Regression: 8250 with 2.6.24

Regression: 8250 with 2.6.24

[ael]

Legacy i386 system with Intel 440BX chip set:
------------------------------------------------------------------------
# lspci
00:00.0 Host bridge: Intel Corporation 440BX/ZX/DX - 82443BX/ZX/DX Host
bridge (rev 02)
00:01.0 PCI bridge: Intel Corporation 440BX/ZX/DX - 82443BX/ZX/DX AGP
bridge (rev 02)
00:07.0 ISA bridge: Intel Corporation 82371AB/EB/MB PIIX4 ISA (rev 02)
00:07.1 IDE interface: Intel Corporation 82371AB/EB/MB PIIX4 IDE (rev 01)
00:07.2 USB Controller: Intel Corporation 82371AB/EB/MB PIIX4 USB (rev 01)
00:07.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 02)
00:08.0 SCSI storage controller: Adaptec AHA-2940U/UW/D / AIC-7881U (rev 01)
00:0a.0 Ethernet controller: Lite-On Communications Inc LNE100TX (rev 20)
01:00.0 VGA compatible controller: Matrox Graphics, Inc. MGA G200 AGP
(rev 01)
----------------------------------------------------------------------

Moving from 2.6.23.14 to 2.6.24, the /dev/ttyS* are incorrectly created:-

In /var/log/messages:-
----------------------------------------------------------------------
Feb  4 17:01:54 conquest3 kernel: serial8250: ttyS0 at I/O 0x3f8 (irq =
4) is a 16550A
Feb  4 17:01:54 conquest3 kernel: serial8250: ttyS1 at I/O 0x2f8 (irq =
3) is a 16550A
Feb  4 17:01:54 conquest3 kernel: 00:0c: ttyS0 at I/O 0x3f8 (irq = 4) is
a 16550A
Feb  4 17:01:54 conquest3 kernel: 00:0f: ttyS1 at I/O 0x2f8 (irq = 3) is
a 16550A
------------------------------------------------------------------------
[which is correct.]

However:-
#  ls -l /sys/devices/platform/serial8250/tty/
total 0
drwxr-xr-x 3 root root 0 Feb  8 18:22 ttyS2
drwxr-xr-x 3 root root 0 Feb  8 18:22 ttyS3
[which is incorrect.]

udev then creates the wrong devices:-
# ls -l /dev/ttyS*
crw-rw---- 1 root dialout 4, 66 Feb  8 18:03 /dev/ttyS2
crw-rw---- 1 root dialout 4, 67 Feb  8 18:03 /dev/ttyS3

Accessing those nodes gives errors:

# statserial /dev/ttyS3
statserial: TIOCMGET failed: Input/output error

Complications:
--------------

1) This is non deterministic: very rarely all four devices appear:
ttyS0,ttyS1, ttyS2 and ttyS3. Accessing ttyS0 and ttys1 then works
correctly.

2) 2.6.23.14 (and probably earlier kernels) also generated 4 devices,
and rarely had a similar problem - but I can't recall exact details.
Thus the bug might have been around for a longish period with some sort
of race triggered almost always in 2.6.24.
-------------------------------------------------------------------------

Anyone throw any light on this?

ael

Re: Regression: 8250 with 2.6.24

[ael]

For the record, this bug seems to be associated with 8250_pnp.

Under 2.6.24.2, if 8250_pnp is blacklisted and 8250_pci is explicity
loaded while booting, then /dev/ttyS{0,1,2,3} are created and accessing
ttyS0 and ttyS1 works.

Given the original nondetermination, I cannot be sure this is reliable,
but it has worked in every test so far.

I am not clear why 8250_pci does not get loaded automatically.

A Lawrence