From patchwork Wed May 4 11:02:42 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Santosh Shilimkar X-Patchwork-Id: 753352 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by demeter1.kernel.org (8.14.4/8.14.3) with ESMTP id p44B3044001956 for ; Wed, 4 May 2011 11:03:01 GMT Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751689Ab1EDLCx (ORCPT ); Wed, 4 May 2011 07:02:53 -0400 Received: from na3sys009aog107.obsmtp.com ([74.125.149.197]:56478 "EHLO na3sys009aog107.obsmtp.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751120Ab1EDLCw (ORCPT ); Wed, 4 May 2011 07:02:52 -0400 Received: from mail-gy0-f179.google.com ([209.85.160.179]) (using TLSv1) by na3sys009aob107.postini.com ([74.125.148.12]) with SMTP ID DSNKTcEyW27h0JNS3r39ij6TEcYYTg+INToC@postini.com; Wed, 04 May 2011 04:02:52 PDT Received: by gyg8 with SMTP id 8so450497gyg.24 for ; Wed, 04 May 2011 04:02:51 -0700 (PDT) Received: by 10.236.193.72 with SMTP id j48mr1139124yhn.486.1304506971055; Wed, 04 May 2011 04:02:51 -0700 (PDT) Received: from [172.24.137.243] (dragon.ti.com [192.94.94.33]) by mx.google.com with ESMTPS id 45sm455493yhl.2.2011.05.04.04.02.45 (version=SSLv3 cipher=OTHER); Wed, 04 May 2011 04:02:48 -0700 (PDT) Message-ID: <4DC13252.2080308@ti.com> Date: Wed, 04 May 2011 16:32:42 +0530 From: Santosh Shilimkar User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.2.15) Gecko/20110303 Thunderbird/3.1.9 MIME-Version: 1.0 To: Catalin Marinas , Will Deacon CC: linux-arm-kernel@lists.infradead.org, linux-omap@vger.kernel.org Subject: Re: [PATCH V3] ARM: GIC: Convert GIC library to use the IO relaxed operations References: <1304058131-8375-1-git-send-email-santosh.shilimkar@ti.com> <1304417481.19196.60.camel@e102109-lin.cambridge.arm.com> <4DBFD59D.50008@ti.com> In-Reply-To: <4DBFD59D.50008@ti.com> Sender: linux-omap-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-omap@vger.kernel.org X-Greylist: IP, sender and recipient auto-whitelisted, not delayed by milter-greylist-4.2.6 (demeter1.kernel.org [140.211.167.41]); Wed, 04 May 2011 11:03:01 +0000 (UTC) On 5/3/2011 3:44 PM, Santosh Shilimkar wrote: > On 5/3/2011 3:41 PM, Catalin Marinas wrote: [...] >> >> Otherwise the patch looks fine (I'll add my ack after you fix the above). >> > Thanks. Will add above comment, drop the readl and repost with your ack. > > Same will push it the patch system > Below is the updated version. I didn't push this to patch system because it is generated on top of Will's GIC series. Will, Can you queue this patch part of your series please? Regards Santosh From 1506abc77b36eb10ae0e3f8711e6ad1b87ca363d Mon Sep 17 00:00:00 2001 From: Santosh Shilimkar Date: Mon, 28 Mar 2011 19:27:46 +0530 Subject: [PATCH] ARM: GIC: Convert GIC library to use the IO relaxed operations The GIC register accesses today make use of readl()/writel() which prove to be very expensive when used along with mandatory barriers. This mandatory barriers also introduces an un-necessary and expensive l2x0_sync() operation. On Cortex-A9 MP cores, GIC IO accesses from CPU are direct and doesn't go through L2X0 write buffer. A DSB before writel_relaxed() in gic_raise_softirq() is