Hello ! I just asked Ralf for an account to deposite the patches (which he offered in May). Indeed there are many changes since the 1st version (by now i wonder why it worked so well ;-), and they are made to 2.4 and 2.6. 2.4 runs astonishing stable, for 2.6 i can't tell too much, since the (my ?) system slowly falls asleep while working through rc.sysinit. (The move to 2.6 was big leap backward, almost like beginning from the start again :-() The former scsi driver definitely didn't work (very long), i had to rewrite sgiwd93.c *very* heavily, the diff is larger than the source file (the same applies to sgiseeq.c). To attack the R10k issues at source code level, one can not do much more than carefully adjust the dma-buffers to the cache, do invalidate or write back in the appropriate places - for details see sgiwd93.c and sgiseeq.c. (Moreover, on IP28/IP26 there is a problem with uncached writes to main memory, see comments for more information) However, without compiler/assembler support the "Side Effects of Speculative Execution" (R10k manual) hang over us like Damocles' sword. It indeed happened here at the border of ram, and could be solved by restricting the available memory. (Inspecting some hundred thousand lines of code to insert cache barriers isn't appropriate...) For now, i shall pack together (some twenty) diff files to preliminary 2.4 patches (one for 2.4.22, one for the current files from cvsweb), without the not yet written readme file, but with exclusion of any warranty ;-) and send these to to you until tonight. with kind regards pf BTW: did the ip22zilog patch put it to live again ? On Wed, 14 Jul 2004, Kumba wrote: > > Date: Wed, 14 Jul 2004 23:07:53 -0400 > > From: Kumba > > To: Peter Fuerst > > Subject: IP28 Patch > > > > > > I got Ladis to send me your 2.4.22 patch, and I managed to get most of > > it applied to 2.4.26 (snapshot from 20040712) and surprisingly, it > > actually boots and makes it to userland (nfsroot). > > > > I was curious to know if you've got anything newer worth trying, either > > for 2.4.x or even 2.6.x. The only bits I wasn't able to sucessfully > > merge from your 2.4.22 patch into 2.4.26 was the bits for the scsi > > driver (things changed too much and I wasn't able to figure out what > > went where). Can't really test this all too much, I wasn't about to > > risk the existing install on the I2's drive (which runs under an R4x00 > > system), so if you have any specific tests you'd like run, or other > > data, lemme know. > > > > Also, what's the basis you used to get around the R10K issues? Another > > gentoo user is interested in working on a similar solution for the SGI > > O2 (IP32) for that machines R10K/R12K processors, and it suffers similar > > issues to the IP28/R10K. > > > > > > --Kumba > > > > -- > > "Such is oft the course of deeds that move the wheels of the world: > > small hands do them because they must, while the eyes of the great are > > elsewhere." --Elrond > > On Wed, 14 Jul 2004, Kumba wrote: > > Date: Wed, 14 Jul 2004 23:07:53 -0400 > > From: Kumba > > To: Peter Fuerst > > Subject: IP28 Patch > > > > > > I got Ladis to send me your 2.4.22 patch, and I managed to get most of > > it applied to 2.4.26 (snapshot from 20040712) and surprisingly, it > > actually boots and makes it to userland (nfsroot). > > > > I was curious to know if you've got anything newer worth trying, either > > for 2.4.x or even 2.6.x. The only bits I wasn't able to sucessfully > > merge from your 2.4.22 patch into 2.4.26 was the bits for the scsi > > driver (things changed too much and I wasn't able to figure out what > > went where). Can't really test this all too much, I wasn't about to > > risk the existing install on the I2's drive (which runs under an R4x00 > > system), so if you have any specific tests you'd like run, or other > > data, lemme know. > > > > Also, what's the basis you used to get around the R10K issues? Another > > gentoo user is interested in working on a similar solution for the SGI > > O2 (IP32) for that machines R10K/R12K processors, and it suffers similar > > issues to the IP28/R10K. > > > > > > --Kumba > > > > -- > > "Such is oft the course of deeds that move the wheels of the world: > > small hands do them because they must, while the eyes of the great are > > elsewhere." --Elrond > >