Message ID | 1346159947-3194-1-git-send-email-svenkatr@ti.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Tue, Aug 28, 2012 at 06:49:06PM +0530, Venkatraman S wrote: > omap hsmmc controller IP has a built in timer that can be programmed to > guard against unresponsive operations. But its range is very narrow, > and the maximum countable time is a few seconds. > > Card maintenance operations like BKOPS and MMC_ERASE and long > stream writes like packed command require timers of order of > several minutes, much beyond the capability of the IP timer. > So get rid of using the IP timer entirely and use kernel's hrtimer > functionality for guarding the device operations. > As part of this change, a workaround that disabled timeouts for > MMC_ERASE command is removed, and the arbitary timing of 100ms > is used only when the timeout is not explicitly specified by core. > > A trivial change to get rid of unnecessary dealiasing of host->data > in omap_hsmmc_do_irq is also included. > > Signed-off-by: Venkatraman S <svenkatr@ti.com> Now it looks good: Reviewed-by: Felipe Balbi <balbi@ti.com> > --- > > v1->v2: > Fix typos in commit message. > Add checks to handle subtle races between MMC IRQ and HRTIMER IRQ > Mark set_guard_timer function as static > (Felipe's comment to use macros for INT_EN_MASK is done as a separate patch ) > drivers/mmc/host/omap_hsmmc.c | 96 ++++++++++++++++++++++--------------------- > 1 file changed, 50 insertions(+), 46 deletions(-) > > diff --git a/drivers/mmc/host/omap_hsmmc.c b/drivers/mmc/host/omap_hsmmc.c > index 9afdd20..57e86a4 100644 > --- a/drivers/mmc/host/omap_hsmmc.c > +++ b/drivers/mmc/host/omap_hsmmc.c > @@ -79,7 +79,7 @@ > #define CLKD_SHIFT 6 > #define DTO_MASK 0x000F0000 > #define DTO_SHIFT 16 > -#define INT_EN_MASK 0x307F0033 > +#define INT_EN_MASK 0x306E0033 > #define BWR_ENABLE (1 << 4) > #define BRR_ENABLE (1 << 5) > #define DTO_ENABLE (1 << 20) > @@ -160,6 +160,7 @@ struct omap_hsmmc_host { > unsigned int dma_sg_idx; > unsigned char bus_mode; > unsigned char power_mode; > + unsigned int ns_per_clk_cycle; > int suspended; > int irq; > int use_dma, dma_ch; > @@ -172,6 +173,7 @@ struct omap_hsmmc_host { > int reqs_blocked; > int use_reg; > int req_in_progress; > + struct hrtimer guard_timer; > struct omap_hsmmc_next next_data; > > struct omap_mmc_platform_data *pdata; > @@ -455,10 +457,6 @@ static void omap_hsmmc_enable_irq(struct omap_hsmmc_host *host, > else > irq_mask = INT_EN_MASK; > > - /* Disable timeout for erases */ > - if (cmd->opcode == MMC_ERASE) > - irq_mask &= ~DTO_ENABLE; > - > OMAP_HSMMC_WRITE(host->base, STAT, STAT_CLEAR); > OMAP_HSMMC_WRITE(host->base, ISE, irq_mask); > OMAP_HSMMC_WRITE(host->base, IE, irq_mask); > @@ -508,6 +506,9 @@ static void omap_hsmmc_set_clock(struct omap_hsmmc_host *host) > && time_before(jiffies, timeout)) > cpu_relax(); > > + if (ios->clock) > + host->ns_per_clk_cycle = DIV_ROUND_UP(NSEC_PER_SEC, ios->clock); > + > omap_hsmmc_start_clock(host); > } > > @@ -824,7 +825,7 @@ omap_hsmmc_xfer_done(struct omap_hsmmc_host *host, struct mmc_data *data) > omap_hsmmc_request_done(host, mrq); > return; > } > - > + hrtimer_cancel(&host->guard_timer); > host->data = NULL; > > if (!data->error) > @@ -859,8 +860,11 @@ omap_hsmmc_cmd_done(struct omap_hsmmc_host *host, struct mmc_command *cmd) > cmd->resp[0] = OMAP_HSMMC_READ(host->base, RSP10); > } > } > - if ((host->data == NULL && !host->response_busy) || cmd->error) > + if ((host->data == NULL && !host->response_busy) || cmd->error) { > + if (cmd->error != -ETIMEDOUT) > + hrtimer_cancel(&host->guard_timer); > omap_hsmmc_request_done(host, cmd->mrq); > + } > } > > /* > @@ -992,7 +996,7 @@ static void omap_hsmmc_do_irq(struct omap_hsmmc_host *host, int status) > hsmmc_command_incomplete(host, -EILSEQ); > > end_cmd = 1; > - if (host->data || host->response_busy) { > + if (data || host->response_busy) { > end_trans = 1; > host->response_busy = 0; > } > @@ -1292,41 +1296,35 @@ static int omap_hsmmc_start_dma_transfer(struct omap_hsmmc_host *host, > return 0; > } > > -static void set_data_timeout(struct omap_hsmmc_host *host, > - unsigned int timeout_ns, > - unsigned int timeout_clks) > +static void set_guard_timer(struct omap_hsmmc_host *host, > + unsigned long timeout_ms, unsigned long timeout_ns, > + unsigned int timeout_clks) > { > - unsigned int timeout, cycle_ns; > - uint32_t reg, clkd, dto = 0; > + ktime_t gtime; > + unsigned int sec, nsec; > > - reg = OMAP_HSMMC_READ(host->base, SYSCTL); > - clkd = (reg & CLKD_MASK) >> CLKD_SHIFT; > - if (clkd == 0) > - clkd = 1; > + sec = timeout_ms / MSEC_PER_SEC; > + nsec = (timeout_ms % MSEC_PER_SEC) * NSEC_PER_MSEC + timeout_ns; > > - cycle_ns = 1000000000 / (clk_get_rate(host->fclk) / clkd); > - timeout = timeout_ns / cycle_ns; > - timeout += timeout_clks; > - if (timeout) { > - while ((timeout & 0x80000000) == 0) { > - dto += 1; > - timeout <<= 1; > - } > - dto = 31 - dto; > - timeout <<= 1; > - if (timeout && dto) > - dto += 1; > - if (dto >= 13) > - dto -= 13; > - else > - dto = 0; > - if (dto > 14) > - dto = 14; > - } > + nsec += timeout_clks * host->ns_per_clk_cycle; > + gtime = ktime_set(sec, nsec); > > - reg &= ~DTO_MASK; > - reg |= dto << DTO_SHIFT; > - OMAP_HSMMC_WRITE(host->base, SYSCTL, reg); > + hrtimer_start(&host->guard_timer, gtime, HRTIMER_MODE_REL); > +} > + > +static enum hrtimer_restart omap_hsmmc_timedout(struct hrtimer *gtimer) > +{ > + struct omap_hsmmc_host *host; > + host = container_of(gtimer, struct omap_hsmmc_host, guard_timer); > + if (host->req_in_progress) { > + omap_hsmmc_disable_irq(host); > + hsmmc_command_incomplete(host, -ETIMEDOUT); > + host->response_busy = 0; > + omap_hsmmc_cmd_done(host, host->cmd); > + if (host->data) > + omap_hsmmc_xfer_done(host, host->data); > + } > + return HRTIMER_NORESTART; > } > > /* > @@ -1340,18 +1338,22 @@ omap_hsmmc_prepare_data(struct omap_hsmmc_host *host, struct mmc_request *req) > > if (req->data == NULL) { > OMAP_HSMMC_WRITE(host->base, BLK, 0); > - /* > - * Set an arbitrary 100ms data timeout for commands with > - * busy signal. > - */ > - if (req->cmd->flags & MMC_RSP_BUSY) > - set_data_timeout(host, 100000000U, 0); > + if (req->cmd->cmd_timeout_ms == 0) { > + /* > + * Set an arbitrary 100ms data timeout for commands with > + * busy signal. > + */ > + set_guard_timer(host, 100, 0, 0); > + } else { > + set_guard_timer(host, req->cmd->cmd_timeout_ms, 0, 0); > + } > return 0; > } > > OMAP_HSMMC_WRITE(host->base, BLK, (req->data->blksz) > | (req->data->blocks << 16)); > - set_data_timeout(host, req->data->timeout_ns, req->data->timeout_clks); > + set_guard_timer(host, 0, req->data->timeout_ns, > + req->data->timeout_clks); > > if (host->use_dma) { > ret = omap_hsmmc_start_dma_transfer(host, req); > @@ -1921,6 +1923,8 @@ static int __devinit omap_hsmmc_probe(struct platform_device *pdev) > pdata->suspend = omap_hsmmc_suspend_cdirq; > pdata->resume = omap_hsmmc_resume_cdirq; > } > + hrtimer_init(&host->guard_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); > + host->guard_timer.function = omap_hsmmc_timedout; > > omap_hsmmc_disable_irq(host); > > -- > 1.7.11.1.25.g0e18bef >
