Re: Time to get rid of CPU6 ERRATA on powerpc/8xx ?
christophe leroy writes: > Today, Linux kernel includes a workaround for CPU6 ERRATA on the 8xx > powerpc. > > This ERRATA exists on the 801, the 823, the 855/860 before revision C.0 > It doesn't concern any modern versions of the 8xx, neither the 860 past > and including rev C.0, nor the 866 nor the 885 > > This workaround complicates the TLBmiss and TLBerror handlers and make > the code more and more unreadable. > > Since this workaround addresses very old versions of the 8xx, I'd like > to get rid of it. Do you see any good reason to keep it today ? If not I > will come with a cleanup patch in the coming weeks. I'm happy for you to remove it, and if someone notices and complains loudly we can always put it back. cheers
Re: Time to get rid of CPU6 ERRATA on powerpc/8xx ?
On Sun, 2018-01-07 at 17:23 +0100, christophe leroy wrote: > CAUTION: This email originated from outside of the organization. Do not click > links or open attachments unless you recognize the sender and know the > content is safe. > > > Today, Linux kernel includes a workaround for CPU6 ERRATA on the 8xx > powerpc. > > This ERRATA exists on the 801, the 823, the 855/860 before revision C.0 > It doesn't concern any modern versions of the 8xx, neither the 860 past > and including rev C.0, nor the 866 nor the 885 > > This workaround complicates the TLBmiss and TLBerror handlers and make > the code more and more unreadable. > > Since this workaround addresses very old versions of the 8xx, I'd like > to get rid of it. Do you see any good reason to keep it today ? If not I > will come with a cleanup patch in the coming weeks. > > Another alternative would be to keep that workaround separated from the > rest of the code (ie using different registers to avoid/limit code > nesting), it would add a few cycles but would increase readability while > keeping the ERRATA in, allthought my preference goes to a complete > removal of the workaround. > > So, thanks to let me know your opinion on that. Having done work on TLB Miss/Error, I agree the CPU6 errata is a pain. On all our 8xx (86x/850) we never had to use this errata so I am in favour for removing CPU6 Errata from recent kernels. Jocke
Re: Time to get rid of CPU6 ERRATA on powerpc/8xx ?
On Sun, Jan 07, 2018 at 06:51:26PM +0100, christophe leroy wrote: > Le 07/01/2018 à 17:43, Segher Boessenkool a écrit : > >On Sun, Jan 07, 2018 at 05:23:13PM +0100, christophe leroy wrote: > >>Today, Linux kernel includes a workaround for CPU6 ERRATA on the 8xx > >>powerpc. > >> > >>This ERRATA exists on the 801, the 823, the 855/860 before revision C.0 > >>It doesn't concern any modern versions of the 8xx, neither the 860 past > >>and including rev C.0, nor the 866 nor the 885 > >> > >>This workaround complicates the TLBmiss and TLBerror handlers and make > >>the code more and more unreadable. > >> > >>Since this workaround addresses very old versions of the 8xx, I'd like > >>to get rid of it. Do you see any good reason to keep it today ? If not I > >>will come with a cleanup patch in the coming weeks. > > > >What is "very old"? It'll help if you give some indication. > > CPU6 bug is already announced fixed in rev C.0 in revision 1.4 of the > ERRATA document, issued in August 2000. And that is the newest chip that had the bug? Wow, very old indeed then. > >Removing the workarounds is fine by me, of course... Maybe make it fail > >to boot though, with an error message? Someone *will* try to run it on > >an old system, *especially* if you think no one would. > > Ok, will see how we can do that. It means identifying each revision of > the chip. If it is *that* old you may not have to bother... Up to the maintainers of course. Segher
Re: Time to get rid of CPU6 ERRATA on powerpc/8xx ?
Le 07/01/2018 à 17:43, Segher Boessenkool a écrit : On Sun, Jan 07, 2018 at 05:23:13PM +0100, christophe leroy wrote: Today, Linux kernel includes a workaround for CPU6 ERRATA on the 8xx powerpc. This ERRATA exists on the 801, the 823, the 855/860 before revision C.0 It doesn't concern any modern versions of the 8xx, neither the 860 past and including rev C.0, nor the 866 nor the 885 This workaround complicates the TLBmiss and TLBerror handlers and make the code more and more unreadable. Since this workaround addresses very old versions of the 8xx, I'd like to get rid of it. Do you see any good reason to keep it today ? If not I will come with a cleanup patch in the coming weeks. What is "very old"? It'll help if you give some indication. CPU6 bug is already announced fixed in rev C.0 in revision 1.4 of the ERRATA document, issued in August 2000. Removing the workarounds is fine by me, of course... Maybe make it fail to boot though, with an error message? Someone *will* try to run it on an old system, *especially* if you think no one would. Ok, will see how we can do that. It means identifying each revision of the chip. Thanks Christophe --- L'absence de virus dans ce courrier électronique a été vérifiée par le logiciel antivirus Avast. https://www.avast.com/antivirus
Re: Time to get rid of CPU6 ERRATA on powerpc/8xx ?
On Sun, Jan 07, 2018 at 05:23:13PM +0100, christophe leroy wrote: > Today, Linux kernel includes a workaround for CPU6 ERRATA on the 8xx > powerpc. > > This ERRATA exists on the 801, the 823, the 855/860 before revision C.0 > It doesn't concern any modern versions of the 8xx, neither the 860 past > and including rev C.0, nor the 866 nor the 885 > > This workaround complicates the TLBmiss and TLBerror handlers and make > the code more and more unreadable. > > Since this workaround addresses very old versions of the 8xx, I'd like > to get rid of it. Do you see any good reason to keep it today ? If not I > will come with a cleanup patch in the coming weeks. What is "very old"? It'll help if you give some indication. Removing the workarounds is fine by me, of course... Maybe make it fail to boot though, with an error message? Someone *will* try to run it on an old system, *especially* if you think no one would. Thanks, Segher
Time to get rid of CPU6 ERRATA on powerpc/8xx ?
Today, Linux kernel includes a workaround for CPU6 ERRATA on the 8xx powerpc. This ERRATA exists on the 801, the 823, the 855/860 before revision C.0 It doesn't concern any modern versions of the 8xx, neither the 860 past and including rev C.0, nor the 866 nor the 885 This workaround complicates the TLBmiss and TLBerror handlers and make the code more and more unreadable. Since this workaround addresses very old versions of the 8xx, I'd like to get rid of it. Do you see any good reason to keep it today ? If not I will come with a cleanup patch in the coming weeks. Another alternative would be to keep that workaround separated from the rest of the code (ie using different registers to avoid/limit code nesting), it would add a few cycles but would increase readability while keeping the ERRATA in, allthought my preference goes to a complete removal of the workaround. So, thanks to let me know your opinion on that. Christophe --- L'absence de virus dans ce courrier électronique a été vérifiée par le logiciel antivirus Avast. https://www.avast.com/antivirus
