| Message ID | 20201210132745.v3.1.Iec3430c7d3c2a29262695edef7b82a14aaa567e5@changeid |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v3] mmc: sdhci-msm: Warn about overclocking SD/MMC | expand |
On Thu 10 Dec 15:27 CST 2020, Douglas Anderson wrote: > As talked about in commit 5e4b7e82d497 ("clk: qcom: gcc-sdm845: Use > floor ops for sdcc clks"), most clocks handled by the Qualcomm clock > drivers are rounded _up_ by default instead of down. We should make > sure SD/MMC clocks are always rounded down in the clock drivers. > Let's add a warning in the Qualcomm SDHCI driver to help catch the > problem. > > This would have saved a bunch of time [1]. > > [1] http://lore.kernel.org/r/20201210102234.1.I096779f219625148900fc984dd0084ed1ba87c7f@changeid > > Suggested-by: Stephen Boyd <swboyd@chromium.org> > Signed-off-by: Douglas Anderson <dianders@chromium.org> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> Regards, Bjorn > --- > > Changes in v3: > - Proper printf format code. > > Changes in v2: > - Store rate in unsigned long, not unsigned int. > - Reuse the clk_get_rate() in the later print. > > drivers/mmc/host/sdhci-msm.c | 15 +++++++++++++-- > 1 file changed, 13 insertions(+), 2 deletions(-) > > diff --git a/drivers/mmc/host/sdhci-msm.c b/drivers/mmc/host/sdhci-msm.c > index 3451eb325513..50beb407dbe9 100644 > --- a/drivers/mmc/host/sdhci-msm.c > +++ b/drivers/mmc/host/sdhci-msm.c > @@ -353,6 +353,7 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, > struct sdhci_msm_host *msm_host = sdhci_pltfm_priv(pltfm_host); > struct mmc_ios curr_ios = host->mmc->ios; > struct clk *core_clk = msm_host->bulk_clks[0].clk; > + unsigned long achieved_rate; > int rc; > > clock = msm_get_clock_rate_for_bus_mode(host, clock); > @@ -363,10 +364,20 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, > curr_ios.timing); > return; > } > + > + /* > + * Qualcomm clock drivers by default round clock _up_ if they can't > + * make the requested rate. This is not good for SD. Yell if we > + * encounter it. > + */ > + achieved_rate = clk_get_rate(core_clk); > + if (achieved_rate > clock) > + pr_warn("%s: Card appears overclocked; req %u Hz, actual %lu Hz\n", > + mmc_hostname(host->mmc), clock, achieved_rate); > + > msm_host->clk_rate = clock; > pr_debug("%s: Setting clock at rate %lu at timing %d\n", > - mmc_hostname(host->mmc), clk_get_rate(core_clk), > - curr_ios.timing); > + mmc_hostname(host->mmc), achieved_rate, curr_ios.timing); > } > > /* Platform specific tuning */ > -- > 2.29.2.576.ga3fc446d84-goog >
On 10/12/20 11:27 pm, Douglas Anderson wrote: > As talked about in commit 5e4b7e82d497 ("clk: qcom: gcc-sdm845: Use > floor ops for sdcc clks"), most clocks handled by the Qualcomm clock > drivers are rounded _up_ by default instead of down. We should make > sure SD/MMC clocks are always rounded down in the clock drivers. > Let's add a warning in the Qualcomm SDHCI driver to help catch the > problem. > > This would have saved a bunch of time [1]. > > [1] http://lore.kernel.org/r/20201210102234.1.I096779f219625148900fc984dd0084ed1ba87c7f@changeid > > Suggested-by: Stephen Boyd <swboyd@chromium.org> > Signed-off-by: Douglas Anderson <dianders@chromium.org> > --- > > Changes in v3: > - Proper printf format code. > > Changes in v2: > - Store rate in unsigned long, not unsigned int. > - Reuse the clk_get_rate() in the later print. > > drivers/mmc/host/sdhci-msm.c | 15 +++++++++++++-- > 1 file changed, 13 insertions(+), 2 deletions(-) > > diff --git a/drivers/mmc/host/sdhci-msm.c b/drivers/mmc/host/sdhci-msm.c > index 3451eb325513..50beb407dbe9 100644 > --- a/drivers/mmc/host/sdhci-msm.c > +++ b/drivers/mmc/host/sdhci-msm.c > @@ -353,6 +353,7 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, > struct sdhci_msm_host *msm_host = sdhci_pltfm_priv(pltfm_host); > struct mmc_ios curr_ios = host->mmc->ios; > struct clk *core_clk = msm_host->bulk_clks[0].clk; > + unsigned long achieved_rate; > int rc; > > clock = msm_get_clock_rate_for_bus_mode(host, clock); > @@ -363,10 +364,20 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, > curr_ios.timing); > return; > } > + > + /* > + * Qualcomm clock drivers by default round clock _up_ if they can't > + * make the requested rate. This is not good for SD. Yell if we > + * encounter it. > + */ > + achieved_rate = clk_get_rate(core_clk); > + if (achieved_rate > clock) > + pr_warn("%s: Card appears overclocked; req %u Hz, actual %lu Hz\n", > + mmc_hostname(host->mmc), clock, achieved_rate); How does a warning help? Also, did you consider setting host->mmc->actual_clock > + > msm_host->clk_rate = clock; > pr_debug("%s: Setting clock at rate %lu at timing %d\n", > - mmc_hostname(host->mmc), clk_get_rate(core_clk), > - curr_ios.timing); > + mmc_hostname(host->mmc), achieved_rate, curr_ios.timing); > } > > /* Platform specific tuning */ >
