From: Paolo Bonzini <pbonzini@redhat.com>
To: qemu-devel@nongnu.org
Cc: Tai Yunfang <yunfangtai@tencent.com>,
Xiao Guangrong <xiaoguangrong@tencent.com>
Subject: [Qemu-devel] [PULL 02/20] mc146818rtc: precisely count the clock for periodic timer
Date: Fri, 19 May 2017 13:20:54 +0200 [thread overview]
Message-ID: <1495192872-27667-3-git-send-email-pbonzini@redhat.com> (raw)
In-Reply-To: <1495192872-27667-1-git-send-email-pbonzini@redhat.com>
From: Tai Yunfang <yunfangtai@tencent.com>
There are two issues in current code:
1) If the period is changed by re-configuring RegA, the coalesced
irq will be scaled to reflect the new period, however, it
calculates the new interrupt number like this:
s->irq_coalesced = (s->irq_coalesced * s->period) / period;
There are some clocks will be lost if they are not enough to
be squeezed to a single new period that will cause the VM clock
slower
In order to fix the issue, we calculate the interrupt window
based on the precise clock rather than period, then the clocks
lost during period is scaled can be compensated properly
2) If periodic_timer_update() is called due to RegA reconfiguration,
i.e, the period is updated, current time is not the start point
for the next periodic timer, instead, which should start from the
last interrupt, otherwise, the clock in VM will become slow
This patch takes the clocks from last interrupt to current clock
into account and compensates the clocks for the next interrupt,
especially if a complete interrupt was lost in this window, the
time can be caught up by LOST_TICK_POLICY_SLEW
Signed-off-by: Tai Yunfang <yunfangtai@tencent.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@tencent.com>
Message-Id: <20170510083259.3900-3-xiaoguangrong@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
hw/timer/mc146818rtc.c | 120 +++++++++++++++++++++++++++++++++++++++----------
1 file changed, 97 insertions(+), 23 deletions(-)
diff --git a/hw/timer/mc146818rtc.c b/hw/timer/mc146818rtc.c
index 7d78391..aeb60cc 100644
--- a/hw/timer/mc146818rtc.c
+++ b/hw/timer/mc146818rtc.c
@@ -146,31 +146,100 @@ static void rtc_coalesced_timer(void *opaque)
}
#endif
-/* handle periodic timer */
-static void periodic_timer_update(RTCState *s, int64_t current_time)
+static uint32_t rtc_periodic_clock_ticks(RTCState *s)
{
- int period_code, period;
- int64_t cur_clock, next_irq_clock;
+ int period_code;
+
+ if (!(s->cmos_data[RTC_REG_B] & REG_B_PIE)) {
+ return 0;
+ }
period_code = s->cmos_data[RTC_REG_A] & 0x0f;
- if (period_code != 0
- && (s->cmos_data[RTC_REG_B] & REG_B_PIE)) {
- if (period_code <= 2)
- period_code += 7;
- /* period in 32 Khz cycles */
- period = 1 << (period_code - 1);
-#ifdef TARGET_I386
- if (period != s->period) {
- s->irq_coalesced = (s->irq_coalesced * s->period) / period;
- DPRINTF_C("cmos: coalesced irqs scaled to %d\n", s->irq_coalesced);
- }
- s->period = period;
-#endif
+ if (!period_code) {
+ return 0;
+ }
+
+ if (period_code <= 2) {
+ period_code += 7;
+ }
+
+ /* period in 32 Khz cycles */
+ return 1 << (period_code - 1);
+}
+
+/*
+ * handle periodic timer. @old_period indicates the periodic timer update
+ * is just due to period adjustment.
+ */
+static void
+periodic_timer_update(RTCState *s, int64_t current_time, uint32_t old_period)
+{
+ uint32_t period;
+ int64_t cur_clock, next_irq_clock, lost_clock = 0;
+
+ period = rtc_periodic_clock_ticks(s);
+
+ if (period) {
/* compute 32 khz clock */
cur_clock =
muldiv64(current_time, RTC_CLOCK_RATE, NANOSECONDS_PER_SECOND);
- next_irq_clock = (cur_clock & ~(period - 1)) + period;
+ /*
+ * if the periodic timer's update is due to period re-configuration,
+ * we should count the clock since last interrupt.
