linux-kernel - Re: [PATCH 5/5] usb_debug: EXPERIMENTAL

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 06 May 2009 10:59:32 -0500
From:	Jason Wessel <jason.wessel@...driver.com>
To:	Greg KH <greg@...ah.com>
CC:	Alan Stern <stern@...land.harvard.edu>, linux-usb@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH 5/5] usb_debug: EXPERIMENTAL - poll hcd device to force
 writes

Greg KH wrote:
>>> This patch takes the route of forcibly polling the hcd device to drain
>>> the urb queue and initiate the bulk write call backs.
>>>
>>> NOTE this patch is not signed off, it is a question of what is the
>>> right way to do this?
>>>       
>> This patch is highly questionable.
>>     
>
> I agree.
>
>   

That is why there is no signed-off on this patch.  It is an
RFC/EXPERIMENTAL because it is not clear how to fix the problem.  See
below for more discussion.

>> Is the idea to force the HCD to search for completed URBs before the
>> host controller has issued a completion interrupt?  Wouldn't it be
>> better instead to use more URBs?
>>
>> Besides, why should there be too many outstanding transmit URBs?  A
>> normal serial debugging port running at 115200 baud can transmit no
>> more than 12 bytes per ms.  You should be able to surpass that using
>> only three URBs!  In fact, if you buffer up to 8 bytes per URB then
>> with only two URBs you could send 64 bytes per ms, which is equivalent
>> to 640000 baud.  Do you really need to go more than (say) ten times
>> faster than that?
>>     
>
> Yeah, something seems wrong here.
>
> Jason, why is this really needed?  With your ring buffer, you shouldn't
> need this at all anymore.
>
> Or if you do, just bump up the number outstanding urbs that are
> possible.  Or the urb buffer size.
>
>   

The URB queue has to be unacceptably large to make it work correctly
(>4000)   The issue here stems from the register_console() function.  
As a part of the registration with register_console, the usb_debug
device driver gets a huge flood of printk backlog.  Because the urb
queue is not that deep we loose valuable printk logs.  I would very much
like a way to force it to work correctly.

To answer Alan's question, the rs232 uart drivers don't process the
printk back log asynchronously.  It is a direct write to the hardware
(see serial8250_console_putchar - drivers/serial/8250.c).   That is why
there is no problem with lost printk data in the case of the uart drivers.

I could find no obvious safe way do to this synchronously with the USB
api, which is why I put together some way to "force it to work" with the
poll hook until we can figure out the right way to do this.  The
mechanism for a console write is not the same as the standard tty
service, or at least that is the way it is today in the kernel.

Essentially I am seeking a synchronous write and wait for the usb serial
drivers when used as a system console.  Right now in the usb serial API
there is no distinction between a console write and a tty write. 
Perhaps that is what needs to change, to allow for some synchronous
behavior in usb_console_write()?

The usb_debug driver is not the only usb serial device that loses data
from the printk backlog, the ftdi_sio driver suffers the same problem.

Jason.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/