Re: Skip kbd-chooser when running under Xen
Ian Campbell <[EMAIL PROTECTED]> writes: > On Thu, 2008-05-29 at 15:04 +0200, Ferenc Wagner wrote: >> Ian Campbell <[EMAIL PROTECTED]> writes: >> >>> On Mon, 2008-05-26 at 12:38 +0200, Ferenc Wagner wrote: >>> >>> Will this patch do the right thing if Xen is being installed using the >>> paravirutal framebuffer device present in 2.6.26-rc? It looks to me as >>> if it will not. >> >> Good point, I don't know anything about this new stuff. What do you >> think the best scenario would be? Will that framebuffer be the >> default when there's nothing specified on the kernel command line? > > Unfortunately it is more complex than that... If I understand correctly > (and I'm not sure I do) with recent kernels (i.e. 2.6.26-rcN) the PVFB > will be the default if the device is configured for that domain (i.e. in > the xm config file in domain 0), otherwise hvc0 will be the default. > >> Does it make keyboard configuration sensible or even necessary? > > I'm not *sure* but I think when using PVFB keymap configuration does > make sense. I've found this old thread only. It's muddy, for the least. http://lists.xensource.com/archives/html/xen-devel/2007-01/msg00383.html >> Anyway, would adding a check for console=hvc0 (in conjuction with the >> present Xen check, so not to break the real PowerPC console) suffice? >> >> Or is there a possibility of the hvc console becoming the default? > > Unfortunately yes, it is already the case in the git repo I think. We will need a more generic approach then. But I don't know a way to detect the current console device. Maybe a specific ioctl, which is supported by hvc only? Or is it directly available under /sys? On the other hand, there's not much to gain here and perhaps we'd better wait until these things stabilize. -- Cheers, Feri. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
On Thu, 2008-05-29 at 15:04 +0200, Ferenc Wagner wrote: > Ian Campbell <[EMAIL PROTECTED]> writes: > > > On Mon, 2008-05-26 at 12:38 +0200, Ferenc Wagner wrote: > > > > Will this patch do the right thing if Xen is being installed using the > > paravirutal framebuffer device present in 2.6.26-rc? It looks to me as > > if it will not. > > Good point, I don't know anything about this new stuff. What do you > think the best scenario would be? Will that framebuffer be the > default when there's nothing specified on the kernel command line? Unfortunately it is more complex than that... If I understand correctly (and I'm not sure I do) with recent kernels (i.e. 2.6.26-rcN) the PVFB will be the default if the device is configured for that domain (i.e. in the xm config file in domain 0), otherwise hvc0 will be the default. > Does it make keyboard configuration sensible or even necessary? I'm not *sure* but I think when using PVFB keymap configuration does make sense. > Anyway, would adding a check for console=hvc0 (in conjuction with the > present Xen check, so not to break the real PowerPC console) suffice? > Or is there a possibility of the hvc console becoming the default? Unfortunately yes, it is already the case in the git repo I think. Ian. -- Ian Campbell Current Noise: Isis - Lines Across Eyes If men acted after marriage as they do during courtship, there would be fewer divorces -- and more bankruptcies. -- Frances Rodman -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
Ian Campbell <[EMAIL PROTECTED]> writes: > On Mon, 2008-05-26 at 12:38 +0200, Ferenc Wagner wrote: > > Will this patch do the right thing if Xen is being installed using the > paravirutal framebuffer device present in 2.6.26-rc? It looks to me as > if it will not. Good point, I don't know anything about this new stuff. What do you think the best scenario would be? Will that framebuffer be the default when there's nothing specified on the kernel command line? Does it make keyboard configuration sensible or even necessary? Anyway, would adding a check for console=hvc0 (in conjuction with the present Xen check, so not to break the real PowerPC console) suffice? Or is there a possibility of the hvc console becoming the default? -- Cheers, Feri. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
