Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 8BIT
To: Tomeu Vizoso <tomeu.vizoso@collabora.com>, linux-kernel@vger.kernel.org,
        "Stephen Boyd" <sboyd@codeaurora.org>
From: Mike Turquette <mturquette@linaro.org>
In-Reply-To: <1422011024-32283-4-git-send-email-tomeu.vizoso@collabora.com>
Cc: "Tomeu Vizoso" <tomeu.vizoso@collabora.com>,
        "Paul Walmsley" <paul@pwsan.com>, "Tony Lindgren" <tony@atomide.com>,
        linux-omap@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        t-kristo@ti.com
References: <1422011024-32283-1-git-send-email-tomeu.vizoso@collabora.com>
 <1422011024-32283-4-git-send-email-tomeu.vizoso@collabora.com>
Message-ID: <20150201212432.22722.70917@quantum>
User-Agent: alot/0.3.5
Subject: Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances
Date: Sun, 01 Feb 2015 13:24:32 -0800
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2790
Lines: 67

Quoting Tomeu Vizoso (2015-01-23 03:03:30)
> Moves clock state to struct clk_core, but takes care to change as little API as
> possible.
> 
> struct clk_hw still has a pointer to a struct clk, which is the
> implementation's per-user clk instance, for backwards compatibility.
> 
> The struct clk that clk_get_parent() returns isn't owned by the caller, but by
> the clock implementation, so the former shouldn't call clk_put() on it.
> 
> Because some boards in mach-omap2 still register clocks statically, their clock
> registration had to be updated to take into account that the clock information
> is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and
struct dpll_data, namely this snippet from
arch/arm/mach-omap2/dpll3xxx.c:

        parent = __clk_get_parent(hw->clk);

        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                WARN(parent != dd->clk_bypass,
                                "here0, parent name is %s, bypass name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
                WARN(parent != dd->clk_ref,
                                "here1, parent name is %s, ref name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }

struct dpll_data has members clk_ref and clk_bypass which are struct clk
pointers. This was always a bit of a violation of the clk.h contract
since drivers are not supposed to deref struct clk pointers. Now that we
generate unique pointers for each call to clk_get (clk_ref & clk_bypass
are populated by of_clk_get in ti_clk_register_dpll) then the pointer
comparisons above will never be equal (even if they resolve down to the
same struct clk_core). I added the verbose traces to the WARNs above to
illustrate the point: the names are always the same but the pointers
differ.

AFAICT this doesn't break anything, but booting on OMAP3+ results in
noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers
with a simple integer parent_index. In fact we already have this index.
See how the pointers are populated in ti_clk_register_dpll:


        dd->clk_ref = of_clk_get(node, 0);
        dd->clk_bypass = of_clk_get(node, 1);

Tony, the same problem affects the FAPLL code which copy/pastes some of
the DPLL code.

Thoughts?

Regards,
Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/