From 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 Mon Sep 17 00:00:00 2001 From: Tim Chen Date: Mon, 21 Sep 2026 17:37:22 -0700 Subject: sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling bug alb_break_llc() decides whether to break LLC preference during active load balance. It does so by testing that every runnable fair task on the source rq prefers its LLC: env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable But the two counters cover different sets. nr_pref_llc_running is updated in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued, so it follows queued tasks. h_nr_runnable is updated in set_delayed()/ clear_delayed() and drops delay-dequeued tasks. So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks, alb_break_llc() returns false, and active balance is free to pull a task off its preferred LLC. Active balance only moves runnable tasks, and this is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB skips the per-task test in can_migrate_task(). The runnable set is the one we want. Fix it on the counter side. A task should be counted in nr_pref_llc_running exactly while it is both queued on its preferred LLC (pref_llc_queued) and runnable (!sched_delayed). Define that membership once in task_pref_llc_runnable(), and adjust the counter only through pref_llc_running_inc()/pref_llc_running_dec() from the four sites that change either input: account_llc_enqueue(), account_llc_dequeue(), set_delayed() and clear_delayed(). Gating every update on the same predicate keeps the delay, wake and dequeue paths from double-counting or underflowing; see the comments at those sites for the ordering. nr_llc_running and sd->llc_counts are not touched and stay on queued semantics. Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC") Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/ Reported-by: Zhan Xusheng Suggested-by: Chen Yu Signed-off-by: Tim Chen Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Kayra Cizmeci Cc: # v7.2.x Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com --- Documentation/driver-api/nvmem.rst | 202 +++++++++++++++++++++++++++++++++++++ 1 file changed, 202 insertions(+) create mode 100644 Documentation/driver-api/nvmem.rst (limited to 'Documentation/driver-api/nvmem.rst') diff --git a/Documentation/driver-api/nvmem.rst b/Documentation/driver-api/nvmem.rst new file mode 100644 index 000000000..790e2dc65 --- /dev/null +++ b/Documentation/driver-api/nvmem.rst @@ -0,0 +1,202 @@ +.. SPDX-License-Identifier: GPL-2.0 + +=============== +NVMEM Subsystem +=============== + + Srinivas Kandagatla + +This document explains the NVMEM Framework along with the APIs provided, +and how to use it. + +1. Introduction +=============== +*NVMEM* is the abbreviation for Non Volatile Memory layer. It is used to +retrieve configuration of SOC or Device specific data from non volatile +memories like eeprom, efuses and so on. + +Before this framework existed, NVMEM drivers like eeprom were stored in +drivers/misc, where they all had to duplicate pretty much the same code to +register a sysfs file, allow in-kernel users to access the content of the +devices they were driving, etc. + +This was also a problem as far as other in-kernel users were involved, since +the solutions used were pretty much different from one driver to another, there +was a rather big abstraction leak. + +This framework aims at solve these problems. It also introduces DT +representation for consumer devices to go get the data they require (MAC +Addresses, SoC/Revision ID, part numbers, and so on) from the NVMEMs. + +NVMEM Providers ++++++++++++++++ + +NVMEM provider refers to an entity that implements methods to initialize, read +and write the non-volatile memory. + +2. Registering/Unregistering the NVMEM provider +=============================================== + +A NVMEM provider can register with NVMEM core by supplying relevant +nvmem configuration to nvmem_register(), on success core would return a valid +nvmem_device pointer. + +nvmem_unregister() is used to unregister a previously registered provider. + +For example, a simple nvram case:: + + static int brcm_nvram_probe(struct platform_device *pdev) + { + struct nvmem_config config = { + .name = "brcm-nvram", + .reg_read = brcm_nvram_read, + }; + ... + config.dev = &pdev->dev; + config.priv = priv; + config.size = resource_size(res); + + devm_nvmem_register(&config); + } + +Device drivers can define and register an nvmem cell using the nvmem_cell_info +struct:: + + static const struct nvmem_cell_info foo_nvmem_cell = { + { + .name = "macaddr", + .offset = 0x7f00, + .bytes = ETH_ALEN, + } + }; + + int nvmem_add_one_cell(nvmem, &foo_nvmem_cell); + +Additionally it