On Mon, 2008-05-26 at 12:38 +0200, Ferenc Wagner wrote: > Hi, > > So far I've received no comments on this, may I ask for some again? Will this patch do the right thing if Xen is being installed using the paravirutal framebuffer device present in 2.6.26-rc? It looks to me as if it will not. Ian. > > Thanks, > Feri. > > Ferenc Wagner <[EMAIL PROTECTED]> writes: > > > Frans Pop <[EMAIL PROTECTED]> writes: > > > >> On Thursday 08 May 2008, Ferenc Wagner wrote: > >> > >>> vmlinuz-2.6.18-6-xen-686 suite=etch DEBIAN_FRONTEND=text > >> [...] > >>> kbd-chooser[881]: ERROR **: Couldn't get a file descriptor referring to > >>> the console > >>> main-menu[670]: WARNING **: Configuring 'kbd-chooser' failed with error > >>> code 1 > >> > >> That seems to make fairly clear why kbd-chooser is failing... > > > > Still not fully clear to me, but name it an already fixed Xen bug. > > > >>> vmlinuz-2.6.25-1-xen-686 console=hvc0 suite=etch DEBIAN_FRONTEND=text: > >> > >> And apparently here kbd-chooser _can_ find a file descriptor for the > >> console. > > > > Yes, it looks like hvc0 is a much better console than the earlier > > hacks. Which is not particularly surprising. > > > > * Should we skip running it from a virtualized D-I? (I guess yes, > >especially that it's so simple to achieve.) > >> > >> If I understand you correctly: no, we do not want to add a question for > >> this. > > > > Maybe not. What I'm talking about it the following prompt: > > > > Select a keyboard layout > > > > > > Keymap to use: > > 1. American English [*] 22. Italian > > 2. Belarusian23. Japanese > > [...] > > > > which doesn't appear during serial installation. > > > >>> Yes, testing for functionality would be much better. Like skipping > >>> kbd-chooser if there's no keyboard attached. Is serial installation > >>> handled by checking the kernel command line? > >> > >> I suggest someone starts looking at the kbd-chooser code... > >> For serial console the relevant parts are kbd-chooser.c from line 534 > >> down. > > > > Maybe you want to commit the following fix for a missing hyphen: > > > > Index: packages/kbd-chooser/kbd-chooser.c > > === > > --- packages/kbd-chooser/kbd-chooser.c (revision 53339) > > +++ packages/kbd-chooser/kbd-chooser.c (working copy) > > @@ -552,7 +552,7 @@ > > } > > > > /** > > - * @brief set debian-installer/serial console as to whether we are using a > > serial console > > + * @brief set debian-installer/serial-console as to whether we are using a > > serial console > > * @return 1 if present, 0 if absent, 2 if unknown. > > */ > > sercon_state > > > >> Both UML and serial console are already being detected there. Seems to me > >> that it should be possible to handle Xen in the same or a similar way. > > > > Yes. Line 669 of kbd-chooser.c forces console-tools/archs to > > no-keyboard, just like during UML installation. The latter is > > especially similar to the Xen case, and bases the decision on > > /proc/cpuinfo. This suggests that introducing an analogous function > > which checks for Xen (in the line of [ -d /sys/bus/xen ]) and oring > > that to the above condition would be enough. > > > >> When testing based on hvc0 you should be aware that is also used by > >> PowerPC > >> systems. > > > > This way we wouldn't test for hvc0 at all. Unfortunately, Xen's hvc > > isn't a serial device, so that case can't catch it. > > > >> Bottom line: please implement something that fits _within_ the existing > >> structure of kbd-chooser. Don't add hacks. > > > > What about somethink like this? (Apply before the above!) Totally > > untested, POC only. :) > > > > Index: packages/kbd-chooser/kbd-chooser.c > > === > > --- packages/kbd-chooser/kbd-chooser.c (revision 53339) > > +++ packages/kbd-chooser/kbd-chooser.c (working copy) > > @@ -552,6 +552,26 @@ > > } > > > > /** > > + * @brief set debian-installer/xen-console as to whether we are using a > > Xen console > > + * @return 1 if present, 0 if absent, 2 if unknown. > > + */ > > +sercon_state > > +check_if_xen_console (void) > > +{ > > + sercon_state present = SERIAL_UNKNOWN; > > + struct debconfclient *client = mydebconf_client (); > > + > > + if (!access("/sys/bus/xen", F_OK)) > > + present = SERIAL_PRESENT; > > + else > > + present = SERIAL_ABSENT; > > + > > + debconf_set (client, "debian-installer/xen-console", present ? "true" : > > "false"); > > + di_info ("Setting debian-installer/xen-console to %s", present ? "true" > > : "false"); > > + return present; > > +} > > + > > +/** > > * @brief set debian-installer/serial console as to whether we are using a > > serial console > > * @return 1 if present, 0 if absent, 2 if unknown. > > */ > > @@ -633,6 +653,7 @@ > > int c
