On 15.09.2007 18:27, fedorawiki-noreply@fedoraproject.org wrote:> You have subscribed to a wiki page or wiki category on "Fedora > Project Wiki" for change notification. > > The following page has been changed by PaulFrields: > http://fedoraproject.org/wiki/Docs/Beats/Kernel?action=diff&rev2=112&rev1=111 > [...] > + * The kernel-PAE, for use in 32-bit x86 systems with more > than 4GB of RAM, or with CPUs that have a NX (No eXecute) feature. > This kernel support both uniprocessor and multi-processor systems. > Configured sources are available in the `kernel-PAE-devel` package.This reminded me of: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=c673f1a9d994de501b674b2bb6a48bd5e912afe0 Quoting Patch description:> i386: divorce CONFIG_X86_PAE from CONFIG_HIGHMEM64G > > PAE is useful for more than supporting more than 4GB RAM. It supports > expanded swapspace and NX executable protections. Some users may want NX > or expanded swapspace support without the overhead or instability of > highmem. For these reasons, the following patch divorces CONFIG_X86_PAE > from CONFIG_HIGHMEM64G.Which brings me to the point why I''m writing this mail: should we enable PAE by default after F8 is out and ship a HIGHMEM64G-Kernel instead on x86-32? Then normal users on x86-32 can benefit from NX by default. Or is that considered "not worth the trouble because we ship execshield already"? Further: Does PAE-by-default still make lots of machines unbootable (I doubt that for newer machines, as some-well-known-other-os enables PAE since its service pack 2 by default iirc)? If we suspect it still does: could we fix that by some kind of automatic "if machine is newer than ${year}" check in the kernel? Just wondering. CU knurd
Thorsten Leemhuis wrote:> > Which brings me to the point why I''m writing this mail: should we enable > PAE by default after F8 is out and ship a HIGHMEM64G-Kernel instead on > x86-32? Then normal users on x86-32 can benefit from NX by default. Or > is that considered "not worth the trouble because we ship execshield > already"? > > Further: Does PAE-by-default still make lots of machines unbootable (I > doubt that for newer machines, as some-well-known-other-os enables PAE > since its service pack 2 by default iirc)? If we suspect it still does: > could we fix that by some kind of automatic "if machine is newer than > ${year}" check in the kernel? > > Just wondering.Yes, it does still tend to break machines, especially laptops and older machines (we run into this a lot with Xen, since we only ship a PAE variant there). Unfortunately, PAE is a compile time option, not something that can (currently) be turned on or off at runtime. It would be nice to make it a runtime option (IIRC, OpenSolaris does this), but I don''t know if it is worth the effort given that 32-bit should be becoming less and less common. Chris Lalancette
On 16.09.2007 20:14, Chris Lalancette wrote:> Thorsten Leemhuis wrote: >> Which brings me to the point why I''m writing this mail: should we enable >> PAE by default after F8 is out and ship a HIGHMEM64G-Kernel instead on >> x86-32? Then normal users on x86-32 can benefit from NX by default. Or >> is that considered "not worth the trouble because we ship execshield >> already"? >> >> Further: Does PAE-by-default still make lots of machines unbootable (I >> doubt that for newer machines, as some-well-known-other-os enables PAE >> since its service pack 2 by default iirc)? If we suspect it still does: >> could we fix that by some kind of automatic "if machine is newer than >> ${year}" check in the kernel? >> >> Just wondering. > > Yes, it does still tend to break machines,:-/ -- as expected... Nevertheless thx for sharing your experiences.> [...] Unfortunately, PAE is a compile time option, not > something that can (currently) be turned on or off at runtime. It would > be nice to make it a runtime option (IIRC, OpenSolaris does this), but I > don''t know if it is worth the effort given that 32-bit should be > becoming less and less common.Agreed -- but on the other hand x86-32 still is quite famous even on x86-64 machines. Well, sooner or later that should change :-) Cu knurd