From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tony Lindgren Subject: Re: GPIO regression in Linux next caused by syscon change Date: Mon, 15 Feb 2016 10:06:36 -0800 Message-ID: <20160215180635.GW19432@atomide.com> References: <20160213002643.GQ19432@atomide.com> <20160213004935.GR19432@atomide.com> <20160215160115.GU19432@atomide.com> <1455552961.4899.37.camel@pengutronix.de> <20160215164757.GV19432@atomide.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Content-Disposition: inline In-Reply-To: <20160215164757.GV19432@atomide.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+linux-arm-kernel=m.gmane.org@lists.infradead.org To: Philipp Zabel Cc: Arnd Bergmann , Tero Kristo , Philipp Zabel , linux-omap@vger.kernel.org, Lee Jones , linux-arm-kernel List-Id: linux-omap@vger.kernel.org * Tony Lindgren [160215 08:49]: > * Philipp Zabel [160215 08:17]: > > Am Montag, den 15.02.2016, 08:01 -0800 schrieb Tony Lindgren: > > > * Philipp Zabel [160214 11:24]: > > > > I've just replaced the of_iomap() call with an open coded version, > > > > calling of_address_to_resource() and ioremap() directly. That was > > > > needed so I can use the struct resource returned by > > > > of_address_to_resource() to set the syscon_config.max_register. I > > > > don't see where this could cause resource overlap. Does just setting > > > > syscon_config.max_register to zero again make the problem disappear? > > > > > > Yes commenting out the syscon_config.max_register line in your patch > > > makes things work again. > > > > > > So what does that tell us about the problem? > > > > Maybe some out of bounds writes that previously worked are now catched > > by the max_register check in regmap_writable and regmap_write returns > > -EIO instead of the write being executed. > > Hmm weird that something like that would not produce errors? > > > Is there any omap_ctrl_write?() call with an offset > 0x32c into > > scm_conf? > > Indeed, that's where things go wrong. Adding Tero to Cc, something > is wrong there. Something like this might fix it? Needs to be tested to see if it happens on other omaps. Tero, got any better ideas? Regards, Tony 8< ------------------- --- a/arch/arm/mach-omap2/control.c +++ b/arch/arm/mach-omap2/control.c @@ -167,15 +167,23 @@ u16 omap_ctrl_readw(u16 offset) u32 omap_ctrl_readl(u16 offset) { u32 val; + int err; offset &= 0xfffc; - if (!omap2_ctrl_syscon) - val = readl_relaxed(omap2_ctrl_base + offset); - else - regmap_read(omap2_ctrl_syscon, omap2_ctrl_offset + offset, - &val); + if (omap2_ctrl_syscon) { + err = regmap_read(omap2_ctrl_syscon, + omap2_ctrl_offset + offset, &val); + if (!err) + return val; + } + + if (WARN(!omap2_ctrl_base, + "syscon out of range for offset 0x%x, no omap2_ctrl_base?\n", + offset)) { + return 0; + } - return val; + return readl_relaxed(omap2_ctrl_base + offset); } void omap_ctrl_writeb(u8 val, u16 offset) @@ -206,12 +214,23 @@ void omap_ctrl_writew(u16 val, u16 offset) void omap_ctrl_writel(u32 val, u16 offset) { + int err; + offset &= 0xfffc; - if (!omap2_ctrl_syscon) - writel_relaxed(val, omap2_ctrl_base + offset); - else - regmap_write(omap2_ctrl_syscon, omap2_ctrl_offset + offset, - val); + if (omap2_ctrl_syscon) { + err = regmap_write(omap2_ctrl_syscon, + omap2_ctrl_offset + offset, val); + if (!err) + return; + } + + if (WARN(!omap2_ctrl_base, + "syscon out of range for offset 0x%x, no omap2_ctrl_base?\n", + offset)) { + return; + } + + writel_relaxed(val, omap2_ctrl_base + offset); } #ifdef CONFIG_ARCH_OMAP3 @@ -724,9 +743,6 @@ int __init omap_control_init(void) if (ret) return ret; } - - iounmap(omap2_ctrl_base); - omap2_ctrl_base = NULL; } else { /* No scm_conf found, direct access */ ret = omap2_clk_provider_init(np, data->index, NULL,