page is returned. Push the find_subpage() call into find_lock_entry(), find_get_incore_page() and pagecache_get_page(). [willy@infradead.org: fix oops] Link: https://lkml.kernel.org/r/20200914112738.GM6583@casper.infradead.orgSigned-off-by:
-
Matthew Wilcox (Oracle) authored
i915 does not want to see value entries. Switch it to use find_lock_page() instead, and remove the export of find_lock_entry(). Move find_lock_entry() and find_get_entry() to mm/internal.h to discourage any future use. Signed-off-by:
Johannes Weiner <hannes@cmpxchg.org> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Matthew Auld <matthew.auld@intel.com> Cc: William Kucharski <william.kucharski@oracle.com> Link: https://lkml.kernel.org/r/20200910183318.20139-6-willy@infradead.orgSigned-off-by:
-
Matthew Wilcox (Oracle) authored
Avoid bumping the refcount on pages when we're only interested in the swap entries. Signed-off-by:
-
Matthew Wilcox (Oracle) authored
Instead of calling find_get_entry() for every page index, use an XArray iterator to skip over NULL entries, and avoid calling get_page(), because we only want the swap entries. [willy@infradead.org: fix LTP soft lockups] Link: https://lkml.kernel.org/r/20200914165032.GS6583@casper.infradead.orgSigned-off-by:
-
Matthew Wilcox (Oracle) authored
The current code does not protect against swapoff of the underlying swap device, so this is a bug fix as well as a worthwhile reduction in code complexity. Signed-off-by:
-
Matthew Wilcox (Oracle) authored
Patch series "Return head pages from find_*_entry", v2. This patch series started out as part of the THP patch set, but it has some nice effects along the way and it seems worth splitting it out and submitting separately. Currently find_get_entry() and find_lock_entry() return the page corresponding to the requested index, but the first thing most callers do is find the head page, which we just threw away. As part of auditing all the callers, I found some misuses of the APIs and some plain inefficiencies that I've fixed. The diffstat is unflattering, but I added more kernel-doc and a new wrapper. This patch (of 8); Provide this functionality from the swap cache. It's useful for more than just mincore(). Signed-off-by:
-
John Hubbard authored
Rename head_pincount() --> head_compound_pincount(). These names are more accurate (or less misleading) than the original ones. Signed-off-by:
-
Matthew Wilcox (Oracle) authored
__dump_page() checks i_dentry is fetchable and i_ino is earlier in the struct than i_ino, so it ought to work fine, but it's possible that struct randomisation has reordered i_ino after i_dentry and the pointer is just wild enough that i_dentry is fetchable and i_ino isn't. Also print the inode number if the dentry is invalid. Reported-by:
Vlastimil Babka <vbabka@suse.cz> Signed-off-by:
Mike Rapoport <rppt@linux.ibm.com> Link: https://lkml.kernel.org/r/20200819185710.28180-1-willy@infradead.orgSigned-off-by:
-
Joao Martins authored
Add a sysfs attribute which denotes a range from the dax region to be allocated. It's an write only @mapping sysfs attribute in the format of '<start>-<end>' to allocate a range. @start and @end use hexadecimal values and the @pgoff is implicitly ordered wrt to previous writes to @mapping sysfs e.g. a write of a range of length 1G the pgoff is 0..1G(-4K), a second write will use @pgoff for 1G+4K..<size>. This range mapping interface is useful for: 1) Application which want to implement its own allocation logic, and thus pick the desired ranges from dax_region. 2) For use cases like VMM fast restart[0] where after kexec we want to the same gpa<->phys mappings (as originally created before kexec). [0] https://static.sched.com/hosted_files/kvmforum2019/66/VMM-fast-restart_kvmforum2019.pdfSigned-off-by:Joao Martins <joao.m.martins@oracle.com> Signed-off-by:
Dan Williams <dan.j.williams@intel.com> Signed-off-by:
-
Joao Martins authored
Introduce a new module parameter for dax_hmem which initializes all region devices as free, rather than allocating a pagemap for the region by default. All hmem devices created with dax_hmem.region_idle=1 will have full available size for creating dynamic dax devices. Signed-off-by:
-
Dan Williams authored
Introduce a device align attribute. While doing so, rename the region align attribute to be more explicitly named as so, but keep it named as @align to retain the API for tools like daxctl. Changes on align may not always be valid, when say certain mappings were created with 2M and then we switch to 1G. So, we validate all ranges against the new value being attempted, post resizing. Signed-off-by:
-
Joao Martins authored
Introduce @align to struct dev_dax. When creating a new device, we still initialize to the default dax_region @align. Child devices belonging to a region may wish to keep a different alignment property instead of a global region-defined one. Signed-off-by:
-
Dan Williams authored
In support of interrogating the physical address layout of a device with dis-contiguous ranges, introduce a sysfs directory with 'start', 'end', and 'page_offset' attributes. The alternative is trying to parse /proc/iomem, and that file will not reflect the extent layout until the device is enabled. Signed-off-by:
-
Dan Williams authored
