From mboxrd@z Thu Jan 1 00:00:00 1970 From: Krzysztof Kozlowski Subject: Re: [PATCH] mmc: sdhci-s3c: solve problem with sleeping in atomic context Date: Fri, 16 Jan 2015 15:54:57 +0100 Message-ID: <1421420097.21067.2.camel@AMDC1943> References: <1421418605-457-1-git-send-email-p.osmialowsk@samsung.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from mailout2.w1.samsung.com ([210.118.77.12]:52561 "EHLO mailout2.w1.samsung.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752130AbbAPOzB (ORCPT ); Fri, 16 Jan 2015 09:55:01 -0500 In-reply-to: <1421418605-457-1-git-send-email-p.osmialowsk@samsung.com> Sender: linux-mmc-owner@vger.kernel.org List-Id: linux-mmc@vger.kernel.org To: Paul Osmialowski , Stephen Boyd Cc: Chris Ball , Ulf Hansson , linux-mmc@vger.kernel.org, linux-kernel@vger.kernel.org On pi=C4=85, 2015-01-16 at 15:30 +0100, Paul Osmialowski wrote: > This change addresses following problem: >=20 > [ 2.560726] ------------[ cut here ]------------ > [ 2.565341] WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:274= 4 lockdep_trace_alloc+0xec/0x118() > [ 2.574439] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) > [ 2.579821] Modules linked in: > [ 2.583038] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G W = 3.18.0-next-20141216-00002-g4ff197fc1902-dirty #1318 > [ 2.593796] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) > [ 2.599892] [] (unwind_backtrace) from [] (sho= w_stack+0x10/0x14) > [ 2.607612] [] (show_stack) from [] (dump_stac= k+0x70/0xbc) > [ 2.614822] [] (dump_stack) from [] (warn_slow= path_common+0x74/0xb0) > [ 2.622885] [] (warn_slowpath_common) from [] = (warn_slowpath_fmt+0x30/0x40) > [ 2.631569] [] (warn_slowpath_fmt) from [] (lo= ckdep_trace_alloc+0xec/0x118) > [ 2.640246] [] (lockdep_trace_alloc) from [] (= __kmalloc+0x3c/0x1cc) > [ 2.648240] [] (__kmalloc) from [] (clk_fetch_= parent_index+0xb8/0xd4) > [ 2.656390] [] (clk_fetch_parent_index) from [= ] (clk_calc_new_rates+0xe0/0x1fc) > [ 2.665415] [] (clk_calc_new_rates) from [] (c= lk_calc_new_rates+0x1b4/0x1fc) > [ 2.674181] [] (clk_calc_new_rates) from [] (c= lk_set_rate+0x50/0xc8) > [ 2.682265] [] (clk_set_rate) from [] (sdhci_c= mu_set_clock+0x68/0x16c) > [ 2.690503] [] (sdhci_cmu_set_clock) from [] (= sdhci_do_set_ios+0xf0/0x64c) > [ 2.699095] [] (sdhci_do_set_ios) from [] (sdh= ci_set_ios+0x20/0x2c) > [ 2.707080] [] (sdhci_set_ios) from [] (mmc_po= wer_up+0x118/0x1fc) > [ 2.714889] [] (mmc_power_up) from [] (mmc_sta= rt_host+0x44/0x6c) > [ 2.722615] [] (mmc_start_host) from [] (mmc_a= dd_host+0x58/0x7c) > [ 2.730341] [] (mmc_add_host) from [] (sdhci_a= dd_host+0x968/0xd94) > [ 2.738240] [] (sdhci_add_host) from [] (sdhci= _s3c_probe+0x354/0x52c) > [ 2.746406] [] (sdhci_s3c_probe) from [] (plat= form_drv_probe+0x48/0xa4) > [ 2.754733] [] (platform_drv_probe) from [] (d= river_probe_device+0x13c/0x37c) > [ 2.763585] [] (driver_probe_device) from [] (= __driver_attach+0x94/0x98) > [ 2.772003] [] (__driver_attach) from [] (bus_= for_each_dev+0x54/0x88) > [ 2.780163] [] (bus_for_each_dev) from [] (bus= _add_driver+0xe4/0x200) > [ 2.788322] [] (bus_add_driver) from [] (drive= r_register+0x78/0xf4) > [ 2.796308] [] (driver_register) from [] (do_o= ne_initcall+0xac/0x1f0) > [ 2.804473] [] (do_one_initcall) from [] (kern= el_init_freeable+0x10c/0x1d8) > [ 2.813153] [] (kernel_init_freeable) from [] = (kernel_init+0x28/0x108) > [ 2.821398] [] (kernel_init) from [] (ret_from= _fork+0x14/0x2c) > [ 2.828939] ---[ end trace 03cc00e539849d1f ]--- >=20 +Cc Stephen Boyd This is the fix for issue mentioned here: https://lkml.org/lkml/2014/12/22/125 Best regards, Krzysztof > clk_set_rate() tries to take clk's prepare_lock mutex while being in = atomic > context entered in sdhci_do_set_ios(). > The solution is inspired by similar situation in sdhci_set_power() al= so called > from sdhci_do_set_ios(): >=20 > spin_unlock_irq(&host->lock); > mmc_regulator_set_ocr(mmc, mmc->supply.vmmc, vdd); > spin_lock_irq(&host->lock); >=20 > Note that since sdhci_s3c_set_clock() sets SDHCI_CLOCK_CARD_EN, propo= sed change > first resets this bit. It is reset anyway (by setting SDHCI_CLOCK_INT= _EN bit > only) after call to clk_set_rate() in order to wait for the clock to = stabilize > and is set again as soon as the clock becomes stable. >=20 > Signed-off-by: Paul Osmialowski > --- > drivers/mmc/host/sdhci-s3c.c | 8 ++++++++ > 1 file changed, 8 insertions(+) >=20 > diff --git a/drivers/mmc/host/sdhci-s3c.c b/drivers/mmc/host/sdhci-s3= c.c > index c45b893..115f556 100644 > --- a/drivers/mmc/host/sdhci-s3c.c > +++ b/drivers/mmc/host/sdhci-s3c.c > @@ -12,6 +12,7 @@ > * published by the Free Software Foundation. > */ > =20 > +#include > #include > #include > #include > @@ -312,7 +313,14 @@ static void sdhci_cmu_set_clock(struct sdhci_hos= t *host, unsigned int clock) > =20 > sdhci_s3c_set_clock(host, clock); > =20 > + /* Reset SD Clock Enable */ > + clk =3D sdhci_readw(host, SDHCI_CLOCK_CONTROL); > + clk &=3D ~SDHCI_CLOCK_CARD_EN; > + sdhci_writew(host, clk, SDHCI_CLOCK_CONTROL); > + > + spin_unlock_irq(&host->lock); > ret =3D clk_set_rate(ourhost->clk_bus[ourhost->cur_clk], clock); > + spin_lock_irq(&host->lock); > if (ret !=3D 0) { > dev_err(dev, "%s: failed to set clock rate %uHz\n", > mmc_hostname(host->mmc), clock);