Re: Skip kbd-chooser when running under Xen
Frans Pop <[EMAIL PROTECTED]> writes: > On Thursday 08 May 2008, Ferenc Wagner wrote: > >> vmlinuz-2.6.18-6-xen-686 suite=etch DEBIAN_FRONTEND=text > [...] >> kbd-chooser[881]: ERROR **: Couldn't get a file descriptor referring to >> the console >> main-menu[670]: WARNING **: Configuring 'kbd-chooser' failed with error >> code 1 > > That seems to make fairly clear why kbd-chooser is failing... Still not fully clear to me, but name it an already fixed Xen bug. >> vmlinuz-2.6.25-1-xen-686 console=hvc0 suite=etch DEBIAN_FRONTEND=text: > > And apparently here kbd-chooser _can_ find a file descriptor for the > console. Yes, it looks like hvc0 is a much better console than the earlier hacks. Which is not particularly surprising. * Should we skip running it from a virtualized D-I? (I guess yes, especially that it's so simple to achieve.) > > If I understand you correctly: no, we do not want to add a question for > this. Maybe not. What I'm talking about it the following prompt: Select a keyboard layout Keymap to use: 1. American English [*] 22. Italian 2. Belarusian23. Japanese [...] which doesn't appear during serial installation. >> Yes, testing for functionality would be much better. Like skipping >> kbd-chooser if there's no keyboard attached. Is serial installation >> handled by checking the kernel command line? > > I suggest someone starts looking at the kbd-chooser code... > For serial console the relevant parts are kbd-chooser.c from line 534 down. Maybe you want to commit the following fix for a missing hyphen: Index: packages/kbd-chooser/kbd-chooser.c === --- packages/kbd-chooser/kbd-chooser.c (revision 53339) +++ packages/kbd-chooser/kbd-chooser.c (working copy) @@ -552,7 +552,7 @@ } /** - * @brief set debian-installer/serial console as to whether we are using a serial console + * @brief set debian-installer/serial-console as to whether we are using a serial console * @return 1 if present, 0 if absent, 2 if unknown. */ sercon_state > Both UML and serial console are already being detected there. Seems to me > that it should be possible to handle Xen in the same or a similar way. Yes. Line 669 of kbd-chooser.c forces console-tools/archs to no-keyboard, just like during UML installation. The latter is especially similar to the Xen case, and bases the decision on /proc/cpuinfo. This suggests that introducing an analogous function which checks for Xen (in the line of [ -d /sys/bus/xen ]) and oring that to the above condition would be enough. > When testing based on hvc0 you should be aware that is also used by PowerPC > systems. This way we wouldn't test for hvc0 at all. Unfortunately, Xen's hvc isn't a serial device, so that case can't catch it. > Bottom line: please implement something that fits _within_ the existing > structure of kbd-chooser. Don't add hacks. What about somethink like this? (Apply before the above!) Totally untested, POC only. :) Index: packages/kbd-chooser/kbd-chooser.c === --- packages/kbd-chooser/kbd-chooser.c (revision 53339) +++ packages/kbd-chooser/kbd-chooser.c (working copy) @@ -552,6 +552,26 @@ } /** + * @brief set debian-installer/xen-console as to whether we are using a Xen console + * @return 1 if present, 0 if absent, 2 if unknown. + */ +sercon_state +check_if_xen_console (void) +{ + sercon_state present = SERIAL_UNKNOWN; + struct debconfclient *client = mydebconf_client (); + + if (!access("/sys/bus/xen", F_OK)) + present = SERIAL_PRESENT; + else + present = SERIAL_ABSENT; + + debconf_set (client, "debian-installer/xen-console", present ? "true" : "false"); + di_info ("Setting debian-installer/xen-console to %s", present ? "true" : "false"); + return present; +} + +/** * @brief set debian-installer/serial console as to whether we are using a serial console * @return 1 if present, 0 if absent, 2 if unknown. */ @@ -633,6 +653,7 @@ int choices = 0, first_entry_s = 1, first_entry_t = 1; sercon_state sercon; sercon_state umlcon; + sercon_state xencon; struct debconfclient *client = mydebconf_client (); /* k is returned by a method if it is preferred keyboard. @@ -666,7 +687,8 @@ } sercon = check_if_serial_console(); umlcon = check_if_uml_console(); - if (sercon == SERIAL_PRESENT || umlcon == SERIAL_PRESENT) { + xencon = check_if_xen_console(); + if (sercon == SERIAL_PRESENT || umlcon == SERIAL_PRESENT || xencon == SERIAL_PRESENT) { debconf_metaget(client, "kbd-chooser/no-keyboard", "Description"); arch_descr = strdup(client->value); choices++; @@ -687,7 +709,7 @@ debconf_subst (client, "console-t