is possible to create nvmem cell lookup entries and register +them with the nvmem framework from machine code as shown in the example below:: + + static struct nvmem_cell_lookup foo_nvmem_lookup = { + .nvmem_name = "i2c-eeprom", + .cell_name = "macaddr", + .dev_id = "foo_mac.0", + .con_id = "mac-address", + }; + + nvmem_add_cell_lookups(&foo_nvmem_lookup, 1); + +NVMEM Consumers ++++++++++++++++ + +NVMEM consumers are the entities which make use of the NVMEM provider to +read from and to NVMEM. + +3. NVMEM cell based consumer APIs +================================= + +NVMEM cells are the data entries/fields in the NVMEM. +The NVMEM framework provides 3 APIs to read/write NVMEM cells:: + + struct nvmem_cell *nvmem_cell_get(struct device *dev, const char *name); + struct nvmem_cell *devm_nvmem_cell_get(struct device *dev, const char *name); + + void nvmem_cell_put(struct nvmem_cell *cell); + void devm_nvmem_cell_put(struct device *dev, struct nvmem_cell *cell); + + void *nvmem_cell_read(struct nvmem_cell *cell, ssize_t *len); + int nvmem_cell_write(struct nvmem_cell *cell, void *buf, ssize_t len); + +`*nvmem_cell_get()` apis will get a reference to nvmem cell for a given id, +and nvmem_cell_read/write() can then read or write to the cell. +Once the usage of the cell is finished the consumer should call +`*nvmem_cell_put()` to free all the allocation memory for the cell. + +4. Direct NVMEM device based consumer APIs +========================================== + +In some instances it is necessary to directly read/write the NVMEM. +To facilitate such consumers NVMEM framework provides below apis:: + + struct nvmem_device *nvmem_device_get(struct device *dev, const char *name); + struct nvmem_device *devm_nvmem_device_get(struct device *dev, + const char *name); + struct nvmem_device *nvmem_device_find(void *data, + int (*match)(struct device *dev, const void *data)); + void nvmem_device_put(struct nvmem_device *nvmem); + int nvmem_device_read(struct nvmem_device *nvmem, unsigned int offset, + size_t bytes, void *buf); + int nvmem_device_write(struct nvmem_device *nvmem, unsigned int offset, + size_t bytes, void *buf); + int nvmem_device_cell_read(struct nvmem_device *nvmem, + struct nvmem_cell_info *info, void *buf); + int nvmem_device_cell_write(struct nvmem_device *nvmem, + struct nvmem_cell_info *info, void *buf); + +Before the consumers can read/write NVMEM directly, it should get hold +of nvmem_controller from one of the `*nvmem_device_get()` api. + +The difference between these apis and cell based apis is that these apis always +take nvmem_device as parameter. + +5. Releasing a reference to the NVMEM +===================================== + +When a consumer no longer needs the NVMEM, it has to release the reference +to the NVMEM it has obtained using the APIs mentioned in the above section. +The NVMEM framework provides 2 APIs to release a reference to the NVMEM:: + + void nvmem_cell_put(struct nvmem_cell *cell); + void devm_nvmem_cell_put(struct device *dev, struct nvmem_cell *cell); + void nvmem_device_put(struct nvmem_device *nvmem); + void devm_nvmem_device_put(struct device *dev, struct nvmem_device *nvmem); + +Both these APIs are used to release a reference to the NVMEM and +devm_nvmem_cell_put and devm_nvmem_device_put destroys the devres associated +with this NVMEM. + +Userspace ++++++++++ + +6. Userspace binary interface +============================== + +Userspace can read/write the raw NVMEM file located at:: + + /sys/bus/nvmem/devices/*/nvmem + +ex:: + + hexdump /sys/bus/nvmem/devices/qfprom0/nvmem + + 0000000 0000 0000 0000 0000 0000 0000 0000 0000 + * + 00000a0 db10 2240 0000 e000 0c00 0c00 0000 0c00 + 0000000 0000 0000 0000 0000 0000 0000 0000 0000 + ... + * + 0001000 + +7. DeviceTree Binding +===================== + +See Documentation/devicetree/bindings/nvmem/nvmem.txt + +8. NVMEM layouts +================ + +NVMEM layouts are yet another mechanism to create cells. With the device +tree binding it is possible to specify simple cells by using an offset +and a length. Sometimes, the cells doesn't have a static offset, but +the content is still well defined, e.g. tag-length-values. In this case, +the NVMEM device content has to be first parsed and the cells need to +be added accordingly. Layouts let you read the content of the NVMEM device +and let you add cells dynamically. + +Another use case for layouts is the post processing of cells. With layouts, +it is possible to associate a custom post processing hook to a cell. It +even possible to add this hook to cells not created by the layout itself. + +9. Internal kernel API +====================== + +.. kernel-doc:: drivers/nvmem/core.c + :export: -- cgit v1.3.1