| Message ID | 1522311530-87337-1-git-send-email-shawn.lin@rock-chips.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 29 March 2018 at 10:18, Shawn Lin <shawn.lin@rock-chips.com> wrote: > The intention of this patch is to rework the timeout calculation > for dw_mmc, which isn't ideal enough. It's harmless to add longest > timeout as normally the transfer is fine and the occasional failure > could break out by no matter the Hardware interrupt or software timer. > > However, it doesn't make sense in the tuning process, as it could > trigger more timeout than usual, which is very painful for booting > time. Currently we set TMOUT to 0xffffffff anyway. Assume the max > clock is 200MHz when doing tuning for HS200, the we have: > > MSEC_PER_SEC * 0xffffff / 200MHz = 83ms for hardware timemout interrupt > and 93ms for software timer. > > However, we should note that the cards should guarantee to complete a > sequence of 40 times CMD21 execulations within 150ms per the spec. But > we now have 83ms timeout for each CMD21, which only contains 128Bytes > data playload. > > Maybe the simplest way is to reduce the HW/SW timeout value for tuning > process but it's still worth trying to rework a more accurate timeout. > > Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com> Jaehoon, any comments on this one? Kind regards Uffe > > --- > > drivers/mmc/host/dw_mmc.c | 107 +++++++++++++++++++++++++++++++++------------- > drivers/mmc/host/dw_mmc.h | 4 ++ > 2 files changed, 82 insertions(+), 29 deletions(-) > > diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c > index 623f4d2..c85d703 100644 > --- a/drivers/mmc/host/dw_mmc.c > +++ b/drivers/mmc/host/dw_mmc.c > @@ -378,23 +378,84 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd) > return cmdr; > } > > -static inline void dw_mci_set_cto(struct dw_mci *host) > +static void dw_mci_calc_timeout(struct dw_mci *host) > { > - unsigned int cto_clks; > - unsigned int cto_div; > - unsigned int cto_ms; > - unsigned long irqflags; > + struct mmc_data *data = host->data; > + struct mmc_host *mmc = host->slot->mmc; > + u64 trsf_time_per_blksz; > + u64 idle_time_per_trsf; > + u64 val; > + struct mmc_ios *ios = &mmc->ios; > + unsigned char bus_width = ios->bus_width; > + unsigned int freq, clk_div; > + unsigned int cto_clks, drto_clks; > + > + clk_div = (mci_readl(host, CLKDIV) & 0xff) * 2; > + if (clk_div == 0) > + clk_div = 1; > + > + freq = host->bus_hz / clk_div; > + > + /* > + * The largest 136 bit response in 100KHz is nearly 1ms, > + * and add other transfer parameter per spec, we should > + * allow 2ms maybe. And should also add some spare time > + * to tolerate anything unknown leading to unexpect long > + * latency. > + */ > + host->cmd_timeout_ns = 10 * NSEC_PER_MSEC; > + host->data_timeout_ns = 10 * NSEC_PER_MSEC; > + > + if (data) { > + trsf_time_per_blksz = DIV_ROUND_UP_ULL((u64)data->blksz * > + NSEC_PER_SEC * (8 / bus_width), freq); > + > + /* > + * Especially multiply by 2 to account for write for anything > + * unknown in the cards, for each block. > + */ > + if (data->flags & MMC_DATA_WRITE) > + trsf_time_per_blksz *= 2; > + > + /* Calculate idle time from core basd on spec */ > + idle_time_per_trsf = DIV_ROUND_UP_ULL(data->timeout_ns, > + NSEC_PER_USEC); > + if (data->timeout_clks) { > + /* Align up from cycle of MHz to Hz */ > + val = 1000000ULL * data->timeout_clks; > + if (DIV_ROUND_UP_ULL(val, freq)) > + idle_time_per_trsf++; > + idle_time_per_trsf += val; > + } > + > + /* Calculate timeout for the entire data */ > + host->data_timeout_ns += > + data->blocks * (DIV_ROUND_UP_ULL(idle_time_per_trsf, > + NSEC_PER_USEC) + trsf_time_per_blksz); > > - cto_clks = mci_readl(host, TMOUT) & 0xff; > - cto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; > - if (cto_div == 0) > - cto_div = 1; > + drto_clks = DIV_ROUND_UP_ULL(host->data_timeout_ns * freq, > + NSEC_PER_SEC); > + if (drto_clks > 0xffffff) > + drto_clks = 0xffffff; > > - cto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * cto_clks * cto_div, > - host->bus_hz); > + /* Set accurate hw data timeout */ > + mci_writel(host, TMOUT, drto_clks << 8); > + } > > - /* add a bit spare time */ > - cto_ms += 10; > + host->cmd_timeout_ns += host->cmd->busy_timeout * NSEC_PER_MSEC; > + cto_clks = DIV_ROUND_UP_ULL(host->cmd_timeout_ns * freq, > + NSEC_PER_SEC); > + if (cto_clks > 0xff) > + cto_clks = 0xff; > + > + /* Set accurate cmd timeout */ > + mci_writel(host, TMOUT, cto_clks | > + (mci_readl(host, TMOUT) & 0xffffff00)); > +} > + > +static inline void dw_mci_set_cto(struct dw_mci *host) > +{ > + unsigned long irqflags; > > /* > * The durations we're working with are fairly short so we have to be > @@ -412,7 +473,7 @@ static inline void dw_mci_set_cto(struct dw_mci *host) > spin_lock_irqsave(&host->irq_lock, irqflags); > if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events)) > mod_timer(&host->cto_timer, > - jiffies + msecs_to_jiffies(cto_ms) + 1); > + jiffies + nsecs_to_jiffies(host->cmd_timeout_ns) + 1); > spin_unlock_irqrestore(&host->irq_lock, irqflags); > } > > @@ -424,6 +485,8 @@ static void dw_mci_start_command(struct dw_mci *host, > "start command: ARGR=0x%08x CMDR=0x%08x\n", > cmd->arg, cmd_flags); > > + dw_mci_calc_timeout(host); > + > mci_writel(host, CMDARG, cmd->arg); > wmb(); /* drain writebuffer */ > dw_mci_wait_while_busy(host, cmd_flags); > @@ -1923,26 +1986,12 @@ static int dw_mci_data_complete(struct dw_mci *host, struct mmc_data *data) > > static void dw_mci_set_drto(struct dw_mci *host) > { > - unsigned int drto_clks; > - unsigned int drto_div; > - unsigned int drto_ms; > unsigned long irqflags; > > - drto_clks = mci_readl(host, TMOUT) >> 8; > - drto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; > - if (drto_div == 0) > - drto_div = 1; > - > - drto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * drto_clks * drto_div, > - host->bus_hz); > - > - /* add a bit spare time */ > - drto_ms += 10; > - > spin_lock_irqsave(&host->irq_lock, irqflags); > if (!test_bit(EVENT_DATA_COMPLETE, &host->pending_events)) > mod_timer(&host->dto_timer, > - jiffies + msecs_to_jiffies(drto_ms)); > + jiffies + nsecs_to_jiffies(host->data_timeout_ns)); > spin_unlock_irqrestore(&host->irq_lock, irqflags); > } > > diff --git a/drivers/mmc/host/dw_mmc.h b/drivers/mmc/host/dw_mmc.h > index 46e9f8e..4330128 100644 > --- a/drivers/mmc/host/dw_mmc.h > +++ b/drivers/mmc/host/dw_mmc.h > @@ -126,7 +126,9 @@ struct dw_mci_dma_slave { > * @sdio_id0: Number of slot0 in the SDIO interrupt registers. > * @cmd11_timer: Timer for SD3.0 voltage switch over scheme. > * @cto_timer: Timer for broken command transfer over scheme. > + * @cmd_timeout_ns: Timeout value accociated with cto_timer > * @dto_timer: Timer for broken data transfer over scheme. > + * @data_timeout_ns: Timeout value accociated with dto_timer > * > * Locking > * ======= > @@ -233,7 +235,9 @@ struct dw_mci { > > struct timer_list cmd11_timer; > struct timer_list cto_timer; > + u64 cmd_timeout_ns; > struct timer_list dto_timer; > + u64 data_timeout_ns; > }; > > /* DMA ops for Internal/External DMAC interface */ > -- > 1.9.1 > > -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi all, On 2018/3/29 16:18, Shawn Lin wrote: > The intention of this patch is to rework the timeout calculation > for dw_mmc, which isn't ideal enough. It's harmless to add longest > timeout as normally the transfer is fine and the occasional failure > could break out by no matter the Hardware interrupt or software timer. > > However, it doesn't make sense in the tuning process, as it could > trigger more timeout than usual, which is very painful for booting > time. Currently we set TMOUT to 0xffffffff anyway. Assume the max > clock is 200MHz when doing tuning for HS200, the we have: > > MSEC_PER_SEC * 0xffffff / 200MHz = 83ms for hardware timemout interrupt > and 93ms for software timer. > > However, we should note that the cards should guarantee to complete a > sequence of 40 times CMD21 execulations within 150ms per the spec. But > we now have 83ms timeout for each CMD21, which only contains 128Bytes > data playload. > > Maybe the simplest way is to reduce the HW/SW timeout value for tuning > process but it's still worth trying to rework a more accurate timeout. > Just as a regular share that this patch was backported to 4.4 kernel tree by Ziyuan(cc'ed) and pass the intensive test in QA cycle. I would still like to spin this more time to make sure it work fine by more solid test. Will come back to send a regular patch in coming weeks. Any comments are welcome. > Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com> > > --- > > drivers/mmc/host/dw_mmc.c | 107 +++++++++++++++++++++++++++++++++------------- > drivers/mmc/host/dw_mmc.h | 4 ++ > 2 files changed, 82 insertions(+), 29 deletions(-) > > diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c > index 623f4d2..c85d703 100644 > --- a/drivers/mmc/host/dw_mmc.c > +++ b/drivers/mmc/host/dw_mmc.c > @@ -378,23 +378,84 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd) > return cmdr; > } > > -static inline void dw_mci_set_cto(struct dw_mci *host) > +static void dw_mci_calc_timeout(struct dw_mci *host) > { > - unsigned int cto_clks; > - unsigned int cto_div; > - unsigned int cto_ms; > - unsigned long irqflags; > + struct mmc_data *data = host->data; > + struct mmc_host *mmc = host->slot->mmc; > + u64 trsf_time_per_blksz; > + u64 idle_time_per_trsf; > + u64 val; > + struct mmc_ios *ios = &mmc->ios; > + unsigned char bus_width = ios->bus_width; > + unsigned int freq, clk_div; > + unsigned int cto_clks, drto_clks; > + > + clk_div = (mci_readl(host, CLKDIV) & 0xff) * 2; > + if (clk_div == 0) > + clk_div = 1; > + > + freq = host->bus_hz / clk_div; > + > + /* > + * The largest 136 bit response in 100KHz is nearly 1ms, > + * and add other transfer parameter per spec, we should > + * allow 2ms maybe. And should also add some spare time > + * to tolerate anything unknown leading to unexpect long > + * latency. > + */ > + host->cmd_timeout_ns = 10 * NSEC_PER_MSEC; > + host->data_timeout_ns = 10 * NSEC_PER_MSEC; > + > + if (data) { > + trsf_time_per_blksz = DIV_ROUND_UP_ULL((u64)data->blksz * > + NSEC_PER_SEC * (8 / bus_width), freq); > + > + /* > + * Especially multiply by 2 to account for write for anything > + * unknown in the cards, for each block. > + */ > + if (data->flags & MMC_DATA_WRITE) > + trsf_time_per_blksz *= 2; > + > + /* Calculate idle time from core basd on spec */ > + idle_time_per_trsf = DIV_ROUND_UP_ULL(data->timeout_ns, > + NSEC_PER_USEC); > + if (data->timeout_clks) { > + /* Align up from cycle of MHz to Hz */ > + val = 1000000ULL * data->timeout_clks; > + if (DIV_ROUND_UP_ULL(val, freq)) > + idle_time_per_trsf++; > + idle_time_per_trsf += val; > + } > + > + /* Calculate timeout for the entire data */ > + host->data_timeout_ns += > + data->blocks * (DIV_ROUND_UP_ULL(idle_time_per_trsf, > + NSEC_PER_USEC) + trsf_time_per_blksz); > > - cto_clks = mci_readl(host, TMOUT) & 0xff; > - cto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; > - if (cto_div == 0) > - cto_div = 1; > + drto_clks = DIV_ROUND_UP_ULL(host->data_timeout_ns * freq, > + NSEC_PER_SEC); > + if (drto_clks > 0xffffff) > + drto_clks = 0xffffff; > > - cto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * cto_clks * cto_div, > - host->bus_hz); > + /* Set accurate hw data timeout */ > + mci_writel(host, TMOUT, drto_clks << 8); > + } > > - /* add a bit spare time */ > - cto_ms += 10; > + host->cmd_timeout_ns += host->cmd->busy_timeout * NSEC_PER_MSEC; > + cto_clks = DIV_ROUND_UP_ULL(host->cmd_timeout_ns * freq, > + NSEC_PER_SEC); > + if (cto_clks > 0xff) > + cto_clks = 0xff; > + > + /* Set accurate cmd timeout */ > + mci_writel(host, TMOUT, cto_clks | > + (mci_readl(host, TMOUT) & 0xffffff00)); > +} > + > +static inline void dw_mci_set_cto(struct dw_mci *host) > +{ > + unsigned long irqflags; > > /* > * The durations we're working with are fairly short so we have to be > @@ -412,7 +473,7 @@ static inline void dw_mci_set_cto(struct dw_mci *host) > spin_lock_irqsave(&host->irq_lock, irqflags); > if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events)) > mod_timer(&host->cto_timer, > - jiffies + msecs_to_jiffies(cto_ms) + 1); > + jiffies + nsecs_to_jiffies(host->cmd_timeout_ns) + 1); > spin_unlock_irqrestore(&host->irq_lock, irqflags); > } > > @@ -424,6 +485,8 @@ static void dw_mci_start_command(struct dw_mci *host, > "start command: ARGR=0x%08x CMDR=0x%08x\n", > cmd->arg, cmd_flags); > > + dw_mci_calc_timeout(host); > + > mci_writel(host, CMDARG, cmd->arg); > wmb(); /* drain writebuffer */ > dw_mci_wait_while_busy(host, cmd_flags); > @@ -1923,26 +1986,12 @@ static int dw_mci_data_complete(struct dw_mci *host, struct mmc_data *data) > > static void dw_mci_set_drto(struct dw_mci *host) > { > - unsigned int drto_clks; > - unsigned int drto_div; > - unsigned int drto_ms; > unsigned long irqflags; > > - drto_clks = mci_readl(host, TMOUT) >> 8; > - drto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; > - if (drto_div == 0) > - drto_div = 1; > - > - drto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * drto_clks * drto_div, > - host->bus_hz); > - > - /* add a bit spare time */ > - drto_ms += 10; > - > spin_lock_irqsave(&host->irq_lock, irqflags); > if (!test_bit(EVENT_DATA_COMPLETE, &host->pending_events)) > mod_timer(&host->dto_timer, > - jiffies + msecs_to_jiffies(drto_ms)); > + jiffies + nsecs_to_jiffies(host->data_timeout_ns)); > spin_unlock_irqrestore(&host->irq_lock, irqflags); > } > > diff --git a/drivers/mmc/host/dw_mmc.h b/drivers/mmc/host/dw_mmc.h > index 46e9f8e..4330128 100644 > --- a/drivers/mmc/host/dw_mmc.h > +++ b/drivers/mmc/host/dw_mmc.h > @@ -126,7 +126,9 @@ struct dw_mci_dma_slave { > * @sdio_id0: Number of slot0 in the SDIO interrupt registers. > * @cmd11_timer: Timer for SD3.0 voltage switch over scheme. > * @cto_timer: Timer for broken command transfer over scheme. > + * @cmd_timeout_ns: Timeout value accociated with cto_timer > * @dto_timer: Timer for broken data transfer over scheme. > + * @data_timeout_ns: Timeout value accociated with dto_timer > * > * Locking > * ======= > @@ -233,7 +235,9 @@ struct dw_mci { > > struct timer_list cmd11_timer; > struct timer_list cto_timer; > + u64 cmd_timeout_ns; > struct