Re: Skip kbd-chooser when running under Xen
On Thursday 08 May 2008, Ferenc Wagner wrote: > vmlinuz-2.6.18-6-xen-686 suite=etch DEBIAN_FRONTEND=text [...] > kbd-chooser[881]: ERROR **: Couldn't get a file descriptor referring to > the console > main-menu[670]: WARNING **: Configuring 'kbd-chooser' failed with error > code 1 That seems to make fairly clear why kbd-chooser is failing... > vmlinuz-2.6.25-1-xen-686 console=hvc0 suite=etch DEBIAN_FRONTEND=text: And apparently here kbd-chooser _can_ find a file descriptor for the console. > >> So let me raise the issue again (and risk some bikeshed effect): > >> * Does kbd-chooser make any sense under Xen? (I guess not.) > > > > That's completely up to you (plural). > > > >> * Should we skip running it from a virtualized D-I? (I guess yes, > >>especially that it's so simple to achieve.) > > > > What is the difference with the previous point? > > Maybe sparing a question isn't worth this special-casing. That's a > policy question, not a technical one. If I understand you correctly: no, we do not want to add a question for this. > >> * Would a simple [ -d /sys/bus/xen ] check do? > > > > I guess that would be OK. Unless a different test could also catch > > other virtualized environments. > > Unfortunately I don't know anything about other virtualization > environments. > > > I'd suggest at least printing an info message in the syslog when that > > happens. > > > >From kbd-chooser.isinstallable? That would be easy. No. isinstallable is IMO not the correct place to solve this. See below. On Thursday 08 May 2008, Ferenc Wagner wrote: > Yes, testing for functionality would be much better. Like skipping > kbd-chooser if there's no keyboard attached. Is serial installation > handled by checking the kernel command line? I suggest someone starts looking at the kbd-chooser code... For serial console the relevant parts are kbd-chooser.c from line 534 down. Both UML and serial console are already being detected there. Seems to me that it should be possible to handle Xen in the same or a similar way. When testing based on hvc0 you should be aware that is also used by PowerPC systems. Bottom line: please implement something that fits _within_ the existing structure of kbd-chooser. Don't add hacks. Cheers, FJP signature.asc Description: This is a digitally signed message part.
Re: Skip kbd-chooser when running under Xen
On Thu, 2008-05-08 at 15:52 +0200, Ferenc Wagner wrote: > Ian Campbell <[EMAIL PROTECTED]> writes: > > I'm a bit stuck on getting finish-install.d/90console to create a > > getty on hvc0 without a Xen hack but I'll keep thinking... > > Isn't finding console=hvc0 in /proc/cmdline enough? So long it isn't > made the default... I heard some controversy about this issue, but > didn't find a conclusion. Having it on /proc/cmdline is good enough for grub-installer to include the console on the kernel command line. However it only works if you have "--" before "console=" on the installer's kernel command line, it took me a while to figure that out! Since I worked out the "--" thing I haven't re-investigated the getty issue, it's possible it is fixed because of that change to but I don't think so. I also need to figure out how this will interact with pvfb support when that becomes available in the kernel (it's in 2.6.26). Ian. -- Ian Campbell Current Noise: Entombed - Revel In Flesh "This is a job for BOB VIOLENCE and SCUM, the INCREDIBLY STUPID MUTANT DOG." -- Bob Violence -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
