Re: [PATCH 3/4] mmc: mediatek: Add tune support

[Chaotian Jing]

On Thu, 2015-10-15 at 11:17 +0200, Ulf Hansson wrote:
> [...]
> 
> >> >
> >> >         struct clk *src_clk;    /* msdc source clock */
> >> > +       struct clk *src_clk_parent; /* src_clk's parent */
> >> > +       struct clk *hs400_src;  /* 400Mhz source clock */
> >>
> >> Hmm, so you need to control the upper level clocks. Can you elaborate
> >> on why this is needed?
> >>
> > hs400 is DDR200, in our host design, if the mode is DDR(HS400), if want
> > to get 200Mhz mmc bus clock frequency, the minimum source clock is
> > double of the mmc bus clock, So, we need it for HS400 mode with 200Mhz
> > bus clock.
> 
> Thanks for clarifying.
> 
> [...]
> 
>  >         flags = readl(host->base + MSDC_INTEN);
> >> >         sdr_clr_bits(host->base + MSDC_INTEN, flags);
> >> > -       if (ddr) { /* may need to modify later */
> >> > -               mode = 0x2; /* ddr mode and use divisor */
> >> > +       sdr_clr_bits(host->base + MSDC_CFG, MSDC_CFG_HS400_CK_MODE);
> >> > +       if (timing == MMC_TIMING_UHS_DDR50 ||
> >> > +           timing == MMC_TIMING_MMC_DDR52 ||
> >>
> >> So, no support for HS200?
> >>
> > HS200 is the same with other SDR modes, so it will be handled at else..
> 
> Okay, nice!
> 
> So, your the driver currently supports HS200, but without need for tuning!?
> 

It support and need tuning for HS200, but do not support tuning for
HS400, that's why we fixed the hs400_ds_delay by project.

> [...]
> 
> >> > +static struct msdc_delay_phase get_best_delay(u32 delay)
> >> > +{
> >> > +       int start = 0, len = 0;
> >> > +       int start_final = 0, len_final = 0;
> >> > +       u8 final_phase = 0xff;
> >> > +       struct msdc_delay_phase delay_phase;
> >> > +
> >> > +       if (delay == 0) {
> >> > +               pr_err("phase error: [map:%x]\n", delay);
> >>
> >> Please use dev_err|warn|dbg|info instead.
> >>
> > As you know, this function is just only parse the argument "u32 delay",
> > it do not bind with any device.
> 
> You may just add a msdc_host * as a parameter to the function, that
> would solve this.
> 

Ok, will do it.

> [...]
> 
> >> > +static int msdc_send_tuning(struct mmc_host *host, u32 opcode, int *cmd_error)
> >>
> >> I think you can remove this function and use mmc_send_tuning() instead.
> > Hmm, I also noticed that there was a mmc_send_tuning, but, I need to get
> > the cmd_error when tune cmd response, in this case, do not care the data
> > error.
> 
> Well, if you need to extend the mmc_send_tuning() API to suite your
> needs, let's do that instead of duplicating code.
> 

OK, will extend the mmc_send_tuning, but it need change other vendor's
MMC driver, because it already use the mmc_send_tuning()

> >>
> >> > +{
> >> > +       struct mmc_request mrq = {NULL};
> >> > +       struct mmc_command cmd = {0};
> >> > +       struct mmc_data data = {0};
> >> > +       struct scatterlist sg;
> >> > +       struct mmc_ios *ios = &host->ios;
> >> > +       int size, err = 0;
> >> > +       u8 *data_buf;
> >> > +
> >> > +       if (ios->bus_width == MMC_BUS_WIDTH_8)
> >> > +               size = 128;
> >> > +       else if (ios->bus_width == MMC_BUS_WIDTH_4)
> >> > +               size = 64;
> >> > +       else
> >> > +               return -EINVAL;
> >> > +
> >> > +       data_buf = kzalloc(size, GFP_KERNEL);
> >> > +       if (!data_buf)
> >> > +               return -ENOMEM;
> >> > +
> >> > +       mrq.cmd = &cmd;
> >> > +       mrq.data = &data;
> >> > +
> >> > +       cmd.opcode = opcode;
> >> > +       cmd.flags = MMC_RSP_R1 | MMC_CMD_ADTC;
> >> > +
> >> > +       data.blksz = size;
> >> > +       data.blocks = 1;
> >> > +       data.flags = MMC_DATA_READ;
> >> > +
> >> > +       /*
> >> > +        * According to the tuning specs, Tuning process
> >> > +        * is normally shorter 40 executions of CMD19,
> >> > +        * and timeout value should be shorter than 150 ms
> >> > +        */
> >> > +       data.timeout_ns = 150 * NSEC_PER_MSEC;
> >> > +
> >> > +       data.sg = &sg;
> >> > +       data.sg_len = 1;
> >> > +       sg_init_one(&sg, data_buf, size);
> >> > +
> >> > +       mmc_wait_for_req(host, &mrq);
> >> > +
> >> > +       if (cmd_error)
> >> > +               *cmd_error = cmd.error;
> >> > +
> >> > +       if (cmd.error) {
> >> > +               err = cmd.error;
> >> > +               goto out;
> >> > +       }
> >> > +
> >> > +       if (data.error) {
> >> > +               err = data.error;
> >> > +               goto out;
> >> > +       }
> >> > +
> >> > +out:
> >> > +       kfree(data_buf);
> >> > +       return err;
> >> > +}
> >> > +
> 
> [...]
> 
> >> > +       host->src_clk_parent = clk_get_parent(host->src_clk);
> >>
> >> Don't you need to check the return value here?
> >>
> > will check it.
> >> > +       host->hs400_src = devm_clk_get(&pdev->dev, "400mhz");
> >>
> >> That's a really weird conid for a clock. If it's not too late to
> >> change, please do that!
> >>
> >> > +       if (IS_ERR(host->hs400_src)) {
> >> > +               dev_dbg(&pdev->dev, "Cannot find 400mhz at dts!\n");
> >> > +       } else if (clk_set_parent(host->src_clk_parent, host->hs400_src) < 0) {
> >> > +               dev_err(host->dev, "Failed to set 400mhz source clock!\n");
> >> > +               ret = -EINVAL;
> >>
> >> I think it seems more apropriate to use the return value from
> >> clk_set_parent() instead of inventing your own return value.
> >>
> > OK.
> >> > +               goto host_free;
> >> > +       }
> >>
> >> It seems like you don't need to store the src_clk_parent and the
> >> hs400_src in the host struct, as you are only using it during
> >> ->probe().
> > OK,will remove the member src_clk.
> 
> According to your earlier clarification about the clock source and
> clock rate. I think a more proper solution would be to use the
> clk_set_min_rate() or clk_set_rate_range() API, instead of dealing
> with re-parenting of the clock as above.
> 
> FYI, a clock provider may implement the ->determine_rate() ops to deal
> with re-parenting to find the requested clock rate.
> 

Actually,  at first, I use the clk_set_rate(), but our CCF owner
suggests to use clk_set_parent, because the same clock frequency, may
have several parent clock.
by the way, Sascha suggests to use the "assigned-clocks" and
"assigned-clock-parents", I tried and it works, and ,will move it from
mt8173.dtsi to mt8173-evb.dts, because customer may want use other
parent clock in it's projects.
Thx!

> [...]
> 
> Kind regards
> Uffe


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