- 13 Dec, 2018 9 commits
-
-
Coly Li authored
The option gc_after_writeback is disabled by default, because garbage collection will discard SSD data which drops cached data. Echo 1 into /sys/fs/bcache/<UUID>/internal/gc_after_writeback will enable this option, which wakes up gc thread when writeback accomplished and all cached data is clean. This option is helpful for people who cares writing performance more. In heavy writing workload, all cached data can be clean only happens when writeback thread cleans all cached data in I/O idle time. In such situation a following gc running may help to shrink bcache B+ tree and discard more clean data, which may be helpful for future writing requests. If you are not sure whether this is helpful for your own workload, please leave it as disabled by default. Signed-off-by:
Coly Li <colyli@suse.de> Signed-off-by:
Jens Axboe <axboe@kernel.dk>
-
Coly Li authored
Garbage collection thread starts to work when c->sectors_to_gc is negative value, otherwise nothing will happen even the gc thread is woken up by wake_up_gc(). force_wake_up_gc() sets c->sectors_to_gc to -1 before calling wake_up_gc(), then gc thread may have chance to run if no one else sets c->sectors_to_gc to a positive value before gc_should_run(). This routine can be called where the gc thread is woken up and required to run in force. Signed-off-by:
-
Shenghui Wang authored
"echo 1 > writeback_running" marks writeback_running even if no writeback kthread created as "d_strtoul(writeback_running)" will simply set dc-> writeback_running without checking the existence of dc->writeback_thread. Add check for setting writeback_running via sysfs: if no writeback kthread available, reject setting to 1. v2 -> v3: * Make message on wrong assignment more clear. * Print name of bcache device instead of name of backing device. Signed-off-by:
Shenghui Wang <shhuiw@foxmail.com> Signed-off-by:
-
Shenghui Wang authored
A fresh backing device is not attached to any cache_set, and has no writeback kthread created until first attached to some cache_set. But bch_cached_dev_writeback_init run " dc->writeback_running = true; WARN_ON(test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)); " for any newly formatted backing devices. For a fresh standalone backing device, we can get something like following even if no writeback kthread created: ------------------------ /sys/block/bcache0/bcache# cat writeback_running 1 /sys/block/bcache0/bcache# cat writeback_rate_debug rate: 512.0k/sec dirty: 0.0k target: 0.0k proportional: 0.0k integral: 0.0k change: 0.0k/sec next io: -15427384ms The none ZERO fields are misleading as no alive writeback kthread yet. Set dc->writeback_running false as no writeback thread created in bch_cached_dev_writeback_init(). We have writeback thread created and woken up in bch_cached_dev_writeback _start(). Set dc->writeback_running true before bch_writeback_queue() called, as a writeback thread will check if dc->writeback_running is true before writing back dirty data, and hung if false detected. After the change, we can get the following output for a fresh standalone backing device: ----------------------- /sys/block/bcache0/bcache$ cat writeback_running 0 /sys/block/bcache0/bcache# cat writeback_rate_debug rate: 0.0k/sec dirty: 0.0k target: 0.0k proportional: 0.0k integral: 0.0k change: 0.0k/sec next io: 0ms v1 -> v2: Set dc->writeback_running before bch_writeback_queue() called, Signed-off-by:
-
Shenghui Wang authored
We have struct cached_dev allocated by kzalloc in register_bcache(), which initializes all the fields of cached_dev with 0s. And commit ce4c3e19 ("bcache: Replace bch_read_string_list() by __sysfs_match_string()") has remove the string "default". Update the comment. Signed-off-by:
-
Shenghui Wang authored
commit 220bb38c ("bcache: Break up struct search") introduced changes to struct search and s->iop. bypass/bio are fields of struct data_insert_op now. Update the comment. Signed-off-by:
-
Shenghui Wang authored