Ian Campbell <[EMAIL PROTECTED]> writes: > On Wed, 2008-05-07 at 15:56 +0200, Frans Pop wrote: >> On Tuesday 06 May 2008, Ferenc Wagner wrote: >> > So let me raise the issue again (and risk some bikeshed effect): >> > * Does kbd-chooser make any sense under Xen? (I guess not.) >> >> That's completely up to you (plural). > > I don't think it's necessary to special case kbd-chooser for Xen. I > think the issue was down to the 2.6.18 kernel Ferenc was using since I > never saw the it with 2.6.25 based stuff but I did previously see it > with 2.6.18 when I first played with d-i+xen (ages and ages ago). Yes, the error seems gone for 2.6.25. However, the extra question remained. >>> * Would a simple [ -d /sys/bus/xen ] check do? >> >> I guess that would be OK. Unless a different test could also catch other >> virtualized environments. > > So far I haven't found a need for any Xen specific checks when I've been > working through this stuff (other than obvious things like supporting > xvd devices) and I'd really like to avoid adding any if at all possible. Yes, testing for functionality would be much better. Like skipping kbd-chooser if there's no keyboard attached. Is serial installation handled by checking the kernel command line? > I'm a bit stuck on getting finish-install.d/90console to create a getty > on hvc0 without a Xen hack but I'll keep thinking... Isn't finding console=hvc0 in /proc/cmdline enough? So long it isn't made the default... I heard some controversy about this issue, but didn't find a conclusion. -- Regards, Feri. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
Frans Pop <[EMAIL PROTECTED]> writes: > On Tuesday 06 May 2008, Ferenc Wagner wrote: >> Ian Campbell <[EMAIL PROTECTED]> writes: >> > I've been using console=hvc0 -- I wonder if that has the same effect as >> > console=ttyS0 in terms of skipping the keymap stuff... >> >> Just managed to check this. No, console=hvc0 doesn't skip the keymap >> question, but looks like the kernels which have hvc don't throw and >> error running kbd-chooser. At least that's the case under 2.6.25-xen, >> which I tested a couple of minutes ago. > > What is the output in the syslog from kbd-chooser in both cases (the failing > one and the one with hvc)? vmlinuz-2.6.18-6-xen-686 suite=etch DEBIAN_FRONTEND=text (the kernel messages are those which using libc6-xen would get rid of, usually they cause performance degradation only): May 8 13:35:25 localechooser: info: Falling back to locale 'en_US.UTF-8' May 8 13:35:25 localechooser: info: Set debian-installer/locale = 'en_US.UTF-8' May 8 13:35:25 main-menu[670]: DEBUG: resolver (libc6): package doesn't exist (ignored) May 8 13:35:25 main-menu[670]: INFO: Menu item 'kbd-chooser' selected May 8 13:35:25 main-menu[670]: INFO: Falling back to the package description for console-keymaps-at May 8 13:35:25 main-menu[670]: INFO: Falling back to the package description for console-keymaps-at May 8 13:35:25 kbd-chooser[881]: DEBUG: Mounting usbdevfs to look for kbd May 8 13:35:25 kbd-chooser[881]: DEBUG: Adding generic entry for AT keymaps May 8 13:35:25 kbd-chooser[881]: INFO: keyboard type at: present: false May 8 13:35:25 kbd-chooser[881]: INFO: keyboard type at: present: unknown May 8 13:35:25 kbd-chooser[881]: INFO: Setting debian-installer/serial-console to false May 8 13:35:25 kbd-chooser[881]: INFO: Setting debian-installer/uml-console to false May 8 13:35:25 kbd-chooser[881]: INFO: kbd-chooser: arch at selected May 8 13:35:35 kernel: printk: 1245861 messages suppressed. May 8 13:35:35 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e8ca2c May 8 13:35:35 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e8ca3a May 8 13:35:35 kbd-chooser[881]: INFO: choose_keymap: keymap = us May 8 13:35:35 kbd-chooser[881]: INFO: kbd_chooser: setting keymap us May 8 13:35:35 kbd-chooser[881]: ERROR **: Couldn't get a file descriptor referring to the console May 8 13:35:35 main-menu[670]: WARNING **: Configuring 'kbd-chooser' failed with error code 1 May 8 13:35:35 main-menu[670]: WARNING **: Menu item 'kbd-chooser' failed. May 8 13:35:46 kernel: printk: 520 messages suppressed. May 8 13:35:46 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e736e0 May 8 13:35:46 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e8af09 May 8 13:35:46 main-menu[670]: INFO: Modifying debconf priority limit from 'high' to 'medium' May 8 13:35:46 debconf: Setting debconf/priority to medium May 8 13:35:46 main-menu[670]: DEBUG: resolver (libc6): package doesn't exist (ignored) May 8 13:36:03 kernel: printk: 5574 messages suppressed. May 8 13:36:03 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e6cc68 May 8 13:36:03 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e8af09 May 8 13:36:03 