user/sven/linux.git/include/linux/file.h, branch v2.6.16.45

[PATCH] fix file counting

2006-03-08T22:14:01Z

I have benchmarked this on an x86_64 NUMA system and see no significant performance difference on kernbench. Tested on both x86_64 and powerpc. The way we do file struct accounting is not very suitable for batched freeing. For scalability reasons, file accounting was constructor/destructor based. This meant that nr_files was decremented only when the object was removed from the slab cache. This is susceptible to slab fragmentation. With RCU based file structure, consequent batched freeing and a test program like Serge's, we just speed this up and end up with a very fragmented slab - llm22:~ # cat /proc/sys/fs/file-nr 587730 0 758844 At the same time, I see only a 2000+ objects in filp cache. The following patch I fixes this problem. This patch changes the file counting by removing the filp_count_lock. Instead we use a separate percpu counter, nr_files, for now and all accesses to it are through get_nr_files() api. In the sysctl handler for nr_files, we populate files_stat.nr_files before returning to user. Counting files as an when they are created and destroyed (as opposed to inside slab) allows us to correctly count open files with RCU. Signed-off-by: Dipankar Sarma Cc: "Paul E. McKenney" Cc: "David S. Miller" Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

[PATCH] reorder struct files_struct

2005-11-14T02:14:10Z

The file_lock spinlock sits close to mostly read fields of 'struct files_struct' In SMP (and NUMA) environments, each time a thread wants to open or close a file, it has to acquire the spinlock, thus invalidating the cache line containing this spinlock on other CPUS. So other threads doing read()/write()/... calls that use RCU to access the file table are going to ask further memory (possibly NUMA) transactions to read again this memory line. Move the spinlock to another cache line, so that concurrent threads can share the cache line containing 'count' and 'fdt' fields. It's worth up to 9% on a microbenchmark using a 4-thread 2-package x86 machine. See http://marc.theaimsgroup.com/?l=linux-kernel&m=112680448713342&w=2 Signed-off-by: Eric Dumazet Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

[PATCH] mm: rename kmem_cache_s to kmem_cache

2005-11-07T15:53:24Z

This patch renames struct kmem_cache_s to kmem_cache so we can start using it instead of kmem_cache_t typedef. Signed-off-by: Pekka Enberg Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

[PATCH] files: files struct with RCU

2005-09-09T20:57:55Z

Patch to eliminate struct files_struct.file_lock spinlock on the reader side and use rcu refcounting rcuref_xxx api for the f_count refcounter. The updates to the fdtable are done by allocating a new fdtable structure and setting files->fdt to point to the new structure. The fdtable structure is protected by RCU thereby allowing lock-free lookup. For fd arrays/sets that are vmalloced, we use keventd to free them since RCU callbacks can't sleep. A global list of fdtable to be freed is not scalable, so we use a per-cpu list. If keventd is already handling the current cpu's work, we use a timer to defer queueing of that work. Since the last publication, this patch has been re-written to avoid using explicit memory barriers and use rcu_assign_pointer(), rcu_dereference() premitives instead. This required that the fd information is kept in a separate structure (fdtable) and updated atomically. Signed-off-by: Dipankar Sarma Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

[PATCH] files: break up files struct

2005-09-09T20:57:55Z

In order for the RCU to work, the file table array, sets and their sizes must be updated atomically. Instead of ensuring this through too many memory barriers, we put the arrays and their sizes in a separate structure. This patch takes the first step of putting the file table elements in a separate structure fdtable that is embedded withing files_struct. It also changes all the users to refer to the file table using files_fdtable() macro. Subsequent applciation of RCU becomes easier after this. Signed-off-by: Dipankar Sarma Signed-Off-By: David Howells Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

[PATCH] file_table:expand_files() code cleanup

2005-01-21T00:03:48Z

expand_files() cleanup: Make expand_files() common code for fork.c:fork/copy_files(), open.c:open/get_unused_fd() and fcntl.c:dup/locate_fd(). expand_files() does both expand_fd_array and expand_fd_set based on the need. This is used in dup(). open() and fork() duplicates the work of expand files. At all places we check for expanding fd array, we also check for expanding fdset. There is no need of checking and calling them seperately. This change also moves the expand_files to file.c from fcntl.c, and makes the expand_fd_array and expand_fd_set local to that file. Signed-off-by: Prasanna Meda Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

[PATCH] get_files_struct cleanup

2004-04-13T02:22:25Z

From: Russell King Cleanup the 4 duplicate "get_files_struct" implementations into one get_files_struct() function to compliment put_files_struct().

[PATCH] reduced overheads in fget/fput

2003-05-07T15:19:50Z

From: Dipankar Sarma fget() shows up on profiles, especially on SMP. Dipankar's patch special-cases the situation wherein there are no sharers of current->files. In this situation we know that no other process can close this file, so it is not necessary to increment the file's refcount. It's ugly as sin, but makes a substantial difference. The test is dd if=/dev/zero of=foo bs=1 count=1M On 4CPU P3 xeon with 1MB L2 cache and 512MB ram: kernel sys time std-dev ------------ -------- ------- UP - vanilla 2.104 0.028 UP - file 1.867 0.019 SMP - vanilla 2.976 0.023 SMP - file 2.719 0.026

[PATCH] convert file_lock to a spinlock

2003-04-12T19:56:37Z

Time to write a 2M file, one byte at a time: Before: 1.09s user 4.92s system 99% cpu 6.014 total 0.74s user 5.28s system 99% cpu 6.023 total 1.03s user 4.97s system 100% cpu 5.991 total After: 0.79s user 5.17s system 99% cpu 5.993 total 0.79s user 5.17s system 100% cpu 5.957 total 0.84s user 5.11s system 100% cpu 5.942 total

[PATCH] file_list_lock contention fixes

2003-03-18T05:15:03Z

Patch from Manfred Spraul Fixes the lock contention over file_list_lock by simply removing the unneeded anon_list. So filp allocation does not need to take a global lock any more. The use of a spinlock to protect files_stat.nr_files is a bit awkward, but the alternative is a custom sysctl handler, and isn't much more efficient anyway.