linux-kernel - Re: [PATCH] drm/tilcdc: fix LCD pixel clock setting

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <8adc90ef6ec8cb1ec3b8fdbdad0233cf@iki.fi>
Date:   Fri, 19 Mar 2021 14:57:31 +0200
From:   Jyri Sarha <jyri.sarha@....fi>
To:     Dario Binacchi <dariobin@...ero.it>
Cc:     Tomi Valkeinen <tomba@...nel.org>, linux-kernel@...r.kernel.org,
        Daniel Vetter <daniel@...ll.ch>,
        David Airlie <airlied@...ux.ie>,
        dri-devel@...ts.freedesktop.org
Subject: Re: [PATCH] drm/tilcdc: fix LCD pixel clock setting

On 2021-03-18 23:47, Dario Binacchi wrote:
>> Il 17/03/2021 09:19 Tomi Valkeinen <tomba@...nel.org> ha scritto:
>> 
>> 
>> On 14/03/2021 17:13, Dario Binacchi wrote:
>> > As reported by TI spruh73x RM, the LCD pixel clock (LCD_PCLK) frequency
>> > is obtained by dividing LCD_CLK, the LCD controller reference clock,
>> > for CLKDIV:
>> >
>> > LCD_PCLK = LCD_CLK / CLKDIV
>> >
>> > where CLKDIV must be greater than 1.
>> >
>> > Therefore LCD_CLK must be set to 'req_rate * CLKDIV' instead of req_rate
>> 
>> The above doesn't make sense, the code already sets LCD_CLK to 
>> 'req_rate
>> * clkdiv', not req_rate.
>> 
>> > and the real LCD_CLK rate must be compared with 'req_rate * CLKDIV' and
>> > not with req_rate.
>> 
>> This is true, the code looks at the wrong value.
>> 
>> > Passing req_rate instead of 'req_rate * CLKDIV' to the tilcdc_pclk_diff
>> > routine caused it to fail even if LCD_CLK was properly set.
>> >
>> > Signed-off-by: Dario Binacchi <dariobin@...ero.it>
>> >
>> > ---
>> >
>> >   drivers/gpu/drm/tilcdc/tilcdc_crtc.c | 9 +++++----
>> >   1 file changed, 5 insertions(+), 4 deletions(-)
>> >
>> > diff --git a/drivers/gpu/drm/tilcdc/tilcdc_crtc.c b/drivers/gpu/drm/tilcdc/tilcdc_crtc.c
>> > index 30213708fc99..02f56c9a5da5 100644
>> > --- a/drivers/gpu/drm/tilcdc/tilcdc_crtc.c
>> > +++ b/drivers/gpu/drm/tilcdc/tilcdc_crtc.c
>> > @@ -203,7 +203,7 @@ static void tilcdc_crtc_set_clk(struct drm_crtc *crtc)
>> >   	struct drm_device *dev = crtc->dev;
>> >   	struct tilcdc_drm_private *priv = dev->dev_private;
>> >   	struct tilcdc_crtc *tilcdc_crtc = to_tilcdc_crtc(crtc);
>> > -	unsigned long clk_rate, real_rate, req_rate;
>> > +	unsigned long clk_rate, real_rate, req_rate, clk_div_rate;
>> >   	unsigned int clkdiv;
>> >   	int ret;
>> >
>> > @@ -211,10 +211,11 @@ static void tilcdc_crtc_set_clk(struct drm_crtc *crtc)
>> >
>> >   	/* mode.clock is in KHz, set_rate wants parameter in Hz */
>> >   	req_rate = crtc->mode.clock * 1000;
>> > -
>> > -	ret = clk_set_rate(priv->clk, req_rate * clkdiv);
>> > +	/* LCD clock divisor input rate */
>> > +	clk_div_rate = req_rate * clkdiv;
>> 
>> "clk_div_rate" sounds a bit odd to me. Why not lcd_fck_rate, as that's
>> the name used later? Or lcd_clk_rate. Or maybe lcd_clk_req_rate...
> 
> I prefer lcd_clk_rate.
> 
> How about adding an additional patch that changes the variable names to 
> make
> the code more readable?
> 
> req_rate -> lcd_pclk_rate
> clk_rate -> real_lcd_clk_rate
> 
> And add a comment to the function which highlights the relationship
> LCD_CLK = LCD_PCLK * CLDIV ?
> 

What about renaming current req_rate to pclk_rate (for pixel clock 
rate), and calling pclk_rate * clkdiv = req_rate, as that is the rate we 
need to request from the input clock? Adding lcd to local variable names 
here is quite redundant after all. In any case req_rate is bit 
misleading name here and probably part of the reason why the bug exists 
in the first place.

Best regards,
Jyri



>> 
>> > +	ret = clk_set_rate(priv->clk, clk_div_rate);
>> >   	clk_rate = clk_get_rate(priv->clk);
>> > -	if (ret < 0 || tilcdc_pclk_diff(req_rate, clk_rate) > 5) {
>> > +	if (ret < 0 || tilcdc_pclk_diff(clk_div_rate, clk_rate) > 5) {
>> >   		/*
>> >   		 * If we fail to set the clock rate (some architectures don't
>> >   		 * use the common clock framework yet and may not implement
>> >
>> 
>> I think this fix is fine, but looking at the current code, it's 
>> calling
>> tilcdc_pclk_diff(), but doesn't actually provide pixel clocks to the
>> function, but fclk.
> 
> Yes, I agree.
> 
> Thanks and regards,
> Dario
> 
>> 
>>   Tomi