Re: 2.6.0-test9-mm3 - AIO test results

[Daniel McNeil <daniel@osdl.org>]

Suparna,

Good news and bad news.  Your patch does fix the non-power of two i/o
size problems where AIO previously did not complete:

$ ./aiodio_sparse  -s 1751k -r 18k -w 11k
$ aiodio_sparse -i 9 -dd -s 180k -r 18k -w 18k  
io_submit() return 9
aiodio_sparse: 9 i/o in flight
aiodio_sparse: offset 165888 filesize 184320 inflight 9
aiodio_sparse: io_getevent() returned 1
aiodio_sparse: io_getevent() res 18432 res2 0
io_submit() return 1
AIO DIO write done unlinking file
dio_sparse done writing, kill children
aiodio_sparse 0 children had errors

But when testing using aiocp using O_DIRECT to copy a file to
an already allocated file, the aiocp process hangs.  I used i/o
size of 4k and that compeleted.  Using i/o size of 1k and 2k,
the aiocp process hung during io_sumbit() and are unkillable.
Here are the stack traces:

# ps -fu daniel | grep aiocp
daniel    1920     1  0 16:45 ?        00:00:07 aiocp -b 1k -n 1 -f DIRECT glibc-2.3.2.tar ff2
daniel    2083  2037  0 17:00 pts/2    00:00:03 aiocp -dd -b 1k -n 8 -f DIRECT glibc-2.3.2.tar ff2


aiocp         D 00000001  1920      1                1902 (NOTLB)
e70abd04 00200086 c18dbc80 00000001 00000003 c02897fc 00000060 00200246
       f7cdb8b4 c16522f0 c18dbc80 0000309c 640a05eb 0000008b e6d9e660
c0289a16
       f7cdb8b4 e87e95cc c18dbc80 00000000 00000001 e70abd10 c0123712
e70aa000
Call Trace:
 [<c02897fc>] generic_unplug_device+0x50/0xbd
 [<c0289a16>] blk_run_queues+0xa9/0x15c
 [<c0123712>] io_schedule+0x26/0x30
 [<c0192242>] direct_io_worker+0x376/0x5ab
 [<c014840f>] generic_file_direct_IO+0x70/0x89
 [<c019264a>] __blockdev_direct_IO+0x1d3/0x2d5
 [<c01ac73e>] ext3_direct_io_get_blocks+0x0/0xbf
 [<c01ad72d>] ext3_direct_IO+0xc0/0x1e1
 [<c01ac73e>] ext3_direct_io_get_blocks+0x0/0xbf
 [<c014840f>] generic_file_direct_IO+0x70/0x89
 [<c0145e11>] __generic_file_aio_read+0xfb/0x1ff
 [<c0121b70>] schedule+0x3ac/0x7ef
 [<c0145f48>] generic_file_aio_read+0x33/0x37
 [<c0194ad3>] aio_pread+0x34/0x5f
 [<c0193bec>] aio_run_iocb+0xa6/0x1ed
 [<c019316f>] __aio_get_req+0x27/0x158
 [<c0194a9f>] aio_pread+0x0/0x5f
 [<c0194f62>] io_submit_one+0x1ea/0x2b7
 [<c0195110>] sys_io_submit+0xe1/0x194
 [<c03c29a7>] syscall_call+0x7/0xb
 [<c03c007b>] rpc_depopulate+0x1aa/0x24b


aiocp         D 366EDC94  2083   2037                     (NOTLB)
e758bd04 00200082 f71ba000 366edc94 00000161 c02897fc 00000060 366edc94
       00000161 f71ba000 c18d3c80 000069a9 366f5a0e 00000161 e8d4acc0 c0289a16
       f7cdb8b4 e960465c c18d3c80 00000000 00000001 e758bd10 c0123712 e758a000
Call Trace:
 [<c02897fc>] generic_unplug_device+0x50/0xbd
 [<c0289a16>] blk_run_queues+0xa9/0x15c
 [<c0123712>] io_schedule+0x26/0x30
 [<c0192242>] direct_io_worker+0x376/0x5ab
 [<c019264a>] __blockdev_direct_IO+0x1d3/0x2d5
 [<c01ac73e>] ext3_direct_io_get_blocks+0x0/0xbf
 [<c01ad72d>] ext3_direct_IO+0xc0/0x1e1
 [<c01ac73e>] ext3_direct_io_get_blocks+0x0/0xbf
 [<c014840f>] generic_file_direct_IO+0x70/0x89
 [<c0145e11>] __generic_file_aio_read+0xfb/0x1ff
 [<c0259d3e>] write_chan+0x165/0x21e
 [<c0145f48>] generic_file_aio_read+0x33/0x37
 [<c0194ad3>] aio_pread+0x34/0x5f
 [<c0193bec>] aio_run_iocb+0xa6/0x1ed
 [<c019316f>] __aio_get_req+0x27/0x158
 [<c0194a9f>] aio_pread+0x0/0x5f
 [<c02532ab>] tty_write+0x1e8/0x3b2
 [<c0194f62>] io_submit_one+0x1ea/0x2b7
 [<c0195110>] sys_io_submit+0xe1/0x194
 [<c03c29a7>] syscall_call+0x7/0xb
 [<c03c007b>] rpc_depopulate+0x1aa/0x24b



