Date: Tue, 22 Sep 2020 14:27:27 -0700 From: Mark Millard <marklmi@yahoo.com> To: Brandon Bergren <bdragon@FreeBSD.org> Cc: FreeBSD PowerPC ML <freebsd-ppc@freebsd.org> Subject: Re: head -r365932 on PowerMac G5 (2 dual-core sockets): Crashes before login prompt if powerd is enabled in /etc/rc.conf Message-ID: <D6DC1C60-E50D-4A12-9F3B-6CEBA6A76748@yahoo.com> In-Reply-To: <92f57afb-33cb-4d38-be17-4b8c81cb9c89@www.fastmail.com> References: <52783D16-5DCA-45BC-9238-2518326454A1@yahoo.com> <6E99EE39-D2B8-415A-A5BF-823C0F0C22D6@yahoo.com> <cd9d2b72-219f-4550-a437-4ac3aa1da66d@www.fastmail.com> <AF27169A-00FC-4984-83C2-307EA885D7A1@yahoo.com> <11fe573a-24c3-47be-95ed-c601ec54f168@www.fastmail.com> <E23CF7F6-B69C-44C7-B356-4AB65E011268@yahoo.com> <B4D16405-1B0C-4500-866F-75F32FED235A@yahoo.com> <92f57afb-33cb-4d38-be17-4b8c81cb9c89@www.fastmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2020-Sep-22, at 12:45, Brandon Bergren <bdragon at FreeBSD.org> = wrote: >=20 > On Tue, Sep 22, 2020, at 2:35 PM, Mark Millard wrote: >>=20 >> Seems to target 32 bit powerpc but the context I'm reporting is >> for powerpc64 overall. May be the mis-compiles are not limited >> to 32-bit powerpc variants? >=20 > DOH! You're right,=20 >=20 > Try a build with _MINUS_O forced to -O2 and see if that fixes things. >=20 > FWIW, there is a differential to change the default across the board @ = https://reviews.freebsd.org/D26471 because the whole "inlines are not = inlining" problem affects everything. >=20 > The change boils down to LLVM 11 changing their default of what -O = "means" to be -O1 instead of -O2, which breaks a bunch of kernel = assumptions regarding inlining. >=20 On the 2 dual-core socket G5: # sysctl dev.cpu.0.freq dev.cpu.0.freq: 2500 based on using: # svnlite diff /usr/src/sys/conf/kern.pre.mk Index: /usr/src/sys/conf/kern.pre.mk =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D --- /usr/src/sys/conf/kern.pre.mk (revision 365932) +++ /usr/src/sys/conf/kern.pre.mk (working copy) @@ -51,15 +51,7 @@ SIZE?=3D size =20 .if defined(DEBUG) -.if ${MACHINE_ARCH} =3D=3D "powerpc" || ${MACHINE_ARCH} =3D=3D = "powerpcspe" -# Work around clang 11 miscompile on 32 bit powerpc. -_MINUS_O=3D -O2 -.else -_MINUS_O=3D -O -.endif CTFFLAGS+=3D -g -.else -_MINUS_O=3D -O2 .endif .if ${MACHINE_CPUARCH} =3D=3D "amd64" .if ${COMPILER_TYPE} =3D=3D "clang" @@ -68,7 +60,7 @@ COPTFLAGS?=3D-O2 -frename-registers -pipe .endif .else -COPTFLAGS?=3D${_MINUS_O} -pipe +COPTFLAGS?=3D-O2 -pipe .endif .if !empty(COPTFLAGS:M-O[23s]) && = empty(COPTFLAGS:M-fno-strict-aliasing) COPTFLAGS+=3D -fno-strict-aliasing [I doubt that this had anything to do with the extremely early 2-socket G4 boot failure, given what you had for "powerpc": it should have been -O2 already.] =3D=3D=3D Mark Millard marklmi at yahoo.com ( dsl-only.net went away in early 2018-Mar)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?D6DC1C60-E50D-4A12-9F3B-6CEBA6A76748>