Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Thu, 20 Sep 2007, Charles N Wyble wrote: > > Satyam Sharma wrote: > > > > On Tue, 18 Sep 2007, Charles N Wyble wrote: > >> Andi Kleen wrote: > >>> Besides it's bad taste and taste is very important. > >> Well it's bad taste for you (one person). > > > > FWIW, my opinion is the same as Andi's here. Please, let's avoid this > > disease -- unless *absolutely* required, stuff shouldn't be "default y". > > I am curious why you think it shouldn't be default why? Bad taste? Can > you provide data about how it will harm things? Clear CONFIG_PARAVIRT from your .config and try "make oldconfig", and find out for yourself :-) You'll end up having to answer questions for all other config options below it, that you never cared about, that you didn't know even existed. You'd obviously (I would, at least) feel annoyed by it all ... this is almost a regression :-) And for those who automate the oldconfig process using "yes", well, this "default y" *will* end up introducing a kernel text size regression -- all those config options they never had earlier, suddenly got enabled. Besides, like Andi said, selecting or enabling CONFIG_PARAVIRT is quite pointless in the first place -- it's just a menuconfig symbol, not a "real" kconfig symbol that _actually_ controls code in the Makefiles. Satyam - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Satyam Sharma wrote: > > On Tue, 18 Sep 2007, Charles N Wyble wrote: >> Andi Kleen wrote: >>> Besides it's bad taste and taste is very important. >> Well it's bad taste for you (one person). > > FWIW, my opinion is the same as Andi's here. Please, let's avoid this > disease -- unless *absolutely* required, stuff shouldn't be "default y". > I am curious why you think it shouldn't be default why? Bad taste? Can you provide data about how it will harm things? -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG8ptLkQPZV56XDBMRApxNAJwLOv2dabO6KD6t4Z/Fkffelh1IxwCdEzPM 1JEuIQvlkiojgiMy0tnroqk= =yCiZ -END PGP SIGNATURE- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 18 Sep 2007, Charles N Wyble wrote: > > Andi Kleen wrote: > > > > Besides it's bad taste and taste is very important. > > Well it's bad taste for you (one person). FWIW, my opinion is the same as Andi's here. Please, let's avoid this disease -- unless *absolutely* required, stuff shouldn't be "default y". - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 18 Sep 2007, Charles N Wyble wrote: Andi Kleen wrote: Besides it's bad taste and taste is very important. Well it's bad taste for you (one person). FWIW, my opinion is the same as Andi's here. Please, let's avoid this disease -- unless *absolutely* required, stuff shouldn't be default y. - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Satyam Sharma wrote: On Tue, 18 Sep 2007, Charles N Wyble wrote: Andi Kleen wrote: Besides it's bad taste and taste is very important. Well it's bad taste for you (one person). FWIW, my opinion is the same as Andi's here. Please, let's avoid this disease -- unless *absolutely* required, stuff shouldn't be default y. I am curious why you think it shouldn't be default why? Bad taste? Can you provide data about how it will harm things? -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG8ptLkQPZV56XDBMRApxNAJwLOv2dabO6KD6t4Z/Fkffelh1IxwCdEzPM 1JEuIQvlkiojgiMy0tnroqk= =yCiZ -END PGP SIGNATURE- - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Thu, 20 Sep 2007, Charles N Wyble wrote: Satyam Sharma wrote: On Tue, 18 Sep 2007, Charles N Wyble wrote: Andi Kleen wrote: Besides it's bad taste and taste is very important. Well it's bad taste for you (one person). FWIW, my opinion is the same as Andi's here. Please, let's avoid this disease -- unless *absolutely* required, stuff shouldn't be default y. I am curious why you think it shouldn't be default why? Bad taste? Can you provide data about how it will harm things? Clear CONFIG_PARAVIRT from your .config and try make oldconfig, and find out for yourself :-) You'll end up having to answer questions for all other config options below it, that you never cared about, that you didn't know even existed. You'd obviously (I would, at least) feel annoyed by it all ... this is almost a regression :-) And for those who automate the oldconfig process using yes, well, this default y *will* end up introducing