[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAPLW+4nqSntq=vEY4JL1=YZ+3Hb5EAeOB0aob+B6WGs97Zh3rw@mail.gmail.com>
Date: Fri, 3 Oct 2025 13:48:22 -0500
From: Sam Protsenko <semen.protsenko@...aro.org>
To: André Draszik <andre.draszik@...aro.org>
Cc: Krzysztof Kozlowski <krzk@...nel.org>, Sylwester Nawrocki <s.nawrocki@...sung.com>,
Chanwoo Choi <cw00.choi@...sung.com>, Alim Akhtar <alim.akhtar@...sung.com>,
Michael Turquette <mturquette@...libre.com>, Stephen Boyd <sboyd@...nel.org>,
Peter Griffin <peter.griffin@...aro.org>, Tudor Ambarus <tudor.ambarus@...aro.org>,
Will McVicker <willmcvicker@...gle.com>, kernel-team@...roid.com,
linux-samsung-soc@...r.kernel.org, linux-clk@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] clk: samsung: clk-pll: simplify samsung_pll_lock_wait()
On Wed, Oct 1, 2025 at 10:13 AM André Draszik <andre.draszik@...aro.org> wrote:
>
> readl_relaxed_poll_timeout_atomic() has been updated in 2023 in
> commit 7349a69cf312 ("iopoll: Do not use timekeeping in
> read_poll_timeout_atomic()") to avoid usage of timekeeping APIs. It
> also never used udelay() when no delay was given.
>
> With the implementation avoiding timekeeping APIs, and with a caller
> not passing a delay, the timeout argument simply becomes a loop
> counter.
>
> Therefore the code here can be simplified to unconditionally use
> readl_relaxed_poll_timeout_atomic(). The difference being the last
> argument, the timeout (loop counter). Simply adjust it to pass the
> more generous counter in all cases.
>
> Signed-off-by: André Draszik <andre.draszik@...aro.org>
> ---
Reviewed-by: Sam Protsenko <semen.protsenko@...aro.org>
> drivers/clk/samsung/clk-pll.c | 31 ++++++++++---------------------
> 1 file changed, 10 insertions(+), 21 deletions(-)
>
> diff --git a/drivers/clk/samsung/clk-pll.c b/drivers/clk/samsung/clk-pll.c
> index 7bea7be1d7e45c32f0b303ffa55ce9cde4a4f71d..a7e693f6983ec073bedd633ed8da7efafc1a20bb 100644
> --- a/drivers/clk/samsung/clk-pll.c
> +++ b/drivers/clk/samsung/clk-pll.c
> @@ -17,8 +17,7 @@
> #include "clk.h"
> #include "clk-pll.h"
>
> -#define PLL_TIMEOUT_US 20000U
> -#define PLL_TIMEOUT_LOOPS 1000000U
> +#define PLL_TIMEOUT_LOOPS 20000U
>
> struct samsung_clk_pll {
> struct clk_hw hw;
> @@ -84,7 +83,7 @@ arch_initcall(samsung_pll_disable_early_timeout);
> static int samsung_pll_lock_wait(struct samsung_clk_pll *pll,
> unsigned int reg_mask)
> {
> - int i, ret;
> + int ret;
> u32 val;
>
> /*
> @@ -93,25 +92,15 @@ static int samsung_pll_lock_wait(struct samsung_clk_pll *pll,
> * initialized, another when the timekeeping is suspended. udelay() also
> * cannot be used when the clocksource is not running on arm64, since
> * the current timer is used as cycle counter. So a simple busy loop
> - * is used here in that special cases. The limit of iterations has been
> - * derived from experimental measurements of various PLLs on multiple
> - * Exynos SoC variants. Single register read time was usually in range
> - * 0.4...1.5 us, never less than 0.4 us.
> + * is used here.
> + * The limit of iterations has been derived from experimental
> + * measurements of various PLLs on multiple Exynos SoC variants. Single
> + * register read time was usually in range 0.4...1.5 us, never less than
> + * 0.4 us.
> */
> - if (pll_early_timeout || timekeeping_suspended) {
> - i = PLL_TIMEOUT_LOOPS;
> - while (i-- > 0) {
> - if (readl_relaxed(pll->con_reg) & reg_mask)
> - return 0;
> -
> - cpu_relax();
> - }
> - ret = -ETIMEDOUT;
> - } else {
> - ret = readl_relaxed_poll_timeout_atomic(pll->con_reg, val,
> - val & reg_mask, 0, PLL_TIMEOUT_US);
> - }
> -
> + ret = readl_relaxed_poll_timeout_atomic(pll->con_reg, val,
> + val & reg_mask, 0,
> + PLL_TIMEOUT_LOOPS);
> if (ret < 0)
> pr_err("Could not lock PLL %s\n", clk_hw_get_name(&pll->hw));
>
>
> ---
> base-commit: 3b9b1f8df454caa453c7fb07689064edb2eda90a
> change-id: 20251001-samsung-clk-pll-simplification-3e02f8912122
>
> Best regards,
> --
> André Draszik <andre.draszik@...aro.org>
>
>
Powered by blists - more mailing lists