debugfs_remove and debugfs_remove_recursive will check if the dentry pointer is NULL or ERR, and will do nothing in that case. Remove the check in cache_set_free and bch_debug_init. Signed-off-by:
-
Shenghui Wang authored
We have the following define for btree iterator: struct btree_iter { size_t size, used; #ifdef CONFIG_BCACHE_DEBUG struct btree_keys *b; #endif struct btree_iter_set { struct bkey *k, *end; } data[MAX_BSETS]; }; We can see that the length of data[] field is static MAX_BSETS, which is defined as 4 currently. But a btree node on disk could have too many bsets for an iterator to fit on the stack - maybe far more that MAX_BSETS. Have to dynamically allocate space to host more btree_iter_sets. bch_cache_set_alloc() will make sure the pool cache_set->fill_iter can allocate an iterator equipped with enough room that can host (sb.bucket_size / sb.block_size) btree_iter_sets, which is more than static MAX_BSETS. bch_btree_node_read_done() will use that pool to allocate one iterator, to host many bsets in one btree node. Add more comment around cache_set->fill_iter to make code less confusing. Signed-off-by: -
Dennis Zhou authored
Between v3 [1] and v4 [2] of the blkg association series, the association point moved from generic_make_request_checks(), which is called after the request enters the queue, to bio_set_dev(), which is when the bio is formed before submit_bio(). When the request_queue goes away, the blkgs supporting the request_queue are destroyed and then the q->root_blkg is set to %NULL. This patch adds a %NULL check to blkg_tryget_closest() to prevent the NPE caused by the above. It also adds a guard to see if the request_queue is dying when creating a blkg to prevent creating a blkg for a dead request_queue. [1] https://lore.kernel.org/lkml/20180911184137.35897-1-dennisszhou@gmail.com/ [2] https://lore.kernel.org/lkml/20181126211946.77067-1-dennis@kernel.org/ Fixes: 5cdf2e3f ("blkcg: associate blkg when associating a device") Reported-and-tested-by:
Ming Lei <ming.lei@redhat.com> Reviewed-by:
Bart Van Assche <bvanassche@acm.org> Signed-off-by:
Dennis Zhou <dennis@kernel.org> Signed-off-by:
-
- 12 Dec, 2018 2 commits
-
-
Ming Lei authored
rwb_enabled() can't be changed when there is any inflight IO. wbt_disable_default() may set rwb->wb_normal as zero, however the blk_stat timer may still be pending, and the timer function will update wrb->wb_normal again. This patch introduces blk_stat_deactivate() and applies it in wbt_disable_default(), then the following IO hang triggered when running parted & switching io scheduler can be fixed: [ 369.937806] INFO: task parted:3645 blocked for more than 120 seconds. [ 369.938941] Not tainted 4.20.0-rc6-00284-g906c801e5248 #498 [ 369.939797] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 369.940768] parted D 0 3645 3239 0x00000000 [ 369.941500] Call Trace: [ 369.941874] ? __schedule+0x6d9/0x74c [ 369.942392] ? wbt_done+0x5e/0x5e [ 369.942864] ? wbt_cleanup_cb+0x16/0x16 [ 369.943404] ? wbt_done+0x5e/0x5e [ 369.943874] schedule+0x67/0x78 [ 369.944298] io_schedule+0x12/0x33 [ 369.944771] rq_qos_wait+0xb5/0x119 [ 369.945193] ? karma_partition+0x1c2/0x1c2 [ 369.945691] ? wbt_cleanup_cb+0x16/0x16 [ 369.946151] wbt_wait+0x85/0xb6 [ 369.946540] __rq_qos_throttle+0x23/0x2f [ 369.947014] blk_mq_make_request+0xe6/0x40a [ 369.947518] generic_make_request+0x192/0x2fe [ 369.948042] ? submit_bio+0x103/0x11f [ 369.948486] ? __radix_tree_lookup+0x35/0xb5 [ 369.949011] submit_bio+0x103/0x11f [ 369.949436] ? blkg_lookup_slowpath+0x25/0x44 [ 369.949962] submit_bio_wait+0x53/0x7f [ 369.950469] blkdev_issue_flush+0x8a/0xae [ 369.951032] blkdev_fsync+0x2f/0x3a [ 369.951502] do_fsync+0x2e/0x47 [ 369.951887] __x64_sys_fsync+0x10/0x13 [ 369.952374] do_syscall_64+0x89/0x149 [ 369.952819] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 369.953492] RIP: 0033:0x7f95a1e729d4 [ 369.953996] Code: Bad RIP value. [ 369.954456] RSP: 002b:00007ffdb570dd48 EFLAGS: 00000246 ORIG_RAX: 000000000000004a [ 369.955506] RAX: ffffffffffffffda RBX: 000055c2139c6be0 RCX: 00007f95a1e729d4 [ 369.956389] RDX: 0000000000000001 RSI: 0000000000001261 RDI: 0000000000000004 [ 369.957325] RBP: 0000000000000002 R08: 0000000000000000 R09: 000055c2139c6ce0 [ 369.958199] R10: 0000000000000000 R11: 0000000000000246 R12: 000055c2139c0380 [ 369.959143] R13: 0000000000000004 R14: 0000000000000100 R15: 0000000000000008 Cc: stable@vger.kernel.org Cc: Paolo Valente <paolo.valente@linaro.org> Signed-off-by:
-
Jens Axboe authored
We're missing a deferred clear off the shallow get, which can cause a hang. Additionally, when we resize the sbitmap, we should also flush deferred clears for good measure. Ensure we have full coverage on batch clears, even for paths where we would not be doing deferred clear. This makes it less error prone for future additions. Reported-by:
-
- 11 Dec, 2018 24 commits
-
-
Igor Konopko authored
Ehen using pblk with 0 sized metadata both ppa list and meta list points to the same memory since pblk_dma_meta_size() returns 0 in that case. This patch fix that issue by ensuring that pblk_dma_meta_size() always returns space equal to sizeof(struct pblk_sec_meta) and thus ppa list and meta list points to different memory address. Even that in that case drive does not really care about meta_list pointer, this is the easiest way to fix that issue without introducing changes in many places in the code just for 0 sized metadata case. The same approach needs to be also done for pblk_get_sec_meta() since we also cannot point to the same memory address in meta buffer when we are using it for pblk recovery process Reported-by:
Hans Holmberg <hans.holmberg@cnexlabs.com> Tested-by:
Igor Konopko <igor.j.konopko@intel.com> Signed-off-by:
Matias Bjørling <mb@lightnvm.io> Signed-off-by:
-
Igor Konopko authored
pblk performs recovery of open lines by storing the LBA in the per LBA metadata field. Recovery therefore only works for drives that has this field. This patch adds support for packed metadata, which store l2p mapping for open lines in last sector of every write unit and enables drives without per IO metadata to recover open lines. After this patch, drives with OOB size <16B will use packed metadata and metadata size larger than16B will continue to use the device per IO metadata. Reviewed-by:
Javier González <javier@cnexlabs.com> Signed-off-by:
-
Igor Konopko authored
Currently pblk only check the size of I/O metadata and does not take into account if this metadata is in a separate buffer or interleaved in a single metadata buffer. In reality only the first scenario is supported, where second mode will break pblk functionality during any IO operation. This patch prevents pblk to be instantiated in case device only supports interleaved metadata. Reviewed-by:
-
Igor Konopko authored
Currently lightnvm and pblk uses single DMA pool, for which the entry size always is equal to PAGE_SIZE. The contents of each entry allocated from the DMA pool consists of a PPA list (8bytes * 64), leaving 56bytes * 64 space for metadata. Since the metadata field can be bigger, such as 128 bytes, the static size does not cover this use-case. This patch adds support for I/O metadata above 56 bytes by changing DMA pool size based on device meta size and allows pblk to use OOB metadata >=16B. Reviewed-by:
-
Igor Konopko authored
pblk currently assumes that size of OOB metadata on drive is always equal to size of pblk_sec_meta struct. This commit add helpers which will allow to handle different sizes of OOB metadata on drive in the future. After this patch only OOB metadata equal to 16 bytes is supported. Reviewed-by:
-
Igor Konopko authored
Currently DMA allocated memory is reused on partial read for lba_list_mem and lba_list_media arrays. In preparation for dynamic DMA pool sizes we need to move this arrays into pblk_pr_ctx structures. Reviewed-by:
-
Javier González authored