a kernel text size regression -- all those config options they never had earlier, suddenly got enabled. Besides, like Andi said, selecting or enabling CONFIG_PARAVIRT is quite pointless in the first place -- it's just a menuconfig symbol, not a real kconfig symbol that _actually_ controls code in the Makefiles. Satyam - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 2007-09-18 at 20:59 -0700, Jeremy Fitzhardinge wrote: > Andi Kleen wrote: > > At least the Xen port seems to have specific requirements > > and essentially only work on xen-unstable (?) [or at least > > some very new Xen version] which probably very few > > people use. > > > > Only on 64-bit hosts, because of bugs in the 64-bit compat layer. > 32-on-32 and 64-on-64 (when its done) should work fine. All the required fixes for 32on64 are queued for Xen 3.1.1 too. The proposed updates are at http://xenbits.xensource.com/xen-3.1-testing.pq.hg Ian. -- Ian Campbell Current Noise: Detonation - Sword-Carved Skin The executioner is, I hear, very expert, and my neck is very slender. -- Anne Boleyn - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 2007-09-18 at 20:59 -0700, Jeremy Fitzhardinge wrote: Andi Kleen wrote: At least the Xen port seems to have specific requirements and essentially only work on xen-unstable (?) [or at least some very new Xen version] which probably very few people use. Only on 64-bit hosts, because of bugs in the 64-bit compat layer. 32-on-32 and 64-on-64 (when its done) should work fine. All the required fixes for 32on64 are queued for Xen 3.1.1 too. The proposed updates are at http://xenbits.xensource.com/xen-3.1-testing.pq.hg Ian. -- Ian Campbell Current Noise: Detonation - Sword-Carved Skin The executioner is, I hear, very expert, and my neck is very slender. -- Anne Boleyn - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
Andi Kleen wrote: > At least the Xen port seems to have specific requirements > and essentially only work on xen-unstable (?) [or at least > some very new Xen version] which probably very few > people use. > Only on 64-bit hosts, because of bugs in the 64-bit compat layer. 32-on-32 and 64-on-64 (when its done) should work fine. BTW, what does "xm info" say on your system that fails? I'll try to put a more graceful failure in there. J - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 2007-09-18 at 23:52 +0200, Andi Kleen wrote: > On Tuesday 18 September 2007 23:34, Rusty Russell wrote: > > How about a "select" based on Xen, lguest or VMI? There's no other > > reason to enable it, after all. > > I did an patch to do that recently because the current setup > is indeed unobvious. > > But I had to drop it again because > it ended up with Kconfig warnings. about undefined symbols > on x86-64. The problem is that lguest > is visible in Kconfig for all architectures and it warns > if you select something that doesn't exist on all architectures. I think that's fixed as a side-effect of this cleanup. At least, it works for me on x86-64. Patch below: if you agree, I'll re-xmit all three. > > > Also I would still consider it experimental. > > > > After 9 months in mainline and three kernel versions, > > Well it changed a lot each release. Well, the biggest change was the patching code getting enhanced in 2.6.22 (to cover all calls, not just 5). The 22 -> 23 changes were fairly trivial. So I think 2.6.24 is a reasonable time to remove EXPERIMENTAL. > > I'd hope not. > > It's been pretty damn stable (ok, you broke it once, but maybe that's > > because you consider it experimental). > > Is there a significant user base? It's enabled in Ubuntu Feisty (2.6.20). > At least the Xen port seems to have specific requirements > and essentially only work on xen-unstable (?) [or at least > some very new Xen version] which probably very few > people use. Sure, and that might well still be experimental (Jeremy?). But that's not CONFIG_PARAVIRT. Hope that helps, Rusty. == Andi points out that PARAVIRT is an option best selected when needed. We introduce PARAVIRT_GUEST for the menu itself, and select PARAVIRT if they ask for anything which needs it. This also makes PARAVIRT non-experimental. Signed-off-by: Rusty Russell <[EMAIL PROTECTED]> diff -r 8efa5fdb22d8 arch/i386/Kconfig --- a/arch/i386/Kconfig Wed Sep 19 11:23:18 2007 +1000 +++ b/arch/i386/Kconfig Wed Sep 19 11:33:59 2007 +1000 @@ -214,24 +214,30 @@ config X86_ES7000 endchoice -menuconfig PARAVIRT +config PARAVIRT + bool + depends on !