timer_list dto_timer; > + u64 data_timeout_ns; > }; > > /* DMA ops for Internal/External DMAC interface */ > -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Dear Ulf, On 04/20/2018 04:14 PM, Ulf Hansson wrote: > On 29 March 2018 at 10:18, Shawn Lin <shawn.lin@rock-chips.com> wrote: >> The intention of this patch is to rework the timeout calculation >> for dw_mmc, which isn't ideal enough. It's harmless to add longest >> timeout as normally the transfer is fine and the occasional failure >> could break out by no matter the Hardware interrupt or software timer. >> >> However, it doesn't make sense in the tuning process, as it could >> trigger more timeout than usual, which is very painful for booting >> time. Currently we set TMOUT to 0xffffffff anyway. Assume the max >> clock is 200MHz when doing tuning for HS200, the we have: >> >> MSEC_PER_SEC * 0xffffff / 200MHz = 83ms for hardware timemout interrupt >> and 93ms for software timer. >> >> However, we should note that the cards should guarantee to complete a >> sequence of 40 times CMD21 execulations within 150ms per the spec. But >> we now have 83ms timeout for each CMD21, which only contains 128Bytes >> data playload. >> >> Maybe the simplest way is to reduce the HW/SW timeout value for tuning >> process but it's still worth trying to rework a more accurate timeout. >> >> Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com> > > Jaehoon, any comments on this one? Sorry for late. Actually, drto/crto timeout value are something wrong when clock is 400KHz. It should be set to huge value. And it will be also overflowed. I will test this patch on my target. Will share the result until tomorrow. Best Regards, Jaehoon Chung > > Kind regards > Uffe > >> >> --- >> >> drivers/mmc/host/dw_mmc.c | 107 +++++++++++++++++++++++++++++++++------------- >> drivers/mmc/host/dw_mmc.h | 4 ++ >> 2 files changed, 82 insertions(+), 29 deletions(-) >> >> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c >> index 623f4d2..c85d703 100644 >> --- a/drivers/mmc/host/dw_mmc.c >> +++ b/drivers/mmc/host/dw_mmc.c >> @@ -378,23 +378,84 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd) >> return cmdr; >> } >> >> -static inline void dw_mci_set_cto(struct dw_mci *host) >> +static void dw_mci_calc_timeout(struct dw_mci *host) >> { >> - unsigned int cto_clks; >> - unsigned int cto_div; >> - unsigned int cto_ms; >> - unsigned long irqflags; >> + struct mmc_data *data = host->data; >> + struct mmc_host *mmc = host->slot->mmc; >> + u64 trsf_time_per_blksz; >> + u64 idle_time_per_trsf; >> + u64 val; >> + struct mmc_ios *ios = &mmc->ios; >> + unsigned char bus_width = ios->bus_width; >> + unsigned int freq, clk_div; >> + unsigned int cto_clks, drto_clks; >> + >> + clk_div = (mci_readl(host, CLKDIV) & 0xff) * 2; >> + if (clk_div == 0) >> + clk_div = 1; >> + >> + freq = host->bus_hz / clk_div; >> + >> + /* >> + * The largest 136 bit response in 100KHz is nearly 1ms, >> + * and add other transfer parameter per spec, we should >> + * allow 2ms maybe. And should also add some spare time >> + * to tolerate anything unknown leading to unexpect long >> + * latency. >> + */ >> + host->cmd_timeout_ns = 10 * NSEC_PER_MSEC; >> + host->data_timeout_ns = 10 * NSEC_PER_MSEC; >> + >> + if (data) { >> + trsf_time_per_blksz = DIV_ROUND_UP_ULL((u64)data->blksz * >> + NSEC_PER_SEC * (8 / bus_width), freq); >> + >> + /* >> + * Especially multiply by 2 to account for write for anything >> + * unknown in the cards, for each block. >> + */ >> + if (data->flags & MMC_DATA_WRITE) >> + trsf_time_per_blksz *= 2; >> + >> + /* Calculate idle time from core basd on spec */ >> + idle_time_per_trsf = DIV_ROUND_UP_ULL(data->timeout_ns, >> + NSEC_PER_USEC); >> + if (data->timeout_clks) { >> + /* Align up from cycle of MHz to Hz */ >> + val = 1000000ULL * data->timeout_clks; >> + if (DIV_ROUND_UP_ULL(val, freq)) >> + idle_time_per_trsf++; >> + idle_time_per_trsf += val; >> + } >> + >> + /* Calculate timeout for the entire data */ >> + host->data_timeout_ns += >> + data->blocks * (DIV_ROUND_UP_ULL(idle_time_per_trsf, >> + NSEC_PER_USEC) + trsf_time_per_blksz); >> >> - cto_clks = mci_readl(host, TMOUT) & 0xff; >> - cto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; >> - if (cto_div == 0) >> - cto_div = 1; >> + drto_clks = DIV_ROUND_UP_ULL(host->data_timeout_ns * freq, >> + NSEC_PER_SEC); >> + if (drto_clks > 0xffffff) >> + drto_clks = 0xffffff; >> >> - cto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * cto_clks * cto_div, >> - host->bus_hz); >> + /* Set accurate hw data timeout */ >> + mci_writel(host, TMOUT, drto_clks << 8); >> + } >> >> - /* add a bit spare time */ >> - cto_ms += 10; >> + host->cmd_timeout_ns += host->cmd->busy_timeout * NSEC_PER_MSEC; >> + cto_clks = DIV_ROUND_UP_ULL(host->cmd_timeout_ns * freq, >> + NSEC_PER_SEC); >> + if (cto_clks > 0xff) >> + cto_clks = 0xff; >> + >> + /* Set accurate cmd timeout */ >> + mci_writel(host, TMOUT, cto_clks | >> + (mci_readl(host, TMOUT) & 0xffffff00)); >> +} >> + >> +static inline void dw_mci_set_cto(struct dw_mci *host) >> +{ >> + unsigned long irqflags; >> >> /* >> * The durations we're working with are fairly short so we have to be >> @@ -412,7 +473,7 @@ static inline void dw_mci_set_cto(struct dw_mci *host) >> spin_lock_irqsave(&host->irq_lock, irqflags); >> if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events)) >> mod_timer(&host->cto_timer, >> - jiffies + msecs_to_jiffies(cto_ms) + 1); >> + jiffies + nsecs_to_jiffies(host->cmd_timeout_ns) + 1); >> spin_unlock_irqrestore(&host->irq_lock, irqflags); >> } >> >> @@ -424,6 +485,8 @@ static void dw_mci_start_command(struct dw_mci *host, >> "start command: ARGR=0x%08x CMDR=0x%08x\n", >> cmd->arg, cmd_flags); >> >> + dw_mci_calc_timeout(host); >> + >> mci_writel(host, CMDARG, cmd->arg); >> wmb(); /* drain writebuffer */ >> dw_mci_wait_while_busy(host, cmd_flags); >> @@ -1923,26 +1986,12 @@ static int dw_mci_data_complete(struct dw_mci *host, struct mmc_data *data) >> >> static void dw_mci_set_drto(struct dw_mci *host) >> { >> - unsigned int drto_clks; >> - unsigned int drto_div; >> - unsigned int drto_ms; >> unsigned long irqflags; >> >> - drto_clks = mci_readl(host, TMOUT) >> 8; >> - drto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; >> - if (drto_div == 0) >> - drto_div = 1; >> - >> - drto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * drto_clks * drto_div, >> - host->bus_hz); >> - >> - /* add a bit spare time */ >> - drto_ms += 10; >> - >> spin_lock_irqsave(&host->irq_lock, irqflags); >> if (!test_bit(EVENT_DATA_COMPLETE, &host->pending_events)) >> mod_timer(&host->dto_timer, >> - jiffies + msecs_to_jiffies(drto_ms)); >> + jiffies + nsecs_to_jiffies(host->data_timeout_ns)); >> spin_unlock_irqrestore(&host->irq_lock, irqflags); >> } >> >> diff --git a/drivers/mmc/host/dw_mmc.h b/drivers/mmc/host/dw_mmc.h >> index 46e9f8e..4330128 100644 >> --- a/drivers/mmc/host/dw_mmc.h >> +++ b/drivers/mmc/host/dw_mmc.h >> @@ -126,7 +126,9 @@ struct dw_mci_dma_slave { >> * @sdio_id0: Number of slot0 in the SDIO interrupt registers. >> * @cmd11_timer: Timer for SD3.0 voltage switch over