| Message ID | 20230628191243.3632401-1-linus.walleij@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mmc: mmci: Improve ux500 debug prints | expand |
On Wed, 28 Jun 2023 at 21:12, Linus Walleij <linus.walleij@linaro.org> wrote: > > To conclude the ux500 busy timeout fixes, this improves the debug and > error prints so we can see a bit what is going on. Here is a typical > dmesg with these new debug messages enabled: > > [ 2.648864] mmci-pl18x 80005000.mmc: mmc2: PL180 manf 80 rev4 > at 0x80005000 irq 81,0 (pio) > [ 2.662750] mmci-pl18x 80005000.mmc: DMA channels RX dma0chan4, TX dma0chan5 > [ 3.480407] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 > [ 3.487457] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 > [ 3.998321] mmci-pl18x 80005000.mmc: timeout in state waiting for end IRQ > waiting for busy CMD06 > [ 3.998535] mmc2: new DDR MMC card at address 0001 > [ 4.000030] mmcblk2: mmc2:0001 M4G1YC 3.69 GiB > [ 4.008361] mmcblk2: p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12 p13 p14 p15 > p16 p17 p18 p19 p20 p21 p22 p23 p24 p25 > [ 4.017700] mmcblk2boot0: mmc2:0001 M4G1YC 2.00 MiB > [ 4.020477] mmcblk2boot1: mmc2:0001 M4G1YC 2.00 MiB > [ 4.022125] mmcblk2rpmb: mmc2:0001 M4G1YC 128 KiB, chardev (246:0) > [ 5.791381] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 > [ 10.938568] mmci-pl18x 80005000.mmc: timeout in state waiting for end IRQ > waiting for busy CMD06 > [ 17.982849] mmci-pl18x 80005000.mmc: lost busy status when waiting for > busy start IRQ CMD06 > [ 18.683563] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 > [ 19.385437] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 > [ 20.493652] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 > > We see a lot of lost IRQs and the timeout always occur while waiting for > the end IRQ, and then the busy status is *low* meaning the busy indication > is already de-asserted. > > So busy signalling is missed in various ways for various reasons, > sometimes it appears that IRQs are simply lost. > > One hypothesis is that this happens because the events happen so fast > that they are transient, and since the MMCI state machine in effect is > handling an edge trigger (rising or falling signal on DAT0) the > internal logic will miss the event, because the state machine in the > hardware is sampling the line, and will at times detect only the first > event but miss the second, fireing only one IRQ. > > We print the second timeout error with dev_err() since it is pretty > serious, the other events are so common and simple to handle that we > can keep them at dev_dbg() level. > > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Applied for next, thanks! Kind regards Uffe > --- > drivers/mmc/host/mmci.c | 37 ++++++++++++++++++++++++++++++------- > 1 file changed, 30 insertions(+), 7 deletions(-) > > diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c > index 4fd28eaadc07..5d05269a8567 100644 > --- a/drivers/mmc/host/mmci.c > +++ b/drivers/mmc/host/mmci.c > @@ -734,7 +734,8 @@ static bool ux500_busy_complete(struct mmci_host *host, struct mmc_command *cmd, > } > retries--; > } > - dev_dbg(mmc_dev(host->mmc), "no busy signalling in time\n"); > + dev_dbg(mmc_dev(host->mmc), > + "no busy signalling in time CMD%02x\n", cmd->opcode); > ux500_busy_clear_mask_done(host); > break; > > @@ -756,7 +757,8 @@ static bool ux500_busy_complete(struct mmci_host *host, struct mmc_command *cmd, > host->busy_state = MMCI_BUSY_WAITING_FOR_END_IRQ; > } else { > dev_dbg(mmc_dev(host->mmc), > - "lost busy status when waiting for busy start IRQ\n"); > + "lost