Venkatraman S <svenkatr@ti.com> writes: > omap hsmmc controller IP has a built in timer that can be programmed to > guard against unresponsive operations. But its range is very narrow, > and the maximum countable time is a few seconds. > > Card maintenance operations like BKOPS and MMC_ERASE and long > stream writes like packed command require timers of order of > several minutes, much beyond the capability of the IP timer. > So get rid of using the IP timer entirely and use kernel's hrtimer > functionality for guarding the device operations. > As part of this change, a workaround that disabled timeouts for > MMC_ERASE command is removed, and the arbitary timing of 100ms > is used only when the timeout is not explicitly specified by core. > > A trivial change to get rid of unnecessary dealiasing of host->data > in omap_hsmmc_do_irq is also included. > > Signed-off-by: Venkatraman S <svenkatr@ti.com> Dumb question: if the timers needed are on the order of minutes, why do you need to use high-resolution timers? I'm guessing the normal kernel-internal timers should suffice here (see <linux/timer.h, init_timer().) Also, if they're on the order of minutes, I assume the actual firing doesn't have to be precise, so deferrable timers can probably be used (and thus help PM by coalescing wakeup events.) If my assumptions are true, it might be worth considering using the normal timers, and use init_timer_deferrable(). Kevin -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sat, Sep 8, 2012 at 3:29 AM, Kevin Hilman <khilman@deeprootsystems.com> wrote: > Venkatraman S <svenkatr@ti.com> writes: > >> omap hsmmc controller IP has a built in timer that can be programmed to >> guard against unresponsive operations. But its range is very narrow, >> and the maximum countable time is a few seconds. >> >> Card maintenance operations like BKOPS and MMC_ERASE and long >> stream writes like packed command require timers of order of >> several minutes, much beyond the capability of the IP timer. >> So get rid of using the IP timer entirely and use kernel's hrtimer >> functionality for guarding the device operations. >> As part of this change, a workaround that disabled timeouts for >> MMC_ERASE command is removed, and the arbitary timing of 100ms >> is used only when the timeout is not explicitly specified by core. >> >> A trivial change to get rid of unnecessary dealiasing of host->data >> in omap_hsmmc_do_irq is also included. >> >> Signed-off-by: Venkatraman S <svenkatr@ti.com> > > Dumb question: if the timers needed are on the order of minutes, why do > you need to use high-resolution timers? I'm guessing the normal > kernel-internal timers should suffice here (see <linux/timer.h, > init_timer().) > Both sub ms and long duration timers are needed, based on the type of operation. The eMMC standard has evolved to define new commands which require such long duration operations. The short ("control") commands do need high precision. > Also, if they're on the order of minutes, I assume the actual firing > doesn't have to be precise, so deferrable timers can probably be used > (and thus help PM by coalescing wakeup events.) > > If my assumptions are true, it might be