scheme. >> * @cto_timer: Timer for broken command transfer over scheme. >> + * @cmd_timeout_ns: Timeout value accociated with cto_timer >> * @dto_timer: Timer for broken data transfer over scheme. >> + * @data_timeout_ns: Timeout value accociated with dto_timer >> * >> * Locking >> * ======= >> @@ -233,7 +235,9 @@ struct dw_mci { >> >> struct timer_list cmd11_timer; >> struct timer_list cto_timer; >> + u64 cmd_timeout_ns; >> struct timer_list dto_timer; >> + u64 data_timeout_ns; >> }; >> >> /* DMA ops for Internal/External DMAC interface */ >> -- >> 1.9.1 >> >> > -- > To unsubscribe from this list: send the line "unsubscribe linux-mmc" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > > > -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c index 623f4d2..c85d703 100644 --- a/drivers/mmc/host/dw_mmc.c +++ b/drivers/mmc/host/dw_mmc.c @@ -378,23 +378,84 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd) return cmdr; } -static inline void dw_mci_set_cto(struct dw_mci *host) +static void dw_mci_calc_timeout(struct dw_mci *host) { - unsigned int cto_clks; - unsigned int cto_div; - unsigned int cto_ms; - unsigned long irqflags; + struct mmc_data *data = host->data; + struct mmc_host *mmc = host->slot->mmc; + u64 trsf_time_per_blksz; + u64 idle_time_per_trsf; + u64 val; + struct mmc_ios *ios = &mmc->ios; + unsigned char bus_width = ios->bus_width; + unsigned int freq, clk_div; + unsigned int cto_clks, drto_clks; + + clk_div = (mci_readl(host, CLKDIV) & 0xff) * 2; + if (clk_div == 0) + clk_div = 1; + + freq = host->bus_hz / clk_div; + + /* + * The largest 136 bit response in 100KHz is nearly 1ms, + * and add other transfer parameter per spec, we should + * allow 2ms maybe. And should also add some spare time + * to tolerate anything unknown leading to unexpect long + * latency. + */ + host->cmd_timeout_ns = 10 * NSEC_PER_MSEC; + host->data_timeout_ns = 10 * NSEC_PER_MSEC; + + if (data) { + trsf_time_per_blksz = DIV_ROUND_UP_ULL((u64)data->blksz * + NSEC_PER_SEC * (8 / bus_width), freq); + + /* + * Especially multiply by 2 to account for write for anything + * unknown in the cards, for each block. + */ + if (data->flags & MMC_DATA_WRITE) + trsf_time_per_blksz *= 2; + + /* Calculate idle time from core basd on spec */ + idle_time_per_trsf = DIV_ROUND_UP_ULL(data->timeout_ns, + NSEC_PER_USEC); + if (data->timeout_clks) { + /* Align up from cycle of MHz to Hz */ + val = 1000000ULL * data->timeout_clks; + if (DIV_ROUND_UP_ULL(val, freq)) + idle_time_per_trsf++; + idle_time_per_trsf += val; + } + + /* Calculate timeout for the entire data */ + host->data_timeout_ns += + data->blocks * (DIV_ROUND_UP_ULL(idle_time_per_trsf, + NSEC_PER_USEC) + trsf_time_per_blksz); - cto_clks = mci_readl(host, TMOUT) & 0xff; - cto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; - if (cto_div == 0) - cto_div = 1; + drto_clks = DIV_ROUND_UP_ULL(host->data_timeout_ns * freq, + NSEC_PER_SEC); + if (drto_clks > 0xffffff) + drto_clks = 0xffffff; - cto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * cto_clks * cto_div, - host->bus_hz); + /* Set accurate hw data timeout */ + mci_writel(host, TMOUT, drto_clks << 8); + } - /* add a bit spare time */ - cto_ms += 10; + host->cmd_timeout_ns += host->cmd->busy_timeout * NSEC_PER_MSEC; + cto_clks = DIV_ROUND_UP_ULL(host->cmd_timeout_ns * freq, + NSEC_PER_SEC); + if (cto_clks > 0xff) + cto_clks = 0xff; + + /* Set accurate cmd timeout */ + mci_writel(host, TMOUT, cto_clks | + (mci_readl(host, TMOUT) & 0xffffff00)); +} + +static inline void dw_mci_set_cto(struct dw_mci *host) +{ + unsigned