Hi, On Thu, Dec 10, 2020 at 9:56 PM Adrian Hunter <adrian.hunter@intel.com> wrote: > > On 10/12/20 11:27 pm, Douglas Anderson wrote: > > As talked about in commit 5e4b7e82d497 ("clk: qcom: gcc-sdm845: Use > > floor ops for sdcc clks"), most clocks handled by the Qualcomm clock > > drivers are rounded _up_ by default instead of down. We should make > > sure SD/MMC clocks are always rounded down in the clock drivers. > > Let's add a warning in the Qualcomm SDHCI driver to help catch the > > problem. > > > > This would have saved a bunch of time [1]. > > > > [1] http://lore.kernel.org/r/20201210102234.1.I096779f219625148900fc984dd0084ed1ba87c7f@changeid > > > > Suggested-by: Stephen Boyd <swboyd@chromium.org> > > Signed-off-by: Douglas Anderson <dianders@chromium.org> > > --- > > > > Changes in v3: > > - Proper printf format code. > > > > Changes in v2: > > - Store rate in unsigned long, not unsigned int. > > - Reuse the clk_get_rate() in the later print. > > > > drivers/mmc/host/sdhci-msm.c | 15 +++++++++++++-- > > 1 file changed, 13 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/mmc/host/sdhci-msm.c b/drivers/mmc/host/sdhci-msm.c > > index 3451eb325513..50beb407dbe9 100644 > > --- a/drivers/mmc/host/sdhci-msm.c > > +++ b/drivers/mmc/host/sdhci-msm.c > > @@ -353,6 +353,7 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, > > struct sdhci_msm_host *msm_host = sdhci_pltfm_priv(pltfm_host); > > struct mmc_ios curr_ios = host->mmc->ios; > > struct clk *core_clk = msm_host->bulk_clks[0].clk; > > + unsigned long achieved_rate; > > int rc; > > > > clock = msm_get_clock_rate_for_bus_mode(host, clock); > > @@ -363,10 +364,20 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, > > curr_ios.timing); > > return; > > } > > + > > + /* > > + * Qualcomm clock drivers by default round clock _up_ if they can't > > + * make the requested rate. This is not good for SD. Yell if we > > + * encounter it. > > + */ > > + achieved_rate = clk_get_rate(core_clk); > > + if (achieved_rate > clock) > > + pr_warn("%s: Card appears overclocked; req %u Hz, actual %lu Hz\n", > > + mmc_hostname(host->mmc), clock, achieved_rate); > > How does a warning help? The actual fixes are made in the clock drivers, but this seems to be a common problem in Qualcomm clock drivers. ...so while this doesn't actually fix the problem, it hopefully calls enough attention that someone will know to fix it in the clock driver. > Also, did you consider setting host->mmc->actual_clock That makes sense, though I think logically it's a separate change. I'll add it as patch #2 in the series. It's not as trivial as I would have thought. ;-) -Doug > > > + > > msm_host->clk_rate = clock; > > pr_debug("%s: Setting clock at rate %lu at timing %d\n", > > - mmc_hostname(host->mmc), clk_get_rate(core_clk), > > - curr_ios.timing); > > + mmc_hostname(host->mmc), achieved_rate, curr_ios.timing); > > } > > > > /* Platform specific tuning */ > > >
diff --git a/drivers/mmc/host/sdhci-msm.c b/drivers/mmc/host/sdhci-msm.c index 3451eb325513..50beb407dbe9 100644 --- a/drivers/mmc/host/sdhci-msm.c +++ b/drivers/mmc/host/sdhci-msm.c @@ -353,6 +353,7 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, struct sdhci_msm_host *msm_host = sdhci_pltfm_priv(pltfm_host); struct mmc_ios curr_ios = host->mmc->ios; struct clk *core_clk = msm_host->bulk_clks[0].clk; + unsigned long achieved_rate; int rc; clock = msm_get_clock_rate_for_bus_mode(host, clock); @@ -363,10 +364,20 @@ static void msm_set_clock_rate_for_bus_mode(struct sdhci_host *host, curr_ios.timing); return; } + + /* + * Qualcomm clock drivers by default round clock _up_ if they can't + * make the requested rate. This is not good for SD. Yell if we + * encounter it. + */ + achieved_rate = clk_get_rate(core_clk); + if (achieved_rate > clock) + pr_warn("%s: Card appears overclocked; req %u Hz, actual %lu Hz\n", + mmc_hostname(host->mmc), clock, achieved_rate); + msm_host->clk_rate = clock; pr_debug("%s: Setting clock at rate %lu at timing %d\n", - mmc_hostname(host->mmc), clk_get_rate(core_clk), - curr_ios.timing); + mmc_hostname(host->mmc), achieved_rate, curr_ios.timing); } /* Platform specific tuning */
As talked about in commit 5e4b7e82d497 ("clk: qcom: gcc-sdm845: Use floor ops for sdcc clks"), most clocks handled by the Qualcomm clock drivers are rounded _up_ by default instead of down. We should make sure SD/MMC clocks are always rounded down in the clock drivers. Let's add a warning in the Qualcomm SDHCI driver to help catch the problem. This would have saved a bunch of time [1]. [1] http://lore.kernel.org/r/20201210102234.1.I096779f219625148900fc984dd0084ed1ba87c7f@changeid Suggested-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Douglas Anderson <dianders@chromium.org> --- Changes in v3: - Proper printf format code. Changes in v2: - Store rate in unsigned long, not unsigned int. - Reuse the clk_get_rate() in the later print. drivers/mmc/host/sdhci-msm.c | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)