added to be compliant with the Barrier Litmus document - the mailbox scenario. Signed-off-by: Santosh Shilimkar Acked-by: Catalin Marinas Cc: Will Deacon --- arch/arm/common/gic.c | 54 +++++++++++++++++++++++++++--------------------- 1 files changed, 30 insertions(+), 24 deletions(-) gic_arch_extn.irq_mask(d); spin_unlock(&irq_controller_lock); @@ -102,7 +102,7 @@ static void gic_unmask_irq(struct irq_data *d) spin_lock(&irq_controller_lock); if (gic_arch_extn.irq_unmask) gic_arch_extn.irq_unmask(d); - writel(mask, gic_dist_base(d) + GIC_DIST_ENABLE_SET + (gic_irq(d) / 32) * 4); + writel_relaxed(mask, gic_dist_base(d) + GIC_DIST_ENABLE_SET + (gic_irq(d) / 32) * 4); spin_unlock(&irq_controller_lock); } @@ -114,7 +114,7 @@ static void gic_eoi_irq(struct irq_data *d) spin_unlock(&irq_controller_lock); } - writel(gic_irq(d), gic_cpu_base(d) + GIC_CPU_EOI); + writel_relaxed(gic_irq(d), gic_cpu_base(d) + GIC_CPU_EOI); } static int gic_set_type(struct irq_data *d, unsigned int type) @@ -140,7 +140,7 @@ static int gic_set_type(struct irq_data *d, unsigned int type) if (gic_arch_extn.irq_set_type) gic_arch_extn.irq_set_type(d, type); - val = readl(base + GIC_DIST_CONFIG + confoff); + val = readl_relaxed(base + GIC_DIST_CONFIG + confoff); if (type == IRQ_TYPE_LEVEL_HIGH) val &= ~confmask; else if (type == IRQ_TYPE_EDGE_RISING) @@ -150,15 +150,15 @@ static int gic_set_type(struct irq_data *d, unsigned int type) * As recommended by the spec, disable the interrupt before changing * the configuration */ - if (readl(base + GIC_DIST_ENABLE_SET + enableoff) & enablemask) { - writel(enablemask, base + GIC_DIST_ENABLE_CLEAR + enableoff); + if (readl_relaxed(base + GIC_DIST_ENABLE_SET + enableoff) & enablemask) { + writel_relaxed(enablemask, base + GIC_DIST_ENABLE_CLEAR + enableoff); enabled = true; } - writel(val, base + GIC_DIST_CONFIG + confoff); + writel_relaxed(val, base + GIC_DIST_CONFIG + confoff); if (enabled) - writel(enablemask, base + GIC_DIST_ENABLE_SET + enableoff); + writel_relaxed(enablemask, base + GIC_DIST_ENABLE_SET + enableoff); spin_unlock(&irq_controller_lock); @@ -190,8 +190,8 @@ static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val, spin_lock(&irq_controller_lock); d->node = cpu; - val = readl(reg) & ~mask; - writel(val | bit, reg); + val = readl_relaxed(reg) & ~mask; + writel_relaxed(val | bit, reg); spin_unlock(&irq_controller_lock); return 0; @@ -223,7 +223,7 @@ static void gic_handle_cascade_irq(unsigned int irq, struct irq_desc *desc) chained_irq_enter(chip, desc); spin_lock(&irq_controller_lock); - status = readl(chip_data->cpu_base + GIC_CPU_INTACK); + status = readl_relaxed(chip_data->cpu_base + GIC_CPU_INTACK); spin_unlock(&irq_controller_lock); gic_irq = (status & 0x3ff); @@ -272,13 +272,13 @@ static void __init gic_dist_init(struct gic_chip_data *gic, cpumask |= cpumask << 8; cpumask |= cpumask << 16; - writel(0, base + GIC_DIST_CTRL); + writel_relaxed(0, base + GIC_DIST_CTRL); /* * Find out how many interrupts are supported. * The GIC only supports up to 1020 interrupt sources. */ - gic_irqs = readl(base + GIC_DIST_CTR) & 0x1f; + gic_irqs = readl_relaxed(base + GIC_DIST_CTR) & 0x1f; gic_irqs = (gic_irqs + 1) * 32; if (gic_irqs > 1020) gic_irqs = 1020; @@ -287,26 +287,26 @@ static void __init gic_dist_init(struct gic_chip_data *gic, * Set all global interrupts to be level triggered, active low. */ for (i = 32; i < gic_irqs; i += 16) - writel(0, base + GIC_DIST_CONFIG + i * 4 / 16); + writel_relaxed(0, base + GIC_DIST_CONFIG + i * 4 / 16); /* * Set all global interrupts to this CPU only. */ for (i = 32; i < gic_irqs; i += 4) - writel(cpumask, base + GIC_DIST_TARGET + i * 4 / 4); + writel_relaxed(cpumask, base + GIC_DIST_TARGET + i * 4 / 4); /* * Set priority on all global interrupts. */ for (i = 32; i < gic_irqs; i += 4) - writel(0xa0a0a0a0, base + GIC_DIST_PRI + i * 4 / 4); + writel_relaxed(0xa0a0a0a0, base + GIC_DIST_PRI + i * 4 / 4); /* * Disable all interrupts. Leave the PPI and SGIs alone * as these enables are banked registers. */ for (i = 32; i < gic_irqs; i += 32) - writel(0xffffffff, base + GIC_DIST_ENABLE_CLEAR + i * 4 / 32); + writel_relaxed(0xffffffff, base + GIC_DIST_ENABLE_CLEAR + i * 4 / 32); /* * Limit number of interrupts registered to the platform maximum @@ -324,7 +324,7 @@ static void __init gic_dist_init(struct gic_chip_data *gic, set_irq_flags(i, IRQF_VALID | IRQF_PROBE); } - writel(1, base + GIC_DIST_CTRL); + writel_relaxed(1, base + GIC_DIST_CTRL); } static void __cpuinit gic_cpu_init(struct gic_chip_data *gic) @@ -337,17 +337,17 @@ static void __cpuinit gic_cpu_init(struct gic_chip_data *gic) * Deal with the banked PPI and SGI interrupts - disable all * PPI interrupts, ensure all SGI interrupts are enabled. */ - writel(0xffff0000, dist_base + GIC_DIST_ENABLE_CLEAR); - writel(0x0000ffff, dist_base + GIC_DIST_ENABLE_SET); + writel_relaxed(0xffff0000, dist_base + GIC_DIST_ENABLE_CLEAR); + writel_relaxed(0x0000ffff, dist_base + GIC_DIST_ENABLE_SET); /* * Set priority on PPI and SGI interrupts */ for (i = 0; i < 32; i += 4) - writel(0xa0a0a0a0, dist_base + GIC_DIST_PRI + i * 4 / 4); + writel_relaxed(0xa0a0a0a0, dist_base + GIC_DIST_PRI + i * 4 / 4); - writel(0xf0, base + GIC_CPU_PRIMASK); - writel(1, base + GIC_CPU_CTRL); + writel_relaxed(0xf0, base + GIC_CPU_PRIMASK); + writel_relaxed(1, base + GIC_CPU_CTRL); } void __init gic_init(unsigned int gic_nr, unsigned int irq_start, @@ -391,7 +391,13 @@ void gic_raise_softirq(const struct cpumask *mask, unsigned int irq) { unsigned long map = *cpus_addr(*mask); + /* + * Ensure that stores to Normal memory are visible to the + * other CPUs before issuing the IPI. + */ + dsb(); + /* this always happens on GIC0 */ - writel(map << 16 | irq, gic_data[0].dist_base + GIC_DIST_SOFTINT); + writel_relaxed(map << 16 | irq, gic_data[0].dist_base + GIC_DIST_SOFTINT); } #endif diff --git a/arch/arm/common/gic.c b/arch/arm/common/gic.c index e9c2ff8..4ddd0a6 100644 --- a/arch/arm/common/gic.c +++ b/arch/arm/common/gic.c @@ -89,7 +89,7 @@ static void gic_mask_irq(struct irq_data *d) u32 mask = 1 << (d->irq % 32); spin_lock(&irq_controller_lock); - writel(mask, gic_dist_base(d) + GIC_DIST_ENABLE_CLEAR + (gic_irq(d) / 32) * 4); + writel_relaxed(mask, gic_dist_base(d) + GIC_DIST_ENABLE_CLEAR + (gic_irq(d) / 32) * 4); if (gic_arch_extn.irq_mask)