lkml.org 
[lkml]   [2012]   [Mar]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH] tty: hold lock across tty buffer finding and buffer filling
Date


> On 03/16/2012 02:50 PM, Du, Alek wrote:
>> For #1, as I said, we already held the tty refcount. The tty refcount cannot help this crash.
>
> Where exactly? Do you have some local changes?
> $ grep tty_port_tty_ drivers/tty/serial/mfd.c
> $

Yes, we do have local changes not upstream yet. But you should know now, TTY recount is not the key of the spin lock hole.


>
>> For #2, this patch will avoid memcpy to a NULL pointer and avoid a kernel crash.
>
> Ok, could you explain how? Linearized code from moxa.c with your patch
> applied follows:
> ofs = baseAddr + DynPage_addr + bufhead + head;
> size = (tail >= head) ? (tail - head) : (rx_mask + 1 - head);
> size = min(size, count);
> // call to tty_prepare_flip_string
> spin_lock_irqsave(&tty->buf.lock, flags);
> space = __tty_buffer_request_room(tty, size);
>
> tb = tty->buf.tail;
> if (likely(space)) {
> *chars = tb->char_buf_ptr + tb->used;
> memset(tb->flag_buf_ptr + tb->used, TTY_NORMAL, space);
> tb->used += space;
> }
> spin_unlock_irqrestore(&tty->buf.lock, flags);
> // return to MoxaPortReadData
> memcpy_fromio(*chars, ofs, space); <- memcpy without buf.lock

Ok. This is true. Seems we need another patch to remove this prepare interface. And let callers use the protected interface.

Thanks,
Alek


> head = (head + len) & rx_mask;
> ...
>
> thanks,
> --
> js
> suse labs


\
 
 \ /
  Last update: 2012-03-17 02:09    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site