- 29 Apr, 2008 40 commits
-
-
Pierre Peiffer authored
All IPCs make use of an intermetiate *_setbuf structure to handle the IPC_SET command. This is not really needed and, moreover, it complicates a little bit the code. This patch gets rid of the use of it and uses directly the semid64_ds/ msgid64_ds/shmid64_ds structure. In addition of removing one struture declaration, it also simplifies and improves a little bit the common 64-bits path. Signed-off-by:
Pierre Peiffer <pierre.peiffer@bull.net> Acked-by:
Serge Hallyn <serue@us.ibm.com> Cc: Nadia Derbey <Nadia.Derbey@bull.net> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Pierre Peiffer authored
semctl_down() takes one unused parameter: semnum. This patch proposes to get rid of it. Signed-off-by:
-
Pierre Peiffer authored
semctl_down is called with the rwmutex (the one which protects the list of ipcs) taken in write mode. This patch moves this rwmutex taken in write-mode inside semctl_down. This has the advantages of reducing a little bit the window during which this rwmutex is taken, clarifying sys_semctl, and finally of having a coherent behaviour with [shm|msg]ctl_down Signed-off-by:
-
Pierre Peiffer authored
Currently, sys_msgctl is not easy to read. This patch tries to improve that by introducing the msgctl_down function to handle all commands requiring the rwmutex to be taken in write mode (ie IPC_SET and IPC_RMID for now). It is the equivalent function of semctl_down for message queues. This greatly changes the readability of sys_msgctl and also harmonizes the way these commands are handled among all IPCs. Signed-off-by:
-
Pierre Peiffer authored
Currently, the way the different commands are handled in sys_shmctl introduces some duplicated code. This patch introduces the shmctl_down function to handle all the commands requiring the rwmutex to be taken in write mode (ie IPC_SET and IPC_RMID for now). It is the equivalent function of semctl_down for shared memory. This removes some duplicated code for handling these both commands and harmonizes the way they are handled among all IPCs. Signed-off-by:
-
Pierre Peiffer authored
Trivial patch which adds some small locking functions and makes use of them to factorize some part of the code and to make it cleaner. Signed-off-by:
-
Nadia Derbey authored
The enhancement as asked for by Yasunori: if msgmni is set to a negative value, register it back into the ipcns notifier chain. A new interface has been added to the notification mechanism: notifier_chain_cond_register() registers a notifier block only if not already registered. With that new interface we avoid taking care of the states changes in procfs. Signed-off-by:
Nadia Derbey <Nadia.Derbey@bull.net> Cc: Yasunori Goto <y-goto@jp.fujitsu.com> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Mingming Cao <cmm@us.ibm.com> Cc: Pierre Peiffer <pierre.peiffer@bull.net> Signed-off-by:
-
Nadia Derbey authored
Make msgmni not recomputed anymore upon ipc namespace creation / removal or memory add/remove, as soon as it has been set from userland. As soon as msgmni is explicitly set via procfs or sysctl(), the associated callback routine is unregistered from the ipc namespace notifier chain. Signed-off-by:
-
Nadia Derbey authored
Introduce a notification mechanism that aims at recomputing msgmni each time an ipc namespace is created or removed. The ipc namespace notifier chain already defined for memory hotplug management is used for that purpose too. Each time a new ipc namespace is allocated or an existing ipc namespace is removed, the ipcns notifier chain is notified. The callback routine for each registered ipc namespace is then activated in order to recompute msgmni for that namespace. Signed-off-by:
-
Nadia Derbey authored
Make the memory hotplug chain's mutex held for a shorter time: when memory is offlined or onlined a work item is added to the global workqueue. When the work item is run, it notifies the ipcns notifier chain with the IPCNS_MEMCHANGED event. Signed-off-by:
-
Nadia Derbey authored
Introduce the registration of a callback routine that recomputes msg_ctlmni upon memory add / remove. A single notifier block is registered in the hotplug memory chain for all the ipc namespaces. Since the ipc namespaces are not linked together, they have their own notification chain: one notifier_block is defined per ipc namespace. Each time an ipc namespace is created (removed) it registers (unregisters) its notifier block in (from) the ipcns chain. The callback routine registered in the memory chain invokes the ipcns notifier chain with the IPCNS_LOWMEM event. Each callback routine registered in the ipcns namespace, in turn, recomputes msgmni for the owning namespace. Signed-off-by:
-
Nadia Derbey authored
This is a trivial patch that defines the priority of slab_memory_callback in the callback chain as a constant. This is to prepare for next patch in the series. Signed-off-by:
-
Nadia Derbey authored
