- 02 Oct, 2006 40 commits
-
-
Greg Banks authored
Actually implement multiple pools. On NUMA machines, allocate a svc_pool per NUMA node; on SMP a svc_pool per CPU; otherwise a single global pool. Enqueue sockets on the svc_pool corresponding to the CPU on which the socket bh is run (i.e. the NIC interrupt CPU). Threads have their cpu mask set to limit them to the CPUs in the svc_pool that owns them. This is the patch that allows an Altix to scale NFS traffic linearly beyond 4 CPUs and 4 NICs. Incorporates changes and feedback from Neil Brown, Trond Myklebust, and Christoph Hellwig. Signed-off-by:
Greg Banks <gnb@melbourne.sgi.com> Signed-off-by:
Neil Brown <neilb@suse.de> Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Greg Banks authored
Replace the existing list of all nfsd threads with new code using svc_create_pooled(). Signed-off-by:
-
Greg Banks authored
Currently knfsd keeps its own list of all nfsd threads in nfssvc.c; add a new way of managing the list of all threads in a svc_serv. Add svc_create_pooled() to allow creation of a svc_serv whose threads are managed by the sunrpc code. Add svc_set_num_threads() to manage the number of threads in a service, either per-pool or globally across the service. Signed-off-by:
-
Greg Banks authored
add svc_get() for those occasions when we need to temporarily bump up svc_serv->sv_nrthreads as a pseudo refcount. Signed-off-by:
-
Greg Banks authored
Split out the list of idle threads and pending sockets from svc_serv into a new svc_pool structure, and allocate a fixed number (in this patch, 1) of pools per svc_serv. The new structure contains a lock which takes over several of the duties of svc_serv->sv_lock, which is now relegated to protecting only sv_tempsocks, sv_permsocks, and sv_tmpcnt in svc_serv. The point is to move the hottest fields out of svc_serv and into svc_pool, allowing a following patch to arrange for a svc_pool per NUMA node or per CPU. This is a major step towards making the NFS server NUMA-friendly. Signed-off-by:
-
Greg Banks authored
The SK_BUSY bit in svc_sock->sk_flags ensures that we do not attempt to enqueue a socket twice. Currently, setting and clearing the bit is protected by svc_serv->sv_lock. As I intend to reduce the data that the lock protects so it's not held when svc_sock_enqueue() tests and sets SK_BUSY, that test and set needs to be atomic. Signed-off-by:
-
Greg Banks authored
Convert the svc_sock->sk_reserved variable from an int protected by svc_serv->sv_lock, to an atomic. This reduces (by 1) the number of places we need to take the (effectively global) svc_serv->sv_lock. Signed-off-by:
-
Greg Banks authored
Protect the svc_sock->sk_deferred list with a new lock svc_sock->sk_defer_lock instead of svc_serv->sv_lock. Using the more fine-grained lock reduces the number of places we need to take the svc_serv lock. Signed-off-by:
-
Greg Banks authored
Convert the svc_sock->sk_inuse counter from an int protected by svc_serv->sv_lock, to an atomic. This reduces the number of places we need to take the (effectively global) svc_serv->sv_lock. Signed-off-by:
-
Greg Banks authored
Following are 11 patches from Greg Banks which combine to make knfsd more Numa-aware. They reduce hitting on 'global' data structures, and create some data-structures that can be node-local. knfsd threads are bound to a particular node, and the thread to handle a new request is chosen from the threads that are attach to the node that received the interrupt. The distribution of threads across nodes can be controlled by a new file in the 'nfsd' filesystem, though the default approach of an even spread is probably fine for most sites. Some (old) numbers that show the efficacy of these patches: N == number of NICs == number of CPUs == nmber of clients. Number of NUMA nodes == N/2 N Throughput, MiB/s CPU usage, % (max=N*100) Before After Before After --- ------ ---- ----- ----- 4 312 435 350 228 6 500 656 501 418 8 562 804 690 589 This patch: Move the aging of RPC/TCP connection sockets from the main svc_recv() loop to a timer which uses a mark-and-sweep algorithm every 6 minutes. This reduces the amount of work that needs to be done in the main RPC loop and the length of time we need to hold the (effectively global) svc_serv->sv_lock. [akpm@osdl.org: cleanup] Signed-off-by:
-
NeilBrown authored
If lockd_up fails - what should we expect? Do we have to later call lockd_down? Well the nfs client thinks "no", the nfs server thinks "yes". lockd thinks "yes". The only answer that really makes sense is "no" !! So: Make lockd_up only increment nlmsvc_users on success. Make nfsd handle errors from lockd_up properly. Make sure lockd_up(0) never fails when lockd is running so that the 'reclaimer' call to lockd_up doesn't need to be error checked. Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: -
NeilBrown authored
Thus it is printed for any path that leads to failure (make_socks is called from two places). Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by:
-
NeilBrown authored
We should be checking the return value of lockd_up when adding a new socket to nfsd. So move the lockd_up before the svc_addsock and check the return value. The move is because lockd_down is easy, but there is no easy way to remove a recently added socket. Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by:
-
NeilBrown authored
It isn't needed as it is available in rqstp->rq_server, and dropping it allows some local vars to be dropped. [akpm@osdl.org: build fix] Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by:
-
Josh Triplett authored
e_start acquires svc_export_cache.hash_lock, and e_stop releases it. Add lock annotations to these two functions so that sparse can check callers for lock pairing, and so that sparse will not complain about these functions since they intentionally use locks in this manner. Signed-off-by:
Josh Triplett <josh@freedesktop.org> Cc: Neil Brown <neilb@cse.unsw.edu.au> Signed-off-by:
-
Greg Banks authored
Signed-off-by:
-
NeilBrown authored
Userspace should create and bind a socket (but not connectted) and write the 'fd' to portlist. This will cause the nfs server to listen on that socket. To close a socket, the name of the socket - as read from 'portlist' can be written to 'portlist' with a preceding '-'. Signed-off-by:
-
NeilBrown authored
This file will list all ports that nfsd has open. Default when TCP enabled will be ipv4 udp 0.0.0.0 2049 ipv4 tcp 0.0.0.0 2049 Later, the list of ports will be settable. 'portlist' chosen rather than 'ports', to avoid unnecessary confusion with non-mainline patches which created 'ports' with different semantics. [akpm@osdl.org: cleanups, build fix] Signed-off-by:
-
NeilBrown authored
Separate out the code for creating a new service, and for creating initial sockets. Some of these new functions will have multiple callers soon. [akpm@osdl.org: cleanups] Signed-off-by:
-
NeilBrown authored
We have an array 'nfsd_version' which lists the available versions of nfsd, and 'nfsd_versions' (poor choice there :-() which lists the currently active versions. Then we have a bitmap - nfsd_versbits which says which versions are wanted. The bits in this bitset cause content to be copied from nfsd_version to nfsd_versions when nfsd starts. This patch removes nfsd_versbits and moves information directly from nfsd_version to nfsd_versions when requests for version changes arrive. Note that this doesn't make it possible to change versions while the server is running. This is because serv->sv_xdrsize is calculated when a service is created, and used when threads are created, and xdrsize depends on the active versions. Signed-off-by:
-
NeilBrown authored
Currently lockd listens on UDP always, and TCP if CONFIG_NFSD_TCP is set. However as lockd performs services of the client as well, this is a problem. If CONFIG_NfSD_TCP is not set, and a tcp mount is used, the server will not be able to call back to lockd. So: - add an option to lockd_up saying which protocol is needed - Always open sockets for which an explicit port was given, otherwise only open a socket of the type required - Change nfsd to do one lockd_up per socket rather than one per thread. This - removes the dependancy on CONFIG_NFSD_TCP - means that lockd may open sockets other than at startup - means that lockd will *not* listen on UDP if the only mounts are TCP mount (and nfsd hasn't started). The latter is the only one that concerns me at all - I don't know if this might be a problem with some servers. Signed-off-by:
-
NeilBrown authored
nfsd has some cleanup that it wants to do when the last thread exits, and there will shortly be some more. So collect this all into one place and define a callback for an rpc service to call when the service is about to be destroyed. [akpm@osdl.org: cleanups, build fix] Signed-off-by:
-
Greg Banks authored
Signed-off-by:
-
Greg Banks authored
Signed-off-by:
-
Greg Banks authored
Signed-off-by:
-
Greg Banks authored
cpumask: ensure that node_to_cpumask() is available to modules for all supported combinations of architecture and CONFIG_NUMA. Signed-off-by:
-
Greg Banks authored
cpumask: ensure that the cpu_online_map and cpu_possible_map bitmasks, and hence all the macros in <linux/cpumask.h> that require them, are available to modules for all supported combinations of architecture and CONFIG_SMP. Signed-off-by:
-
Greg Banks authored
cpumask: add highest_possible_node_id(), analogous to highest_possible_processor_id(). [pj@sgi.com: fix typo] Signed-off-by:
Paul Jackson <pj@sgi.com> Signed-off-by:
-
Andrew Morton authored
As reported in http://bugzilla.kernel.org/show_bug.cgi?id=6970, ISDN can issue excessively-long udelays, which triggers a build-time error on ARM. This is very sucky of ISDN, but I doubt if anyone is going to suddenly fix it. So change the macro to do the microsecond counting itself. Cc: <tch@wpkg.org> Cc: Karsten Keil <kkeil@suse.de> Signed-off-by:
-
Tilman Schmidt authored
This patch to the Siemens Gigaset driver fixes the compile warning "ignoring return value of 'class_device_create_file', declared with attribute warn_unused_result" appearing with CONFIG_ENABLE_MUST_CHECK=y in release 2.6.18-rc1-mm1. Signed-off-by:
Tilman Schmidt <tilman@imap.cc> Acked-by:
Hansjoerg Lipp <hjlipp@web.de> Signed-off-by:
-
Ankita Garg authored
A simple module to test Linux Kernel Dump mechanism. This module uses jprobes to install/activate pre-defined crash points. At different crash points, various types of crashing scenarios are created like a BUG(), panic(), exception, recursive loop and stack overflow. The user can activate a crash point with specific type by providing parameters at the time of module insertion. Please see the file header for usage information. The module is based on the Linux Kernel Dump Test Tool by Fernando <http://lkdtt.sourceforge.net>. This module could be merged with mainline. Jprobes is used here so that the context in which crash point is hit, could be maintained. This implements all the crash points as done by LKDTT except the one in the middle of tasklet_action(). Signed-off-by:
Ankita Garg <ankita@in.ibm.com> Signed-off-by:
-
bibo,mao authored
kprobe_flush_task() possibly calls kfree function during holding kretprobe_lock spinlock, if kfree function is probed by kretprobe that will incur spinlock deadlock. This patch moves kfree function out scope of kretprobe_lock. Signed-off-by:
bibo, mao <bibo.mao@intel.com> Signed-off-by:
Ananth N Mavinakayanahalli <ananth@in.ibm.com> Signed-off-by:
-
bibo,mao authored
When kprobe is re-entered, the re-entered kprobe kernel path will will call atomic_notifier_call_chain function, if this function is kprobed that will incur numerous kprobe recursive fault. This patch disallows kprobes on atomic_notifier_call_chain function. Signed-off-by:
-
bibo,mao authored
Whitespace is used to indent, this patch cleans up these sentences by kernel coding style. Signed-off-by:
-
Ananth N Mavinakayanahalli authored
Documentation/kprobes.txt updated to reflect: o In-kernel symbol resolution o CONFIG_KALLSYMS dependency o Usage of JPROBE_ENTRY o Addition of regs_return_value() Also update the references list and usage examples to use correct module interfaces. Signed-off-by:
Jim Keniston <jkenisto@us.ibm.com> Signed-off-by:
-
Ananth N Mavinakayanahalli authored
Add the regs_return_value() macro to extract the return value in an architecture agnostic manner, given the pt_regs. Other architecture maintainers may want to add similar helpers. Signed-off-by:
Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> Signed-off-by:
Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by:
-
Ananth N Mavinakayanahalli authored
kallsyms_lookup_name() allows for <module:symbol> style specification for looking up symbol addresses. Handle the case where the user specifies <module:.symbol> on powerpc, given that 64-bit powerpc uses function descriptors. Signed-off-by:
-
Ananth N Mavinakayanahalli authored
In an effort to make kprobe modules more portable, here is a patch that: o Introduces the "symbol_name" field to struct kprobe. The symbol->address resolution now happens in the kernel in an architecture agnostic manner. 64-bit powerpc users no longer have to specify the ".symbols" o Introduces the "offset" field to struct kprobe to allow a user to specify an offset into a symbol. o The legacy mechanism of specifying the kprobe.addr is still supported. However, if both the kprobe.addr and kprobe.symbol_name are specified, probe registration fails with an -EINVAL. o The symbol resolution code uses kallsyms_lookup_name(). So CONFIG_KPROBES now depends on CONFIG_KALLSYMS o Apparantly kprobe modules were the only legitimate out-of-tree user of the kallsyms_lookup_name() EXPORT. Now that the symbol resolution happens in-kernel, remove the EXPORT as suggested by Christoph Hellwig o Modify tcp_probe.c that uses the kprobe interface so as to make it work on multiple platforms (in its earlier form, the code wouldn't work, say, on powerpc) Signed-off-by:
Prasanna S Panchamukhi <prasanna@in.ibm.com> Signed-off-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
-
Cedric Le Goater authored
Replaces the pid_t value with a struct pid to avoid pid wrap around problems. Signed-off-by:
Cedric Le Goater <clg@fr.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Acked-by:
Eric W. Biederman <ebiederm@xmission.com> Signed-off-by:
-
Eric W. Biederman authored
The problem with remembering a user space process by its pid is that it is possible that the process will exit, pid wrap around will occur. Converting to a struct pid avoid that problem, and paves the way for implementing a pid namespace. Also since usb is the only user of kill_proc_info_as_uid rename kill_proc_info_as_uid to kill_pid_info_as_uid and have the new version take a struct pid. Signed-off-by:
Greg Kroah-Hartman <gregkh@suse.de> Signed-off-by:
-