Break the requirement that device-dax instances are physically contiguous. With this constraint removed it allows fragmented available capacity to be fully allocated. This capability is useful to mitigate the "noisy neighbor" problem with memory-side-cache management for virtual machines, or any other scenario where a platform address boundary also designates a performance boundary. For example a direct mapped memory side cache might rotate cache colors at 1GB boundaries. With dis-contiguous allocations a device-dax instance could be configured to contain only 1 cache color. It also satisfies Joao's use case (see link) for partitioning memory for exclusive guest access. It allows for a future potential mode where the host kernel need not allocate 'struct page' capacity up-front. Reported-by:
-
Dan Williams authored
In support of device-dax growing the ability to front physically dis-contiguous ranges of memory, update devm_memremap_pages() to track multiple ranges with a single reference counter and devm instance. Convert all [devm_]memremap_pages() users to specify the number of ranges they are mapping in their 'struct dev_pagemap' instance. Signed-off-by:
-
Dan Williams authored
The 'struct resource' in 'struct dev_pagemap' is only used for holding resource span information. The other fields, 'name', 'flags', 'desc', 'parent', 'sibling', and 'child' are all unused wasted space. This is in preparation for introducing a multi-range extension of devm_memremap_pages(). The bulk of this change is unwinding all the places internal to libnvdimm that used 'struct resource' unnecessarily, and replacing instances of 'struct dev_pagemap'.res with 'struct dev_pagemap'.range. P2PDMA had a minor usage of the resource flags field, but only to report failures with "%pR". That is replaced with an open coded print of the range. [dan.carpenter@oracle.com: mm/hmm/test: use after free in dmirror_allocate_chunk()] Link: https://lkml.kernel.org/r/20200926121402.GA7467@kadamSigned-off-by:
-
Dan Williams authored
Make the device-dax 'size' attribute writable to allow capacity to be split between multiple instances in a region. The intended consumers of this capability are users that want to split a scarce memory resource between device-dax and System-RAM access, or users that want to have multiple security domains for a large region. By default the hmem instance provider allocates an entire region to the first instance. The process of creating a new instance (assuming a region-id of 0) is find the region and trigger the 'create' attribute which yields an empty instance to configure. For example: cd /sys/bus/dax/devices echo dax0.0 > dax0.0/driver/unbind echo $new_size > dax0.0/size echo 1 > $(readlink -f dax0.0)../dax_region/create seed=$(cat $(readlink -f dax0.0)../dax_region/seed) echo $new_size > $seed/size echo dax0.0 > ../drivers/{device_dax,kmem}/bind echo dax0.1 > ../drivers/{device_dax,kmem}/bind Instances can be destroyed by: echo $device > $(readlink -f $device)../dax_region/delete Signed-off-by: -
Dan Williams authored
Use sysfs_streq() in device_find_child_by_name() to allow it to use a sysfs input string that might contain a trailing newline. The other "device by name" interfaces, {bus,driver,class}_find_device_by_name(), already account for sysfs strings. Signed-off-by:Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Ben Skeggs <bskeggs@redhat.com> Cc: Bjorn Helgaas <bhelgaas@google.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Brice Goglin <Brice.Goglin@inria.fr> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Dave Jiang <dave.jiang@intel.com> Cc: David Airlie <airlied@linux.ie> Cc: David Hildenbrand <david@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Hulk Robot <hulkci@huawei.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@mellanox.com> Cc: Jason Yan <yanaijie@huawei.com> Cc: Jeff Moyer <jmoyer@redhat.com> Cc: "Jérôme Glisse" <jglisse@redhat.com> Cc: Jia He <justin.he@arm.com> Cc: Joao Martins <joao.m.martins@oracle.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Juergen Gross <jgross@suse.com> Cc: kernel test robot <lkp@intel.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Paul Mackerras <paulus@ozlabs.org> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Tom Lendacky <thomas.lendacky@amd.com> Cc: Vishal Verma <vishal.l.verma@intel.com> Cc: Vivek Goyal <vgoyal@redhat.com> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: Will Deacon <will@kernel.org> Link: https://lkml.kernel.org/r/159643102106.4062302.12229802117645312104.stgit@dwillia2-desk3.amr.corp.intel.com Link: https://lkml.kernel.org/r/160106114576.30709.2960091665444712180.stgit@dwillia2-desk3.amr.corp.intel.comSigned-off-by:
-
Dan Williams authored
Add a seed device concept for dynamic dax regions to be able to split the region amongst multiple sub-instances. The seed device, similar to libnvdimm seed devices, is a device that starts with zero capacity allocated and unbound to a driver. In contrast to libnvdimm seed devices explicit 'create' and 'delete' interfaces are added to the region to trigger seeds to be created and unused devices to be reclaimed. The explicit create and delete replaces implicit create as a side effect of probe and implicit delete when writing 0 to the size that libnvdimm implements. Delete can be performed on any 0-sized and idle device. This avoids the gymnastics of needing to move device_unregister() to its own async context. Specifically, it avoids the deadlock of deleting a device via one of its own attributes. It is also less surprising to userspace which never sees an extra device it did not request. For now just add the device creation, teardown, and ->probe() prevention. A later patch will arrange for the 'dax/size' attribute to be writable to allocate capacity from the region. Signed-off-by:
-
Dan Williams authored
In preparation for introducing seed devices the dax-bus core needs to be able to intercept ->probe() and ->remove() operations. Towards that end arrange for the bus and drivers to switch from raw 'struct device' driver operations to 'struct dev_dax' typed operations. Reported-by:
Hulk Robot <hulkci@huawei.com> Signed-off-by:
-
Dan Williams authored
In preparation for a facility that enables dax regions to be sub-divided, introduce infrastructure to track and allocate region capacity. The new dax_region/available_size attribute is only enabled for volatile hmem devices, not pmem devices that are defined by nvdimm namespace boundaries. This is per Jeff's feedback the last time dynamic device-dax capacity allocation support was discussed. Signed-off-by:
-
Dan Williams authored
Towards removing the mode specific @dax_kmem_res attribute from the generic 'struct dev_dax', and preparing for multi-range support, change the kmem driver to use the idiomatic release_mem_region() to pair with the initial request_mem_region(). This also eliminates the need to open code the release of the resource allocated by request_mem_region(). As there are no more dax_kmem_res users, delete this struct member. Signed-off-by:
-
Dan Williams authored
Towards removing the mode specific @dax_kmem_res attribute from the generic 'struct dev_dax', and preparing for multi-range support, move resource name tracking to driver data. The memory for the resource name needs to have its own lifetime separate from the device bind lifetime for cases where the driver is unbound, but the kmem range could not be unplugged from the page allocator. Signed-off-by:
-
Dan Williams authored
Towards removing the mode specific @dax_kmem_res attribute from the generic 'struct dev_dax', and preparing for multi-range support, teach the driver to calculate the hotplug range from the device range. The hotplug range is the trivially calculated memory-block-size aligned version of the device range. Signed-off-by:
-
Dan Williams authored
The passed in dev_pagemap is only required in the pmem case as the libnvdimm core may have reserved a vmem_altmap for dev_memremap_pages() to place the memmap in pmem directly. In the hmem case there is no agent reserving an altmap so it can all be handled by a core internal default. Pass the resource range via a new @range property of 'struct dev_dax_data'. Signed-off-by:
-
Dan Williams authored
In preparation for adding more parameters to instance creation, move existing parameters to a new struct. Signed-off-by:
-
Dan Williams authored
All callers specify the same flags to alloc_dax_region(), so there is no need to allow for anything other than PFN_DEV|PFN_MAP, or carry a ->pfn_flags around on the region. Device-dax instances are always page backed. Signed-off-by:
-
Dan Williams authored
The hmem enabling in commit cf8741ac ("ACPI: NUMA: HMAT: Register "soft reserved" memory as an "hmem" device") only registered ranges to the hmem driver for each soft-reservation that also appeared in the HMAT. While this is meant to encourage platform firmware to "do the right thing" and publish an HMAT, the corollary is that platforms that fail to publish an accurate HMAT will strand memory from Linux usage. Additionally, the "efi_fake_mem" kernel command line option enabling will strand memory by default without an HMAT. Arrange for "soft reserved" memory that goes unclaimed by HMAT entries to be published as raw resource ranges for the hmem driver to consume. Include a module parameter to disable either this fallback behavior, or the hmat enabling from creating hmem devices. The module parameter requires the hmem device enabling to have unique name in the module namespace: "device_hmem". The driver depends on the architecture providing phys_to_target_node() which is only x86 via numa_meminfo() and arm64 via a generic memblock implementation. [joao.m.martins@oracle.com: require NUMA_KEEP_MEMINFO for phys_to_target_node()] Link: https://lkml.kernel.org/r/aaae71a7-4846-f5cc-5acf-cf05fdb1f2dc@oracle.comSigned-off-by:
-
Dan Williams authored
In preparation to set a fallback value for dev_dax->target_node, introduce generic fallback helpers for phys_to_target_node() A generic implementation based on node-data or memblock was proposed, but as noted by Mike: "Here again, I would prefer to add a weak default for phys_to_target_node() because the "generic" implementation is not really generic. The fallback to reserved ranges is x86 specfic because on x86 most of the reserved areas is not in memblock.memory. AFAIK, no other architecture does this." The info message in the generic memory_add_physaddr_to_nid() implementation is fixed up to properly reflect that memory_add_physaddr_to_nid() communicates "online" node info and phys_to_target_node() indicates "target / to-be-onlined" node info. [akpm@linux-foundation.org: fix CONFIG_MEMORY_HOTPLUG=n build] Link: https://lkml.kernel.org/r/202008252130.7YrHIyMI%25lkp@intel.comSigned-off-by:
-