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e8ca2c May 8 13:36:03 kernel: 4gb seg fixup, process debconf (pid 669), cs:ip 73:b7e8ca3a May 8 13:36:03 main-menu[670]: INFO: Menu item 'ethdetect' selected vmlinuz-2.6.25-1-xen-686 console=hvc0 suite=etch DEBIAN_FRONTEND=text: May 8 13:30:05 localechooser: info: Falling back to locale 'en_US.UTF-8' May 8 13:30:05 localechooser: info: Set debian-installer/locale = 'en_US.UTF-8' May 8 13:30:05 main-menu[653]: DEBUG: resolver (libc6): package doesn't exist (ignored) May 8 13:30:05 main-menu[653]: INFO: Menu item 'kbd-chooser' selected May 8 13:30:05 main-menu[653]: INFO: Falling back to the package description for console-keymaps-at May 8 13:30:05 main-menu[653]: INFO: Falling back to the package description for console-keymaps-at May 8 13:30:05 kbd-chooser[864]: DEBUG: Mounting usbdevfs to look for kbd May 8 13:30:05 kbd-chooser[864]: DEBUG: Adding generic entry for AT keymaps May 8 13:30:05 kbd-chooser[864]: INFO: keyboard type at: present: false May 8 13:30:05 kbd-chooser[864]: INFO: keyboard type at: present: unknown May 8 13:30:05 kbd-chooser[864]: INFO: Setting debian-installer/serial-console to false May 8 13:30:05 kbd-chooser[864]: INFO: Setting debian-installer/uml-console to false May 8 13:30:05 kbd-chooser[864]: INFO: kbd-chooser: arch at selected May 8 13:30:25 kbd-chooser[864]: INFO: choose_keymap: keymap = us May 8 13:30:25 kbd-chooser[864]: INFO: kbd_chooser: setting keymap us May 8 13:30:25 main-menu[653]: DEBUG: resolver (libc6): package doesn't exist (ignored) May 8 13:30:25 main-menu[653]: INFO: Menu item 'ethdetect' selected >> So let me raise the issue again (and risk some bikeshed effect): >> * Does kbd-
Re: Skip kbd-chooser when running under Xen
On Wed, 2008-05-07 at 15:56 +0200, Frans Pop wrote: > On Tuesday 06 May 2008, Ferenc Wagner wrote: > > So let me raise the issue again (and risk some bikeshed effect): > > * Does kbd-chooser make any sense under Xen? (I guess not.) > > That's completely up to you (plural). I don't think it's necessary to special case kbd-chooser for Xen. I think the issue was down to the 2.6.18 kernel Ferenc was using since I never saw the it with 2.6.25 based stuff but I did previously see it with 2.6.18 when I first played with d-i+xen (ages and ages ago). > > * Would a simple [ -d /sys/bus/xen ] check do? > > I guess that would be OK. Unless a different test could also catch other > virtualized environments. So far I haven't found a need for any Xen specific checks when I've been working through this stuff (other than obvious things like supporting xvd devices) and I'd really like to avoid adding any if at all possible. If one does become necessary a check on /sys/bus/xen seems like the best current option since neither /proc/xen nor /sys/hypervisor exist in 2.6.25. I'm a bit stuck on getting finish-install.d/90console to create a getty on hvc0 without a Xen hack but I'll keep thinking... Ian. -- Ian Campbell Hold the MAYO & pass the COSMIC AWARENESS ... signature.asc Description: This is a digitally signed message part
Re: Skip kbd-chooser when running under Xen
On Tuesday 06 May 2008, Ferenc Wagner wrote: > Ian Campbell <[EMAIL PROTECTED]> writes: > > I've been using console=hvc0 -- I wonder if that has the same effect as > > console=ttyS0 in terms of skipping the keymap stuff... > > Just managed to check this. No, console=hvc0 doesn't skip the keymap > question, but looks like the kernels which have hvc don't throw and > error running kbd-chooser. At least that's the case under 2.6.25-xen, > which I tested a couple of minutes ago. What is the output in the syslog from kbd-chooser in both cases (the failing one and the one with hvc)? > So let me raise the issue again (and risk some bikeshed effect): > * Does kbd-chooser make any sense under Xen? (I guess not.) That's completely up to you (plural). > * Should we skip running it from a virtualized D-I? (I guess yes, >especially that it's so simple to achieve.) What is the difference with the previous point? > * Would a simple [ -d /sys/bus/xen ] check do? I guess that would be OK. Unless a different test could also catch other virtualized environments. I'd suggest at least printing an info message in the syslog when that happens. Cheers, FJP signature.asc Description: This is a digitally signed message part.
