- 12 Mar, 2024 12 commits
-
-
Dave Jiang authored
For the numa nodes that are not created by SRAT, no memory_target is allocated and is not managed by the HMAT_REPORTING code. Therefore hmat_callback() memory hotplug notifier will exit early on those NUMA nodes. The CXL memory hotplug notifier will need to call node_set_perf_attrs() directly in order to setup the access sysfs attributes. In acpi_numa_init(), the last proximity domain (pxm) id created by SRAT is stored. Add a helper function acpi_node_backed_by_real_pxm() in order to check if a NUMA node id is defined by SRAT or created by CFMWS. node_set_perf_attrs() symbol is exported to allow update of perf attribs for a node. The sysfs path of /sys/devices/system/node/nodeX/access0/initiators/* is created by node_set_perf_attrs() for the various attributes where nodeX is matched to the NUMA node of the CXL region. Cc: Rafael J. Wysocki <rafael@kernel.org> Reviewed-by:
Alison Schofield <alison.schofield@intel.com> Reviewed-by:
Jonathan Cameron <Jonathan.Cameron@huawei.com> Tested-by:
Dave Jiang <dave.jiang@intel.com> Link: https://lore.kernel.org/r/20240308220055.2172956-13-dave.jiang@intel.comSigned-off-by:
Dan Williams <dan.j.williams@intel.com>
-
Dave Jiang authored
When the CXL region is formed, the driver computes the performance data for the region. However this data is not available at the node data collection that has been populated by the HMAT during kernel initialization. Add a memory hotplug notifier to update the access coordinates to the 'struct memory_target' context kept by the HMAT_REPORTING code. Add CXL_CALLBACK_PRI for a memory hotplug callback priority. Set the priority number to be called before HMAT_CALLBACK_PRI. The CXL update must happen before hmat_callback(). A new HMAT_REPORTING helper hmat_update_target_coordinates() is added in order to allow CXL to update the memory_target access coordinates. A new ext_updated member is added to the memory_target to indicate that the access coordinates within the memory_target has been updated by an external agent such as CXL. This prevents data being overwritten by the hmat_update_target_attrs() triggered by hmat_callback(). Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Rafael J. Wysocki <rafael@kernel.org> Reviewed-by:
Huang, Ying <ying.huang@intel.com> Reviewed-by:
-
Dave Jiang authored
Add read/write latencies and bandwidth sysfs attributes for the enabled CXL region. The bandwidth is the aggregated bandwidth of all devices that contribute to the CXL region. The latency is the worst latency of the device amongst all the devices that contribute to the CXL region. Reviewed-by:
-
Dave Jiang authored
Calculate and store the performance data for a CXL region. Find the worst read and write latency for all the included ranges from each of the devices that attributes to the region and designate that as the latency data. Sum all the read and write bandwidth data for each of the device region and that is the total bandwidth for the region. The perf list is expected to be constructed before the endpoint decoders are registered and thus there should be no early reading of the entries from the region assemble action. The calling of the region qos calculate function is under the protection of cxl_dpa_rwsem and will ensure that all DPA associated work has completed. Reviewed-by:
-
Dave Jiang authored
Move setting of cxlmd->endpoint to before calling add_device() on the port device. Otherwise when referencing cxlmd->endpoint in region discovery code that is triggered by the port driver probe function, the endpoint port pointer is not valid. Current code does not hit this issue yet since cxlmd->endpoint is not being referenced during region discovery. However follow on code that does performance calculations will. Tested-by:
Wonjae Lee <wj28.lee@samsung.com> Reviewed-by:
-
Dave Jiang authored
Retrieve the qos_class (QTG ID) using the access coordinates from the nearest CPU rather than the nearst initiator that may not be a CPU. This may be the more appropriate number that applications care about. For most cases, access0 and access1 have the same values. Link: https://lore.kernel.org/linux-cxl/20240112113023.00006c50@Huawei.com/Suggested-by:
-
Dave Jiang authored
The difference between access class 0 and access class 1 for 'struct access_coordinate', if any, is that class 0 is for the distance from the target to the closest initiator and that class 1 is for the distance from the target to the closest CPU. For CXL memory, the nearest initiator may not necessarily be a CPU node. The performance path from the CXL endpoint to the host bridge should remain the same. However, the numbers extracted and stored from HMAT is the difference for the two access classes. Split out the performance numbers for the host bridge (generic target) from the calculation of the entire path in order to allow calculation of both access classes for a CXL region. Reviewed-by:
-
Dave Jiang authored
Refactor the common code of combining coordinates in order to reduce code. Create a new function cxl_cooordinates_combine() it combine two 'struct access_coordinate'. Reviewed-by:
-
Dave Jiang authored
Update acpi_get_genport_coordinates() to allow retrieval of both access classes of the 'struct access_coordinate' for a generic target. The update will allow CXL code to compute access coordinates for both access class. Cc: Rafael J. Wysocki <rafael@kernel.org> Reviewed-by:
-
Dave Jiang authored
In order to compute access0 and access1 classes for CXL memory, 2 levels of generic port information must be stored. Access0 will indicate the generic port access coordinates to the closest initiator and access1 will indicate the generic port access coordinates to the cloest CPU. Cc: Rafael J. Wysocki <rafael@kernel.org> Reviewed-by:
-
Dave Jiang authored
Both generic node and HMAT handling code have been using magic numbers to indicate access classes for 'struct access_coordinate'. Introduce enums to enumerate the access0 and access1 classes shared by the two subsystems. Update the function parameters and callers as appropriate to utilize the new enum. Access0 is named to ACCESS_COORDINATE_LOCAL in order to indicate that the access class is for 'struct access_coordinate' between a target node and the nearest initiator node. Access1 is named to ACCESS_COORDINATE_CPU in order to indicate that the access class is for 'struct access_coordinate' between a target node and the nearest CPU node. Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Rafael J. Wysocki <rafael@kernel.org> Reviewed-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by:
-
Dave Jiang authored
For generic targets, there's no reason to call register_memory_node_under_compute_node() with the access levels that are only visible to HMAT handling code. Only update the attributes and rename hmat_register_generic_target_initiators() to hmat_update_generic_target(). The original call path ends up triggering register_memory_node_under_compute_node(). Although the access level would be "3" and not impact any current node arrays, it introduces unwanted data into the numa node access_coordinate array. Fixes: a3a3e341 ("acpi: numa: Add setting of generic port system locality attributes") Cc: Rafael J. Wysocki <rafael@kernel.org> Reviewed-by:
-
- 25 Feb, 2024 28 commits
-
-
Linus Torvalds authored
-
https://evilpiepirate.org/git/bcachefsLinus Torvalds authored
Pull bcachefs fixes from Kent Overstreet: "Some more mostly boring fixes, but some not User reported ones: - the BTREE_ITER_FILTER_SNAPSHOTS one fixes a really nasty performance bug; user reported an untar initially taking two seconds and then ~2 minutes - kill a __GFP_NOFAIL in the buffered read path; this was a leftover from the trickier fix to kill __GFP_NOFAIL in readahead, where we can't return errors (and have to silently truncate the read ourselves). bcachefs can't use GFP_NOFAIL for folio state unlike iomap based filesystems because our folio state is just barely too big, 2MB hugepages cause us to exceed the 2 page threshhold for GFP_NOFAIL. additionally, the flags argument was just buggy, we weren't supplying GFP_KERNEL previously (!)" * tag 'bcachefs-2024-02-25' of https://evilpiepirate.org/git/bcachefs: bcachefs: fix bch2_save_backtrace() bcachefs: Fix check_snapshot() memcpy bcachefs: Fix bch2_journal_flush_device_pins() bcachefs: fix iov_iter count underflow on sub-block dio read bcachefs: Fix BTREE_ITER_FILTER_SNAPSHOTS on inodes btree bcachefs: Kill __GFP_NOFAIL in buffered read path bcachefs: fix backpointer_to_text() when dev does not exist -
Kent Overstreet authored
Missed a call in the previous fix. Signed-off-by:Kent Overstreet <kent.overstreet@linux.dev>
-
git://git.lwn.net/linuxLinus Torvalds authored
Pull two documentation build fixes from Jonathan Corbet: - The XFS online fsck documentation uses incredibly deeply nested subsection and list nesting; that broke the PDF docs build. Tweak a parameter to tell LaTeX to allow the deeper nesting. - Fix a 6.8 PDF-build regression * tag 'docs-6.8-fixes3' of git://git.lwn.net/linux: docs: translations: use attribute to store current language docs: Instruct LaTeX to cope with deeper nesting
-
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usbLinus Torvalds authored