(X86_VISWS || X86_VOYAGER) + help + This changes the kernel so it can modify itself when it is run + under a hypervisor, potentially improving performance significantly + over full virtualization. However, when run without a hypervisor + the kernel is theoretically slower and slightly larger. + +menuconfig PARAVIRT_GUEST - bool "Paravirtualized guest support (EXPERIMENTAL)" - depends on EXPERIMENTAL + bool "Paravirtualized guest support" - depends on !(X86_VISWS || X86_VOYAGER) - help - Paravirtualization is a way of running multiple instances of - Linux on the same machine, under a hypervisor. This option - changes the kernel so it can modify itself when it is run - under a hypervisor, improving performance significantly. - However, when run without a hypervisor the kernel is - theoretically slower. If in doubt, say N. - -if PARAVIRT + help + Say Y here to get to see options related to running Linux under + various hypervisors. This option alone does not add any kernel code. + + If you say N, all options in this submenu will be skipped and disabled. + +if PARAVIRT_GUEST source "arch/i386/xen/Kconfig" config VMI bool "VMI Guest support" + select PARAVIRT help VMI provides a paravirtualized interface to the VMware ESX server (it could be used by other hypervisors in theory too, but is not @@ -239,6 +246,7 @@ config VMI config LGUEST_GUEST bool "Lguest guest support" + select PARAVIRT depends on !X86_PAE help Lguest is a tiny in-kernel hypervisor. Selecting this will diff -r 8efa5fdb22d8 arch/i386/xen/Kconfig --- a/arch/i386/xen/Kconfig Wed Sep 19 11:23:18 2007 +1000 +++ b/arch/i386/xen/Kconfig Wed Sep 19 11:25:07 2007 +1000 @@ -4,6 +4,7 @@ config XEN bool "Xen guest support" + select PARAVIRT depends on X86_CMPXCHG && X86_TSC && !NEED_MULTIPLE_NODES help This is the Linux Xen port. Enabling this will allow the - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tuesday 18 September 2007 23:34, Rusty Russell wrote: > On Tue, 2007-09-18 at 12:57 +0200, Andi Kleen wrote: > > On Friday 14 September 2007 07:21, Rusty Russell wrote: > > > It's pretty widely used, > > > > Is it? By whom? > > Hi Andi, > > Please stop asking for facts! It's was easy claim to make, and hard to > disprove 8) > > > > and the distributions will turn it on. > > > > That's no reason to make it default y. Please undo that. default y > > is near always a bad idea. > > How about a "select" based on Xen, lguest or VMI? There's no other > reason to enable it, after all. I did an patch to do that recently because the current setup is indeed unobvious. But I had to drop it again because it ended up with Kconfig warnings. about undefined symbols on x86-64. The problem is that lguest is visible in Kconfig for all architectures and it warns if you select something that doesn't exist on all architectures. The only workaround would have been to define PARAVIRT for all architectures, which I considered too ugly. I think Sam stated recently he wanted to remove that warning but it needed some infrastructure work. > > Also I would still consider it experimental. > > After 9 months in mainline and three kernel versions, Well it changed a lot each release. > I'd hope not. > It's been pretty damn stable (ok, you broke it once, but maybe that's > because you consider it experimental). Is there a significant user base? At least the Xen port seems to have specific requirements and essentially only work on xen-unstable (?) [or at least some very new Xen version] which probably very few people use. -Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 2007-09-18 at 12:57 +0200, Andi Kleen wrote: > On Friday 14 September 2007 07:21, Rusty Russell wrote: > > It's pretty widely used, > > Is it? By whom? Hi Andi, Please stop asking for facts! It's was easy claim to make, and hard to disprove 8) > > and the distributions will turn it on. > > That's no reason to make it default y. Please undo that. default y > is near always a bad idea. How about a "select" based on Xen, lguest or VMI? There's no other reason to enable it, after all. > Also I would still consider it experimental. After 9 months in mainline and three kernel versions, I'd hope not. It's been pretty damn stable (ok, you broke it once, but maybe that's because you consider it experimental). Cheers, Rusty. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
