- 24 Aug, 2004 40 commits
-
-
Neil Brown authored
1/ rationalise read_balance and "map" in raid1. Discard map and tidyup the interface to read_balance so it can be used instead. 2/ use offsetof rather than a caclulation to find the size of an structure with a var-length array at the end. 3/ remove some meaningless #defines 4/ use printk_ratelimit to limit reports of failed sectors being redirected. Signed-off-by:
Neil Brown <neilb@cse.unsw.edu.au> Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Neil Brown authored
1/ Introduce "mddev->resync_max_sectors" so that an md personality can ask for resync to cover a different address range than that of a single drive. raid10 will use this. 2/ fix is_mddev_idle so that if there seem to be a negative number of events, it doesn't immediately assume activity. 3/ make "sync_io" (the count of IO sectors used for array resync) an atomic_t to avoid SMP races. 4/ Pass md_sync_acct a "block_device" rather than the containing "rdev", as the whole rdev isn't needed. Also make this an inline function. 5/ Make sure recovery gets interrupted on any error. Signed-off-by:
-
William Lee Irwin III authored
During the kernel summit, some discussion was had about the support requirements for a userspace program loader that loads executables into hugetlb on behalf of a major application (Oracle). In order to support this in a robust fashion, the cleanup of the hugetlb must be robust in the presence of disorderly termination of the programs (e.g. kill -9). Hence, the cleanup semantics are those of System V shared memory, but Linux' System V shared memory needs one critical extension for this use: executability. The following microscopic patch enables this major application to provide robust hugetlb cleanup. Signed-off-by:
-
William Lee Irwin III authored
PAE is artificially limited in terms of swapspace to the same bitsplit as ordinary i386, a 5/24 split (32 swapfiles, 64GB max swapfile size), when a 5/27 split (32 swapfiles, 512GB max swapfile size) is feasible. This patch transparently removes that limitation by using more of the space available in PAE's wider ptes for swap ptes. While this is obviously not likely to be used directly, it is important from the standpoint of strict non-overcommit, where the swapspace must be potentially usable in order to be reserved for non-overcommit. There are workloads with Committed_AS of over 256GB on ia32 PAE wanting strict non-overcommit to prevent being OOM killed. Signed-off-by:
-
Zwane Mwaikambo authored
This was broken when the mwait stuff went in since it executes after the initial idle_setup() has already selected an idle routine and overrides it with default_idle. Signed-off-by:
Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by:
Zwane Mwaikambo <zwane@linuxpower.ca> Signed-off-by:
-
Manfred Spraul authored
Michael Kerrisk found a bug in the shm accounting code: sysv shm allows to create SHMMNI+1 shared memory segments, instead of SHMMNI segments. The +1 is probably from the first shared anonymous mapping implementation that used the sysv code to implement shared anon mappings. The implementation got replaced, it's now the other way around (sysv uses the shared anon code), but the +1 remained. Signed-off-by:
Manfred Spraul <manfred@colorfullife.com> Signed-off-by:
-
Zwane Mwaikambo authored
The incorrect mask was being used when writing back to PMNC write-only-zero bits as well as only ticking the CCNT every 64 processor cycles. Tested on IOP331 and PXA270, i'm still looking for XScale1 users... Signed-off-by:
Luca Rossato <l.rossato@tiscali.it> Signed-off-by:
Zwane Mwaikambo <zwane@arm.linux.org.uk> Signed-off-by:
-
William Lee Irwin III authored
The sole remaining usage of CLONE_IDLETASK is to determine whether pid allocation should be performed in copy_process(). This patch eliminates that last branch on CLONE_IDLETASK in the normal process creation path, removes the masking of CLONE_IDLETASK from clone_flags as it's now ignored under all circumstances, and furthermore eliminates the symbol CLONE_IDLETASK entirely. From: William Lee Irwin III <wli@holomorphy.com> Fix the fork-idle consolidation. During that consolidation, the generic code was made to pass a pointer to on-stack pt_regs that had been memset() to 0. ia64, however, requires a NULL pt_regs pointer argument and dispatches on that in its copy_thread() function to do SMP trampoline-specific RSE -related setup. Passing pointers to zeroed pt_regs resulted in SMP wakeup -time deadlocks and exceptions. Signed-off-by:
-
William Lee Irwin III authored
Every arch now bears the burden of sanitizing CLONE_IDLETASK out of the clone_flags passed to do_fork() by userspace. This patch hoists the masking of CLONE_IDLETASK out of the system call entrypoints into do_fork(), and thereby removes some small overheads from do_fork(), as do_fork() may now assume that CLONE_IDLETASK has been cleared. Signed-off-by:
-
Josh Aas authored
Attached is a patch that greatly improves the speed of freeing boot memory. On ia64 machines with 2GB or more memory (I didn't test with less, but I can't imagine there being a problem), the speed improvement is about 75% for the function free_all_bootmem_core. This translates to savings on the order of 1 minute / TB of memory during boot time. That number comes from testing on a machine with 512GB, and extrapolating based on profiling of an unpatched 4TB machine. For 4 and 8 TB machines, the time spent in this function is about 1 minutes/TB, which is painful especially given that there is no indication of what is going on put to the console (this issue to possibly be addressed later). The basic idea is to free higher order pages instead of going through every single one. Also, some unnecessary atomic operations are done away with and replaced with non-atomic equivalents, and prefetching is done where it helps the most. For a more in-depth discusion of this patch, please see the linux-ia64 archives (topic is "free bootmem feedback patch"). The patch is originally Tony Luck's, and I added some further optimizations (non-atomic ops improvements and prefetching). Signed-off-by:
Tony Luck <tony.luck@intel.com> Signed-off-by:
Josh Aas <josha@sgi.com> Signed-off-by:
-
Badari Pulavarty authored
The problem is, if we increase our readhead size arbitrarily (say 2M), we call mpage_readpages() with 2M and when it tries to allocated a bio enough to fit 2M it fails, then we kick it back to "confused" code - which does 4K at a time. The fix is to ask for the maxium the driver can handle. Signed-off-by:
-
Roland Dreier authored
Looks like arch/i386/kernel/doublefault.c is one place in the code that hardcodes the assumption that PAGE_OFFSET == 0xC0000000. Here's a patch that fixes that. Signed-off-by:
Roland Dreier <roland@topspin.com> Signed-off-by:
-
James Courtier-Dutton authored
Rui Sousa has been unreachable for a long time now, so I have taken over the emu10k1 project on sf.net. Signed-off-by:
-
Andrea Arcangeli authored
There's some minor bug in the d_path handling (the nfsd one may not the the correct fix, there's no failure path for it, so I just terminate the string, and the last one in the audit subsystem is just a robustness cleanup if somebody will extend d_path in the future, right now it's a noop). Signed-off-by:
-
Andrew Morton authored
Fix up the logic which decides when the caller can dip into page reserves. - If the caller has realtime scheduling policy, or if the caller cannot run direct reclaim, then allow the caller to use up to a quarter of the page reserves. - If the caller has __GFP_HIGH then allow the caller to use up to half of the page reserves. - If the caller has PF_MEMALLOC then the caller can use 100% of the page reserves. Signed-off-by:
-
Nick Piggin authored
Previously the ->protection[] logic was broken. It was difficult to follow and basically didn't use the asynch reclaim watermarks (pages_min, pages_low, pages_high) properly. Now use ->protection *only* for lower-zone protection. So the allocator now explicitly uses the ->pages_low, ->pages_min watermarks and adds ->protection on top of that, instead of trying to use ->protection for everything. Pages are allocated down to (->pages_low + ->protection), once this is reached, kswapd the background reclaim is started; after this, we can allocate down to (->pages_min + ->protection) without blocking; the memory below pages_min is reserved for __GFP_HIGH and PF_MEMALLOC allocations. kswapd attempts to reclaim memory until ->pages_high is reached. Signed-off-by:
Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by:
-
Nick Piggin authored
Slightly change the writeout watermark calculations so we keep background and synchronous writeout watermarks in the same ratios after adjusting them for the amout of mapped memory. This ensures we should always attempt to start background writeout before synchronous writeout and preserves the admin's desired background-versus-forground ratios after we've auto-adjusted one of them. Signed-off-by:
Nick Piggin <nickpiggin@cyberone.com.au> Signed-off-by:
-
Hugh Dickins authored
A tmpfs user reported increasingly slow directory reads when repeatedly creating and unlinking in a mkstemp-like way. The negative dentries accumulate alarmingly (until memory pressure finally frees them), and are just a hindrance to any in-memory filesystem. simple_lookup set d_op to arrange for negative dentries to be deleted immediately. (But I failed to discover how it is that on-disk filesystems seem to keep their negative dentries within manageable bounds: this effect was gross with tmpfs or ramfs, but no problem at all with extN or reiser.) Signed-off-by:
Hugh Dickins <hugh@veritas.com> Signed-off-by:
-
Hugh Dickins authored
Clarify mmgrab by collapsing it into get_task_mm (in fork.c not inline), and commenting on the special case it is guarding against: when use_mm in an AIO daemon temporarily adopts the mm while it's on its way out. Signed-off-by:
-
Marcelo Tosatti authored
Back when we were discussing the need for a memory barrier in sync_page(), it came to me (thanks Andrea!) that the bit operations can be perfectly reordered on architectures other than x86. I think the commentary on i386 bitops.h is misleading, its worth to note that that these operations are not guaranteed not to be reordered on different architectures. clear_bit() already does that: * clear_bit() is atomic and may not be reordered. However, it does * not contain a memory barrier, so if it is used for locking purposes, * you should call smp_mb__before_clear_bit() and/or smp_mb__after_clear_bit() * in order to ensure changes are visible on other processors. Signed-off-by:
-
Hugh Dickins authored
Swapoff can make good use of a page's anon_vma and index, while it's still left in swapcache, or once it's brought back in and the first pte mapped back: unuse_vma go directly to just one page of only those vmas with the same anon_vma. And unuse_process can skip any vmas without an anon_vma (extending the hugetlb check: hugetlb vmas have no anon_vma). This just hacks in on top of the existing procedure, still going through all the vmas of all the mms in mmlist. A more elegant procedure might replace mmlist by a list of anon_vmas: but that would be more work to implement, with apparently more overhead in the common paths. Signed-off-by:
-
Hugh Dickins authored
With page_map_lock out of the way, there's no need for page_referenced and try_to_unmap to use trylocks - provided we switch anon_vma->lock and mm->page_table_lock around in anon_vma_prepare. Though I suppose it's possible that we'll find that vmscan makes better progress with trylocks than spinning - we're free to choose trylocks again if so. Try to update the mm lock ordering documentation in filemap.c. But I still find it confusing, and I've no idea of where to stop. So add an mm lock ordering list I can understand to rmap.c. Signed-off-by:
-
Hugh Dickins authored
With page_map_lock gone, how to stabilize page->mapping's anon_vma while acquiring anon_vma->lock in page_referenced_anon and try_to_unmap_anon? The page cannot actually be freed (vmscan holds reference), but however much we check page_mapped (which guarantees that anon_vma is in use - or would guarantee that if we added suitable barriers), there's no locking against page becoming unmapped the instant after, then anon_vma freed. It's okay to take anon_vma->lock after it's freed, so long as it remains a struct anon_vma (its list would become empty, or perhaps reused for an unrelated anon_vma: but no problem since we always check that the page located is the right one); but corruption if that memory gets reused for some other purpose. This is not unique: it's liable to be problem whenever the kernel tries to approach a structure obliquely. It's generally solved with an atomic reference count; but one advantage of anon_vma over anonmm is that it does not have such a count, and it would be a backward step to add one. Therefore... implement SLAB_DESTROY_BY_RCU flag, to guarantee that such a kmem_cache_alloc'ed structure cannot get freed to other use while the rcu_read_lock is held i.e. preempt disabled; and use that for anon_vma. Fix concerns raised by Manfred: this flag is incompatible with poisoning and destructor, and kmem_cache_destroy needs to synchronize_kernel. I hope SLAB_DESTROY_BY_RCU may be useful elsewhere; but though it's safe for little anon_vma, I'd be reluctant to use it on any caches whose immediate shrinkage under pressure is important to the system. Signed-off-by:
-
Hugh Dickins authored
The pte_chains rmap used pte_chain_lock (bit_spin_lock on PG_chainlock) to lock its pte_chains. We kept this (as page_map_lock: bit_spin_lock on PG_maplock) when we moved to objrmap. But the file objrmap locks its vma tree with mapping->i_mmap_lock, and the anon objrmap locks its vma list with anon_vma->lock: so isn't the page_map_lock superfluous? Pretty much, yes. The mapcount was protected by it, and needs to become an atomic: starting at -1 like page _count, so nr_mapped can be tracked precisely up and down. The last page_remove_rmap can't clear anon page mapping any more, because of races with page_add_rmap; from which some BUG_ONs must go for the same reason, but they've served their purpose. vmscan decisions are naturally racy, little change there beyond removing page_map_lock/unlock. But to stabilize the file-backed page->mapping against truncation while acquiring i_mmap_lock, page_referenced_file now needs page lock to be held even for refill_inactive_zone. There's a similar issue in acquiring anon_vma->lock, where page lock doesn't help: which this patch pretends to handle, but actually it needs the next. Roughly 10% cut off lmbench fork numbers on my 2*HT*P4. Must confess my testing failed to show the races even while they were knowingly exposed: would benefit from testing on racier equipment. Signed-off-by:
-
Hugh Dickins authored
First of a batch of five patches to eliminate rmap's page_map_lock, replace its trylocking by spinlocking, and use anon_vma to speed up swapoff. Patches updated from the originals against 2.6.7-mm7: nothing new so I won't spam the list, but including Manfred's SLAB_DESTROY_BY_RCU fixes, and omitting the unuse_process mmap_sem fix already in 2.6.8-rc3. This patch: Replace the PG_anon page->flags bit by setting the lower bit of the pointer in page->mapping when it's anon_vma: PAGE_MAPPING_ANON bit. We're about to eliminate the locking which kept the flags and mapping in synch: it's much easier to work on a local copy of page->mapping, than worry about whether flags and mapping are in synch (though I imagine it could be done, at greater cost, with some barriers). Signed-off-by:
-
Roger Luethi authored
I really wanted /proc/pid/statm to die and I still believe the reasoning is valid. As it doesn't look like that is going to happen, though, I offer this fix for the respective documentation. Note: lrs/drs fields are switched. Signed-off-by:
Roger Luethi <rl@hellgate.ch> Signed-off-by:
-
Arjan van de Ven authored
This enables apic=bigsmp automatically on some big HP machines that need it. This makes them boot without kernel parameters on a generic arch kernel. Signed-off-by:
-
William Lee Irwin III authored
We need to be able to dereference struct device in include/asm-ia64/dma-mapping.h. Signed-off-by:
-
Andi Kleen authored
Prevents some noise during boot up when no MD volumes are found. I think I picked it up from someone else, but I cannot remember from whom (sorry) Cc: Neil Brown <neilb@cse.unsw.edu.au> Signed-off-by:
-
Pete Zaitcev authored
We at Red Hat shipped a larger number of arguments for quite some time, it was required for installations on IBM mainframe (s390), which doesn't have a good way to pass arguments. There are a number of reasonable situations that go past the current limits of 8. One that comes to mind is when you want to perform a manual vnc install on a headless machine using anaconda. This requires passing in a number of parameters to get anaconda past the initial (no-gui) loader screens. Signed-off-by:
-
Suparna Bhattacharya authored
From: Chris Mason I compared the 2.6 pipetest results with the 2.4 suse kernel, and 2.6 was roughly 40% slower. During the pipetest run, 2.6 generates ~600,000 context switches per second while 2.4 generates 30 or so. aio-context-switch (attached) has a few changes that reduces our context switch rate, and bring performance back up to 2.4 levels. These have only really been tested against pipetest, they might make other workloads worse. The basic theory behind the patch is that it is better for the userland process to call run_iocbs than it is to schedule away and let the worker thread do it. 1) on io_submit, use run_iocbs instead of run_iocb 2) on io_getevents, call run_iocbs if no events were available. 3) don't let two procs call run_iocbs for the same context at the same time. They just end up bouncing on spinlocks. The first three optimizations got me down to 360,000 context switches per second, and they help build a little structure to allow optimization #4, which uses queue_delayed_work(HZ/10) instead of queue_work. That brings down the number of context switches to 2.4 levels. Adds aio_run_all_iocbs so that normal processes can run all the pending retries on the run list. This allows worker threads to keep using list splicing, but regular procs get to run the list until it stays empty. The end result should be less work for the worker threads. I was able to trigger short stalls (1sec) with aio-stress, and with the current patch they are gone. Could be wishful thinking on my part though, please let me know how this works for you. Signed-off-by: -
