From: <se...@ex...> - 2005-03-30 22:35:10
|
Hi, how to build cryptoloop support in 2.6.9-bs7 ? if I build uml I see things like "-U__i386__ -Ui386 -D__arch_um__ -DSUBARCH="i386"" is the uml binary not optimized for i686 ? How to do that ? is it possible to use the aes-i586 (asm) implementation under uml ? thank you for your response. /sebastian |
From: Blaisorblade <bla...@ya...> - 2005-04-01 19:04:35
|
On Thursday 31 March 2005 00:37, Sebastian B=F6hm wrote: > Hi, > > how to build cryptoloop support in 2.6.9-bs7 ? Well, cryptoloop should work without problem, the right question is for=20 aes-i586, right? > if I build uml I see things like "-U__i386__ -Ui386 -D__arch_um__ > -DSUBARCH=3D"i386"" > > is the uml binary not optimized for i686 ? How to do that ? Good question... I'll look into this soon. > is it possible to use the aes-i586 (asm) implementation under uml ? Well, it should be possible to enable this - I'll do a patch ASAP (remind m= e=20 if I forget this). > thank you for your response. > > /sebastian =2D-=20 Paolo Giarrusso, aka Blaisorblade Linux registered user n. 292729 http://www.user-mode-linux.org/~blaisorblade |
From: <se...@ex...> - 2005-04-04 01:22:32
|
Am 01.04.2005 um 21:03 schrieb Blaisorblade: > On Thursday 31 March 2005 00:37, Sebastian B=F6hm wrote: > >> Hi, >> >> how to build cryptoloop support in 2.6.9-bs7 ? >> > Well, cryptoloop should work without problem, the right question is =20= > for > aes-i586, right? not completely, because: in menuconfig the cryptoloop option should =20 be right next to the loop option, but is not (is there some depedency =20= I have to activate first ?) > >> if I build uml I see things like "-U__i386__ -Ui386 -D__arch_um__ >> -DSUBARCH=3D"i386"" >> >> is the uml binary not optimized for i686 ? How to do that ? >> > Good question... I'll look into this soon. Sounds great. (would that make uml substancially faster ? probably =20 not, cause most of the code that gets executed is outside the uml =20 binary, not ? ) > >> is it possible to use the aes-i586 (asm) implementation under uml ? >> > Well, it should be possible to enable this - I'll do a patch ASAP =20 > (remind me > if I forget this). That would be very nice, as the asm implementation should be alot =20 faster. /sebastian |
From: Blaisorblade <bla...@ya...> - 2005-04-04 18:40:22
Attachments:
uml-crypto-i586-dirty.patch
|
On Monday 04 April 2005 03:22, Sebastian B=F6hm wrote: > Am 01.04.2005 um 21:03 schrieb Blaisorblade: > > On Thursday 31 March 2005 00:37, Sebastian B=F6hm wrote: > >> Hi, > >> > >> how to build cryptoloop support in 2.6.9-bs7 ? > > > > Well, cryptoloop should work without problem, the right question is > > for > > aes-i586, right? > > not completely, because: in menuconfig the cryptoloop option should > be right next to the loop option, but is not (is there some depedency > I have to activate first ?) Obviously there is the dependency on Loop support itself, as you already kn= ow.=20 I forgot that I had to fix this bug. Upgrade to 2.6.11 will help for cryptoloop itself, I've fixed this recently. > >> if I build uml I see things like "-U__i386__ -Ui386 -D__arch_um__ > >> -DSUBARCH=3D"i386"" > >> > >> is the uml binary not optimized for i686 ? How to do that ? > > > > Good question... I'll look into this soon. > > Sounds great. (would that make uml substancially faster ? probably > not, cause most of the code that gets executed is outside the uml > binary, not ? ) More or less it will make the same difference between a kernel compiled for= a=20 386 processor and one compiled for Pentium II, or whatever processor you=20 take, because even for Linux the majority of the executed code is made up b= y=20 userspace binaries. However, the UML code is written in different ways, so I don't know if=20 optimizations will give more or less than on the host, and how much. > >> is it possible to use the aes-i586 (asm) implementation under uml ? > > > > Well, it should be possible to enable this - I'll do a patch ASAP > > (remind me > > if I forget this). > That would be very nice, as the asm implementation should be alot > faster. Ok, I'm writing the patch. Use the attached one on 2.6.11 (and/or -bs1), no= r=20 on 2.6.9! I'm writing a better one for merge, which works well also for AMD64 users=20 (which can't use aes-i586). Also, a question: from what I see, it seems that aes-i586 can be enabled al= so=20 for older chipset. Do you know if that's reasonable (i.e. it works on the=20 older ones but is optimised for i586 and above) or it's a bug? =2D-=20 Paolo Giarrusso, aka Blaisorblade Linux registered user n. 292729 http://www.user-mode-linux.org/~blaisorblade |
From: Blaisorblade <bla...@ya...> - 2005-04-05 16:36:20
Attachments:
uml-crypto-i586-dirty.patch
|
On Monday 04 April 2005 20:39, Blaisorblade wrote: > Ok, I'm writing the patch. Use the attached one on 2.6.11 (and/or -bs1), > nor on 2.6.9! > I'm writing a better one for merge, which works well also for AMD64 users > (which can't use aes-i586). Just a note - the patch I sent yesterday in a hurry was bogus, it didn't allow to enable the config option because I forgot deleting a line. The attached one actually allow to enable the option and it is compiled in. Let me know of the result, so I can make it work for next release. This patch will be in the next 2.6.11-bb/-bs release. -- Paolo Giarrusso, aka Blaisorblade Linux registered user n. 292729 http://www.user-mode-linux.org/~blaisorblade |