From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from e8.ny.us.ibm.com (e8.ny.us.ibm.com [32.97.182.138]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client CN "e8.ny.us.ibm.com", Issuer "Equifax" (verified OK)) by ozlabs.org (Postfix) with ESMTPS id 73556B7D7B for ; Wed, 19 May 2010 07:53:03 +1000 (EST) Received: from d01relay05.pok.ibm.com (d01relay05.pok.ibm.com [9.56.227.237]) by e8.ny.us.ibm.com (8.14.3/8.13.1) with ESMTP id o4ILgNTj027676 for ; Tue, 18 May 2010 17:42:23 -0400 Received: from d01av02.pok.ibm.com (d01av02.pok.ibm.com [9.56.224.216]) by d01relay05.pok.ibm.com (8.13.8/8.13.8/NCO v10.0) with ESMTP id o4ILqsVB160924 for ; Tue, 18 May 2010 17:52:54 -0400 Received: from d01av02.pok.ibm.com (loopback [127.0.0.1]) by d01av02.pok.ibm.com (8.14.3/8.13.1/NCO v10.0 AVout) with ESMTP id o4ILqr6H018651 for ; Tue, 18 May 2010 18:52:54 -0300 Message-ID: <4BF30C32.1020403@linux.vnet.ibm.com> Date: Tue, 18 May 2010 16:52:50 -0500 From: Brian King MIME-Version: 1.0 To: dvhltc@linux.vnet.ibm.com Subject: Re: [PATCH RT] ehea: make receive irq handler non-threaded (IRQF_NODELAY) References: <4BF30793.5070300@us.ibm.com> In-Reply-To: <4BF30793.5070300@us.ibm.com> Content-Type: text/plain; charset=ISO-8859-1 Cc: Jan-Bernd Themann , linux-kernel@vger.kernel.org, Will Schmidt , niv@linux.vnet.ibm.com, Thomas Gleixner , Doug Maxey , linuxppc-dev@lists.ozlabs.org, Michael Ellerman List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Is IRQF_NODELAY something specific to the RT kernel? I don't see it in mainline... -Brian On 05/18/2010 04:33 PM, dvhltc@linux.vnet.ibm.com wrote: >>>From ad81664794e33d785f533c5edee37aaba20dd92d Mon Sep 17 00:00:00 2001 > From: Darren Hart > Date: Tue, 18 May 2010 11:07:13 -0700 > Subject: [PATCH RT] ehea: make receive irq handler non-threaded (IRQF_NODELAY) > > The underlying hardware is edge triggered but presented by XICS as level > triggered. The edge triggered interrupts are not reissued after masking. This > is not a problem in mainline which does not mask the interrupt (relying on the > EOI mechanism instead). The threaded interrupts in PREEMPT_RT do mask the > interrupt, and can lose interrupts that occurred while masked, resulting in a > hung ethernet interface. > > The receive handler simply calls napi_schedule(), as such, there is no > significant additional overhead in making this non-threaded, since we either > wakeup the threaded irq handler to call napi_schedule(), or just call > napi_schedule() directly to wakeup the softirqs. As the receive handler is > lockless, there is no need to convert any of the ehea spinlock_t's to > atomic_spinlock_t's. > > Without this patch, a simple scp file copy loop would fail quickly (usually > seconds). We have over two hours of sustained scp activity with the patch > applied. > > Credit goes to Will Schmidt for lots of instrumentation and tracing which > clarified the scenario and to Thomas Gleixner for the incredibly simple > solution. > > Signed-off-by: Darren Hart > Acked-by: Will Schmidt > Cc: Thomas Gleixner > Cc: Jan-Bernd Themann > Cc: Nivedita Singhvi > Cc: Brian King > Cc: Michael Ellerman > Cc: Doug Maxey > --- > drivers/net/ehea/ehea_main.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/drivers/net/ehea/ehea_main.c b/drivers/net/ehea/ehea_main.c > index 977c3d3..2c53df2 100644 > --- a/drivers/net/ehea/ehea_main.c > +++ b/drivers/net/ehea/ehea_main.c > @@ -1263,7 +1263,7 @@ static int ehea_reg_interrupts(struct net_device *dev) > "%s-queue%d", dev->name, i); > ret = ibmebus_request_irq(pr->eq->attr.ist1, > ehea_recv_irq_handler, > - IRQF_DISABLED, pr->int_send_name, > + IRQF_DISABLED | IRQF_NODELAY, pr->int_send_name, > pr); > if (ret) { > ehea_error("failed registering irq for ehea_queue " -- Brian King Linux on Power Virtualization IBM Linux Technology Center