| Message ID | alpine.DEB.2.00.1108191647430.9948@utopia.booyaka.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On Fri, Aug 19, 2011 at 04:48:52PM -0600, Paul Walmsley wrote: > > After commit 665d001338b494d6d62810aa99b4c0fa1a0884b9 ("OMAP2+: hwmod: > Follow the recommended PRCM module enable sequence"), device drivers > for OMAP IP blocks that do not use runtime PM can cause oopses or > kernel instability[1][2]. > > This is because those non-runtime PM drivers do not use the hwmod > code, which implements the correct IP block enable and disable > sequence. > > Several options for dealing with this problem have been proposed: > > 1. Add a new field to the OMAP struct clk to mark clocks that are > currently used by non-runtime PM drivers. Modify the clock code to > use the old clockdomain sequence for these marked clocks. As > drivers are converted to use runtime PM, remove the annotation from > the clocks. > > 2. Similar to #1, but associate the flag with the struct omap_clk > instead. > > 3. Add IDLEST wait support to the OMAP4 clock code, similar to the way > it is implemented for OMAP2/3, and enable it in each struct clk > currently used by non-runtime PM drivers. As drivers are converted > to use runtime PM, remove the annotation from the clocks. > > 4. Do nothing; leave the problem to those responsible for the > unconverted drivers. > > 5. Re-enable clock-based clockdomain control in the OMAP4 clock code. > This would revert back to the behavior of Linux 3.0, simply with a > slightly longer module enable/disable latency. > > Unfortunately, no approach seemed particularly good. Options 1 > through 3 seemed unwise due to the following reasons: > > A. The OMAP struct clks are intended primarily to describe hardware > clock nodes, and the intention is that no driver-specific data > should be stored there (applies to #1) > > B. The resulting patch would have been quite large for the -rc series > (applies to #1, #2, #3) > > C. The patch would have been a new, yet temporary hack; and similar fixes > have drawn negative comments in the recent past (see for example [3]) > > Option 4 is undesirable because commit > 665d001338b494d6d62810aa99b4c0fa1a0884b9 ("OMAP2+: hwmod: Follow the > recommended PRCM module enable sequence") has resulted in a less > stable kernel; and kernel stability is more important than OMAP4 power > management. > > Option 5 is the approach taken in this patch. This seemed to be the > least intrusive approach for 3.1-rc. > > The approach in this patch was originally proposed by Ohad Ben-Cohen > <ohad@wizery.com>. I'm simply writing the commit message and passing > it along. > > ... > > Thanks to Luciano Coelho <coelho@ti.com> for reporting the problem. > Thanks to Ohad Ben-Cohen <ohad@wizery.com> for tracking the problem > down, generating a temporary workaround, and proposing a patch to deal > with the problem. Thanks to Rajendra Nayak <rnayak@ti.com> for > proposing another patch to deal with the problem. > > This patch is intended for the 3.1-rc fixes series. > > 1. Coelho, Luciano <coelho@ti.com>. _Re: Oops on ehci_hcd when > booting 3.0.0-rc2 on panda_. Tue, 09 Aug 2011 14:26:08 +0300. > Posted to the <linux-omap@vger.kernel.org> mailing list. Available > from (among others) > http://www.spinics.net/linux/lists/linux-omap/msg55213.html > > 2. Munegowda, Keshava <keshava_mgowda@ti.com>. _Re: Oops on ehci_hcd > when booting 3.0.0-rc2 on panda_. Thu, 11 Aug 2011 13:51:05 +0530. > Posted to the <linux-omap@vger.kernel.org> mailing list. Available > from (among others) > http://www.spinics.net/linux/lists/linux-omap/msg55371.html > > 3. King, Russell <linux@arm.linux.org.uk>. _Re: [PATCH 5/8] OMAP4: > PM: TEMP: Prevent l3init from idling/force sleep_. Thu, 23 Jun > 2011 16:22:49 +0100. Posted to the <linux-omap@vger.kernel.org> > mailing list. Available from (among others) > http://www.mail-archive.com/linux-omap@vger.kernel.org/msg51392.html > > Signed-off-by: Paul Walmsley <paul@pwsan.com> > Cc: Luciano Coelho <coelho@ti.com> > Cc: Ohad Ben-Cohen <ohad@wizery.com> > Cc: Rajendra Nayak <rnayak@ti.com> > Cc: Benoît Cousson <b-cousson@ti.com> > --- > arch/arm/mach-omap2/clock44xx_data.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/arch/arm/mach-omap2/clock44xx_data.c b/arch/arm/mach-omap2/clock44xx_data.c > index 2af0e3f..fe6508d 100644 > --- a/arch/arm/mach-omap2/clock44xx_data.c > +++ b/arch/arm/mach-omap2/clock44xx_data.c > @@ -3379,7 +3379,7 @@ int __init omap4xxx_clk_init(void) > } > > clk_init(&omap2_clk_functions); maybe add a comment here stating why this line is commented out ? > - omap2_clk_disable_clkdm_control(); > + /* omap2_clk_disable_clkdm_control(); */ > > for (c = omap44xx_clks; c < omap44xx_clks + ARRAY_SIZE(omap44xx_clks); > c++)