Since all the namespaces see the same amount of memory (the total one) this patch introduces a new variable that counts the ipc namespaces and divides msg_ctlmni by this counter. Signed-off-by:
-
Nadia Derbey authored
On large systems we'd like to allow a larger number of message queues. In some cases up to 32K. However simply setting MSGMNI to a larger value may cause problems for smaller systems. The first patch of this series introduces a default maximum number of message queue ids that scales with the amount of lowmem. Since msgmni is per namespace and there is no amount of memory dedicated to each namespace so far, the second patch of this series scales msgmni to the number of ipc namespaces too. Since msgmni depends on the amount of memory, it becomes necessary to recompute it upon memory add/remove. In the 4th patch, memory hotplug management is added: a notifier block is registered into the memory hotplug notifier chain for the ipc subsystem. Since the ipc namespaces are not linked together, they have their own notification chain: one notifier_block is defined per ipc namespace. Each time an ipc namespace is created (removed) it registers (unregisters) its notifier block in (from) the ipcns chain. The callback routine registered in the memory chain invokes the ipcns notifier chain with the IPCNS_MEMCHANGE event. Each callback routine registered in the ipcns namespace, in turn, recomputes msgmni for the owning namespace. The 5th patch makes it possible to keep the memory hotplug notifier chain's lock for a lesser amount of time: instead of directly notifying the ipcns notifier chain upon memory add/remove, a work item is added to the global workqueue. When activated, this work item is the one who notifies the ipcns notifier chain. Since msgmni depends on the number of ipc namespaces, it becomes necessary to recompute it upon ipc namespace creation / removal. The 6th patch uses the ipc namespace notifier chain for that purpose: that chain is notified each time an ipc namespace is created or removed. This makes it possible to recompute msgmni for all the namespaces each time one of them is created or removed. When msgmni is explicitely set from userspace, we should avoid recomputing it upon memory add/remove or ipcns creation/removal. This is what the 7th patch does: it simply unregisters the ipcns callback routine as soon as msgmni has been changed from procfs or sysctl(). Even if msgmni is set by hand, it should be possible to make it back automatically recomputed upon memory add/remove or ipcns creation/removal. This what is achieved in patch 8: if set to a negative value, msgmni is added back to the ipcns notifier chain, making it automatically recomputed again. This patch: Compute msg_ctlmni to make it scale with the amount of lowmem. msg_ctlmni is now set to make the message queues occupy 1/32 of the available lowmem. Some cleaning has also been done for the MSGPOOL constant: the msgctl man page says it's not used, but it also defines it as a size in bytes (the code expresses it in Kbytes). Signed-off-by:
-
Pierre Peiffer authored
By continuing to consolidate a little the IPC code, each id can be built directly in ipc_addid() instead of having it built from each callers of ipc_addid() And I also remove shm_addid() in order to have, as much as possible, the same code for shm/sem/msg. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
-
Randy Dunlap authored
Fix kernel bugzilla #10388. DMA-API.txt has wrong argument type for some functions. It uses struct device but should use struct pci_dev. Signed-off-by:
Randy Dunlap <randy.dunlap@oracle.com> Acked-by:
James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by:
-
Arthur Kepner authored
Add a new parameter, dmasync, to the ib_umem_get() prototype. Use dmasync = 1 when mapping user-allocated CQs with ib_umem_get(). Signed-off-by:
Arthur Kepner <akepner@sgi.com> Cc: Tony Luck <tony.luck@intel.com> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> Cc: Jes Sorensen <jes@sgi.com> Cc: Randy Dunlap <randy.dunlap@oracle.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Cc: David Miller <davem@davemloft.net> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Grant Grundler <grundler@parisc-linux.org> Cc: Michael Ellerman <michael@ellerman.id.au> Signed-off-by:
-
Arthur Kepner authored
Change all ia64 machvecs to use the new dma_*map*_attrs() interfaces. Implement the old dma_*map_*() interfaces in terms of the corresponding new interfaces. For ia64/sn, make use of one dma attribute, DMA_ATTR_WRITE_BARRIER. Introduce swiotlb_*map*_attrs() functions. Signed-off-by:
-
Arthur Kepner authored
Document the new dma_*map*_attrs() functions. [markn@au1.ibm.com: fix up for dma-add-dma_map_attrs-interfaces and update docs] Signed-off-by:
David S. Miller <davem@davemloft.net> Cc: Tony Luck <tony.luck@intel.com> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> Cc: Jes Sorensen <jes@sgi.com> Cc: Randy Dunlap <randy.dunlap@oracle.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Grant Grundler <grundler@parisc-linux.org> Cc: Michael Ellerman <michael@ellerman.id.au> Signed-off-by:
Mark Nelson <markn@au1.ibm.com> Signed-off-by:
-
Arthur Kepner authored
Introduce new interfaces, dma_*map*_attrs(), for passing architecture-specific attributes when memory is mapped and unmapped for DMA. Give the interfaces default implementations which ignore attributes. Also introduce the dma_{set|get}_attr() interfaces for setting and retrieving individual attributes. Define one attribute, DMA_ATTR_WRITE_BARRIER, in anticipation of its use by ia64/sn. Select whether architectures implement arch-specific versions of the dma_*map*_attrs() interfaces via HAVE_DMA_ATTRS in Kconfig. [markn@au1.ibm.com: dma_{set,get}_attr() have to be static inline] Signed-off-by: -
Oleg Nesterov authored
cpu_hotplug_begin() must be always called under cpu_add_remove_lock, this means that only one process can be cpu_hotplug.active_writer. So we don't need the cpu_hotplug.writer_queue, we can wake up the ->active_writer directly. Also, fix the comment. Signed-off-by:
Oleg Nesterov <oleg@tv-sign.ru> Cc: Dipankar Sarma <dipankar@in.ibm.com> Acked-by:
Gautham R Shenoy <ego@in.ibm.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com> Signed-off-by:
-
Oleg Nesterov authored
cleanup_workqueue_thread() doesn't need the second argument, remove it. Signed-off-by:
-
Oleg Nesterov authored
When cpu_populated_map was introduced, it was supposed that cwq->thread can survive after CPU_DEAD, that is why we never shrink cpu_populated_map. This is not very nice, we can safely remove the already dead CPU from the map. The only required change is that destroy_workqueue() must hold the hotplug lock until it destroys all cwq->thread's, to protect the cpu_populated_map. We could make the local copy of cpu mask and drop the lock, but sizeof(cpumask_t) may be very large. Also, fix the comment near queue_work(). Unless _cpu_down() happens we do guarantee the cpu-affinity of the work_struct, and we have users which rely on this. [akpm@linux-foundation.org: repair comment] Signed-off-by:
-
Paul Menage authored
This flag provides the hardwalling properties of mem_exclusive, without enforcing the exclusivity. Either mem_hardwall or mem_exclusive is sufficient to prevent GFP_KERNEL allocations from passing outside the cpuset's assigned nodes. Signed-off-by:
Paul Menage <menage@google.com> Acked-by:
Paul Jackson <pj@sgi.com> Signed-off-by:
-
Paul Menage authored
Currently the cpusets mem_exclusive flag is overloaded to mean both "no-overlapping" and "no GFP_KERNEL allocations outside this cpuset". These patches add a new mem_hardwall flag with just the allocation restriction part of the mem_exclusive semantics, without breaking backwards-compatibility for those who continue to use just mem_exclusive. Additionally, the cgroup control file registration for cpusets is cleaned up to reduce boilerplate. This patch: This change tidies up the cpusets control file definitions, and reduces the amount of boilerplate required to add/change control files in the future. Signed-off-by:
Li Zefan <lizf@cn.fujitsu.com> Acked-by:
-
Adrian Bunk authored
Make the following needlessly global functions static: - cpuset_test_cpumask() - cpuset_change_cpumask() - cpuset_do_move_task() Signed-off-by:
Adrian Bunk <bunk@kernel.org> Acked-by:
-
Li Zefan authored
*mem has been zeroed, that means mem->info has already been filled with 0. Signed-off-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by:
Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
On ia64, this kmalloc() requires order-4 pages. But this is not necessary to be physically contiguous. For big mem_cgroup, vmalloc is better. For small ones, kmalloc is used. [akpm@linux-foundation.org: simplification] Signed-off-by:
-
Balbir Singh authored
This patch makes the memory controller more responsive on my desktop. 1. Set all cached pages as inactive. We were by default marking all pages as active, thus forcing us to go through two passes for reclaiming pages 2. Remove congestion_wait(), since we already have that logic in do_try_to_free_pages() Signed-off-by:
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp> Cc: Paul Menage <menage@google.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
remove_list/add_list uses page_cgroup_zoneinfo() in it. So, it's called twice before and after lock. mz = page_cgroup_zoneinfo(); lock(); mz = page_cgroup_zoneinfo(); .... unlock(); And address of mz never changes. This is not good. This patch fixes this behavior. Signed-off-by:
-
Pavel Emelyanov authored
This is a very common requirement from people using the resource accounting facilities (not only memcgroup but also OpenVZ beancounters). They want to put the cgroup in an initial state without re-creating it. For example after re-configuring a group people want to observe how this new configuration fits the group needs without saving the previous failcnt value. Merge two resets into one mem_cgroup_reset() function to demonstrate how multiplexing work. Besides, I have plans to move the files, that correspond to res_counter to the res_counter.c file and somehow "import" them into controller. I don't know how to make it gracefully yet, but merging resets of max_usage and failcnt in one function will be there for sure. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
