| Message ID | 1348589142-11983-2-git-send-email-t-kristo@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, 25 Sep 2012, Tero Kristo wrote: > Previously the code only acquired spinlock after increasing / decreasing > the usecount value, which is wrong. This leaves a small window where > a task switch may occur between the check of the usecount and the actual > wakeup / sleep of the domain. Fixed by moving the spinlock locking before > the usecount access. Left the usecount as atomic_t if someone wants an > easy access to the parameter through atomic_read. > > Signed-off-by: Tero Kristo <t-kristo@ti.com> Thanks, queued for 3.7-rc fixes. - Paul
On Tue, 25 Sep 2012, Tero Kristo wrote: > Previously the code only acquired spinlock after increasing / decreasing > the usecount value, which is wrong. This leaves a small window where > a task switch may occur between the check of the usecount and the actual > wakeup / sleep of the domain. Fixed by moving the spinlock locking before > the usecount access. Left the usecount as atomic_t if someone wants an > easy access to the parameter through atomic_read. > > Signed-off-by: Tero Kristo <t-kristo@ti.com> Acked-by: Paul Walmsley <paul@pwsan.com> - Paul
* Paul Walmsley <paul@pwsan.com> [121016 19:26]: > On Tue, 25 Sep 2012, Tero Kristo wrote: > > > Previously the code only acquired spinlock after increasing / decreasing > > the usecount value, which is wrong. This leaves a small window where > > a task switch may occur between the check of the usecount and the actual > > wakeup / sleep of the domain. Fixed by moving the spinlock locking before > > the usecount access. Left the usecount as atomic_t if someone wants an > > easy access to the parameter through atomic_read. > > > > Signed-off-by: Tero Kristo <t-kristo@ti.com> > > Acked-by: Paul Walmsley <paul@pwsan.com> OK I'll apply this into omap-for-v3.7-rc1/fixes-take4. Regards, Tony
diff --git a/arch/arm/mach-omap2/clockdomain.c b/arch/arm/mach-omap2/clockdomain.c index 8664f5a..173905d 100644 --- a/arch/arm/mach-omap2/clockdomain.c +++ b/arch/arm/mach-omap2/clockdomain.c @@ -914,15 +914,18 @@ static int _clkdm_clk_hwmod_enable(struct clockdomain *clkdm) if (!clkdm || !arch_clkdm || !arch_clkdm->clkdm_clk_enable) return -EINVAL; + spin_lock_irqsave(&clkdm->lock, flags); + /* * For arch's with no autodeps, clkcm_clk_enable * should be called for every clock instance or hwmod that is * enabled, so the clkdm can be force woken up. */ - if ((atomic_inc_return(&clkdm->usecount) > 1) && autodeps) + if ((atomic_inc_return(&clkdm->usecount) > 1) && autodeps) { + spin_unlock_irqrestore(&clkdm->lock, flags); return 0; + } - spin_lock_irqsave(&clkdm->lock, flags); arch_clkdm->clkdm_clk_enable(clkdm); pwrdm_state_switch(clkdm->pwrdm.ptr); spin_unlock_irqrestore(&clkdm->lock, flags); @@ -939,15 +942,19 @@ static int _clkdm_clk_hwmod_disable(struct clockdomain *clkdm) if (!clkdm || !arch_clkdm || !arch_clkdm->clkdm_clk_disable) return -EINVAL; + spin_lock_irqsave(&clkdm->lock, flags); + if (atomic_read(&clkdm->usecount) == 0) { + spin_unlock_irqrestore(&clkdm->lock, flags); WARN_ON(1); /* underflow */ return -ERANGE; } - if (atomic_dec_return(&clkdm->usecount) > 0) + if (atomic_dec_return(&clkdm->usecount) > 0) { + spin_unlock_irqrestore(&clkdm->lock, flags); return 0; + } - spin_lock_irqsave(&clkdm->lock, flags); arch_clkdm->clkdm_clk_disable(clkdm); pwrdm_state_switch(clkdm->pwrdm.ptr); spin_unlock_irqrestore(&clkdm->lock, flags);
Previously the code only acquired spinlock after increasing / decreasing the usecount value, which is wrong. This leaves a small window where a task switch may occur between the check of the usecount and the actual wakeup / sleep of the domain. Fixed by moving the spinlock locking before the usecount access. Left the usecount as atomic_t if someone wants an easy access to the parameter through atomic_read. Signed-off-by: Tero Kristo <t-kristo@ti.com> --- arch/arm/mach-omap2/clockdomain.c | 15 +++++++++++---- 1 files changed, 11 insertions(+), 4 deletions(-)