busy status when waiting for busy start IRQ CMD%02x\n", > + cmd->opcode); > cancel_delayed_work(&host->ux500_busy_timeout_work); > ux500_busy_clear_mask_done(host); > } > @@ -770,13 +772,14 @@ static bool ux500_busy_complete(struct mmci_host *host, struct mmc_command *cmd, > ux500_busy_clear_mask_done(host); > } else { > dev_dbg(mmc_dev(host->mmc), > - "busy status still asserted when handling busy end IRQ - will keep waiting\n"); > + "busy status still asserted when handling busy end IRQ - will keep waiting CMD%02x\n", > + cmd->opcode); > } > break; > > default: > - dev_dbg(mmc_dev(host->mmc), "fell through on state %d\n", > - host->busy_state); > + dev_sbg(mmc_dev(host->mmc), "fell through on state %d, CMD%02x\n", > + host->busy_state, cmd->opcode); > break; > } > > @@ -1503,6 +1506,20 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd, > } > } > > +static char *ux500_state_str(struct mmci_host *host) > +{ > + switch (host->busy_state) { > + case MMCI_BUSY_WAITING_FOR_START_IRQ: > + return "waiting for start IRQ"; > + case MMCI_BUSY_WAITING_FOR_END_IRQ: > + return "waiting for end IRQ"; > + case MMCI_BUSY_DONE: > + return "not waiting for IRQs"; > + default: > + return "unknown"; > + } > +} > + > /* > * This busy timeout worker is used to "kick" the command IRQ if a > * busy detect IRQ fails to appear in reasonable time. Only used on > @@ -1522,9 +1539,15 @@ static void ux500_busy_timeout_work(struct work_struct *work) > /* If we are still busy, let's tag on timeout error. */ > if (status & host->variant->busy_detect_flag) { > status |= MCI_CMDTIMEOUT; > - dev_dbg(mmc_dev(host->mmc), "timeout waiting for busy\n"); > + dev_err(mmc_dev(host->mmc), > + "timeout in state %s still busy with CMD%02x\n", > + ux500_state_str(host), > + host->cmd->opcode); > } else { > - dev_dbg(mmc_dev(host->mmc), "timeout waiting for busy IRQ\n"); > + dev_err(mmc_dev(host->mmc), > + "timeout in state %s waiting for busy CMD%02x\n", > + ux500_state_str(host), > + host->cmd->opcode); > } > > mmci_cmd_irq(host, host->cmd, status); > -- > 2.40.1 >
diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c index 4fd28eaadc07..5d05269a8567 100644 --- a/drivers/mmc/host/mmci.c +++ b/drivers/mmc/host/mmci.c @@ -734,7 +734,8 @@ static bool ux500_busy_complete(struct mmci_host *host, struct mmc_command *cmd, } retries--; } - dev_dbg(mmc_dev(host->mmc), "no busy signalling in time\n"); + dev_dbg(mmc_dev(host->mmc), + "no busy signalling in time CMD%02x\n", cmd->opcode); ux500_busy_clear_mask_done(host); break; @@ -756,7 +757,8 @@ static bool ux500_busy_complete(struct mmci_host *host, struct mmc_command *cmd, host->busy_state = MMCI_BUSY_WAITING_FOR_END_IRQ; } else { dev_dbg(mmc_dev(host->mmc), - "lost busy status when waiting for busy start IRQ\n"); + "lost busy status when waiting for busy start IRQ CMD%02x\n", + cmd->opcode); cancel_delayed_work(&host->ux500_busy_timeout_work); ux500_busy_clear_mask_done(host); } @@ -770,13 +772,14 @@ static bool ux500_busy_complete(struct mmci_host *host, struct mmc_command *cmd, ux500_busy_clear_mask_done(host); } else { dev_dbg(mmc_dev(host->mmc), - "busy status still asserted when handling busy end IRQ - will keep waiting\n"); + "busy status still asserted when handling busy end IRQ - will keep waiting CMD%02x\n", + cmd->opcode); } break; default: - dev_dbg(mmc_dev(host->mmc), "fell through on state %d\n", - host->busy_state); + dev_sbg(mmc_dev(host->mmc), "fell through on state %d, CMD%02x\n", + host->busy_state, cmd->opcode); break; } @@ -1503,6 +1506,20 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd, } } +static