Re: [RFCv3 3/3] interconnect: imx: Add platform driver for imx8mm

[Leonard Crestez <leonard.crestez@nxp.com>]

On 16.10.2019 17:09, Angus Ainslie wrote:
> On 2019-10-15 07:05, Leonard Crestez wrote:
>> On 10.10.2019 17:43, Angus Ainslie wrote:
>>>
>>> I've integrated your u-boot and ATF on our board and I have a couple
>>> of questions. Our board is running imx8mq B0 (Rev 2.0) silicon.
>>>
>>> It looks like this line limits the training frequencies to 800 MHz and
>>> 166 MHz
>>
>> Yes! This is due to a hardware errata which was solved in B1: DRAM pll
>> can't be disabled. This means that instead of 25/100/800 freqs are
>> 166/800, and this requires code changes.
>>
>>> Does 100 MHz and 25 MHz not work on B0 ?
>>
>> No, lower rates require dram clk from a composite slice (dram_alt_root)
>>
>>> I added the ddrc_and noc opp as well as the 166MHz opp
>>>
>>> I also added the interconnects ( do we need them on imx8mq ? )
>>
>> The interconnect stuff is not required to switch dram frequency, it's
>> for device to make minimum bandwidth requests. It an additional feature
>> on top.
>>
>> As a hack I configured FEC to do so but a saner example would be to
>> request bandwidth based on display resolution and color depth.
>>
>>> I had to add a hack as the PM QoS was limiting the bus speed to 399MHz,
>>> if you have any ideas why that would be appreciated.
>>
>> You probably need to set ethernet down (which is awkward) or better
>> just drop the interconnect properties and test using the devfreq userspace
>> governor.
>>
>>> The driver is probing
>>>
>>> [   12.136537] bus: 'platform': driver_probe_device: matched device
>>> 3d400000.dram-controller with driver imx-ddrc-devfrq
>>> [   12.147259] bus: 'platform': really_probe: probing driver
>>> imx-ddrc-devfreq with device 3d400000.dram-controller
>>> [   12.157382] imx-ddrc-devfreq 3d400000.dram-controller: no pinctrl
>>> handle
>>> [   12.164197] arm_smcc rate 0 800000000
>>> [   12.167880] arm_smcc rate 1 166750000
>>> [   12.171778] of: _opp_add_static_v2: turbo:0 rate:25000000 uv:0
>>> uvmin:0 uvmax:0 latency:0
>>> [   12.179994] of: _opp_add_static_v2: turbo:0 rate:100000000 uv:0
>>> uvmin:0 uvmax:0 latency:0
>>> [   12.188311] of: _opp_add_static_v2: turbo:0 rate:166750000 uv:0
>>> uvmin:0 uvmax:0 latency:0
>>> [   12.196606] of: _opp_add_static_v2: turbo:0 rate:800000000 uv:0
>>> uvmin:0 uvmax:0 latency:0
>>> [   12.204930] imx-ddrc-devfreq 3d400000.dram-controller: events from
>>> pmu imx8_ddr0
>>> [   12.212403] Added freq 0 25000000
>>> [   12.215742] Added freq 1 100000000
>>> [   12.219177] Added freq 2 166750000
>>> [   12.222648] Added freq 3 800000000
>>> [   12.226105] device: 'devfreq0': device_add
>>> [   12.230287] PM: Adding info for No Bus:devfreq0
>>> [   12.234864] driver: 'imx-ddrc-devfreq': driver_bound: bound to
>>> device
>>> '3d400000.dram-controller'
>>> [   12.243699] bus: 'platform': really_probe: bound device
>>> 3d400000.dram-controller to driver imx-ddrc-devfreq
>>>
>>> Add seems to run correctly until it tries to adjust the clock to
>>> 166MHz
>>>
>>> [   19.555482] ddrc checking rate 800000000 166750000
>>> [   19.555489] ddrc checking rate 166750000 166750000
>>> [   19.560442] bus: 'usb-serial': really_probe: bound device ttyUSB0
>>> to
>>> driver option1
>>> [   19.568751] imx-ddrc-devfreq 3d400000.dram-controller: ddrc about
>>> to
>>> change freq 800000000 to 166750000
>>>
>>> And the board hangs there. Any ideas on how to get past this ?
>>
>> Please try this ATF patch:
> 
> Ok applied this to the tree we're using
> 
>> I tested switching on imx8mq-evk with B0 SoC but a few additional
>> changes are required in kernel to support switching between rates which
>> are both backed by PLL:
>>
>> * Mark the PLL CLK_GET_RATE_NOCACHE
> 
> Is this what you meant ?
> 
> diff --git a/drivers/clk/imx/clk-imx8mq.c b/drivers/clk/imx/clk-imx8mq.c
> index 2813884f69c1..e5f50cf8a264 100644
> --- a/drivers/clk/imx/clk-imx8mq.c
> +++ b/drivers/clk/imx/clk-imx8mq.c
> @@ -345,7 +345,7 @@ static int imx8mq_clocks_probe(struct
> platform_device *pdev)
>           clks[IMX8MQ_SYS1_PLL_OUT] = imx_clk_sccg_pll("sys1_pll_out",
> sys1_pll_out_sels, ARRAY_SIZE(sys1_pll_out_sels), 0, 0, 0, base + 0x30,
> CLK_IS_CRITICAL);
>           clks[IMX8MQ_SYS2_PLL_OUT] = imx_clk_sccg_pll("sys2_pll_out",
> sys2_pll_out_sels, ARRAY_SIZE(sys2_pll_out_sels), 0, 0, 1, base + 0x3c,
> CLK_IS_CRITICAL);
>           clks[IMX8MQ_SYS3_PLL_OUT] = imx_clk_sccg_pll("sys3_pll_out",
> sys3_pll_out_sels, ARRAY_SIZE(sys3_pll_out_sels), 0, 0, 1, base + 0x48,
> CLK_IS_CRITICAL);
> -       clks[IMX8MQ_DRAM_PLL_OUT] = imx_clk_sccg_pll("dram_pll_out",
> dram_pll_out_sels, ARRAY_SIZE(dram_pll_out_sels), 0, 0, 0, base + 0x60,
> CLK_IS_CRITICAL);
> +       clks[IMX8MQ_DRAM_PLL_OUT] = imx_clk_sccg_pll("dram_pll_out",
> dram_pll_out_sels, ARRAY_SIZE(dram_pll_out_sels), 0, 0, 0, base + 0x60,
> CLK_IS_CRITICAL|CLK_GET_RATE_NOCACHE);