> Actually if I understand the functionality of paravirt correctly that is > not correct. I believe that will turn on the paravirt bits which allow > it to run under things such as VMI or Xen. You don't understand it correctly then. -Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Andi Kleen wrote: > >> Why is making something default y a bad idea? >> Those most likely to care can turn it off. Is there a harmful effect >> from leaving it on if its not being used? > > Running yes "" | make oldconfig to upgrade kernel configs is standard practice > and you definitely don't want to have all kinds of random new unnecessary > features > be turned on then. H. I disagree its a standard practice. I thought the whole point of make oldconfig was to give you just the delta in configuration options and was targeted at manual review? Also how many people are building there own kernels these days? Expanding on what I said in the original e-mail those who are likely to care "CAN TURN IT OFF". Those who care about such things should be REVIEWING CHANGES anyway. Thats what I do when looking at building custom kernels. What changed that makes me want to move to a new version? I generally try to stick with the distro kernel when possible, but part of being an early and aggressive adopter of virtualization technology involves running non distro kernels and patches. Most people don't really care. > > Besides paravirt by itself is pretty useless; you need typically quite > complex other options set to do any meaningfull virtualization. Actually if I understand the functionality of paravirt correctly that is not correct. I believe that will turn on the paravirt bits which allow it to run under things such as VMI or Xen. > > The only reason to use default y is in options that are not user visible > and have a reasonable default or things that cause direct boot failures > when upgrading old configurations. That all doesn't apply here. Again you need to think about the target audience here. A distro kernel you don't have to worry about this stuff. A user compiling there own kernel should already be able to handle this. > > Besides it's bad taste and taste is very important. Well it's bad taste for you (one person). Taste is highly subjective. So be careful in making broad ranging statements like this. :) > > -Andi > -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG8DSnkQPZV56XDBMRAtZcAJ4rtRXGW14b70YRIBKyHCsaKTdO/wCeOdoM AUc4YGUaqs5DmDDbov7X980= =UA4y -END PGP SIGNATURE- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
> Why is making something default y a bad idea? > Those most likely to care can turn it off. Is there a harmful effect > from leaving it on if its not being used? Running yes "" | make oldconfig to upgrade kernel configs is standard practice and you definitely don't want to have all kinds of random new unnecessary features be turned on then. Besides paravirt by itself is pretty useless; you need typically quite complex other options set to do any meaningfull virtualization. The only reason to use default y is in options that are not user visible and have a reasonable default or things that cause direct boot failures when upgrading old configurations. That all doesn't apply here. Besides it's bad taste and taste is very important. -Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Andi Kleen wrote: > On Friday 14 September 2007 07:21, Rusty Russell wrote: >> It's pretty widely used, > > Is it? By whom? > >> and the distributions will turn it on. > > That's no reason to make it default y. Please undo that. default y > is near always a bad idea. Andi, What do you base this on? Why is making something default y a bad idea? Those most likely to care can turn it off. Is there a harmful effect from leaving it on if its not being used? -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG8C9QkQPZV56XDBMRAowcAJ9fkxKAjjsvGd6hOPkvHXU9ThpoQwCdGUI8 mjpUh6G+ZyzL0CW3sYNS97g= =ZgiJ -END PGP SIGNATURE- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Friday 14 September 2007 07:21, Rusty Russell wrote: > It's pretty widely used, Is it? By whom? > and the distributions will turn it on. That's no reason to make it default y. Please undo that. default y is near always a