From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753510AbbIRKnG (ORCPT ); Fri, 18 Sep 2015 06:43:06 -0400 Received: from foss.arm.com ([217.140.101.70]:45677 "EHLO foss.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750853AbbIRKnD (ORCPT ); Fri, 18 Sep 2015 06:43:03 -0400 Date: Fri, 18 Sep 2015 11:42:56 +0100 From: Marc Zyngier To: Oleksij Rempel Cc: , , , Oleksij Rempel Subject: Re: Possible Spam [PATCH v2 2/2] ARM: irqchip: mxs: add Alpascale ASM9260 support Message-ID: <20150918114256.74ab823a@arm.com> In-Reply-To: <1442567922-14538-3-git-send-email-linux@rempel-privat.de> References: <1442567922-14538-1-git-send-email-linux@rempel-privat.de> <1442567922-14538-3-git-send-email-linux@rempel-privat.de> Organization: ARM Ltd X-Mailer: Claws Mail 3.11.1 (GTK+ 2.24.25; arm-unknown-linux-gnueabihf) MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, 18 Sep 2015 11:18:42 +0200 Oleksij Rempel wrote: > From: Oleksij Rempel > > Freescale iMX23/iMX28 and Alphascale ASM9260 have similar Is it Alphascale or Alpascale? You may need to fix the patch title. > interrupt collectors. It makes easy to reuse irq-mxs code for ASM9260. > Differences between this devices are fallowing: > - different register offsets > - different count of intterupt lines per register > - ASM9260 don't provide reset bit > - ASM9260 don't support FIQ. > > Signed-off-by: Oleksij Rempel > --- > drivers/irqchip/Kconfig | 5 ++ > drivers/irqchip/Makefile | 2 +- > drivers/irqchip/alphascale_asm9260-icoll.h | 109 +++++++++++++++++++++++++++++ > drivers/irqchip/irq-mxs.c | 106 +++++++++++++++++++++++++++- > 4 files changed, 220 insertions(+), 2 deletions(-) > create mode 100644 drivers/irqchip/alphascale_asm9260-icoll.h > [...] > diff --git a/drivers/irqchip/irq-mxs.c b/drivers/irqchip/irq-mxs.c > index 14374de..1470087 100644 > --- a/drivers/irqchip/irq-mxs.c > +++ b/drivers/irqchip/irq-mxs.c > @@ -1,5 +1,7 @@ > /* > * Copyright (C) 2009-2010 Freescale Semiconductor, Inc. All Rights Reserved. > + * Copyright (C) 2014 Oleksij Rempel > + * Add Alphascale ASM9260 support. > * > * This program is free software; you can redistribute it and/or modify > * it under the terms of the GNU General Public License as published by > @@ -28,6 +30,8 @@ > #include > #include > > +#include "alphascale_asm9260-icoll.h" > + > /* > * this device provide 4 offsets for each register: > * 0x0 - plain read write mode > @@ -49,6 +53,11 @@ > > #define ICOLL_NUM_IRQS 128 > > +enum icoll_type { > + ICOLL, > + ASM9260_ICOLL, > +}; > + > struct icoll_priv { > void __iomem *vector; > void __iomem *levelack; > @@ -58,10 +67,38 @@ struct icoll_priv { > /* number of interrupts per register */ > int ; > void __iomem *clear; > + enum icoll_type type; > }; > > static struct icoll_priv icoll_priv; > static struct irq_domain *icoll_domain; > +static DEFINE_RAW_SPINLOCK(icoll_lock); > + > +/* calculate bit offset depending on number of intterupt per register */ > +static u32 icoll_intr_bitshift(struct irq_data *d, u32 bit) > +{ > + /* > + * We expect intr_per_reg to be 4 or 1, it means > + * "n" will be 3 or 0. > + */ > + int n = icoll_priv.intr_per_reg - 1; > + > + /* > + * If n = 0, "bit" is never shifted. > + * If n = 3, mask lower part of hwirq to convert it > + * in 0, 1, 2 or 3 and then multiply it by 8 (or shift by 3) > + */ > + return bit << ((d->hwirq & n) << n); > +} > + > +/* calculate mem offset depending on number of intterupt per register */ > +static void __iomem *icoll_intr_reg(struct irq_data *d) > +{ > + int n = icoll_priv.intr_per_reg >> 1; > + > + /* offset = hwirq / intr_per_reg * 0x10 */ > + return icoll_priv.intr + ((d->hwirq >> n) * 0x10); > +} Please correct me if I'm wrong, but it looks like these function are only useful when used on ams9260. So why do we need intr_per_reg at all? MXS doesn't need it (always 1), and ams9260 always need it (always 4). Save yourself some previous cycles and simplify the whole thing. > > static void icoll_ack_irq(struct irq_data *d) > { > @@ -86,12 +123,38 @@ static void icoll_unmask_irq(struct irq_data *d) > icoll_priv.intr + SET_REG + HW_ICOLL_INTERRUPTn(d->hwirq)); > } > > +static void asm9260_mask_irq(struct irq_data *d) > +{ > + raw_spin_lock(&icoll_lock); > + __raw_writel(icoll_intr_bitshift(d, BM_ICOLL_INTR_ENABLE), > + icoll_intr_reg(d) + CLR_REG); > + raw_spin_unlock(&icoll_lock); > +} > + > +static void asm9260_unmask_irq(struct irq_data *d) > +{ > + raw_spin_lock(&icoll_lock); > + __raw_writel(ASM9260_BM_CLEAR_BIT(d->hwirq), > + icoll_priv.clear + > + ASM9260_HW_ICOLL_CLEARn(d->hwirq)); > + > + __raw_writel(icoll_intr_bitshift(d, BM_ICOLL_INTR_ENABLE), > + icoll_intr_reg(d) + SET_REG); > + raw_spin_unlock(&icoll_lock); > +} Can you please explain the rational for this lock? mask/unmask use different registers, and it is not obvious to me what race you are trying to avoid here. Thanks, M. -- Jazz is not dead. It just smells funny.