long irqflags; /* * The durations we're working with are fairly short so we have to be @@ -412,7 +473,7 @@ static inline void dw_mci_set_cto(struct dw_mci *host) spin_lock_irqsave(&host->irq_lock, irqflags); if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events)) mod_timer(&host->cto_timer, - jiffies + msecs_to_jiffies(cto_ms) + 1); + jiffies + nsecs_to_jiffies(host->cmd_timeout_ns) + 1); spin_unlock_irqrestore(&host->irq_lock, irqflags); } @@ -424,6 +485,8 @@ static void dw_mci_start_command(struct dw_mci *host, "start command: ARGR=0x%08x CMDR=0x%08x\n", cmd->arg, cmd_flags); + dw_mci_calc_timeout(host); + mci_writel(host, CMDARG, cmd->arg); wmb(); /* drain writebuffer */ dw_mci_wait_while_busy(host, cmd_flags); @@ -1923,26 +1986,12 @@ static int dw_mci_data_complete(struct dw_mci *host, struct mmc_data *data) static void dw_mci_set_drto(struct dw_mci *host) { - unsigned int drto_clks; - unsigned int drto_div; - unsigned int drto_ms; unsigned long irqflags; - drto_clks = mci_readl(host, TMOUT) >> 8; - drto_div = (mci_readl(host, CLKDIV) & 0xff) * 2; - if (drto_div == 0) - drto_div = 1; - - drto_ms = DIV_ROUND_UP_ULL((u64)MSEC_PER_SEC * drto_clks * drto_div, - host->bus_hz); - - /* add a bit spare time */ - drto_ms += 10; - spin_lock_irqsave(&host->irq_lock, irqflags); if (!test_bit(EVENT_DATA_COMPLETE, &host->pending_events)) mod_timer(&host->dto_timer, - jiffies + msecs_to_jiffies(drto_ms)); + jiffies + nsecs_to_jiffies(host->data_timeout_ns)); spin_unlock_irqrestore(&host->irq_lock, irqflags); } diff --git a/drivers/mmc/host/dw_mmc.h b/drivers/mmc/host/dw_mmc.h index 46e9f8e..4330128 100644 --- a/drivers/mmc/host/dw_mmc.h +++ b/drivers/mmc/host/dw_mmc.h @@ -126,7 +126,9 @@ struct dw_mci_dma_slave { * @sdio_id0: Number of slot0 in the SDIO interrupt registers. * @cmd11_timer: Timer for SD3.0 voltage switch over scheme. * @cto_timer: Timer for broken command transfer over scheme. + * @cmd_timeout_ns: Timeout value accociated with cto_timer * @dto_timer: Timer for broken data transfer over scheme. + * @data_timeout_ns: Timeout value accociated with dto_timer * * Locking * ======= @@ -233,7 +235,9 @@ struct dw_mci { struct timer_list cmd11_timer; struct timer_list cto_timer; + u64 cmd_timeout_ns; struct timer_list dto_timer; + u64 data_timeout_ns; }; /* DMA ops for Internal/External DMAC interface */
The intention of this patch is to rework the timeout calculation for dw_mmc, which isn't ideal enough. It's harmless to add longest timeout as normally the transfer is fine and the occasional failure could break out by no matter the Hardware interrupt or software timer. However, it doesn't make sense in the tuning process, as it could trigger more timeout than usual, which is very painful for booting time. Currently we set TMOUT to 0xffffffff anyway. Assume the max clock is 200MHz when doing tuning for HS200, the we have: MSEC_PER_SEC * 0xffffff / 200MHz = 83ms for hardware timemout interrupt and 93ms for software timer. However, we should note that the cards should guarantee to complete a sequence of 40 times CMD21 execulations within 150ms per the spec. But we now have 83ms timeout for each CMD21, which only contains 128Bytes data playload. Maybe the simplest way is to reduce the HW/SW timeout value for tuning process but it's still worth trying to rework a more accurate timeout. Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com> --- drivers/mmc/host/dw_mmc.c | 107 +++++++++++++++++++++++++++++++++------------- drivers/mmc/host/dw_mmc.h | 4 ++ 2 files changed, 82 insertions(+), 29 deletions(-)