char *ux500_state_str(struct mmci_host *host) +{ + switch (host->busy_state) { + case MMCI_BUSY_WAITING_FOR_START_IRQ: + return "waiting for start IRQ"; + case MMCI_BUSY_WAITING_FOR_END_IRQ: + return "waiting for end IRQ"; + case MMCI_BUSY_DONE: + return "not waiting for IRQs"; + default: + return "unknown"; + } +} + /* * This busy timeout worker is used to "kick" the command IRQ if a * busy detect IRQ fails to appear in reasonable time. Only used on @@ -1522,9 +1539,15 @@ static void ux500_busy_timeout_work(struct work_struct *work) /* If we are still busy, let's tag on timeout error. */ if (status & host->variant->busy_detect_flag) { status |= MCI_CMDTIMEOUT; - dev_dbg(mmc_dev(host->mmc), "timeout waiting for busy\n"); + dev_err(mmc_dev(host->mmc), + "timeout in state %s still busy with CMD%02x\n", + ux500_state_str(host), + host->cmd->opcode); } else { - dev_dbg(mmc_dev(host->mmc), "timeout waiting for busy IRQ\n"); + dev_err(mmc_dev(host->mmc), + "timeout in state %s waiting for busy CMD%02x\n", + ux500_state_str(host), + host->cmd->opcode); } mmci_cmd_irq(host, host->cmd, status);
To conclude the ux500 busy timeout fixes, this improves the debug and error prints so we can see a bit what is going on. Here is a typical dmesg with these new debug messages enabled: [ 2.648864] mmci-pl18x 80005000.mmc: mmc2: PL180 manf 80 rev4 at 0x80005000 irq 81,0 (pio) [ 2.662750] mmci-pl18x 80005000.mmc: DMA channels RX dma0chan4, TX dma0chan5 [ 3.480407] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 [ 3.487457] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 [ 3.998321] mmci-pl18x 80005000.mmc: timeout in state waiting for end IRQ waiting for busy CMD06 [ 3.998535] mmc2: new DDR MMC card at address 0001 [ 4.000030] mmcblk2: mmc2:0001 M4G1YC 3.69 GiB [ 4.008361] mmcblk2: p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12 p13 p14 p15 p16 p17 p18 p19 p20 p21 p22 p23 p24 p25 [ 4.017700] mmcblk2boot0: mmc2:0001 M4G1YC 2.00 MiB [ 4.020477] mmcblk2boot1: mmc2:0001 M4G1YC 2.00 MiB [ 4.022125] mmcblk2rpmb: mmc2:0001 M4G1YC 128 KiB, chardev (246:0) [ 5.791381] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 [ 10.938568] mmci-pl18x 80005000.mmc: timeout in state waiting for end IRQ waiting for busy CMD06 [ 17.982849] mmci-pl18x 80005000.mmc: lost busy status when waiting for busy start IRQ CMD06 [ 18.683563] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 [ 19.385437] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 [ 20.493652] mmci-pl18x 80005000.mmc: no busy signalling in time CMD06 We see a lot of lost IRQs and the timeout always occur while waiting for the end IRQ, and then the busy status is *low* meaning the busy indication is already de-asserted. So busy signalling is missed in various ways for various reasons, sometimes it appears that IRQs are simply lost. One hypothesis is that this happens because the events happen so fast that they are transient, and since the MMCI state machine in effect is handling an edge trigger (rising or falling signal on DAT0) the internal logic will miss the event, because the state machine in the hardware is sampling the line, and will at times detect only the first event but miss the second, fireing only one IRQ. We print the second timeout error with dev_err() since it is pretty serious, the other events are so common and simple to handle that we can keep them at dev_dbg() level. Signed-off-by: Linus Walleij <linus.walleij@linaro.org> --- drivers/mmc/host/mmci.c | 37 ++++++++++++++++++++++++++++++------- 1 file changed, 30 insertions(+), 7 deletions(-)