Repository navigation
Closed
Task
18 / 1818 of 18 issues completed
Copy link
Description
Objective. OpenVMM's emulated PIC, PIT, CMOS RTC, and IOAPIC behave like the hardware they model, never panic on a guest access or a restored state, and keep their timing across a snapshot. The microVM's fixed chipset holds exactly these devices beside the portb, shutdown, and snapshot-request devices (openvmm/openvmm_defs/src/microvm.rs), so each bug here reaches microVMs.
Children.
| Device | Issue | Scope | Blocked by |
|---|---|---|---|
| PIC | #1062, #1139 | A poll-mode command read leaves the cascade and ready outputs stale, so the next INTA panics. | — |
| PIT | #1063, #1146 | Low/high access mode keeps the buffered low byte, so the next count is wrong. | — |
| PIT | #1065, #1162 | evaluate drops sub-100 ns time, so the PIT runs faster the more often the guest reads it. |
— |
| PIT | #1066, #1147 | Restore accepts a control word with RW = 0, and the next counter access reaches unreachable!. |
— |
| PIT | #1067 | Timer::evaluate takes time proportional to the elapsed periods after idle time. |
— |
| RTC | #1143 | 12-hour mode reports noon as AM and rejects 12 PM writes. | — |
| RTC | #1151 | Wide port reads leave stale upper bytes. | — |
| RTC | #1152 | The alarm calculation misdecodes wildcards and wraps field borrows. | — |
| RTC | #1164 | Periodic catch-up after downtime takes one iteration per missed period. | — |
| RTC | #1163 | Save and restore lose sub-second deadlines, the UIP phase, and stopped-boundary alignment. | — |
| RTC | #1376 | Restoring a clock time near i64::MIN or i64::MAX, or a downtime that brings it there, overflows the RTC clock's offset from host time. |
— |
| IOAPIC | #1140 | 64-bit MMIO reads panic in IoApicDevice::mmio_read. |
— |
| IOAPIC | #1172 | Restore accepts an id above 15, and the next ID-register read panics. | — |
| IOAPIC | #1400 | On MSHV, the control console's interrupt sometimes never reaches the guest, so interrupt-map fails; a lost edge on its edge-triggered input while the virtio-mmio line stays high is the likely cause. |
— |
Order. No child blocks another. Outside this umbrella, #516 (the one-year downtime limit) is blocked by #1164.
Notes.
- openvmm: RTC alarm calculation misdecodes wildcards and wraps time-field borrows #1152, openvmm: CMOS RTC snapshot save/restore omits sub-second timer deadlines, UIP phase, and stopped-boundary alignment #1163, and openvmm: RTC periodic catch-up stalls restore with billions of iterations #1164 all change the RTC's alarm and timer scheduling, and openvmm: RTC periodic catch-up stalls restore with billions of iterations #1164 leaves the timers' phase across a restore to openvmm: CMOS RTC snapshot save/restore omits sub-second timer deadlines, UIP phase, and stopped-boundary alignment #1163; land them one at a time.
- openvmm: CMOS RTC snapshot save/restore omits sub-second timer deadlines, UIP phase, and stopped-boundary alignment #1163 and openvmm: restoring an extreme snapshot RTC time overflows the RTC clock #1376 both change the RTC's restore: openvmm: CMOS RTC snapshot save/restore omits sub-second timer deadlines, UIP phase, and stopped-boundary alignment #1163 what it saves, and openvmm: restoring an extreme snapshot RTC time overflows the RTC clock #1376 the clock times that it accepts and its downtime advance (
advance_clock); land them one at a time. - Four pairs are duplicates, filed on 2026-10-31 and again on 2026-11-01: openvmm: a secondary PIC poll leaves a stale cascade request and the next INTA can panic #1139 duplicates chipset: PIC poll-mode read does not re-sync outputs; the next INTA panics with "pic1 ready was set" #1062, openvmm: consecutive PIT low/high writes reuse the preceding count's low byte #1146 duplicates chipset: PIT LSB/MSB access mode keeps the buffered LSB after the MSB write #1063, openvmm: frequent PIT reads make the counter run about 4.75% fast #1162 duplicates PIT
evaluatedrops sub-100ns time, so the PIT runs fast and same-instant reads differ #1065, and openvmm: PIT restore accepts RW=0 and subsequent counter I/O can panic #1147 duplicates chipset: PIT restore accepts a control word with rw = 0, then hitsunreachable!()on the next access #1066. One fix closes each pair. - At NVX's OpenVMM pin,
Timer::evaluatealready skips whole periods in modes 2 and 3 (testlong_elapsed_interval_fast_forwards_periodic_pit), so check whether PITTimer::evaluateis O(elapsed ticks / count); one PIO access after idle time can take hundreds of ms #1067 still reproduces onfeature/microvm/dev.
Metadata
Metadata
Assignees
Labels
No labels
Activity
evaluatedrops sub-100ns time, so the PIT runs fast and same-instant reads differ #1065unreachable!()on the next access #1066Timer::evaluateis O(elapsed ticks / count); one PIO access after idle time can take hundreds of ms #1067ppenna commentedon Nov 14, 2026
Closing: all child issues are resolved, and their fixes are present in origin/dev. NVX PRs #1551, #1557, #1558, #1563, #1565, #1568, #1569, #1570, #1578, and #1410 are merged. The current OpenVMM pin includes upstream fixes microsoft/openvmm#4586, #4838, #4959, #4963, and #4971. #516 remains open, but was explicitly listed as outside this umbrella's scope.