Re: [PATCH v3] clk: at91: PLL recalc_rate() now using cached MUL+DIV values

From: Boris Brezillon
Date: Sun Apr 29 2018 - 09:19:48 EST


Hi Marcin,

On Fri, 27 Apr 2018 13:56:09 -0400
Marcin Ziemianowicz <marcin@xxxxxxxxxxxxxxxx> wrote:

> Stephen Boyd <sboyd@xxxxxxxxxx>,
> linux-clk@xxxxxxxxxxxxxxx,
> linux-arm-kernel@xxxxxxxxxxxxxxxxxxx,
> linux-kernel@xxxxxxxxxxxxxxx
> Bcc:
> Subject: [PATCH v3] clk: at91: PLL recalc_rate() now using cached MUL and DIV
> values
> Reply-To:

Hm, I don't know how you prepared and sent your patch, but you shouldn't
have these fields in the body of your email. Please use git format-patch
to prepare the patch and then git send-email to send it.

>
> When a USB device is connected to the USB host port on the SAM9N12 then
> you get "-62" error which seems to indicate USB replies from the device
> are timing out. Based on a logic sniffer, I saw the USB bus was running
> at half speed.
>
> The PLL code uses cached MUL and DIV values which get set in set_rate()
> and applied in prepare(), but the recalc_rate() function instead
> queries the hardware instead of using these cached values. Therefore,
> if recalc_rate() is called between a set_rate() and prepare(), the
> wrong frequency is calculated and later the USB clock divider for the
> SAM9N12 SOC will be configured for an incorrect clock.
>
> In my case, the PLL hardware was set to 96 Mhz before the OHCI
> driver loads, and therefore the usb clock divider was being set
> to /2 even though the OHCI driver set the PLL to 48 Mhz.
>
> As an alternative explanation, I noticed this was fixed in the past:
> http://lists.infradead.org/pipermail/linux-arm-kernel/2014-September/283502.html
> but was later changed back via a large patch (maybe by mistake?):
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=1bdf02326b71eae7e9b4b335b881856aaf9d1af6

Yep, probably by mistake. I started this rework long before it has been
submitted to the ML, so I probably messed something up when rebasing.

Also, prefer commit IDs to links to the ML archive. The above would
sentence would give:

"
As an alternative explanation, I noticed this was fixed in the past by
87e2ed338f1b ("clk: at91: fix recalc_rate implementation of PLL
driver") but the bug was later re-introduced by 1bdf02326b71 ("clk:
at91: make use of syscon/regmap internally").
"


The following comment and the changelog should be placed after the
'---' line, so that it's not part of the commit message.

> Thank you for bearing with me about this Boris.
>
> Changes since V2:
> Removed all logging/debug messages I added
> > Comment by Boris Brezillon about my fix being wrong addressed
> Changes since V1:
> Added patch set cover letter
> Shortened lines which were over >80 characters long
> > Comment by Greg Kroah-Hartman about "from" field in email addressed
> > Comment by Alan Stern about redundant debug lines addressed
>

You should add Fixes and Cc-stable tags so that the fix is backported
to stable branches:

Fixes: 1bdf02326b71 ("clk: at91: make use of syscon/regmap internally)
Cc: <stable@xxxxxxxxxxxxxxx>

> Signed-off-by: Marcin Ziemianowicz <marcin@xxxxxxxxxxxxxxxx>
> ---
> drivers/clk/at91/clk-pll.c | 13 +------------
> 1 file changed, 1 insertion(+), 12 deletions(-)
>
> diff --git a/drivers/clk/at91/clk-pll.c b/drivers/clk/at91/clk-pll.c
> index 7d3223fc..cc6e0364 100644
> --- a/drivers/clk/at91/clk-pll.c
> +++ b/drivers/clk/at91/clk-pll.c
> @@ -132,19 +132,8 @@ static unsigned long clk_pll_recalc_rate(struct clk_hw *hw,
> unsigned long parent_rate)
> {
> struct clk_pll *pll = to_clk_pll(hw);
> - unsigned int pllr;
> - u16 mul;
> - u8 div;
> -
> - regmap_read(pll->regmap, PLL_REG(pll->id), &pllr);
> -
> - div = PLL_DIV(pllr);
> - mul = PLL_MUL(pllr, pll->layout);
> -
> - if (!div || !mul)
> - return 0;
>
> - return (parent_rate / div) * (mul + 1);
> + return return (parent_rate / pll->div) * (pll->mul + 1);

The fix looks good. Let me know if you struggle with git
format-patch/send-email and I'll try to help you (or send the patch for
you if you don't care learning the process, but I think it's better if
you learn how to submit patches).

Thanks,

Boris

> }
>
> static long clk_pll_get_best_div_mul(struct clk_pll *pll, unsigned long rate,