Pull USB fixes from Greg KH: "Here are some small USB fixes for 6.8-rc6 to resolve some reported problems. These include: - regression fixes with typec tpcm code as reported by many - cdnsp and cdns3 driver fixes - usb role setting code bugfixes - build fix for uhci driver - ncm gadget driver bugfix - MAINTAINERS entry update All of these have been in linux-next all week with no reported issues and there is at least one fix in here that is in Thorsten's regression list that is being tracked" * tag 'usb-6.8-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: usb: typec: tpcm: Fix issues with power being removed during reset MAINTAINERS: Drop myself as maintainer of TYPEC port controller drivers usb: gadget: ncm: Avoid dropping datagrams of properly parsed NTBs Revert "usb: typec: tcpm: reset counter when enter into unattached state after try role" usb: gadget: omap_udc: fix USB gadget regression on Palm TE usb: dwc3: gadget: Don't disconnect if not started usb: cdns3: fix memory double free when handle zero packet usb: cdns3: fixed memory use after free at cdns3_gadget_ep_disable() usb: roles: don't get/set_role() when usb_role_switch is unregistered usb: roles: fix NULL pointer issue when put module's reference usb: cdnsp: fixed issue with incorrect detecting CDNSP family controllers usb: cdnsp: blocked some cdns3 specific code usb: uhci-grlib: Explicitly include linux/platform_device.h
-
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/ttyLinus Torvalds authored
Pull tty/serial driver fixes from Greg KH: "Here are three small serial/tty driver fixes for 6.8-rc6 that resolve the following reported errors: - riscv hvc console driver fix that was reported by many - amba-pl011 serial driver fix for RS485 mode - stm32 serial driver fix for RS485 mode All of these have been in linux-next all week with no reported problems" * tag 'tty-6.8-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty: serial: amba-pl011: Fix DMA transmission in RS485 mode serial: stm32: do not always set SER_RS485_RX_DURING_TX if RS485 is enabled tty: hvc: Don't enable the RISC-V SBI console by default
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull x86 fixes from Borislav Petkov: - Make sure clearing CPU buffers using VERW happens at the latest possible point in the return-to-userspace path, otherwise memory accesses after the VERW execution could cause data to land in CPU buffers again * tag 'x86_urgent_for_v6.8_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: KVM/VMX: Move VERW closer to VMentry for MDS mitigation KVM/VMX: Use BT+JNC, i.e. EFLAGS.CF to select VMRESUME vs. VMLAUNCH x86/bugs: Use ALTERNATIVE() instead of mds_user_clear static key x86/entry_32: Add VERW just before userspace transition x86/entry_64: Add VERW just before userspace transition x86/bugs: Add asm helpers for executing VERW
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull irq fixes from Borislav Petkov: - Make sure GICv4 always gets initialized to prevent a kexec-ed kernel from silently failing to set it up - Do not call bus_get_dev_root() for the mbigen irqchip as it always returns NULL - use NULL directly - Fix hardware interrupt number truncation when assigning MSI interrupts - Correct sending end-of-interrupt messages to disabled interrupts lines on RISC-V PLIC * tag 'irq_urgent_for_v6.8_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: irqchip/gic-v3-its: Do not assume vPE tables are preallocated irqchip/mbigen: Don't use bus_get_dev_root() to find the parent PCI/MSI: Prevent MSI hardware interrupt number truncation irqchip/sifive-plic: Enable interrupt if needed before EOI
-
git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofsLinus Torvalds authored
Pull erofs fix from Gao Xiang: - Fix page refcount leak when looking up specific inodes introduced by metabuf reworking * tag 'erofs-for-6.8-rc6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs: erofs: fix refcount on the metabuf used for inode lookup
-
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfsLinus Torvalds authored