+ */
+ if (old_period) {
+ int64_t last_periodic_clock, next_periodic_clock;
+
+ next_periodic_clock = muldiv64(s->next_periodic_time,
+ RTC_CLOCK_RATE, NANOSECONDS_PER_SECOND);
+ last_periodic_clock = next_periodic_clock - old_period;
+ lost_clock = cur_clock - last_periodic_clock;
+ assert(lost_clock >= 0);
+ }
+
+#ifdef TARGET_I386
+ /*
+ * s->irq_coalesced can change for two reasons:
+ *
+ * a) if one or more periodic timer interrupts have been lost,
+ * lost_clock will be more that a period.
+ *
+ * b) when the period may be reconfigured, we expect the OS to
+ * treat delayed tick as the new period. So, when switching
+ * from a shorter to a longer period, scale down the missing,
+ * because the OS will treat past delayed ticks as longer
+ * (leftovers are put back into lost_clock). When switching
+ * to a shorter period, scale up the missing ticks since the
+ * OS handler will treat past delayed ticks as shorter.
+ */
+ if (s->lost_tick_policy == LOST_TICK_POLICY_SLEW) {
+ uint32_t old_irq_coalesced = s->irq_coalesced;
+
+ s->period = period;
+ lost_clock += old_irq_coalesced * old_period;
+ s->irq_coalesced = lost_clock / s->period;
+ lost_clock %= s->period;
+ if (old_irq_coalesced != s->irq_coalesced ||
+ old_period != s->period) {
+ DPRINTF_C("cmos: coalesced irqs scaled from %d to %d, "
+ "period scaled from %d to %d\n", old_irq_coalesced,
+ s->irq_coalesced, old_period, s->period);
+ rtc_coalesced_timer_update(s);
+ }
+ } else
+#endif
+ {
+ /*
+ * no way to compensate the interrupt if LOST_TICK_POLICY_SLEW
+ * is not used, we should make the time progress anyway.
+ */
+ lost_clock = MIN(lost_clock, period);
+ }
+
+ assert(lost_clock >= 0 && lost_clock <= period);
+
+ next_irq_clock = cur_clock + period - lost_clock;
s->next_periodic_time = muldiv64(next_irq_clock, NANOSECONDS_PER_SECOND,
RTC_CLOCK_RATE) + 1;
timer_mod(s->periodic_timer, s->next_periodic_time);
@@ -186,7 +255,7 @@ static void rtc_periodic_timer(void *opaque)
{
RTCState *s = opaque;
- periodic_timer_update(s, s->next_periodic_time);
+ periodic_timer_update(s, s->next_periodic_time, 0);
s->cmos_data[RTC_REG_C] |= REG_C_PF;
if (s->cmos_data[RTC_REG_B] & REG_B_PIE) {
s->cmos_data[RTC_REG_C] |= REG_C_IRQF;
@@ -391,6 +460,7 @@ static void cmos_ioport_write(void *opaque, hwaddr addr,
uint64_t data, unsigned size)
{
RTCState *s = opaque;
+ uint32_t old_period;
bool update_periodic_timer;
if ((addr & 1) == 0) {
@@ -425,6 +495,7 @@ static void cmos_ioport_write(void *opaque, hwaddr addr,
break;
case RTC_REG_A:
update_periodic_timer = (s->cmos_data[RTC_REG_A] ^ data) & 0x0f;
+ old_period = rtc_periodic_clock_ticks(s);
if ((data & 0x60) == 0x60) {
if (rtc_running(s)) {
@@ -450,7 +521,8 @@ static void cmos_ioport_write(void *opaque, hwaddr addr,
(s->cmos_data[RTC_REG_A] & REG_A_UIP);
if (update_periodic_timer) {
- periodic_timer_update(s, qemu_clock_get_ns(rtc_clock));
+ periodic_timer_update(s, qemu_clock_get_ns(rtc_clock),
+ old_period);
}
check_update_timer(s);
@@ -458,6 +530,7 @@ static void cmos_ioport_write(void *opaque, hwaddr addr,
case RTC_REG_B:
update_periodic_timer = (s->cmos_data[RTC_REG_B] ^ data)
& REG_B_PIE;
+ old_period = rtc_periodic_clock_ticks(s);
if (data & REG_B_SET) {
/* update cmos to when the rtc was stopping */
@@ -487,7 +560,8 @@ static void cmos_ioport_write(void *opaque, hwaddr addr,
s->cmos_data[RTC_REG_B] = data;
if (update_periodic_timer) {
- periodic_timer_update(s, qemu_clock_get_ns(rtc_clock));
+ periodic_timer_update(s, qemu_clock_get_ns(rtc_clock),
+ old_period);
}