Re: Skip kbd-chooser when running under Xen
Ian Campbell <[EMAIL PROTECTED]> writes: > On Sat, 2008-04-26 at 11:46 +0200, Ferenc Wagner wrote: > >> Guido Günther <[EMAIL PROTECTED]> writes: >> >>> On Thu, Apr 24, 2008 at 10:47:50AM +0200, Ferenc Wagner wrote: >>> When running the installer in a Xen paravirtualised environment, kbd-chooser errors out after selecting a keymap: >>> >>> I vaguely remember that we passed 'console=ttyS0' to the kernel so d-i >>> would end up doing the right thing. Could you try that - would probably >>> be easier than skipping kbd detection? > > (I didn't see Guido's original mail so I'll reply here instead) > > I've been using console=hvc0 -- I wonder if that has the same effect as > console=ttyS0 in terms of skipping the keymap stuff... Just managed to check this. No, console=hvc0 doesn't skip the keymap question, but looks like the kernels which have hvc don't throw and error running kbd-chooser. At least that's the case under 2.6.25-xen, which I tested a couple of minutes ago. So let me raise the issue again (and risk some bikeshed effect): * Does kbd-chooser make any sense under Xen? (I guess not.) * Should we skip running it from a virtualized D-I? (I guess yes, especially that it's so simple to achieve.) * Would a simple [ -d /sys/bus/xen ] check do? -- Regards, Feri. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
On Sat, 2008-04-26 at 11:46 +0200, Ferenc Wagner wrote: > Guido Günther <[EMAIL PROTECTED]> writes: > > > On Thu, Apr 24, 2008 at 10:47:50AM +0200, Ferenc Wagner wrote: > > > >> When running the installer in a Xen paravirtualised environment, > >> kbd-chooser errors out after selecting a keymap: > > > > I vaguely remember that we passed 'console=ttyS0' to the kernel so d-i > > would end up doing the right thing. Could you try that - would probably > > be easier than skipping kbd detection? (I didn't see Guido's original mail so I'll reply here instead) I've been using console=hvc0 -- I wonder if that has the same effect as console=ttyS0 in terms of skipping the keymap stuff... Ian > > Hi Guido, > > I did several serial installs already and kbd detection really isn't a > problem there. It's easy to work around anyway (one can even preseed > it in the Lenny installer), I've just raised the issue so that we can > eliminate a superfluous workaround. Running kbd-chooser in a virtual > machine just doesn't make sense after all. At least I think so... > But maybe newer Xens solve this issue anyway. -- Ian Campbell You cannot achieve the impossible without attempting the absurd. signature.asc Description: This is a digitally signed message part
Re: Skip kbd-chooser when running under Xen
Guido Günther <[EMAIL PROTECTED]> writes: > On Thu, Apr 24, 2008 at 10:47:50AM +0200, Ferenc Wagner wrote: > >> When running the installer in a Xen paravirtualised environment, >> kbd-chooser errors out after selecting a keymap: > > I vaguely remember that we passed 'console=ttyS0' to the kernel so d-i > would end up doing the right thing. Could you try that - would probably > be easier than skipping kbd detection? Hi Guido, I did several serial installs already and kbd detection really isn't a problem there. It's easy to work around anyway (one can even preseed it in the Lenny installer), I've just raised the issue so that we can eliminate a superfluous workaround. Running kbd-chooser in a virtual machine just doesn't make sense after all. At least I think so... But maybe newer Xens solve this issue anyway. -- Thanks, Feri.
Re: Skip kbd-chooser when running under Xen
Ian Campbell <[EMAIL PROTECTED]> writes: > On Thu, 2008-04-24 at 13:00 +0200, Ferenc Wagner wrote: >> Ian Campbell <[EMAIL PROTECTED]> writes: >> >>> On Thu, 2008-04-24 at 10:47 +0200, Ferenc Wagner wrote: >>> When running the installer in a Xen paravirtualised environment, kbd-chooser errors out after selecting a keymap: >>> >>> I don't see this with latest installer with a pv_ops 2.6.25 kernel (from >>> pkg-kernel). Which kernel are you using? Since you mention /proc I guess >>> you are using a 2.6.18 based thing? >> >> Yes, exactly. I'm trying to run vmlinuz-2.6.24-1-xen-686 from Sid, >> but that just freezes under this Xen 3.0.3 (stock Etch). > > Just to clarify -- you are seeing the kbd error with 2.6.18 and you > tried 2.6.24-1-xen-686 from sid but it didn't work for you at all? Yes. > 3.0.3 is a bit long in the tooth I can well imagine there are bugs which > prevent the pv_ops kernel from working there -- is there any reason not > to upgrade to 3.1 or 3.2? I simply don't know how to, without risking too much the stability of this system, which I suddenly inherited. I'll probably try the 3.2 hypervisor from Backports shortly. >>> It might just be my priority level isn't picking up this question -- >>> I'll try later with all questions on. >> >> It doesn't matter wheter the question is asked, the error comes from >> kbd-chooser when it tries to do something. > > OK, so it isn't affecting me then for some reason, whether that is down > to kernel version or not I don't know. Right, so I'll try a newer version and see it that also gives me this error. -- Thanks, Feri. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