worth considering using the > normal timers, and use init_timer_deferrable(). > I'll certainly look into this. One way is to use the IP timer for short commands and deferrable timers for long ones - but I reckon it would be messy. I've actually put this patch on hold - as someone pointed out to me that this violates some sections of eMMC spec. > Kevin -- To unsubscribe from this list: send the line "unsubscribe linux-omap" 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/omap_hsmmc.c b/drivers/mmc/host/omap_hsmmc.c index 9afdd20..57e86a4 100644 --- a/drivers/mmc/host/omap_hsmmc.c +++ b/drivers/mmc/host/omap_hsmmc.c @@ -79,7 +79,7 @@ #define CLKD_SHIFT 6 #define DTO_MASK 0x000F0000 #define DTO_SHIFT 16 -#define INT_EN_MASK 0x307F0033 +#define INT_EN_MASK 0x306E0033 #define BWR_ENABLE (1 << 4) #define BRR_ENABLE (1 << 5) #define DTO_ENABLE (1 << 20) @@ -160,6 +160,7 @@ struct omap_hsmmc_host { unsigned int dma_sg_idx; unsigned char bus_mode; unsigned char power_mode; + unsigned int ns_per_clk_cycle; int suspended; int irq; int use_dma, dma_ch; @@ -172,6 +173,7 @@ struct omap_hsmmc_host { int reqs_blocked; int use_reg; int req_in_progress; + struct hrtimer guard_timer; struct omap_hsmmc_next next_data; struct omap_mmc_platform_data *pdata; @@ -455,10 +457,6 @@ static void omap_hsmmc_enable_irq(struct omap_hsmmc_host *host, else irq_mask = INT_EN_MASK; - /* Disable timeout for erases */ - if (cmd->opcode == MMC_ERASE) - irq_mask &= ~DTO_ENABLE; - OMAP_HSMMC_WRITE(host->base, STAT, STAT_CLEAR); OMAP_HSMMC_WRITE(host->base, ISE, irq_mask); OMAP_HSMMC_WRITE(host->base, IE, irq_mask); @@ -508,6 +506,9 @@ static void omap_hsmmc_set_clock(struct omap_hsmmc_host *host) && time_before(jiffies, timeout)) cpu_relax(); + if (ios->clock) + host->ns_per_clk_cycle = DIV_ROUND_UP(NSEC_PER_SEC, ios->clock); + omap_hsmmc_start_clock(host); } @@ -824,7 +825,7 @@ omap_hsmmc_xfer_done(struct omap_hsmmc_host *host, struct mmc_data *data) omap_hsmmc_request_done(host, mrq); return; } - + hrtimer_cancel(&host->guard_timer); host->data = NULL; if (!data->error) @@ -859,8 +860,11 @@ omap_hsmmc_cmd_done(struct omap_hsmmc_host *host, struct mmc_command *cmd) cmd->resp[0] = OMAP_HSMMC_READ(host->base, RSP10); } } - if ((host->data == NULL && !host->response_busy) || cmd->error) + if ((host->data == NULL && !host->response_busy) || cmd->error) { + if (cmd->error != -ETIMEDOUT) + hrtimer_cancel(&host->guard_timer); omap_hsmmc_request_done(host, cmd->mrq); + } } /* @@ -992,7 +996,7 @@ static void omap_hsmmc_do_irq(struct omap_hsmmc_host *host, int status) hsmmc_command_incomplete(host, -EILSEQ); end_cmd = 1; - if (host->data || host->response_busy) { + if (data || host->response_busy) { end_trans = 1; host->response_busy = 0; } @@ -1292,41 +1296,35 @@ static int omap_hsmmc_start_dma_transfer(struct omap_hsmmc_host *host, return 0; } -static void set_data_timeout(struct omap_hsmmc_host *host, - unsigned int timeout_ns, - unsigned int timeout_clks) +static void set_guard_timer(struct omap_hsmmc_host *host, + unsigned long timeout_ms, unsigned long timeout_ns, + unsigned int timeout_clks) { - unsigned int timeout, cycle_ns; - uint32_t reg, clkd, dto = 0; + ktime_t gtime; + unsigned int sec, nsec; - reg = OMAP_HSMMC_READ(host->base, SYSCTL); - clkd = (reg & CLKD_MASK) >> CLKD_SHIFT; - if (clkd == 0) - clkd = 1; + sec = timeout_ms / MSEC_PER_SEC; + nsec = (timeout_ms % MSEC_PER_SEC) * NSEC_PER_MSEC + timeout_ns; - cycle_ns = 1000000000 / (clk_get_rate(host->fclk) / clkd); - timeout = timeout_ns / cycle_ns; - timeout += timeout_clks; - if (timeout) { - while ((timeout & 0x80000000) == 0) { - dto += 1; - timeout <<= 1; - } - dto = 31 - dto; - timeout <<= 1; - if (timeout && dto) - dto += 1; - if (dto >= 13) - dto -= 13; - else - dto = 0; - if (dto > 14) - dto = 14; - } + nsec += timeout_clks * host->ns_per_clk_cycle; + gtime = ktime_set(sec, nsec); - reg &= ~DTO_MASK; - reg |= dto << DTO_SHIFT; - OMAP_HSMMC_WRITE(host->base, SYSCTL, reg); + hrtimer_start(&host->guard_timer, gtime, HRTIMER_MODE_REL); +} + +static enum hrtimer_restart omap_hsmmc_timedout(struct hrtimer *gtimer) +{ + struct omap_hsmmc_host *host; + host = container_of(gtimer, struct omap_hsmmc_host, guard_timer); + if (host->req_in_progress) { + omap_hsmmc_disable_irq(host); + hsmmc_command_incomplete(host, -ETIMEDOUT); + host->response_busy = 0; + omap_hsmmc_cmd_done(host, host->cmd); + if (host->data) + omap_hsmmc_xfer_done(host, host->data); + } + return HRTIMER_NORESTART; } /* @@ -1340,18 +1338,22 @@ omap_hsmmc_prepare_data(struct omap_hsmmc_host *host, struct mmc_request *req) if (req->data == NULL) { OMAP_HSMMC_WRITE(host->base, BLK, 0); - /* - * Set an arbitrary 100ms data timeout for commands with - * busy signal. - */ - if (req->cmd->flags & MMC_RSP_BUSY) - set_data_timeout(host, 100000000U, 0); + if (req->cmd->cmd_timeout_ms == 0) { + /* + * Set an arbitrary 100ms data timeout for commands with + * busy signal. + */ + set_guard_timer(host, 100, 0, 0); + } else { + set_guard_timer(host, req->cmd->cmd_timeout_ms, 0, 0); + } return 0; } OMAP_HSMMC_WRITE(host->base, BLK, (req->data->blksz) | (req->data->blocks << 16)); - set_data_timeout(host, req->data->timeout_ns, req->data->timeout_clks); + set_guard_timer(host, 0, req->data->timeout_ns, + req->data->timeout_clks); if (host->use_dma) { ret = omap_hsmmc_start_dma_transfer(host, req); @@ -1921,6 +1923,8 @@ static int __devinit omap_hsmmc_probe(struct platform_device *pdev) pdata->suspend = omap_hsmmc_suspend_cdirq; pdata->resume = omap_hsmmc_resume_cdirq; } + hrtimer_init(&host->guard_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + host->guard_timer.function = omap_hsmmc_timedout; omap_hsmmc_disable_irq(host);
omap hsmmc controller IP has a built in timer that can be programmed to guard against unresponsive operations. But its range is very narrow, and the maximum countable time is a few seconds. Card maintenance operations like BKOPS and MMC_ERASE and long stream writes like packed command require timers of order of several minutes, much beyond the capability of the IP timer. So get rid of using the IP timer entirely and use kernel's hrtimer functionality for guarding the device operations. As part of this change, a workaround that disabled timeouts for MMC_ERASE command is removed, and the arbitary timing of 100ms is used only when the timeout is not explicitly specified by core. A trivial change to get rid of unnecessary dealiasing of host->data in omap_hsmmc_do_irq is also included. Signed-off-by: Venkatraman S <svenkatr@ti.com> --- v1->v2: Fix typos in commit message. Add checks to handle subtle races between MMC IRQ and HRTIMER IRQ Mark set_guard_timer function as static (Felipe's comment to use macros for INT_EN_MASK is done as a separate patch ) drivers/mmc/host/omap_hsmmc.c | 96 ++++++++++++++++++++++--------------------- 1 file changed, 50 insertions(+), 46 deletions(-)