Yes.

>> * Set the rate to 166935483 exactly (to match clk_get_rate)
> 
> Okay I hacked that in
> 
> diff --git a/drivers/devfreq/imx-ddrc.c b/drivers/devfreq/imx-ddrc.c
> index 4eed6f50bb8d..a832768a865f 100644
> --- a/drivers/devfreq/imx-ddrc.c
> +++ b/drivers/devfreq/imx-ddrc.c
> @@ -436,6 +436,10 @@ static int imx_ddrc_init_freq_info(struct device
> *dev)
>                           return -ENODEV;
>                   }
> 
> +               /* B0 hack */
> +               if ( freq->rate == 166750000 )
> +                       freq->rate = 166935483;
> +inserting 
>                   pr_err( "arm_smcc rate %d %lu\n", index, freq->rate );
>           }

A nicer solution would be to keep imx_ddrc_freq.rate in MT/s as reported 
by firmware and divide by 25000 in imx_ddrc_find_freq instead.

> --- a/arch/arm64/boot/dts/freescale/imx8mq.dtsi
> +++ b/arch/arm64/boot/dts/freescale/imx8mq.dtsi
> @@ -211,7 +211,7 @@
>                           opp-hz = /bits/ 64 <100000000>;
>                   };
>                   opp-166M {
> -                       opp-hz = /bits/ 64 <166750000>;
> +                       opp-hz = /bits/ 64 <166935483>;
>                   };
>                   opp-800M {
>                           opp-hz = /bits/ 64 <800000000>;

Yes, this is the precise clock rate in Hz.

>> * Make the rounding in imx-ddrc more generous.
> 
> Sorry I don't understand what you mean by this

I meant to make imx_ddrc_find_freq find 667 MT/s for an OPP of 166935483:

         /*
          * Firmware reports values in MT/s, so we round-down from Hz
          * Rounding is extra generous to ensure a match.
          */
         rate = DIV_ROUND_CLOSEST(rate, 250000);
         for (i = 0; i < priv->freq_count; ++i) {
                 struct imx_ddrc_freq *freq = &priv->freq_table[i];
                 if (freq->rate == rate ||
                                 freq->rate + 1 == rate ||
                                 freq->rate - 1 == rate)
                         return freq;
         }

But your B0 hack above should also work.

> Adding the other changes the board no longer hangs when trying to change
> frequencies but it also doesn't seem to actually change the frequency.
> 
> [    3.076426] ddrc checking rate 800000000 166935483
> [    3.081290] ddrc checking rate 166935483 166935483
> [    3.086225] imx-ddrc-devfreq 3d400000.dram-controller: ddrc about to
> change freq 800000000 to 166935483
> [    3.086891] imx-ddrc-devfreq 3d400000.dram-controller: ddrc changed
> freq 800000000 to 166935483
> 
> root@pureos:~# cat /sys/class/devfreq/devfreq0/cur_freq
> 800000000
> root@pureos:~# cat /sys/class/devfreq/devfreq0/target_freq
> 166935483

The target_freq value is from clk_get_rate(dram_core) but it is 
dram_core's parent which gets updated. It seems that a clk mux ignores 
CLK_GET_RATE_NOCACHE on the parent.

An update can be forced by adding `clk_get_rate(new_dram_core_parent);` 
at the end of imx_ddrc_set_freq.

You should also be able to check by looking at clk_summary or
/sys/kernel/debug/clk/dram_core_clk/clk_rate
/sys/kernel/debug/clk/dram_pll_out/clk_rate

--
Regards,
Leonard