Honor CBAUD/CIBAUD in TCGETS, TCGETS2 and TCSETS

[jotpalch]

Note

Piece 1 of #310, sent separately per the discussion there.

Summary

Plain TCSETS dropped the baud rate and TCGETS reported B0. TCGETS/TCGETS2/TCSETS now encode and decode CBAUD/CIBAUD like the Linux tty core, so tcsetattr(B115200) from a musl or glibc <= 2.41 guest reaches the host line and cfgetospeed() round-trips.

Important

What to look at:

1. apply_linux_cflag_speeds(): the two values that must not touch the host speed: B0 in CBAUD and a bare BOTHER. The kernel resolves BOTHER on a plain TCSETS from the tty's currentc_ospeed (drivers/tty/tty_ioctl.cset_termios starts from tty->termios; drivers/tty/tty_baudrate.ctty_termios_baud_rate). An earlier draft decoded it to 0 and collapsed a 74880-baud line to B0 on the next tcgetattr/tcsetattr pair (see evidence).
2. linux_cflag_speed_bits(): CIBAUD is set only when input != output, and TCGETS2 now uses the same encoding instead of always BOTHER, matching tty_termios_encode_baud_rate().
3. TCSETS2 decodes CIBAUD symmetrically (B0 = same as output, BOTHER = numeric c_ispeed, else table): since TCGETS2 now reports a split rate as CBAUD+CIBAUD indexes rather than the old blanket BOTHER, an unchanged tcgetattr/tcsetattr pair from glibc 2.42 would otherwise collapse the input rate to the output rate (found in self-review on the staging fork, Honor CBAUD/CIBAUD in TCGETS, TCGETS2 and TCSETS jotpalch/elfuse#1).

Problem

	observed
src/syscall/io.c TCSETS arm (io.c:2289-2316 at bffd6bd)	never calls cfsetispeed/cfsetospeed; only the TCSETS2 arm did
TCGETS	never encodes the host rate into CBAUD, so cfgetospeed() == 0
who is affected	musl (always) and glibc <= 2.41 issue plain TCSETS with the rate in CBAUD; glibc 2.42+ switched to TCGETS2/TCSETS2 unconditionally (sysdeps/unix/sysv/linux/tcsetattr.c, kernel-features.h__ASSUME_TERMIOS2)
why it stayed hidden	on a host pty the rate is meaningless; it becomes visible the moment a guest opens a real serial port, which the /dev passthrough already allows (/dev/cu.usbmodem*)

Change

• io.c: linux_cflag_speed_bits() (host termios -> CBAUD/CIBAUD bits) and apply_linux_cflag_speeds() (Linux c_cflag -> cfsetospeed/cfsetispeed), used by the TCGETS, TCGETS2 and TCSETS/TCSETSW/TCSETSF arms. LINUX_IBSHIFT added next to LINUX_CBAUD.
• tests/test-pty.c: slave-side block (5 checks): B115200 round trip; unchanged termios keeps the rate; 74880 set via raw TCSETS2 survives a plain tcgetattr/tcsetattr pair; split rate through CIBAUD and re-unification; TCGETS/TCGETS2 agree.

Kernel references (Linux v7.2): include/uapi/asm-generic/termbits.h (CBAUD 0x100f, BOTHER 0x1000, CIBAUD = CBAUD << 16, B115200 0x1002), drivers/tty/tty_baudrate.c (tty_termios_baud_rate, tty_termios_input_baud_rate, tty_termios_encode_baud_rate), drivers/tty/tty_ioctl.c (set_termios, get_termios). x86_64 (Rosetta) guests share the asm-generic layout.