bad idea. Also I would still consider it experimental. -Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Friday 14 September 2007 07:21, Rusty Russell wrote: It's pretty widely used, Is it? By whom? and the distributions will turn it on. That's no reason to make it default y. Please undo that. default y is near always a bad idea. Also I would still consider it experimental. -Andi - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Andi Kleen wrote: On Friday 14 September 2007 07:21, Rusty Russell wrote: It's pretty widely used, Is it? By whom? and the distributions will turn it on. That's no reason to make it default y. Please undo that. default y is near always a bad idea. Andi, What do you base this on? Why is making something default y a bad idea? Those most likely to care can turn it off. Is there a harmful effect from leaving it on if its not being used? -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG8C9QkQPZV56XDBMRAowcAJ9fkxKAjjsvGd6hOPkvHXU9ThpoQwCdGUI8 mjpUh6G+ZyzL0CW3sYNS97g= =ZgiJ -END PGP SIGNATURE- - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
Why is making something default y a bad idea? Those most likely to care can turn it off. Is there a harmful effect from leaving it on if its not being used? Running yes | make oldconfig to upgrade kernel configs is standard practice and you definitely don't want to have all kinds of random new unnecessary features be turned on then. Besides paravirt by itself is pretty useless; you need typically quite complex other options set to do any meaningfull virtualization. The only reason to use default y is in options that are not user visible and have a reasonable default or things that cause direct boot failures when upgrading old configurations. That all doesn't apply here. Besides it's bad taste and taste is very important. -Andi - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Andi Kleen wrote: Why is making something default y a bad idea? Those most likely to care can turn it off. Is there a harmful effect from leaving it on if its not being used? Running yes | make oldconfig to upgrade kernel configs is standard practice and you definitely don't want to have all kinds of random new unnecessary features be turned on then. H. I disagree its a standard practice. I thought the whole point of make oldconfig was to give you just the delta in configuration options and was targeted at manual review? Also how many people are building there own kernels these days? Expanding on what I said in the original e-mail those who are likely to care CAN TURN IT OFF. Those who care about such things should be REVIEWING CHANGES anyway. Thats what I do when looking at building custom kernels. What changed that makes me want to move to a new version? I generally try to stick with the distro kernel when possible, but part of being an early and aggressive adopter of virtualization technology involves running non distro kernels and patches. Most people don't really care. Besides paravirt by itself is pretty useless; you need typically quite complex other options set to do any meaningfull virtualization. Actually if I understand the functionality of paravirt correctly that is not correct. I believe that will turn on the paravirt bits which allow it to run under things such as VMI or Xen. The only reason to use default y is in options that are not user visible and have a reasonable default or things that cause direct boot failures when upgrading old configurations. That all doesn't apply here. Again you need to think about the target audience here. A distro kernel you don't have to worry about this stuff. A user compiling there own kernel should already be able to handle this. Besides it's bad taste and taste is very important. Well it's bad taste for you (one person). Taste is highly subjective. So be careful in making broad ranging statements like this. :) -Andi -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG8DSnkQPZV56XDBMRAtZcAJ4rtRXGW14b70YRIBKyHCsaKTdO/wCeOdoM AUc4YGUaqs5DmDDbov7X980= =UA4y -END PGP SIGNATURE- - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
