Re: [PATCH] reorder struct files_struct

[Eric Dumazet <dada1@cosmosbay.com>]

Dipankar Sarma a écrit :
> On Wed, Sep 14, 2005 at 11:17:42PM +0200, Eric Dumazet wrote:
> 
>>In SMP (and NUMA) environnements, 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.
>>
>>Please consider applying this patch. It moves the spinlock to another cache 
>>line, so that concurrent threads can share the cache line containing 
>>'count' and 'fdt' fields.
>>
>>--- linux-2.6.14-rc1/include/linux/file.h	2005-09-13 05:12:09.000000000 +0200
>>+++ linux-2.6.14-rc1-ed/include/linux/file.h	2005-09-15 01:09:13.000000000 +0200
>>@@ -34,12 +34,12 @@
>>  */
>> struct files_struct {
>>         atomic_t count;
>>-        spinlock_t file_lock;     /* Protects all the below members.  Nests inside tsk->alloc_lock */
>> 	struct fdtable *fdt;
>> 	struct fdtable fdtab;
>>         fd_set close_on_exec_init;
>>         fd_set open_fds_init;
>>         struct file * fd_array[NR_OPEN_DEFAULT];
>>+	spinlock_t file_lock;     /* Protects concurrent writers.  Nests inside tsk->alloc_lock */
>> };
>> 
>> #define files_fdtable(files) (rcu_dereference((files)->fdt))
> 
> 
> For most apps without too many open fds, the embedded fd_sets
> are going to be used. Wouldn't that mean that open()/close() will
> invalidate the cache line containing fdt, fdtab by updating
> the fd_sets ? If so, you optimization really doesn't help.
> 

If the embedded struct fdtable is used, then the only touched field is 
'next_fd', so we could also move this field at the end of 'struct fdtable'

But I wonder if 'next_fd' really has to be in 'struct fdtable', maybe it could 
be moved to 'struct files_struct' close to file_lock ?

If yes, the whole embedded struct fdtable is readonly.

Eric