Evidence

Harness: open a macOS pty pair, preset the host side to 9600, hand the slave to a static aarch64 guest that does tcsetattr(B115200) and reads it back.

glibc 2.41 guest, before (bffd6bd) vs after

before:

== 3. baud rate round-trip (TCSETS/TCSETS2 depending on libc) ==
tcsetattr(B115200) = 0
cfgetospeed -> 0 (want 4098)
== 4. line / modem-control ioctls ==
TIOCMGET = -1Inappropriate ioctl for device bits=0x0
TIOCMBIS(DTR|RTS)= -1Inappropriate ioctl for device
TIOCMBIC(DTR|RTS)= -1Inappropriate ioctl for device
tcflush/TCFLSH = -1Inappropriate ioctl for device
tcdrain/TCSBRK = -1Inappropriate ioctl for device
TIOCOUTQ = -1Inappropriate ioctl for device outq=-1
TIOCEXCL = -1Inappropriate ioctl for device
host-side speed after guest tcsetattr(B115200): ispeed=9600 ospeed=9600

after:

== 3. baud rate round-trip (TCSETS/TCSETS2 depending on libc) ==
tcsetattr(B115200) = 0
cfgetospeed -> 4098 (want 4098)
== 4. line / modem-control ioctls ==
TIOCMGET = -1Inappropriate ioctl for device bits=0x0
TIOCMBIS(DTR|RTS)= -1Inappropriate ioctl for device
TIOCMBIC(DTR|RTS)= -1Inappropriate ioctl for device
tcflush/TCFLSH = 0
tcdrain/TCSBRK = 0
TIOCOUTQ = 0 outq=0
TIOCEXCL = 0
host-side speed after guest tcsetattr(B115200): ispeed=115200 ospeed=115200

musl 1.2.5 guest, after

== 3. baud rate round-trip (TCSETS/TCSETS2 depending on libc) ==
tcsetattr(B115200) = 0
cfgetospeed -> 4098 (want 4098)
== 4. line / modem-control ioctls ==
TIOCMGET = -1Not a tty bits=0x0
TIOCMBIS(DTR|RTS)= -1Not a tty
TIOCMBIC(DTR|RTS)= -1Not a tty
tcflush/TCFLSH = 0
tcdrain/TCSBRK = 0
TIOCOUTQ = 0 outq=0
TIOCEXCL = 0
host-side speed after guest tcsetattr(B115200): ispeed=115200 ospeed=115200

74880 (BOTHER) regression: draft vs final

draft (decoded bare BOTHER to 0):

host pty slave = /dev/ttys002, host speed preset to ispeed=74880 ospeed=74880
guest tcgetattr: cfgetospeed=4096 (BOTHER=0x1000) c_cflag&CBAUD=0x1000
guest tcsetattr(unchanged speed) = 0 TCFLSH(TCIFLUSH=0) = 0 TCFLSH(3 invalid) = -1 Invalid argument (want EINVAL)
TCXONC(TCOON=1) = 0 TCXONC(4 invalid) = -1 Invalid argument (want EINVAL)
TCSBRKP(0) = 0 TIOCNXCL = 0 host-side speed after guest round-trip: ispeed=0 ospeed=0 (want 74880)

final:

host pty slave = /dev/ttys002, host speed preset to ispeed=74880 ospeed=74880
guest tcgetattr: cfgetospeed=4096 (BOTHER=0x1000) c_cflag&CBAUD=0x1000
guest tcsetattr(unchanged speed) = 0 TCFLSH(TCIFLUSH=0) = 0 TCFLSH(3 invalid) = -1 Invalid argument (want EINVAL)
TCXONC(TCOON=1) = 0 TCXONC(4 invalid) = -1 Invalid argument (want EINVAL)
TCSBRKP(0) = 0 TIOCNXCL = 0 host-side speed after guest round-trip: ispeed=74880 ospeed=74880 (want 74880)

native macOS build of the same program, same pty (control)

== 3. baud rate round-trip (TCSETS/TCSETS2 depending on libc) ==
tcsetattr(B115200) = 0
cfgetospeed -> 115200 (want 115200)
== 4. line / modem-control ioctls ==
TIOCMGET = -1Inappropriate ioctl for device bits=0x0
TIOCMBIS(DTR|RTS)= -1Inappropriate ioctl for device
TIOCMBIC(DTR|RTS)= -1Inappropriate ioctl for device
tcflush/TCFLSH = 0
tcdrain/TCSBRK = 0
TIOCOUTQ = 0 outq=0
TIOCEXCL = 0
host pty /dev/ttys002
host-side speed after native tcsetattr: 115200 115200

real hardware: ESP32-S3 USB-Serial/JTAG (/dev/cu.usbmodem101), before vs after

before:

== 2. open(/dev/cu.usbmodem101) ==
open = 3 ok
== 3. baud rate round-trip (TCSETS/TCSETS2 depending on libc) ==
tcsetattr(B115200) = 0
cfgetospeed -> 0 (want 4098)
== 4. line / modem-control ioctls ==
TIOCMGET = -1Inappropriate ioctl for device bits=0x0
TIOCMBIS(DTR|RTS)= -1Inappropriate ioctl for device
TIOCMBIC(DTR|RTS)= -1Inappropriate ioctl for device
tcflush/TCFLSH = -1Inappropriate ioctl for device
tcdrain/TCSBRK = -1Inappropriate ioctl for device
TIOCOUTQ = -1Inappropriate ioctl for device outq=-1
TIOCEXCL = -1Inappropriate ioctl for device

after (this PR + piece 2):

== 2. open(/dev/cu.usbmodem101) ==
open = 3 ok
== 3. baud rate round-trip (TCSETS/TCSETS2 depending on libc) ==
tcsetattr(B115200) = 0
cfgetospeed -> 4098 (want 4098)
== 4. line / modem-control ioctls ==
TIOCMGET = 0 bits=0x6
TIOCMBIS(DTR|RTS)= 0
TIOCMBIC(DTR|RTS)= 0
tcflush/TCFLSH = 0
tcdrain/TCSBRK = 0
TIOCOUTQ = 0 outq=0
TIOCEXCL = 0

Tests

• build/elfuse build/test-pty: 63 passed, 0 failed (includes the split-rate TCGETS2/TCSETS2 regression test)
• make check BAREMETAL_CROSS=aarch64-elf- LINUX_TOOLCHAIN=/opt/homebrew/opt/aarch64-unknown-linux-gnu -j8: rc=0, All 77 tests passed (fixture-dependent musl/dyn-glibc cases skipped as on a stock checkout)
• clang-format --dry-run -Werror clean

Not in this PR

• The line/modem ioctls (TCFLSH, TCSBRK, TIOCM*, ...) -> piece 2 (stacked PR).
• B0 hang-up (drop DTR/RTS) is not emulated; documented in the comment.

---

Summary by cubic

Honor CBAUD/CIBAUD in TCGETS, TCGETS2, and TCSETS so baud rates round-trip correctly. Previously TCSETS ignored baud bits and TCGETS reported B0; now standard, custom (BOTHER), and split input rates are preserved.

• TCGETS/TCGETS2: encode host ospeed/ispeed into CBAUD and CIBAUD like the Linux tty core; TCGETS2 no longer forces BOTHER and still fills numeric speeds.
• TCSETS/TCSETS2: apply CBAUD/CIBAUD; CIBAUD B0 means “same as output”; B0 in CBAUD and a bare BOTHER leave the host speed unchanged; Linux’s B0 hang-up is not emulated.
• Tests: cover B115200 round-trip via plain TCSETS, preserving a BOTHER-only rate (74880) across tcgetattr/tcsetattr, split-rate round-trip and re-unification, and TCGETS vs TCGETS2 agreement.

^{Written for commit 3a6c880. Summary will update on new commits.}

[jserv]

Rebase latest main branch as #313 installs Git hooks and a CI gate to enforce commit‑message rules and staged checks.

[jotpalch]

Rebased onto main at 7cc000b (#313). Both commits pass the new gate (scripts/check-commit-log.sh) and scripts/test-git-hooks.sh locally; comment reflow clean, make check 77/77, test-pty 70/70 on the stacked pair.

[jserv]

Thank @jotpalch for contributing!

[jotpalch]

Thank you! This is my first contribution to your projects and I'm thrilled it landed. The review gates here (commentflow, the commit-log hooks, the EINTR contract checker) taught me a lot along the way. Piece 2 of #310 follows next.