Suparna Bhattacharya authored
This patch tries be a little fairer across multiple io contexts in handling retries, helping make sure progress happens uniformly across different io contexts (especially if they are acting on independent queues). It splices the ioctx runlist before processing it in __aio_run_iocbs. If new iocbs get added to the ctx in meantime, it queues a fresh workqueue entry instead of handling them righaway, so that other ioctxs' retries get a chance to be processed before the newer entries in the queue. This might make a difference in a situation where retries are getting queued very fast on one ioctx, while the workqueue entry for another ioctx is stuck behind it. I've only seen this occasionally earlier and can't recreate it consistently, but may be worth including anyway. Signed-off-by:
-
Suparna Bhattacharya authored
From: Daniel McNeil <daniel@osdl.org> From: Chris Mason <mason@suse.com> AIO: retry infrastructure fixes and enhancements Reorganises, comments and fixes the AIO retry logic. Fixes and enhancements include: - Split iocb setup and execution in io_submit (also fixes io_submit error reporting) - Use aio workqueue instead of keventd for retries - Default high level retry methods - Subtle use_mm/unuse_mm fix - Code commenting - Fix aio process hang on EINVAL (Daniel McNeil) - Hold the context lock across unuse_mm - Acquire task_lock in use_mm() - Allow fops to override the retry method with their own - Elevated ref count for AIO retries (Daniel McNeil) - set_fs needed when calling use_mm - Flush workqueue on __put_ioctx (Chris Mason) - Fix io_cancel to work with retries (Chris Mason) - Read-immediate option for socket/pipe retry support Note on default high-level retry methods support ================================================ High-level retry methods allows an AIO request to be executed as a series of non-blocking iterations, where each iteration retries the remaining part of the request from where the last iteration left off, by reissuing the corresponding AIO fop routine with modified arguments representing the remaining I/O. The retries are "kicked" via the AIO waitqueue callback aio_wake_function() which replaces the default wait queue entry used for blocking waits. The high level retry infrastructure is responsible for running the iterations in the mm context (address space) of the caller, and ensures that only one retry instance is active at a given time, thus relieving the fops themselves from having to deal with potential races of that sort. Signed-off-by: -
Bjorn Helgaas authored
Add pci_enable_device()/pci_disable_device(). In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. Signed-off-by:
Bjorn Helgaas <bjorn.helgaas@hp.com> Signed-off-by:
-
Bjorn Helgaas authored
Add pci_enable_device()/pci_disable_device(). In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. Signed-off-by:
-
Bjorn Helgaas authored
Add pci_enable_device()/pci_disable_device(). In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. Signed-off-by:
-
Bjorn Helgaas authored
Add pci_enable_device()/pci_disable_device(). In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. Signed-off-by:
-
Bjorn Helgaas authored
Add pci_enable_device()/pci_disable_device(). In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. Signed-off-by:
-
Bjorn Helgaas authored
Add pci_enable_device()/pci_disable_device(). In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. Signed-off-by:
-
Bjorn Helgaas authored
I don't have this hardware, so this has been compiled but not tested. Add pci_enable_device()/pci_disable_device In the past, drivers often worked without this, but it is now required in order to route PCI interrupts correctly. In addition, this driver incorrectly used the IRQ value from PCI config space rather than the one in the struct pci_dev. Signed-off-by:
-
