| Message ID | 1427988853-9549-2-git-send-email-heiko@sntech.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 04/02/15 08:34, Heiko Stuebner wrote: > From: Doug Anderson <dianders@chromium.org> > > With the existing code, if you find a parent for an orhpan that has > already been prepared / enabled, you won't enable the parent. That > can cause later problems since the clock tree isn't in a consistent > state. Fix by propagating the prepare and enable. > > NOTE: this does bring up the question about whether the enable of the > orphan actually made sense. If the orphan's parent wasn't enabled by > default (by the bootloader or the default state of the hardware) then > the original enable of the orphan probably didn't do what the caller > though it would. Some users of the orphan might have preferred an > EPROBE_DEFER be returned until we had a full path to a root clock. > This patch doesn't address those concerns and really just syncs up the > state. > > Tested on rk3288-evb-rk808 by temporarily considering "sclk_tsadc" as > a critical clock (to simulate a driver enabling it at bootup). > > Before: > > clock enable_cnt prepare_cnt rate accuracy phase > ---------------------------------------------------------------------------------------- > xin32k 0 0 32768 0 0 > sclk_hdmi_cec 0 0 32768 0 0 > sclk_otg_adp 0 0 32768 0 0 > sclk_tsadc 1 1 993 0 0 > > After: > > clock enable_cnt prepare_cnt rate accuracy phase > ---------------------------------------------------------------------------------------- > xin32k 1 1 32768 0 0 > sclk_hdmi_cec 0 0 32768 0 0 > sclk_otg_adp 0 0 32768 0 0 > sclk_tsadc 1 1 993 0 0 > > Note that xin32k on rk808 is a clock that cannot be disabled in > hardware (it's an always on clock), so really all we needed to do was > to sync up the state. > > Signed-off-by: Doug Anderson <dianders@chromium.org> > > Adapted to recent clk->clk_core changes and move orphan handling to > a separate function as the main clk_core_reparent function is already > also used in other contexts. > Signed-off-by: Heiko Stuebner <heiko@sntech.de> > --- Sorry I don't get this patch at all. We shouldn't be handing out clocks to consumers if the clock is an orphan. We also have a problem syncing enable state in the framework with what is there in hardware when we boot up. These aren't the same problem but they're related. Let's not conflate the two.
diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c index f85c8e2..512323f 100644 --- a/drivers/clk/clk.c +++ b/drivers/clk/clk.c @@ -2011,6 +2011,27 @@ static void clk_core_reparent(struct clk_core *clk, __clk_recalc_rates(clk, POST_RATE_CHANGE); } +static void clk_core_reparent_orphan(struct clk_core *clk, + struct clk_core *new_parent) +{ + clk_core_reparent(clk, new_parent); + + if (clk->prepare_count) { + unsigned long flags; + + if (clk_core_prepare(new_parent)) { + pr_err("%s: could not prepare new parent %s\n", + __func__, clk->name); + return; + } + + flags = clk_enable_lock(); + if (clk->enable_count) + clk_core_enable(new_parent); + clk_enable_unlock(flags); + } +} + void clk_hw_reparent(struct clk_hw *hw, struct clk_hw *new_parent) { if (!hw) @@ -2404,13 +2425,13 @@ static int __clk_init(struct device *dev, struct clk *clk_user) if (orphan->num_parents && orphan->ops->get_parent) { i = orphan->ops->get_parent(orphan->hw); if (!strcmp(clk->name, orphan->parent_names[i])) - clk_core_reparent(orphan, clk); + clk_core_reparent_orphan(orphan, clk); continue; } for (i = 0; i < orphan->num_parents; i++) if (!strcmp(clk->name, orphan->parent_names[i])) { - clk_core_reparent(orphan, clk); + clk_core_reparent_orphan(orphan, clk); break; } }