Actually if I understand the functionality of paravirt correctly that is not correct. I believe that will turn on the paravirt bits which allow it to run under things such as VMI or Xen. You don't understand it correctly then. -Andi - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 2007-09-18 at 12:57 +0200, Andi Kleen wrote: On Friday 14 September 2007 07:21, Rusty Russell wrote: It's pretty widely used, Is it? By whom? Hi Andi, Please stop asking for facts! It's was easy claim to make, and hard to disprove 8) and the distributions will turn it on. That's no reason to make it default y. Please undo that. default y is near always a bad idea. How about a select based on Xen, lguest or VMI? There's no other reason to enable it, after all. Also I would still consider it experimental. After 9 months in mainline and three kernel versions, I'd hope not. It's been pretty damn stable (ok, you broke it once, but maybe that's because you consider it experimental). Cheers, Rusty. - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tuesday 18 September 2007 23:34, Rusty Russell wrote: On Tue, 2007-09-18 at 12:57 +0200, Andi Kleen wrote: On Friday 14 September 2007 07:21, Rusty Russell wrote: It's pretty widely used, Is it? By whom? Hi Andi, Please stop asking for facts! It's was easy claim to make, and hard to disprove 8) and the distributions will turn it on. That's no reason to make it default y. Please undo that. default y is near always a bad idea. How about a select based on Xen, lguest or VMI? There's no other reason to enable it, after all. I did an patch to do that recently because the current setup is indeed unobvious. But I had to drop it again because it ended up with Kconfig warnings. about undefined symbols on x86-64. The problem is that lguest is visible in Kconfig for all architectures and it warns if you select something that doesn't exist on all architectures. The only workaround would have been to define PARAVIRT for all architectures, which I considered too ugly. I think Sam stated recently he wanted to remove that warning but it needed some infrastructure work. Also I would still consider it experimental. After 9 months in mainline and three kernel versions, Well it changed a lot each release. I'd hope not. It's been pretty damn stable (ok, you broke it once, but maybe that's because you consider it experimental). Is there a significant user base? At least the Xen port seems to have specific requirements and essentially only work on xen-unstable (?) [or at least some very new Xen version] which probably very few people use. -Andi - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
On Tue, 2007-09-18 at 23:52 +0200, Andi Kleen wrote: On Tuesday 18 September 2007 23:34, Rusty Russell wrote: How about a select based on Xen, lguest or VMI? There's no other reason to enable it, after all. I did an patch to do that recently because the current setup is indeed unobvious. But I had to drop it again because it ended up with Kconfig warnings. about undefined symbols on x86-64. The problem is that lguest is visible in Kconfig for all architectures and it warns if you select something that doesn't exist on all architectures. I think that's fixed as a side-effect of this cleanup. At least, it works for me on x86-64. Patch below: if you agree, I'll re-xmit all three. Also I would still consider it experimental. After 9 months in mainline and three kernel versions, Well it changed a lot each release. Well, the biggest change was the patching code getting enhanced in 2.6.22 (to cover all calls, not just 5). The 22 - 23 changes were fairly trivial. So I think 2.6.24 is a reasonable time to remove EXPERIMENTAL. I'd hope not. It's been pretty damn stable (ok, you broke it once, but maybe that's because you consider it experimental). Is there a significant user base? It's enabled in Ubuntu Feisty (2.6.20). At least the Xen port seems to have specific requirements and essentially only work on xen-unstable (?) [or at least some very new Xen version] which probably very few people use. Sure, and that might well still be experimental (Jeremy?). But that's not CONFIG_PARAVIRT. Hope that helps, Rusty. == Andi points out that PARAVIRT is an option best selected when needed. We introduce PARAVIRT_GUEST for the menu itself, and select PARAVIRT if they ask for anything which needs it. This also makes PARAVIRT non-experimental. Signed-off-by: Rusty Russell [EMAIL PROTECTED] diff -r 8efa5fdb22d8 arch/i386/Kconfig --- a/arch/i386/Kconfig Wed Sep 19 11:23:18 2007 +1000 +++ b/arch/i386/Kconfig Wed Sep 19 11:33:59 2007 +1000 @@ -214,24 +214,30 @@ config X86_ES7000 endchoice -menuconfig PARAVIRT +config PARAVIRT + bool + depends on !