Pavel Emelyanov <xemul@openvz.org> Acked-by:
-
Pavel Emelyanov authored
These two files are essentially event callbacks. They do not care about the contents of the string, but only about the fact of the write itself. Signed-off-by:
-
Balbir Singh authored
Move the memory controller data structure page_cgroup to its own slab cache. It saves space on the system, allocations are not necessarily pushed to order of 2 and should provide performance benefits. Users who disable the memory controller can also double check that the memory controller is not allocating page_cgroup's. NOTE: Hugh Dickins brought up the issue of whether we want to mark page_cgroup as __GFP_MOVABLE or __GFP_RECLAIMABLE. I don't think there is an easy answer at the moment. page_cgroup's are associated with user pages, they can be reclaimed once the user page has been reclaimed, so it might make sense to mark them as __GFP_RECLAIMABLE. For now, I am leaving the marking to default values that the slab allocator uses. Signed-off-by:
-
Pavel Emelyanov authored
The resource counter is supposed to facilitate the resource accounting of arbitrary resource (and it already does this for memory controller). However, it is about to be used in other resources controllers (swap, kernel memory, networking, etc), so provide a doc describing how to work with it. This will eliminate all the possible future duplications in the appropriate controllers' docs. Fixed errors pointed out by Randy. [akpm@linux-foundation.org: fix documentation tpyo] Signed-off-by:
-
Pavel Emelyanov authored
This field is the maximal value of the usage one since the counter creation (or since the latest reset). To reset this to the usage value simply write anything to the appropriate cgroup file. Signed-off-by:
-
Balbir Singh authored
Remove the mem_cgroup member from mm_struct and instead adds an owner. This approach was suggested by Paul Menage. The advantage of this approach is that, once the mm->owner is known, using the subsystem id, the cgroup can be determined. It also allows several control groups that are virtually grouped by mm_struct, to exist independent of the memory controller i.e., without adding mem_cgroup's for each controller, to mm_struct. A new config option CONFIG_MM_OWNER is added and the memory resource controller selects this config option. This patch also adds cgroup callbacks to notify subsystems when mm->owner changes. The mm_cgroup_changed callback is called with the task_lock() of the new task held and is called just prior to changing the mm->owner. I am indebted to Paul Menage for the several reviews of this patchset and helping me make it lighter and simpler. This patch was tested on a powerpc box, it was compiled with both the MM_OWNER config turned on and off. After the thread group leader exits, it's moved to init_css_state by cgroup_exit(), thus all future charges from runnings threads would be redirected to the init_css_set's subsystem. Signed-off-by:
Pekka Enberg <penberg@cs.helsinki.fi> Reviewed-by:
-
Serge E. Hallyn authored
Introduce a read_seq() helper in cftype, which uses seq_file to print out lists. Use it in the devices cgroup. Also split devices.allow into two files, so now devices.deny and devices.allow are the ones to use to manipulate the whitelist, while devices.list outputs the cgroup's current whitelist. Signed-off-by:
-
Li Zefan authored
Now we can run through the hash table instead of running through the linked-list. Signed-off-by:
-
Li Zefan authored
We are at system boot and there is only 1 cgroup group (i,e, init_css_set), so we don't need to run through the css_set linked list. Neither do we need to run through the task list, since no processes have been created yet. Also referring to a comment in cgroup.h: struct css_set { ... /* * Set of subsystem states, one for each subsystem. This array * is immutable after creation apart from the init_css_set * during subsystem registration (at boot time). */ struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; } Signed-off-by: -
Li Zefan authored
When we attach a process to a different cgroup, the css_set linked-list will be run through to find a suitable existing css_set to use. This patch implements a hash table for better performance. The following benchmarks have been tested: For N in 1, 5, 10, 50, 100, 500, 1000, create N cgroups with one sleeping task in each, and then move an additional task through each cgroup in turn. Here is a test result: N Loop orig - Time(s) hash - Time(s) ---------------------------------------------- 1 10000 1.201231728 1.196311177 5 2000 1.065743872 1.040566424 10 1000 0.991054735 0.986876440 50 200 0.976554203 0.969608733 100 100 0.998504680 0.969218270 500 20 1.157347764 0.962602963 1000 10 1.619521852 1.085140172 Signed-off-by:
-