Pull RCU pathwalk fixes from Al Viro: "We still have some races in filesystem methods when exposed to RCU pathwalk. This series is a result of code audit (the second round of it) and it should deal with most of that stuff. Still pending: ntfs3 ->d_hash()/->d_compare() and ceph_d_revalidate(). Up to maintainers (a note for NTFS folks - when documentation says that a method may not block, it *does* imply that blocking allocations are to be avoided. Really)" [ More explanations for people who aren't familiar with the vagaries of RCU path walking: most of it is hidden from filesystems, but if a filesystem actively participates in the low-level path walking it needs to make sure the fields involved in that walk are RCU-safe. That "actively participate in low-level path walking" includes things like having its own ->d_hash()/->d_compare() routines, or by having its own directory permission function that doesn't just use the common helpers. Having a ->d_revalidate() function will also have this issue. Note that instead of making everything RCU safe you can also choose to abort the RCU pathwalk if your operation cannot be done safely under RCU, but that obviously comes with a performance penalty. One common pattern is to allow the simple cases under RCU, and abort only if you need to do something more complicated. So not everything needs to be RCU-safe, and things like the inode etc that the VFS itself maintains obviously already are. But these fixes tend to be about properly RCU-delaying things like ->s_fs_info that are maintained by the filesystem and that got potentially released too early. - Linus ] * tag 'pull-fixes.pathwalk-rcu-2' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: ext4_get_link(): fix breakage in RCU mode cifs_get_link(): bail out in unsafe case fuse: fix UAF in rcu pathwalks procfs: make freeing proc_fs_info rcu-delayed procfs: move dropping pde and pid from ->evict_inode() to ->free_inode() nfs: fix UAF on pathwalk running into umount nfs: make nfs_set_verifier() safe for use in RCU pathwalk afs: fix __afs_break_callback() / afs_drop_open_mmap() race hfsplus: switch to rcu-delayed unloading of nls and freeing ->s_fs_info exfat: move freeing sbi, upcase table and dropping nls into rcu-delayed helper affs: free affs_sb_info with kfree_rcu() rcu pathwalk: prevent bogus hard errors from may_lookup() fs/super.c: don't drop ->s_user_ns until we free struct super_block itself
-
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfsLinus Torvalds authored
Pull vfs fixes from Al Viro: "A couple of fixes - revert of regression from this cycle and a fix for erofs failure exit breakage (had been there since way back)" * tag 'pull-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: erofs: fix handling kern_mount() failure Revert "get rid of DCACHE_GENOCIDE"
-
Al Viro authored
1) errors from ext4_getblk() should not be propagated to caller unless we are really sure that we would've gotten the same error in non-RCU pathwalk. 2) we leak buffer_heads if ext4_getblk() is successful, but bh is not uptodate. Signed-off-by:Al Viro <viro@zeniv.linux.org.uk>
-
Al Viro authored
->d_revalidate() bails out there, anyway. It's not enough to prevent getting into ->get_link() in RCU mode, but that could happen only in a very contrieved setup. Not worth trying to do anything fancy here unless ->d_revalidate() stops kicking out of RCU mode at least in some cases. Reviewed-by:
Christian Brauner <brauner@kernel.org> Acked-by:
Miklos Szeredi <mszeredi@redhat.com> Signed-off-by:
-
Al Viro authored
->permission(), ->get_link() and ->inode_get_acl() might dereference ->s_fs_info (and, in case of ->permission(), ->s_fs_info->fc->user_ns as well) when called from rcu pathwalk. Freeing ->s_fs_info->fc is rcu-delayed; we need to make freeing ->s_fs_info and dropping ->user_ns rcu-delayed too. Signed-off-by: -
Al Viro authored
makes proc_pid_ns() safe from rcu pathwalk (put_pid_ns() is still synchronous, but that's not a problem - it does rcu-delay everything that needs to be) Reviewed-by:
-
Al Viro authored
that keeps both around until struct inode is freed, making access to them safe from rcu-pathwalk Acked-by:
-
Al Viro authored
NFS ->d_revalidate(), ->permission() and ->get_link() need to access some parts of nfs_server when called in RCU mode: server->flags server->caps *(server->io_stats) and, worst of all, call server->nfs_client->rpc_ops->have_delegation (the last one - as NFS_PROTO(inode)->have_delegation()). We really don't want to RCU-delay the entire nfs_free_server() (it would have to be done with schedule_work() from RCU callback, since it can't be made to run from interrupt context), but actual freeing of nfs_server and ->io_stats can be done via call_rcu() just fine. nfs_client part is handled simply by making nfs_free_client() use kfree_rcu(). Acked-by:
-
Al Viro authored