The current kref implementation around pblk global caches triggers a false positive on refcount_inc_checked() (when called) as the kref is initialized to 0. Instead of usint kref_inc() on a 0 reference, which is in principle correct, use kref_init() to avoid the check. This is also more explicit about what actually happens on cache creation. In the process, do a small refactoring to use kref helpers. Fixes: 1864de94 "lightnvm: pblk: stop recreating global caches" Signed-off-by:
-
Matias Bjørling authored
Currently the geometry of an OCSSD is enumerated using a two step approach: First, nvm_register is called, the OCSSD identify command is issued, and second the geometry sos and csecs values are read either from the OCSSD identify if it is a 1.2 drive, or from the NVMe namespace data structure if it is a 2.0 device. This patch recombines it into a single step, such that nvm_register can use the csecs and sos fields independent of which version is used. This enables one to dynamically size the lightnvm subsystem dma pool. Reviewed-by:
-
Javier González authored
pblk's recovery path is single threaded and therefore a number of assumptions regarding concurrency can be made. To avoid confusion, make this explicit with a couple of comments in the code. Signed-off-by:
-
Hua Su authored
Protect the list_add on the pblk_line_init_bb() error path in case this code is used for some other purpose in the future. Signed-off-by:
Hua Su <suhua.tanke@gmail.com> Reviewed-by:
-
Hua Su authored
Signed-off-by:
-
Hans Holmberg authored
Remove the call to pblk_line_replace_data as it returns directly because we have not set l_mg->data_next yet. Signed-off-by:
Javier González <javier@javigon.com> Signed-off-by:
-
Hans Holmberg authored
The chunk metadata is allocated with vmalloc, so we need to use vfree to free it. Fixes: 090ee26f ("lightnvm: use internal allocation for chunk log page") Signed-off-by:
-
Hans Holmberg authored
ADDR_POOL_SIZE is not used anymore, so remove the macro. Signed-off-by:
-
Hans Holmberg authored
In a worst-case scenario (random writes), OP% of sectors in each line will be invalid, and we will then need to move data out of 100/OP% lines to free a single line. So, to prevent the possibility of running out of lines, temporarily block user writes when there is less than 100/OP% free lines. Also ensure that pblk creation does not produce instances with insufficient over provisioning. Insufficient over-provising is not a problem on real hardware, but often an issue when running QEMU simulations (with few lines). 100 lines is enough to create a sane instance with the standard (11%) over provisioning. Signed-off-by:
-
Hans Holmberg authored
If mapping fails (i.e. when running out of lines), handle the error and stop writing. Signed-off-by:
-
Hans Holmberg authored
Lines inflicted with write errors lines might be recovered if they have not been recycled after write error garbage collection. Ensure that the emeta accounting of valid lbas is correct for such lines to avoid recovery inconsistencies. Signed-off-by:
-
Hans Holmberg authored
Make sure we only look up valid lba addresses on the resubmission path. If an lba is invalidated in the write buffer, that sector will be submitted to disk (as it is already mapped to a ppa), and that write might fail, resulting in a crash when trying to look up the lba in the mapping table (as the lba is marked as invalid). Signed-off-by:
-
Hans Holmberg authored
The check for chunk closes suffers from an off-by-one issue, leading to chunk close events not being traced. Fixes: 4c44abf4 ("lightnvm: pblk: add trace events for chunk states") Signed-off-by:
-
Geert Uytterhoeven authored
With gcc 4.1: drivers/lightnvm/core.c: In function ‘nvm_get_bb_meta’: drivers/lightnvm/core.c:977: warning: ‘ret’ may be used uninitialized in this function and drivers/nvme/host/lightnvm.c: In function ‘nvme_nvm_get_chk_meta’: drivers/nvme/host/lightnvm.c:580: warning: ‘ret’ may be used uninitialized in this function Indeed, if (for the former) the number of channels or LUNs is zero, or (for both) the passed number of chunks is zero, ret will be returned uninitialized. Fix this by preinitializing ret to zero. Fixes: aff3fb18 ("lightnvm: move bad block and chunk state logic to core") Fixes: a294c199 ("lightnvm: implement get log report chunk helpers") Signed-off-by:Geert Uytterhoeven <geert@linux-m68k.org> Signed-off-by:
-
Zhoujie Wu authored
The smeta area l2p mapping is empty, and actually the recovery procedure only need to restore data sector's l2p mapping. So ignore the smeta oob scan. Signed-off-by:
Zhoujie Wu <zjwu@marvell.com> Reviewed-by:
-
Mike Snitzer authored
The md->wait waitqueue is used by both bio-based and request-based DM. Commit dbd3bbd2 ("dm rq: leverage blk_mq_queue_busy() to check for outstanding IO") lost sight of the requirement that dm_wait_for_completion() must work with all types of DM devices. Fix md_in_flight() to call the blk-mq or bio-based method accordingly. Fixes: dbd3bbd2 ("dm rq: leverage blk_mq_queue_busy() to check for outstanding IO") Signed-off-by:
Mike Snitzer <snitzer@redhat.com> Signed-off-by:
-
Jens Axboe authored
Guenter reported an boot hang issue on HPPA after we default to 0 poll queues. We have two issues in the queue count calculations: 1) We don't separate the poll queues from the read/write queues. This is important, since the former doesn't need interrupts. 2) The adjust logic is broken. Adjust the poll queue count before doing nvme_calc_io_queues(). The poll queue count is only limited by the IO queue count we were able to get from the controller, not failures in the IRQ allocation loop. This leaves nvme_calc_io_queues() just adjusting the read/write queue map. Reported-by:
Reported-by: Guenter Roeck <linux@roeck-us.net> Reviewed-by:
Christoph Hellwig <hch@lst.de> Reviewed-by:
Sagi Grimberg <sagi@grimberg.me> Signed-off-by:
-
Jens Axboe authored
After switching to percpu inflight counters, the inflight check is totally buggy. It's perfectly valid for some counters to be non-zero while having a total inflight IO count of 0, that's how these kinds of counters work (inc on one CPU, dec on another). Fix the md_in_flight() check to sum all counters before returning a false positive, potentially. While at it, remove the inflight read for IO completion. We don't need it, just wake anyone that's waiting for the IO count to drop to zero. The caller needs to re-check that value anyway when woken, which it does. Fixes: 6f757231 ("dm: remove the pending IO accounting") Acked-by:
-
- 10 Dec, 2018 5 commits
-
-
Jens Axboe authored
For cases where we can only fail with IO in-flight, we should be using BLK_STS_DEV_RESOURCE instead of BLK_STS_RESOURCE. The latter refers to system wide resource constraints. Signed-off-by: -
Arnd Bergmann authored
The "cmd_slot_unal" semaphore is never used in a blocking way but only as an atomic counter. Change the code to using atomic_dec_if_positive() as a better API. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
-
Mikulas Patocka authored
Remove the "pending" atomic counters, that duplicate block-core's in_flight counters, and update md_in_flight() to look at percpu in_flight counters. Signed-off-by:
Mikulas Patocka <mpatocka@redhat.com> Signed-off-by:
-
Mikulas Patocka authored
The previous patches deleted all the code that needed the second value returned from part_in_flight - now the kernel only uses the first value. Consequently, part_in_flight (and blk_mq_in_flight) may be changed so that it only returns one value. This patch just refactors the code, there's no functional change. Signed-off-by:
-
Mikulas Patocka authored
Now when part_round_stats is gone, we can switch to per-cpu in-flight counters. We use the local-atomic type local_t, so that if part_inc_in_flight or part_dec_in_flight is reentrantly called from an interrupt, the value will be correct. The other counters could be corrupted due to reentrant interrupt, but the corruption only results in slight counter skew - the in_flight counter must be exact, so it needs local_t. Signed-off-by:
-
