Date: Mon, 15 Dec 2014 22:04:06 -0800 From: Peter Wemm <peter@wemm.org> To: Adrian Chadd <adrian@freebsd.org> Cc: Alfred Perlstein <bright@mu.org>, "freebsd-stable@freebsd.org" <freebsd-stable@freebsd.org>, Ian Lepore <ian@freebsd.org> Subject: Re: i386 PAE kernel works fine on 10-stable Message-ID: <2036758.o6RQzZpOtp@overcee.wemm.org> In-Reply-To: <CAJ-VmokJ=%2BNwAfzoATzaiWD1RsbWeSbb07DLUeYKwCOSCzDuBg@mail.gmail.com> References: <1418579278.2026.9.camel@freebsd.org> <A33FE253-B9EE-46FE-9229-CFBC21A80024@mu.org> <CAJ-VmokJ=%2BNwAfzoATzaiWD1RsbWeSbb07DLUeYKwCOSCzDuBg@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
--nextPart4420963.u7c8slbCU8 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="us-ascii" On Monday, December 15, 2014 05:12:39 PM Adrian Chadd wrote: > On 15 December 2014 at 16:33, Alfred Perlstein <bright@mu.org> wrote:= > >> On Dec 15, 2014, at 3:42 PM, Peter Wemm <peter@wemm.org> wrote: > >>=20 > >> It's always been a 1GB/3GB split. It was never a problem until ce= rtain > >> scaling defaults were changed to scale solely based on physical ra= m > >> without > >> regard for kva limits. > >=20 > > Hmm the original patch I gave for that only changed scaling for mac= hines > > with 64 bit pointers. Why was it that the 32 bit stuff was made to > > change? > I recall this - I went digging; commit >=20 > 7beb738c8a72cc197d3e898784afe3fba28f1834 >=20 > removed that particular bit of autotuning based on the size of void *= . >=20 > Maybe this stuff is a little busted and we need to add some more > config parameters? >=20 > (Also, there's the vm space shift thing that also needed adjustment > for memory-constrained systems. We've had do it on MIPS and low-memor= y > (64mb, 128mb) RAM i386 systems.) >=20 >=20 >=20 > -adrian There's more to it than just one commit. There's years of general tren= ds that=20 added scaling by physical ram without regard for kva consumption. The=20= starting point for Alfred's commits a few years ago already didn't work= =20 properly on large memory systems. Alfred worked on the scaling for the= 64-bit=20 / direct map systems and tried to leave the caps alone for kva-constrai= ned=20 systems. The caps were already too high. Except for i386, kva-limited systems generally are pretty small and are= =20 explicitly tuned for "small". The tuning considerations for a 32MB or = 64MB=20 mips/arm system is very different than a 16GB i386 PAE system. =2D-=20 Peter Wemm - peter@wemm.org; peter@FreeBSD.org; peter@yahoo-inc.com; KI= 6FJV UTF-8: for when a ' or ... just won\342\200\231t do\342\200\246 --nextPart4420963.u7c8slbCU8 Content-Type: application/pgp-signature; name="signature.asc" Content-Description: This is a digitally signed message part. Content-Transfer-Encoding: 7Bit -----BEGIN PGP SIGNATURE----- Version: GnuPG v2 iQEcBAABAgAGBQJUj8taAAoJEDXWlwnsgJ4EA/EIAItF3mNK02PQvolcgS4y9fJb kQkNUTKo7gzWTaclcTNQ90OmHmQExs9uOMD/i5EN2thTfil7YmcwlrF+cSrJPb/l 5jEBn4irS/31C59bG4LmwmEuq3QW2kQj9i6l6avZuC3mXcMpRNHYkz/8X2YdXkTn o+DFHQOzB+rGwIjl5aRoO2OZnMjdmxqtSI+OVlU7bOLMEu+TcdbcggsI/kqKHR2X iwFiSokL8Gz8fzBJjrdGCorU71xhhXbaycSK0NmJACiHvXCCXyHx2AglTGuojClg tgDQlG6GOkBK7ymhyCwTtpldWUPUnutD/D01Akjre3DY88YxakkHLQfb6EFOiMU= =iOt5 -----END PGP SIGNATURE----- --nextPart4420963.u7c8slbCU8--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?2036758.o6RQzZpOtp>