nfs_set_verifier() relies upon dentry being pinned; if that's the case, grabbing ->d_lock stabilizes ->d_parent and guarantees that ->d_parent points to a positive dentry. For something we'd run into in RCU mode that is *not* true - dentry might've been through dentry_kill() just as we grabbed ->d_lock, with its parent going through the same just as we get to into nfs_set_verifier_locked(). It might get to detaching inode (and zeroing ->d_inode) before nfs_set_verifier_locked() gets to fetching that; we get an oops as the result. That can happen in nfs{,4} ->d_revalidate(); the call chain in question is nfs_set_verifier_locked() <- nfs_set_verifier() <- nfs_lookup_revalidate_delegated() <- nfs{,4}_do_lookup_revalidate(). We have checked that the parent had been positive, but that's done before we get to nfs_set_verifier() and it's possible for memory pressure to pick our dentry as eviction candidate by that time. If that happens, back-to-back attempts to kill dentry and its parent are quite normal. Sure, in case of eviction we'll fail the ->d_seq check in the caller, but we need to survive until we return there... Acked-by: -
Al Viro authored
In __afs_break_callback() we might check ->cb_nr_mmap and if it's non-zero do queue_work(&vnode->cb_work). In afs_drop_open_mmap() we decrement ->cb_nr_mmap and do flush_work(&vnode->cb_work) if it reaches zero. The trouble is, there's nothing to prevent __afs_break_callback() from seeing ->cb_nr_mmap before the decrement and do queue_work() after both the decrement and flush_work(). If that happens, we might be in trouble - vnode might get freed before the queued work runs. __afs_break_callback() is always done under ->cb_lock, so let's make sure that ->cb_nr_mmap can change from non-zero to zero while holding ->cb_lock (the spinlock component of it - it's a seqlock and we don't need to mess with the counter). Acked-by:
-
Al Viro authored
->d_hash() and ->d_compare() use those, so we need to delay freeing them. Reviewed-by:
-
Al Viro authored
That stuff can be accessed by ->d_hash()/->d_compare(); as it is, we have a hard-to-hit UAF if rcu pathwalk manages to get into ->d_hash() on a filesystem that is in process of getting shut down. Besides, having nls and upcase table cleanup moved from ->put_super() towards the place where sbi is freed makes for simpler failure exits. Acked-by:
-
Al Viro authored
one of the flags in it is used by ->d_hash()/->d_compare() Reviewed-by:
-
Al Viro authored
If lazy call of ->permission() returns a hard error, check that try_to_unlazy() succeeds before returning it. That both makes life easier for ->permission() instances and closes the race in ENOTDIR handling - it is possible that positive d_can_lookup() seen in link_path_walk() applies to the state *after* unlink() + mkdir(), while nd->inode matches the state prior to that. Normally seeing e.g. EACCES from permission check in rcu pathwalk means that with some timings non-rcu pathwalk would've run into the same; however, running into a non-executable regular file in the middle of a pathname would not get to permission check - it would fail with ENOTDIR instead. Reviewed-by:
-
Al Viro authored
Avoids fun races in RCU pathwalk... Same goes for freeing LSM shite hanging off super_block's arse. Reviewed-by:
-
Kent Overstreet authored
check_snapshot() copies the bch_snapshot to a temporary to easily handle older versions that don't have all the fields of the current version, but it lacked a min() to correctly handle keys newer and larger than the current version. Signed-off-by: -
Kent Overstreet authored
If a journal write errored, the list of devices it was written to could be empty - we're not supposed to mark an empty replicas list. Signed-off-by: -
Brian Foster authored
bch2_direct_IO_read() checks the request offset and size for sector alignment and then falls through to a couple calculations to shrink the size of the request based on the inode size. The problem is that these checks round up to the fs block size, which runs the risk of underflowing iter->count if the block size happens to be large enough. This is triggered by fstest generic/361 with a 4k block size, which subsequently leads to a crash. To avoid this crash, check that the shorten length doesn't exceed the overall length of the iter. Fixes: Signed-off-by:
Brian Foster <bfoster@redhat.com> Reviewed-by:
Su Yue <glass.su@suse.com> Signed-off-by:
-
Kent Overstreet authored
If we're in FILTER_SNAPSHOTS mode and we start scanning a range of the keyspace where no keys are visible in the current snapshot, we have a problem - we'll scan for a very long time before scanning terminates. Awhile back, this was fixed for most cases with peek_upto() (and assertions that enforce that it's being used). But the fix missed the fact that the inodes btree is different - every key offset is in a different snapshot tree, not just the inode field. Fixes: Signed-off-by:
-