On Thu, 2008-04-24 at 13:00 +0200, Ferenc Wagner wrote: > Ian Campbell <[EMAIL PROTECTED]> writes: > > > On Thu, 2008-04-24 at 10:47 +0200, Ferenc Wagner wrote: > > > >> When running the installer in a Xen paravirtualised environment, > >> kbd-chooser errors out after selecting a keymap: > > > > I don't see this with latest installer with a pv_ops 2.6.25 kernel (from > > pkg-kernel). Which kernel are you using? Since you mention /proc I guess > > you are using a 2.6.18 based thing? > > Yes, exactly. I'm trying to run vmlinuz-2.6.24-1-xen-686 from Sid, > but that just freezes under this Xen 3.0.3 (stock Etch). Just to clarify -- you are seeing the kbd error with 2.6.18 and you tried 2.6.24-1-xen-686 from sid but it didn't work for you at all? 3.0.3 is a bit long in the tooth I can well imagine there are bugs which prevent the pv_ops kernel from working there -- is there any reason not to upgrade to 3.1 or 3.2? > > It might just be my priority level isn't picking up this question -- > > I'll try later with all questions on. > > It doesn't matter wheter the question is asked, the error comes from > kbd-chooser when it tries to do something. OK, so it isn't affecting me then for some reason, whether that is down to kernel version or not I don't know. Ian. -- Ian Campbell Current Noise: Mastodon - Crusher Destroyer This is National Non-Dairy Creamer Week. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
Ian Campbell <[EMAIL PROTECTED]> writes: > On Thu, 2008-04-24 at 10:47 +0200, Ferenc Wagner wrote: > >> When running the installer in a Xen paravirtualised environment, >> kbd-chooser errors out after selecting a keymap: > > I don't see this with latest installer with a pv_ops 2.6.25 kernel (from > pkg-kernel). Which kernel are you using? Since you mention /proc I guess > you are using a 2.6.18 based thing? Yes, exactly. I'm trying to run vmlinuz-2.6.24-1-xen-686 from Sid, but that just freezes under this Xen 3.0.3 (stock Etch). > It might just be my priority level isn't picking up this question -- > I'll try later with all questions on. It doesn't matter wheter the question is asked, the error comes from kbd-chooser when it tries to do something. -- Thanks, Feri. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Skip kbd-chooser when running under Xen
On Thursday 24 April 2008, Ian Campbell wrote: > It might just be my priority level isn't picking up this question -- > I'll try later with all questions on. No. Priority does not affect which questions are "run", it only affects which are displayed to the user. If a question is not displayed, the effect is exactly the same as when it would be displayed and the user selects the current (default or preseeded) value. It does not mean code is getting skipped in any way. Cheers, FJP signature.asc Description: This is a digitally signed message part.
Re: Skip kbd-chooser when running under Xen
On Thu, 2008-04-24 at 10:47 +0200, Ferenc Wagner wrote: > Hi, > > When running the installer in a Xen paravirtualised environment, > kbd-chooser errors out after selecting a keymap: I don't see this with latest installer with a pv_ops 2.6.25 kernel (from pkg-kernel). Which kernel are you using? Since you mention /proc I guess you are using a 2.6.18 based thing? It might just be my priority level isn't picking up this question -- I'll try later with all questions on. Ian. -- Ian Campbell To be is to be related. -- C. J. Keyser. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Skip kbd-chooser when running under Xen
Hi, When running the installer in a Xen paravirtualised environment, kbd-chooser errors out after selecting a keymap: Select a keyboard layout Keymap to use: 1. American English [*] 23. Italian [...] 22. Icelandic Prompt: '?' for help, default=1> !! ERROR: Installation step failed An installation step failed. You can try to run the failing item again from the menu, or skip it and choose something else. The failing step is: Select a keyboard layout [Press enter to continue] I'm not exactly sure why this happens, but the whole thing doesn't make much sense to me, as a virtual computer doesn't rellay have a keyboard attached. Therefore I propose the following patch: Index: packages/kbd-chooser/debian/kbd-chooser.isinstallable === --- packages/kbd-chooser/debian/kbd-chooser.isinstallable (revision 52852) +++ packages/kbd-chooser/debian/kbd-chooser.isinstallable (working copy) @@ -1,4 +1,8 @@ #!/bin/sh . /usr/share/debconf/confmodule db_get auto-install/enable && [ "$RET" = true ] && exit 1 + +# A paravirtualised system has no keyboard attached +[ -d /proc/xen ] && exit 1 + exit 0 which makes kbd-chooser not installable in a Xen paravirtualised environment. At least. Of course several questions arise: * What happens under a fully virtualised environment? Does kbd-chooser error out? Does it make sense? Is /proc/xen present? * Using /proc for such things is deprecated, should we check /sys/bus/xen or /sys/devices/xen instead? * Should we use a more elaborate detection method, like the short C program presented in http://lists.xensource.com/archives/html/xen-changelog/2007-02/msg00075.html for example? * Should this problem be solved by other means, like Subarchitecture specifications, whatever those are? See http://lists.debian.org/debian-boot/2003/09/msg00854.html -- Regards, Feri.