diff --git a/arch/arm/mach-omap2/clock44xx_data.c b/arch/arm/mach-omap2/clock44xx_data.c index 2af0e3f..fe6508d 100644 --- a/arch/arm/mach-omap2/clock44xx_data.c +++ b/arch/arm/mach-omap2/clock44xx_data.c @@ -3379,7 +3379,7 @@ int __init omap4xxx_clk_init(void) } clk_init(&omap2_clk_functions); - omap2_clk_disable_clkdm_control(); + /* omap2_clk_disable_clkdm_control(); */ for (c = omap44xx_clks; c < omap44xx_clks + ARRAY_SIZE(omap44xx_clks); c++)
After commit 665d001338b494d6d62810aa99b4c0fa1a0884b9 ("OMAP2+: hwmod: Follow the recommended PRCM module enable sequence"), device drivers for OMAP IP blocks that do not use runtime PM can cause oopses or kernel instability[1][2]. This is because those non-runtime PM drivers do not use the hwmod code, which implements the correct IP block enable and disable sequence. Several options for dealing with this problem have been proposed: 1. Add a new field to the OMAP struct clk to mark clocks that are currently used by non-runtime PM drivers. Modify the clock code to use the old clockdomain sequence for these marked clocks. As drivers are converted to use runtime PM, remove the annotation from the clocks. 2. Similar to #1, but associate the flag with the struct omap_clk instead. 3. Add IDLEST wait support to the OMAP4 clock code, similar to the way it is implemented for OMAP2/3, and enable it in each struct clk currently used by non-runtime PM drivers. As drivers are converted to use runtime PM, remove the annotation from the clocks. 4. Do nothing; leave the problem to those responsible for the unconverted drivers. 5. Re-enable clock-based clockdomain control in the OMAP4 clock code. This would revert back to the behavior of Linux 3.0, simply with a slightly longer module enable/disable latency. Unfortunately, no approach seemed particularly good. Options 1 through 3 seemed unwise due to the following reasons: A. The OMAP struct clks are intended primarily to describe hardware clock nodes, and the intention is that no driver-specific data should be stored there (applies to #1) B. The resulting patch would have been quite large for the -rc series (applies to #1, #2, #3) C. The patch would have been a new, yet temporary hack; and similar fixes have drawn negative comments in the recent past (see for example [3]) Option 4 is undesirable because commit 665d001338b494d6d62810aa99b4c0fa1a0884b9 ("OMAP2+: hwmod: Follow the recommended PRCM module enable sequence") has resulted in a less stable kernel; and kernel stability is more important than OMAP4 power management. Option 5 is the approach taken in this patch. This seemed to be the least intrusive approach for 3.1-rc. The approach in this patch was originally proposed by Ohad Ben-Cohen <ohad@wizery.com>. I'm simply writing the commit message and passing it along. ... Thanks to Luciano Coelho <coelho@ti.com> for reporting the problem. Thanks to Ohad Ben-Cohen <ohad@wizery.com> for tracking the problem down, generating a temporary workaround, and proposing a patch to deal with the problem. Thanks to Rajendra Nayak <rnayak@ti.com> for proposing another patch to deal with the problem. This patch is intended for the 3.1-rc fixes series. 1. Coelho, Luciano <coelho@ti.com>. _Re: Oops on ehci_hcd when booting 3.0.0-rc2 on panda_. Tue, 09 Aug 2011 14:26:08 +0300. Posted to the <linux-omap@vger.kernel.org> mailing list. Available from (among others) http://www.spinics.net/linux/lists/linux-omap/msg55213.html 2. Munegowda, Keshava <keshava_mgowda@ti.com>. _Re: Oops on ehci_hcd when booting 3.0.0-rc2 on panda_. Thu, 11 Aug 2011 13:51:05 +0530. Posted to the <linux-omap@vger.kernel.org> mailing list. Available from (among others) http://www.spinics.net/linux/lists/linux-omap/msg55371.html 3. King, Russell <linux@arm.linux.org.uk>. _Re: [PATCH 5/8] OMAP4: PM: TEMP: Prevent l3init from idling/force sleep_. Thu, 23 Jun 2011 16:22:49 +0100. Posted to the <linux-omap@vger.kernel.org> mailing list. Available from (among others) http://www.mail-archive.com/linux-omap@vger.kernel.org/msg51392.html Signed-off-by: Paul Walmsley <paul@pwsan.com> Cc: Luciano Coelho <coelho@ti.com> Cc: Ohad Ben-Cohen <ohad@wizery.com> Cc: Rajendra Nayak <rnayak@ti.com> Cc: Benoît Cousson <b-cousson@ti.com> --- arch/arm/mach-omap2/clock44xx_data.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)