(X86_VISWS || X86_VOYAGER) + help + This changes the kernel so it can modify itself when it is run + under a hypervisor, potentially improving performance significantly + over full virtualization. However, when run without a hypervisor + the kernel is theoretically slower and slightly larger. + +menuconfig PARAVIRT_GUEST - bool Paravirtualized guest support (EXPERIMENTAL) - depends on EXPERIMENTAL + bool Paravirtualized guest support - depends on !(X86_VISWS || X86_VOYAGER) - help - Paravirtualization is a way of running multiple instances of - Linux on the same machine, under a hypervisor. This option - changes the kernel so it can modify itself when it is run - under a hypervisor, improving performance significantly. - However, when run without a hypervisor the kernel is - theoretically slower. If in doubt, say N. - -if PARAVIRT + help + Say Y here to get to see options related to running Linux under + various hypervisors. This option alone does not add any kernel code. + + If you say N, all options in this submenu will be skipped and disabled. + +if PARAVIRT_GUEST source arch/i386/xen/Kconfig config VMI bool VMI Guest support + select PARAVIRT help VMI provides a paravirtualized interface to the VMware ESX server (it could be used by other hypervisors in theory too, but is not @@ -239,6 +246,7 @@ config VMI config LGUEST_GUEST bool Lguest guest support + select PARAVIRT depends on !X86_PAE help Lguest is a tiny in-kernel hypervisor. Selecting this will diff -r 8efa5fdb22d8 arch/i386/xen/Kconfig --- a/arch/i386/xen/Kconfig Wed Sep 19 11:23:18 2007 +1000 +++ b/arch/i386/xen/Kconfig Wed Sep 19 11:25:07 2007 +1000 @@ -4,6 +4,7 @@ config XEN bool Xen guest support + select PARAVIRT depends on X86_CMPXCHG X86_TSC !NEED_MULTIPLE_NODES help This is the Linux Xen port. Enabling this will allow the - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
Andi Kleen wrote: At least the Xen port seems to have specific requirements and essentially only work on xen-unstable (?) [or at least some very new Xen version] which probably very few people use. Only on 64-bit hosts, because of bugs in the 64-bit compat layer. 32-on-32 and 64-on-64 (when its done) should work fine. BTW, what does xm info say on your system that fails? I'll try to put a more graceful failure in there. J - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
It's pretty widely used, and the distributions will turn it on. Signed-off-by: Rusty Russell <[EMAIL PROTECTED]> diff -r b546335d7e75 arch/i386/Kconfig --- a/arch/i386/Kconfig Fri Sep 14 13:24:49 2007 +1000 +++ b/arch/i386/Kconfig Fri Sep 14 13:27:13 2007 +1000 @@ -215,8 +215,8 @@ endchoice endchoice menuconfig PARAVIRT - bool "Paravirtualized guest support (EXPERIMENTAL)" - depends on EXPERIMENTAL + bool "Paravirtualized guest support" + default y depends on !(X86_VISWS || X86_VOYAGER) help Paravirtualization is a way of running multiple instances of @@ -224,7 +224,7 @@ menuconfig PARAVIRT changes the kernel so it can modify itself when it is run under a hypervisor, improving performance significantly. However, when run without a hypervisor the kernel is - theoretically slower. If in doubt, say N. + theoretically slower. If in doubt, say Y. if PARAVIRT - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH 3/3] Time to make CONFIG_PARAVIRT non-experimental.
It's pretty widely used, and the distributions will turn it on. Signed-off-by: Rusty Russell [EMAIL PROTECTED] diff -r b546335d7e75 arch/i386/Kconfig --- a/arch/i386/Kconfig Fri Sep 14 13:24:49 2007 +1000 +++ b/arch/i386/Kconfig Fri Sep 14 13:27:13 2007 +1000 @@ -215,8 +215,8 @@ endchoice endchoice menuconfig PARAVIRT - bool Paravirtualized guest support (EXPERIMENTAL) - depends on EXPERIMENTAL + bool Paravirtualized guest support + default y depends on !(X86_VISWS || X86_VOYAGER) help Paravirtualization is a way of running multiple instances of @@ -224,7 +224,7 @@ menuconfig PARAVIRT changes the kernel so it can modify itself when it is run under a hypervisor, improving performance significantly. However, when run without a hypervisor the kernel is - theoretically slower. If in doubt, say N. + theoretically slower. If in doubt, say Y. if PARAVIRT - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
