I have a smatch script (smatch.sf.net) that finds the the size of local variables. I created an allyesconfig with 2.5.48 and tested it. These were the functions that declared local datas with size of 5 digits or more (in bits). Some of them should probably be changed to declare a pointer then use kmalloc() and free().
if (foo) {
int frob[HUGE_NUM];
...
} else {
int bar[HUGE_NUM];
...
}
This would not count against you twice, since the script takes code paths into consideration.
The function with the largest variable is riocontrol(). It is used deliberately for some weird hardware. According to the comment, "it's hardware like this that really gets on [the author's] tits."
regards,
dan carpenter
size file lineno function
16768 drivers/atm/iphase.c 2874 ia_ioctl
11104 drivers/block/cpqarray.c 1090 ida_ioctl
21248 drivers/cdrom/optcd.c 1623 cdromread
524736 drivers/char/rio/rioctrl.c 1784 riocontrol
15968 drivers/ide/ide.c 1893 ide_unregister
16448 drivers/isdn/i4l/isdn_common.c 1271 isdn_statstr
12320 drivers/media/video/meye.c 400 jpeg_quantisation_tables
16552 drivers/media/video/w9966.c 911 w9966_v4l_read
16584 drivers/message/i2o/i2o_proc.c 1094 i2o_proc_read_groups
16640 drivers/message/i2o/i2o_proc.c 2295 i2o_proc_read_lan_mcast_addr
16640 drivers/message/i2o/i2o_proc.c 2525 i2o_proc_read_lan_alt_addr
23072 drivers/message/i2o/i2o_proc.c 889 i2o_proc_read_ddm_table
12224 drivers/mtd/chips/amd_flash.c 751 amd_flash_probe
35744 drivers/net/e100/e100_main.c 2812 e100_load_microcode
16496 drivers/net/wireless/airo.c 5949 readrids
16640 drivers/net/wireless/airo.c 6041 writerids
32896 drivers/scsi/53c700.c 1752 NCR_700_proc_directory_info
11744 drivers/scsi/aic7xxx_old.c 11156 aic7xxx_print_card
16824 drivers/scsi/cpqfcTScontrol.c 1630 CpqTsProcessIMQEntry
16496 drivers/scsi/cpqfcTScontrol.c 648 PeekIMQEntry
24800 drivers/scsi/qlogicfc.c 937 isp2x00_make_portdb
25808 drivers/usb/media/ibmcam.c 3480 ibmcam_model3_setup_after_video_if
16280 drivers/usb/media/ibmcam.c 638 ibmcam_parse_lines
16160 drivers/usb/media/ibmcam.c 945 ibmcam_model3_parse_lines
11104 fs/cifs/smbdes.c 270 dohash
33056 fs/intermezzo/journal.c 1281 presto_copy_kml_tail
33344 fs/intermezzo/journal.c 1633 presto_get_fileid
16512 fs/nfsd/nfsctl.c 214 write_export
16512 fs/nfsd/nfsctl.c 224 write_unexport
10336 fs/xfs/xfs_da_btree.c 599 xfs_da_node_rebalance
27104 fs/xfs/xfs_dir_leaf.c 763 xfs_dir_leaf_to_node
12480 fs/xfs/xfs_fsops.c 381 xfs_growfs_data_private
11552 lib/inflate.c 488 huft_build
10560 lib/inflate.c 906 inflate_dynamic
10336 net/wanrouter/wanmain.c 776 device_new_if
11360 sound/core/oss/pcm_oss.c 453 snd_pcm_oss_change_params
16864 sound/pci/emu10k1/emufx.c 2283 snd_emu10k1_fx8010_ioctl
The script used to generate the data:
#!/usr/bin/perl -w
use smatch;
my $max = 0;
sub merge_rules($$){
my $one = shift;
my $two = shift;
if (int($one) > int($two)){
return $one;
}else{
return $two;
}
}
while ($data = get_data()){
if ($data =~ /func_impl/){
$max = 0;
}
if ($data =~ /end_func/){
print get_filename(), " ", get_lineno(), " ", get_cur_func(), " $max\n";
}
if ($data =~ /decl_size\(integer_cst\((\d+)\)\)/){
set_state("total", get_state("total")+$1);
if ($max < get_state("total")){
$max = get_state("total");
}
}
}
--
_______________________________________________
Sign-up for your own FREE Personalized E-mail at Mail.com
http://www.mail.com/?sr=signup
>>>>> "dan" == dan carpenter <[email protected]> writes:
dan> I have a smatch script (smatch.sf.net) that finds the the
dan> size of local variables. I created an allyesconfig with
dan> 2.5.48 and tested it. These were the functions that declared
dan> local datas with size of 5 digits or more (in bits).
This is a minor complaint, but... why do you report the data sizes in
bits? I find myself forced to mentally divide every size by 8. Every
variable (obviously) has a size that's a whole number of bytes. Why
not just report bytes?
Best,
Roland
On 21 Nov 2002, Roland Dreier wrote:
| >>>>> "dan" == dan carpenter <[email protected]> writes:
|
| dan> I have a smatch script (smatch.sf.net) that finds the the
| dan> size of local variables. I created an allyesconfig with
| dan> 2.5.48 and tested it. These were the functions that declared
| dan> local datas with size of 5 digits or more (in bits).
|
| This is a minor complaint, but... why do you report the data sizes in
| bits? I find myself forced to mentally divide every size by 8. Every
| variable (obviously) has a size that's a whole number of bytes. Why
| not just report bytes?
I second that.
It's not just Roland...
But thanks for doing it anyway.
--
~Randy
dan carpenter writes:
> The function with the largest variable is riocontrol(). It is used
> deliberately for some weird hardware. According to the comment,
> "it's hardware like this that really gets on [the author's] tits."
> 524736 drivers/char/rio/rioctrl.c 1784 riocontrol
That variable is static.
/* It's hardware like this that really gets on my tits. */
static unsigned char copy[sizeof(struct DpRam)];
From: SL Baur <[email protected]>
> dan carpenter writes:
> > The function with the largest variable is riocontrol(). It is used
> > deliberately for some weird hardware. According to the comment,
> > "it's hardware like this that really gets on [the author's] tits."
>
> > 524736 drivers/char/rio/rioctrl.c 1784 riocontrol
>
> That variable is static.
>
Ah... I did not know that matterred.
Probably a lot of the other variables on the list are as well.
It will take me a couple days to create a new list that takes
static variables into consideration.
regards,
dan carpenter
--
_______________________________________________
Sign-up for your own FREE Personalized E-mail at Mail.com
http://www.mail.com/?sr=signup
One click access to the Top Search Engines
http://www.exactsearchbar.com/mailcom