Bug#453749: Fix serial console detection on ia64
Thanks for your responses. As I was writing this report, I began to wonder why exactly the proc trick doesn't work on some ia64 configs - I'm investigating this now. If I can find a workable solution there, we may be able to forego these changes - if not, I'll regenerate new patches based on your comments. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#453749: Fix serial console detection on ia64
Frans Pop [EMAIL PROTECTED] writes: On Friday 30 November 2007, dann frazier wrote: The attached patch to di-utils adds a slightly modified version of I compiled this on amd64 and after stripping it is still 4k. As it is only needed during finish install, I wonder if it should be in di-utils and thus in initrds. Maybe we should create a di-utils-extra udeb that includes utils normally needed only after anna has been run. This could also include in-target for example. Not sure though. I like this idea. -- O T A V I OS A L V A D O R - E-mail: [EMAIL PROTECTED] UIN: 5906116 GNU/Linux User: 239058 GPG ID: 49A5F855 Home Page: http://otavio.ossystems.com.br - Microsoft sells you Windows ... Linux gives you the whole house. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#453749: Fix serial console detection on ia64
On Friday 30 November 2007, dann frazier wrote: # ./serial-console-info DEV=ttyS2 SPEED=115200 One more thing... Why not just: device:\tttyS2 speed:\t115200 IMO that would be more in line with how other utils output stuff. Current output looks a bit like envvars, but isn't. signature.asc Description: This is a digitally signed message part.
Bug#453749: Fix serial console detection on ia64
Package: finish-install Version: 2.13 Severity: important Tags: patch If you boot an ia64 system[1] without specifying a console= parameter, the kernel will try to map /dev/console to whatever settings were detected from EFI (firmware interface). If EFI uses a serial console, /dev/console will be mapped to that same serial device. d-i interacts with the user via /dev/console, so most of the time the installer doesn't really care what the underlying device is. That is, until the system needs to decide where it should run gettys, and where root logins are permitted. This works is done by 90console. 90console attempts to detect the real console device by dereferencing /proc/di-pid/fd/0. If a system was booted w/ console=ttyS0, /proc/di-pid/fd/0 will link to /dev/ttyS0. However, on certain ia64 platforms[2], this link points to the ever-so-helpful /dev/console. The only way we've found to identify the correct device is to use a small C program that issues an ioctl to the device to retrieve the line number and speed (thanks to Bryan Stillwell for this research/work[3]). The workaround for this in etch is to always specify a console= parameter, thereby disabling the automatic console setup. The attached patch to di-utils adds a slightly modified version of Bryan's utility. This utility was originally called getserialdev - I've renamed it to serial-console-info, since I think it better describes the function. Also, getserialdev's output was of the form: # ./getserialdev ttyS2 115200 I've modified this to instead tag the data, so that it should be easier to maintain backwards compatability should this utility provide additional output in the future. e.g.: # ./serial-console-info DEV=ttyS2 SPEED=115200 The attached patch to finish-install adds support for using serial-console-info, if available, to 90console. Processing behaves just as before, until it finds that /proc/di-pid/fd/0 is pointing at /dev/console. It will only then run serial-console-info to see if it can determine that /dev/console is actually a serial device. I briefly considered implementing serial console detection entirely with serial-console-info - but decided against it. The major problem is that serial-console-info assumes that any device it finds has a ttyS prefix, which may not always be correct. [1] Possibly other EFI-based systems as well, I haven't verified [2] an rx1600 w/ an MP for one [3] https://launchpad.net/bugs/48752 -- dann frazier -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#453749: Fix serial console detection on ia64
On Friday 30 November 2007, dann frazier wrote: The attached patch to di-utils adds a slightly modified version of I compiled this on amd64 and after stripping it is still 4k. As it is only needed during finish install, I wonder if it should be in di-utils and thus in initrds. Maybe we should create a di-utils-extra udeb that includes utils normally needed only after anna has been run. This could also include in-target for example. Not sure though. Bryan's utility. This utility was originally called getserialdev - I've renamed it to serial-console-info, since I think it better describes the function. If I run the app on my system, I get an error code 255. It's probably correct that it returns an error, but I wonder if the value 255 is correct? The attached patch to finish-install adds support for using serial-console-info, if available, to 90console. Processing behaves just as before, until it finds that /proc/di-pid/fd/0 is pointing at /dev/console. It will only then run serial-console-info to see if it can determine that /dev/console is actually a serial device. I think that, because of the ttyS assumption this patch may break hvsi/hvc virtualized console support (bottom half of 90console). This should be checked. I don't think there's any need to write the output of s-c-i to a temp file. Just assigning it to a variable should work just as well: +if [ $rawconsole == /dev/console ] \ + [ -x /bin/serial-console-info ] \ + sci=$(/bin/serial-console-info); then + rawconsole=/dev/$(echo $sci | grep ^DEV= | cut -d= -f2) + sci_baud=$(echo $sci | grep ^SPEED= | cut -d= -f2) +fi Cheers, FJP signature.asc Description: This is a digitally signed message part.
