You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
openvmm: wide RTC port reads leave stale bytes in the upper portion of the result #1151
RTC PIO reads wider than one byte return the register byte together with stale upper bytes from the destination buffer, despite the intercept's documented zero-extension behavior.
In vm/devices/chipset/src/cmos_rtc.rs, Rtc::io_read writes only data[0]:
// We assume all accesses are one byte in size. Attempts to// access larger sizes will return a single byte of information// (zero-extended to the size of the access).
data[0] = matchRtcIoPort(io_port){RtcIoPort::ADDR => self.state.addr,RtcIoPort::DATA => self.get_cmos_byte(self.state.addr),
_ => returnIoResult::Err(IoError::InvalidRegister),};IoResult::Ok
For a two- or four-byte destination, the remaining bytes are never overwritten. On a reused PIO exit buffer, those bytes can come from an earlier operation rather than the current RTC read.
Reproduction
Write 0x0d to address port 0x70 to select RTC status D, then read data port 0x71 with a 16- or 32-bit PIO access. The low byte is status D's 0x80, but the upper bytes retain the read buffer's previous contents instead of producing 0x0080 or 0x00000080.
A wide read of address port 0x70 has the same problem: only the selected CMOS index replaces the previous buffer contents.
Fix
fix.patch determines the register value, clears the entire destination, then writes the value to its first byte. Invalid-register handling and one-byte reads remain unchanged.
Patch against OpenVMM 998b9f2402050cd16d1faa897bc831ef1434be48. Save the contents below as fix.patch and run git apply fix.patch.
fix.patch
diff --git a/vm/devices/chipset/src/cmos_rtc.rs b/vm/devices/chipset/src/cmos_rtc.rs
index ae091aec8..44fb21532 100644
--- a/vm/devices/chipset/src/cmos_rtc.rs+++ b/vm/devices/chipset/src/cmos_rtc.rs@@ -897,11 +897,13 @@ fn io_read(&mut self, io_port: u16, data: &mut [u8]) -> IoResult {
// We assume all accesses are one byte in size. Attempts to
// access larger sizes will return a single byte of information
// (zero-extended to the size of the access).
- data[0] = match RtcIoPort(io_port) {+ let value = match RtcIoPort(io_port) {
RtcIoPort::ADDR => self.state.addr,
RtcIoPort::DATA => self.get_cmos_byte(self.state.addr),
_ => return IoResult::Err(IoError::InvalidRegister),
};
+ data.fill(0);+ data[0] = value;
IoResult::Ok
}
@@ -1564,4 +1566,59 @@ fn hms_to_duration(h: u8, m: u8, s: u8) -> Duration {
// TODO: test some more alarm scenarios
}
++ /// Regression test of BUG-07: a read of port 0x70 or 0x71 wider than a+ /// byte returns the byte zero-extended to the access size, as `io_read`'s+ /// comment says. Before the fix only the first byte was written and the+ /// others kept the buffer's content (KVM's shared `kvm_run` PIO area,+ /// which vmotherboard passes as is), so a 16- or 32-bit `in` could return+ /// bytes left by an earlier exit.+ #[test]+ fn rtc_wide_read_is_zero_extended() {+ let (_, _, _, mut rtc) = new_test_rtc();+ let (addr, data_port) = (RtcIoPort::ADDR.0, RtcIoPort::DATA.0);+ for (port, len) in [+ (addr, 2usize),+ (data_port, 2),+ (addr, 4),+ (data_port, 4),+ (addr, 8),+ (data_port, 8),+ ] {+ rtc.io_write(addr, &[CmosReg::STATUS_D.0]).unwrap();+ let mut data = vec![0xa5; len];+ rtc.io_read(port, &mut data).unwrap();+ // The register (the index 0x0d, or status D's VRT bit), then zeros.+ assert_eq!(data[0], if port == addr { 0x0d } else { 0x80 });+ assert!(+ data[1..].iter().all(|&b| b == 0),+ "{port:#x} x{len}: {data:x?}"+ );+ }+ }++ /// BUG-07's fix keeps one-byte reads: port 0x70 returns the index+ /// register, port 0x71 the selected register (RAM bytes as written, status+ /// B as written, status D's VRT bit, status C's flags), as before.+ #[test]+ fn rtc_one_byte_reads_are_unchanged() {+ let (_, _, _, mut rtc) = new_test_rtc();+ for (index, value) in [(0x0eu8, 0x5au8), (0x40, 0xa7), (0x7f, 0x01), (0x0b, 0x06)] {+ set_cmos_data(&mut rtc, CmosReg(index), value);+ let mut data = [0xa5];+ rtc.io_read(RtcIoPort::ADDR.0, &mut data).unwrap();+ assert_eq!(data, [index]);+ rtc.io_read(RtcIoPort::DATA.0, &mut data).unwrap();+ assert_eq!(data, [value], "index {index:#x}");+ }+ rtc.io_write(RtcIoPort::ADDR.0, &[CmosReg::STATUS_D.0])+ .unwrap();+ let mut data = [0xa5];+ rtc.io_read(RtcIoPort::DATA.0, &mut data).unwrap();+ assert_eq!(data, [0x80]);+ rtc.io_write(RtcIoPort::ADDR.0, &[CmosReg::STATUS_C.0])+ .unwrap();+ rtc.io_read(RtcIoPort::DATA.0, &mut data).unwrap();+ assert_eq!(data, [0]);+ }
}
Activity
tangruize commentedon Nov 1, 2026
Proposed patch
Patch against OpenVMM
998b9f2402050cd16d1faa897bc831ef1434be48. Save the contents below asfix.patchand rungit apply fix.patch.fix.patchopenvmm: pin the RTC's zero-extended wide reads
chipset: zero-extend wide reads of the RTC's ports