It was <2021-03-16 wto 13:32>, when Alexandre Belloni wrote: > On 16/03/2021 13:12:08+0100, Lukasz Stelmach wrote: >> It was <2021-03-15 pon 23:01>, when Alexandre Belloni wrote: >> > Hello, >> > >> > On 05/03/2021 18:44:11+0100, Łukasz Stelmach wrote: >> >> For an RTC without an IRQ assigned rtc_update_irq_enable() should >> >> return -EINVAL. It will, when uie_unsupported is set. >> >> >> > >> > I'm surprised this is an issue because the current code seems to cover >> > all cases: >> > >> > - no irq and not wakeup-source => set_alarm should fail >> > - no irq and wakeup-source => uie_unsupported is set >> > - irq => UIE should work >> > >> > Can you elaborate on your failing use case? >> >> I've got ds3231 which supports alarms[1] but is not connected to any >> interrupt line. Hence, client->irq is 0 as well as want_irq[2]. There >> is also no other indirect connection, so I don't set wakeup-source >> property and ds1307_can_wakeup_device remains[3] false. Under these >> conditions >> >> want_irq = 0 >> ds1307_can_wakeup_device = false >> >> uie_unsupported remains[4] false. And this is the problem. >> >> hwclock(8) when setting system clock from rtc (--hctosys) calls >> synchronize_to_clock_tick_rtc()[5]. There goes >> >> ioctl(rtc_fd, RTC_UIE_ON, 0); >> >> which leads us to >> >> rtc_update_irq_enable(rtc, 1); >> >> and finally here [6] >> >> if (rtc->uie_unsupported) { >> err = -EINVAL; >> goto out; >> } >> >> and we keep going (uie_unsupported = 0). All the following operations >> succeed because chip supports alarms. >> > > But then, HAS_ALARM is not set and ds1337_set_alarm should fail which > makes rtc_timer_enqueue return an error. I admit this whole part is a > mess, I'm just trying to understand how you can hit that. OK, you are right. The problem seems to be elsewhere. How about this scnario? We call rtc_update_irq_enable(). We read rtc with __rtc_read_time() and calculate the alarm time. We get through rtc_timer_enqueue() and down to __rtc_set_alarm(). We loose the race condition (I can do it, I've got really slow connection to DS3231) and we return -ETIME from __rtc_set_alarm() if (scheduled <= now) return -ETIME; and 0 from rtc_timer_enqueue() and the very same zero from rtc_update_irq_enable(). The caller of ioctl() thinks they can expect interrupts when, in fact, they won't receive any. The really weird stuff happens in rtc_timer_do_work(). For the timer to be dequeued __rtc_set_alarm() needs to return EINVAL three times in a row. In my setup this doesn't happen and the code keeps running loops around "reporogram" and "again" labels. With my patch we never risk the above race condition between __rtc_read_time() in rtc_update_irq_enable() and the one in __rtc_set_alarm(), because we know rtc doesn't support alarms before we start the race. In fact there is another race between __rtc_read_time() and actually setting the alarm in the chip. IMHO the solution is to introduce RTC_HAS_ALARM flag for struct rtc_device and check it at the very beginning of __rtc_set_alarm() the same way it is being done in ds1337_set_alarm(). What are your thoughts? >> We go back to hwclock(8) and we start waiting[7] for the update from >> interrupt which never arrives instead of calling >> busywiat_for_rtc_clock_tick()[8] (mind the invalid indentation) because >> of EINVAL returned from ioctl() (conf. [6]) >> >> [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/rtc/rtc-ds1307.c?h=v5.11#n1032 >> [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/rtc/rtc-ds1307.c?h=v5.11#n1779 >> [3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/rtc/rtc-ds1307.c?h=v5.11#n1802 >> [4] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/rtc/rtc-ds1307.c?h=v5.11#n1977 >> [5] https://git.kernel.org/pub/scm/utils/util-linux/util-linux.git/tree/sys-utils/hwclock-rtc.c?h=v2.36.2#n252 >> [6] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/rtc/interface.c?h=v5.11#n564 >> [7] https://git.kernel.org/pub/scm/utils/util-linux/util-linux.git/tree/sys-utils/hwclock-rtc.c?h=v2.36.2#n283 >> [8] https://git.kernel.org/pub/scm/utils/util-linux/util-linux.git/tree/sys-utils/hwclock-rtc.c?h=v2.36.2#n297 >> >> >> Signed-off-by: Łukasz Stelmach >> >> --- >> >> drivers/rtc/rtc-ds1307.c | 14 +++++++------- >> >> 1 file changed, 7 insertions(+), 7 deletions(-) >> >> >> >> diff --git a/drivers/rtc/rtc-ds1307.c b/drivers/rtc/rtc-ds1307.c >> >> index cd8e438bc9c4..b08a9736fa77 100644 >> >> --- a/drivers/rtc/rtc-ds1307.c >> >> +++ b/drivers/rtc/rtc-ds1307.c >> >> @@ -1973,13 +1973,6 @@ static int ds1307_probe(struct i2c_client *client, >> >> if (IS_ERR(ds1307->rtc)) >> >> return PTR_ERR(ds1307->rtc); >> >> >> >> - if (ds1307_can_wakeup_device && !want_irq) { >> >> - dev_info(ds1307->dev, >> >> - "'wakeup-source' is set, request for an IRQ is disabled!\n"); >> >> - /* We cannot support UIE mode if we do not have an IRQ line */ >> >> - ds1307->rtc->uie_unsupported = 1; >> >> - } >> >> - >> >> if (want_irq) { >> >> err = devm_request_threaded_irq(ds1307->dev, client->irq, NULL, >> >> chip->irq_handler ?: ds1307_irq, >> >> @@ -1993,6 +1986,13 @@ static int ds1307_probe(struct i2c_client *client, >> >> } else { >> >> dev_dbg(ds1307->dev, "got IRQ %d\n", client->irq); >> >> } >> >> + } else { >> >> + if (ds1307_can_wakeup_device) >> >> + dev_info(ds1307->dev, >> >> + "'wakeup-source' is set, request for an IRQ is disabled!\n"); >> >> + >> > >> > Honestly, just drop this message, it should have been removed by 82e2d43f6315 >> > >> > >> >> Done. >> >> >> + /* We cannot support UIE mode if we do not have an IRQ line */ >> >> + ds1307->rtc->uie_unsupported = 1; >> >> } >> >> >> >> ds1307->rtc->ops = chip->rtc_ops ?: &ds13xx_rtc_ops; >> >> -- >> >> 2.26.2 >> >> >> >> -- >> Łukasz Stelmach >> Samsung R&D Institute Poland >> Samsung Electronics -- Łukasz Stelmach Samsung R&D Institute Poland Samsung Electronics