Daniel

On Sun, 2003-11-16 at 21:25, Suparna Bhattacharya wrote:
> On Thu, Nov 13, 2003 at 02:03:58PM -0800, Daniel McNeil wrote:
> > Andrew,
> > 
> > I'm testing test9-mm3 on a 2-proc Xeon with a ext3 file system.
> > I tested using the test programs aiocp and aiodio_sparse.
> > (see http://developer.osdl.org/daniel/AIO/)
> > 
> > Using aiocp with i/o sizes from 1k to 512k to copy files worked
> > without any errors or kernel debug messages.
> > 
> > With 64k i/o, the aiodio_sparse program complete without any errors.
> > There are no kernel error messages, so that is good.
> > 
> > There are still problems with non power of 2 i/o sizes using AIO and
> > O_DIRECT.  It hangs with aio's that do not seem to complete.  The test
> > does exit when hitting ^c and there are no kernel messages.  Test output
> > below:
> 
> Could you check if the following patch fixes the problem for you ?
> 
> Regards
> Suparna
> 
> --------------------------------------------------------------
> 
> With this patch, when the DIO code falls back to buffered i/o after
> having submitted part of the i/o, then buffered i/o is issued only
> for the remaining part of the request (i.e. the part not already 
> covered by DIO).
> 
> diff -ur pure-mm3/fs/direct-io.c linux-2.6.0-test9-mm3/fs/direct-io.c
> --- pure-mm3/fs/direct-io.c	2003-11-14 09:09:06.000000000 +0530
> +++ linux-2.6.0-test9-mm3/fs/direct-io.c	2003-11-17 09:00:47.000000000 +0530
> @@ -74,6 +74,7 @@
>  					   been performed at the start of a
>  					   write */
>  	int pages_in_io;		/* approximate total IO pages */
> +	size_t	size;			/* total request size (doesn't change)*/
>  	sector_t block_in_file;		/* Current offset into the underlying
>  					   file in dio_block units. */
>  	unsigned blocks_available;	/* At block_in_file.  changes */
> @@ -226,7 +227,7 @@
>  			dio_complete(dio, dio->block_in_file << dio->blkbits,
>  					dio->result);
>  			/* Complete AIO later if falling back to buffered i/o */
> -			if (dio->result != -ENOTBLK) {
> +			if (dio->result >= dio->size || dio->rw == READ) {
>  				aio_complete(dio->iocb, dio->result, 0);
>  				kfree(dio);
>  			} else {
> @@ -889,6 +890,7 @@
>  	dio->blkbits = blkbits;
>  	dio->blkfactor = inode->i_blkbits - blkbits;
>  	dio->start_zero_done = 0;
> +	dio->size = 0;
>  	dio->block_in_file = offset >> blkbits;
>  	dio->blocks_available = 0;
>  	dio->cur_page = NULL;
> @@ -925,7 +927,7 @@
>  
>  	for (seg = 0; seg < nr_segs; seg++) {
>  		user_addr = (unsigned long)iov[seg].iov_base;
> -		bytes = iov[seg].iov_len;
> +		dio->size += bytes = iov[seg].iov_len;
>  
>  		/* Index into the first page of the first block */
>  		dio->first_block_in_page = (user_addr & ~PAGE_MASK) >> blkbits;
> @@ -956,6 +958,13 @@
>  		}
>  	} /* end iovec loop */
>  
> +	if (ret == -ENOTBLK && rw == WRITE) {
> +		/*
> +		 * The remaining part of the request will be 
> +		 * be handled by buffered I/O when we return
> +		 */
> +		ret = 0;
> +	}
>  	/*
>  	 * There may be some unwritten disk at the end of a part-written
>  	 * fs-block-sized block.  Go zero that now.
> @@ -986,19 +995,13 @@
>  	 */
>  	if (dio->is_async) {
>  		if (ret == 0)
> -			ret = dio->result;	/* Bytes written */
> -		if (ret == -ENOTBLK) {
> -			/*
> -			 * The request will be reissued via buffered I/O
> -			 * when we return; Any I/O already issued
> -			 * effectively becomes redundant.
> -			 */
> -			dio->result = ret;
> +			ret = dio->result;
> +		if (ret > 0 && dio->result < dio->size && rw == WRITE) {
>  			dio->waiter = current;
>  		}
>  		finished_one_bio(dio);		/* This can free the dio */
>  		blk_run_queues();
> -		if (ret == -ENOTBLK) {
> +		if (dio->waiter) {
>  			/*
>  			 * Wait for already issued I/O to drain out and
>  			 * release its references to user-space pages
> @@ -1032,7 +1035,8 @@
>  		}
>  		dio_complete(dio, offset, ret);
>  		/* We could have also come here on an AIO file extend */
> -		if (!is_sync_kiocb(iocb) && (ret != -ENOTBLK))
> +		if (!is_sync_kiocb(iocb) && !(rw == WRITE && ret >= 0 && 
> +			dio->result < dio->size))
>  			aio_complete(iocb, ret, 0);
>  		kfree(dio);
>  	}
> diff -ur pure-mm3/mm/filemap.c linux-2.6.0-test9-mm3/mm/filemap.c
> --- pure-mm3/mm/filemap.c	2003-11-14 09:15:08.000000000 +0530
> +++ linux-2.6.0-test9-mm3/mm/filemap.c	2003-11-15 11:11:16.000000000 +0530
> @@ -1895,14 +1895,16 @@
>  		 */
>  		if (written >= 0 && file->f_flags & O_SYNC)
>  			status = generic_osync_inode(inode, mapping, OSYNC_METADATA);
> -		if (written >= 0 && !is_sync_kiocb(iocb))
> +		if (written >= count && !is_sync_kiocb(iocb))
>  			written = -EIOCBQUEUED;
> -		if (written != -ENOTBLK)
> +		if (written < 0 || written >= count)
>  			goto out_status;
>  		/*
>  		 * direct-io write to a hole: fall through to buffered I/O
> +		 * for completing the rest of the request.
>  		 */
> -		written = 0;
> +		pos += written;
> +		count -= written;
>  	}
>  
>  	buf = iov->iov_base;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>