From: Tomi Valkeinen Subject: Re: [RFC PATCH] OMAP: DSS2: RFBI driver: Cleanup "kfifo is full" test Date: Thu, 07 Jan 2010 14:06:19 +0200 Message-ID: <1262865979.2127.34.camel@tubuntu.research.nokia.com> References: <20091214031816.GA1829@zakalwe.fi> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from smtp.nokia.com ([192.100.122.233]:64475 "EHLO mgw-mx06.nokia.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752452Ab0AGMFd (ORCPT ); Thu, 7 Jan 2010 07:05:33 -0500 In-Reply-To: <20091214031816.GA1829@zakalwe.fi> Sender: linux-omap-owner@vger.kernel.org List-Id: linux-omap@vger.kernel.org To: ext Heikki Orsila Cc: "linux-omap@vger.kernel.org" , "linux-fbdev@vger.kernel.org" On Mon, 2009-12-14 at 04:18 +0100, ext Heikki Orsila wrote: > RFBI_CMD_FIFO_LEN_BYTES might not be a power-of-two, > but kfifo() allocates the size up to the next power of two, > which makes (RFBI_CMD_FIFO_LEN_BYTES - kfifo_len()) > look very dubious (possibly a negative value!). But this can never happen, as the code never pushes more than RFBI_CMD_FIFO_LEN_BYTES bytes into the fifo. > > Notes: > > [1] Not tested nor compiled here > > [2] The removed code compares int with a size_t. > It looks buggy, because if the full allocated > size of the fifo is used, the integer comparison > leads to "always empty space in the fifo". > Fortunately, the code can never go there, because > available == 0 will come before negative integer > values. > > [3] Do we want to use the whole allocated space? I think if we allocate space for n bytes, we should use n bytes of the area, even though the underlying mechanism would allocate more. > In that case the change should be: > > - available = RFBI_CMD_FIFO_LEN_BYTES - > - __kfifo_len(rfbi.cmd_fifo); > + available = rfbi.cmd_fifo->size - __kfifo_len(rfbi.cmd_fifo); > > Signed-off-by: Heikki Orsila > --- > drivers/video/omap2/dss/rfbi.c | 5 +---- > 1 files changed, 1 insertions(+), 4 deletions(-) > > diff --git a/drivers/video/omap2/dss/rfbi.c b/drivers/video/omap2/dss/rfbi.c > index d0b3006..1dbe993 100644 > --- a/drivers/video/omap2/dss/rfbi.c > +++ b/drivers/video/omap2/dss/rfbi.c > @@ -1050,14 +1050,11 @@ static void rfbi_push_cmd(struct update_param *p) > > while (1) { > unsigned long flags; > - int available; > > spin_lock_irqsave(rfbi.cmd_fifo->lock, flags); > - available = RFBI_CMD_FIFO_LEN_BYTES - > - __kfifo_len(rfbi.cmd_fifo); > > /* DSSDBG("%d bytes left in fifo\n", available); */ > - if (available < sizeof(struct update_param)) { > + if (__kfifo_len(rfbi.cmd_fifo) >= RFBI_CMD_FIFO_LEN_BYTES) { > DSSDBG("Going to wait because FIFO FULL..\n"); > spin_unlock_irqrestore(rfbi.cmd_fifo->lock, flags); > atomic_inc(&rfbi.cmd_fifo_full); Hmm, I don't think this is correct. The point is to test if there's enough space for one struct update_param in the fifo. You are testing if the fifo is full. Tomi