From mboxrd@z Thu Jan 1 00:00:00 1970 From: Peter Meerwald Subject: Re: OMAP3 ISP capabilities, resizer Date: Tue, 18 Feb 2014 09:08:41 +0100 (CET) Message-ID: References: <1820490.0WrQgPlyuR@avalon> Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Return-path: Received: from ns.pmeerw.net ([87.118.82.44]:36305 "EHLO pmeerw.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753340AbaBRIIm (ORCPT ); Tue, 18 Feb 2014 03:08:42 -0500 In-Reply-To: <1820490.0WrQgPlyuR@avalon> Sender: linux-omap-owner@vger.kernel.org List-Id: linux-omap@vger.kernel.org To: Laurent Pinchart Cc: linux-media@vger.kernel.org, linux-omap@vger.kernel.org, sakari.ailus@iki.fi Hello, > > * I have a test program, http://pmeerw.net/scaler.c, which exercises the > > OMAP3 ISP resizer standalone with the pipeline given below; it crashes the > > system quite reliably on 3.7 and recent kernels :( > > > > the reason for the crash is that the ISP resizer can still be busy and > > doesn't like to be turned off then; a little sleep before > > omap3isp_sbl_disable() helps (or waiting for the ISP resize to become > > idle, see the patch below) > > > > I'm not sure what omap3isp_module_sync_idle() is supposed to do, it > > immediately returns since we are in SINGLESHOT mode and > > isp_pipeline_ready() is false > > The function is supposed to wait until the module becomes idle. I'm not sure > why we don't call it for memory-to-memory pipelines, I need to investigate > that. Sakari, do you remember what we've done there ? > > > with below patch I am happily resizing... > > > > // snip, RFC > > From: Peter Meerwald > > Date: Wed, 12 Feb 2014 15:59:20 +0100 > > Subject: [PATCH] omap3isp: Wait for resizer to become idle before > > disabling > > > > --- > > drivers/media/platform/omap3isp/ispresizer.c | 10 ++++++++++ > > 1 file changed, 10 insertions(+) > > > > diff --git a/drivers/media/platform/omap3isp/ispresizer.c > > b/drivers/media/platform/omap3isp/ispresizer.c > > index d11fb26..fea98f7 100644 > > --- a/drivers/media/platform/omap3isp/ispresizer.c > > +++ b/drivers/media/platform/omap3isp/ispresizer.c > > @@ -1145,6 +1145,7 @@ static int resizer_set_stream(struct v4l2_subdev *sd, > > int enable) struct isp_video *video_out = &res->video_out; > > struct isp_device *isp = to_isp_device(res); > > struct device *dev = to_device(res); > > + unsigned long timeout = 0; > > > > if (res->state == ISP_PIPELINE_STREAM_STOPPED) { > > if (enable == ISP_PIPELINE_STREAM_STOPPED) > > @@ -1176,6 +1177,15 @@ static int resizer_set_stream(struct v4l2_subdev *sd, > > int enable) if (omap3isp_module_sync_idle(&sd->entity, &res->wait, > > &res->stopping)) > > dev_dbg(dev, "%s: module stop timeout.\n", > > sd->name); > > + > > + while (omap3isp_resizer_busy(res)) { > > + if (timeout++ > 1000) { > > + dev_alert(isp->dev, "ISP resizer does not > > become idle\n"); > > + return -ETIMEDOUT; > > + } > > + udelay(100); > > + } > > + > > Let's try to avoid busy loops if possible. Does it help if you comment out the > condition at the top of omap3isp_module_sync_idle() ? > > > omap3isp_sbl_disable(isp, OMAP3_ISP_SBL_RESIZER_READ | > > OMAP3_ISP_SBL_RESIZER_WRITE); > > omap3isp_subclk_disable(isp, OMAP3_ISP_SUBCLK_RESIZER); tried without the check at the top of omap3isp_module_sync_idle() I am not sure yet when and how many interrupts occur in memory-to-memory (singleshot) mode... and the code might be racy: if the resizer is done, we don't want to wait for another interrupt which might never happen; if the resizer is not done (how do we know?), we need to test that and set the stopping flag so that omap3isp_module_sync_is_stopping() wakes us when done omap3isp_module_sync_idle(): if (pipe->stream_state == ISP_PIPELINE_STREAM_STOPPED || (0 && pipe->stream_state == ISP_PIPELINE_STREAM_SINGLESHOT && !isp_pipeline_ready(pipe))) return 0; isp_pipeline_ready() seems to be the wrong check for resizer done; the final interrupt could occur here, before setting 'stopping' /* * atomic_set() doesn't include memory barrier on ARM platform for SMP * scenario. We'll call it here to avoid race conditions. */ atomic_set(stopping, 1); regards, p. -- Peter Meerwald +43-664-2444418 (mobile)