check_update_timer(s);
@@ -757,7 +831,7 @@ static int rtc_post_load(void *opaque, int version_id)
uint64_t now = qemu_clock_get_ns(rtc_clock);
if (now < s->next_periodic_time ||
now > (s->next_periodic_time + get_max_clock_jump())) {
- periodic_timer_update(s, qemu_clock_get_ns(rtc_clock));
+ periodic_timer_update(s, qemu_clock_get_ns(rtc_clock), 0);
}
}
@@ -822,7 +896,7 @@ static void rtc_notify_clock_reset(Notifier *notifier, void *data)
int64_t now = *(int64_t *)data;
rtc_set_date_from_host(ISA_DEVICE(s));
- periodic_timer_update(s, now);
+ periodic_timer_update(s, now, 0);
check_update_timer(s);
#ifdef TARGET_I386
if (s->lost_tick_policy == LOST_TICK_POLICY_SLEW) {
--
1.8.3.1
next prev parent reply other threads:[~2017-05-19 11:21 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-05-19 11:20 [Qemu-devel] [PULL 00/20] Misc patches for 2017-05-19 Paolo Bonzini
2017-05-19 11:20 ` [Qemu-devel] [PULL 01/20] mc146818rtc: update periodic timer only if it is needed Paolo Bonzini
2017-05-19 11:20 ` Paolo Bonzini [this message]
2017-05-19 11:20 ` [Qemu-devel] [PULL 03/20] mc146818rtc: ensure LOST_TICK_POLICY_SLEW is only enabled on TARGET_I386 Paolo Bonzini
2017-05-19 11:20 ` [Qemu-devel] [PULL 04/20] mc146818rtc: drop unnecessary '#ifdef TARGET_I386' Paolo Bonzini
2017-05-19 11:20 ` [Qemu-devel] [PULL 05/20] mc146818rtc: embrace all x86 specific code Paolo Bonzini
2017-05-19 11:20 ` [Qemu-devel] [PULL 06/20] kvm: irqchip: trace changes on msi add/remove Paolo Bonzini
2017-05-19 11:20 ` [Qemu-devel] [PULL 07/20] msix: trace control bit write op Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 08/20] kvm: irqchip: skip update msi when disabled Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 09/20] Check the return value of fcntl in qemu_set_cloexec Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 10/20] nbd: strict nbd_wr_syncv Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 11/20] nbd: read_sync and friends: return 0 on success Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 12/20] nbd: add errp parameter to nbd_wr_syncv() Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 13/20] nbd: add errp to read_sync, write_sync and drop_sync Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 14/20] nbd/client.c: use errp instead of LOG Paolo Bonzini
2017-05-26 11:09 ` [Qemu-devel] [PATCH v2] " Vladimir Sementsov-Ogievskiy
2017-05-26 13:43 ` Eric Blake
2017-05-19 11:21 ` [Qemu-devel] [PULL 15/20] exec: simplify phys_page_find() params Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 16/20] virtio-scsi: Unset hotplug handler when unrealize Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 17/20] vhost-user-scsi: Introduce vhost-user-scsi host device Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 18/20] vhost-user-scsi: Introduce a vhost-user-scsi sample application Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 19/20] target/i386: enable A20 automatically in system management mode Paolo Bonzini
2017-05-19 11:21 ` [Qemu-devel] [PULL 20/20] target/i386: use multiple CPU AddressSpaces Paolo Bonzini
2017-05-19 12:41 ` [Qemu-devel] [PULL 00/20] Misc patches for 2017-05-19 no-reply
2017-05-19 15:51 ` Stefan Hajnoczi
2017-05-19 16:09 ` Paolo Bonzini
2017-05-19 15:49 ` Stefan Hajnoczi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1495192872-27667-3-git-send-email-pbonzini@redhat.com \
--to=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=xiaoguangrong@tencent.com \
--cc=yunfangtai@tencent.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).