Oliver, my understanding is the limit_sem will only cause write() to sleep and
wait for other IOs to finish when there are alreay WRITES_IN_FLIGHT URBs are on
going. I see code line 509: sema_init(&dev->limit_sem,
WRITES_IN_FLIGHT);
My understanding of Blocking IO should block every write() when corresponding
URBs is not finished, am I right?
----------------------------------------
> From: oneu...@suse.de
> To: unicorn_w...@outlook.com
> CC: linux-usb@vger.kernel.org
> Subject: Re: question on skel_read func of usb_skeleton.c
> Date: Fri, 12 Jul 2013 10:30:28 +0200
>
> On Friday 12 July 2013 04:49:54 WangChen wrote:
>> Hi, Oliver,
>> Regarding skel_write, I see your current desgin only refuse its execution
>> when> WRITES_IN_FLIGHT are on the fly, but this is not blocking IO due to
>> write() will not wait before callback returns, right? Do you think it's
>> unnecessary to support blocking IO on write or any special reason else?
>
>
> /*
> * limit the number of URBs in flight to stop a user from using up all
> * RAM
> */
> if (!(file->f_flags & O_NONBLOCK)) {
> if (down_interruptible(&dev->limit_sem)) {
>
> The limit_sem will cause write() to wait for IO to finish,
> if the limit is reached
>
> retval = -ERESTARTSYS;
> goto exit;
> }
> } else {
> if (down_trylock(&dev->limit_sem)) {
> retval = -EAGAIN;
>
> Unless non-blocking IO is used. In that case we return.
>
> goto exit;
> }
> }
>
> Blocking IO is perfectly well supported. To keep under the limit you don't
> have to wait for the IO you started. Any IO will serve the purpose.
>
> Regards
> Oliver
>
> N‹§²وىr¸›yْèڑطb²X¬¶ا§vط^–)ق؛{.nا+‰·¥ٹ{±؛ئâط^n‡r،ِ¦zثپëh™¨èع&¢ّ ®G«éh®(éڑژٹف¢j"ْ¶m§ےïپêنz¹ق–ٹàf£¢·hڑˆ§~ˆmڑ
