Waaay back in October, we saw some issues with various system firmware
talking to the AMD SP PSP devices (the secure processor feature), and so
CONFIG_CRYPTO_DEV_SP_PSP got disabled in all of our kernel trees, as
Laura noted here: https://bugzilla.redhat.com/show_bug.cgi?id=1608242#c44 .
I'm trying to use it, and this obviously makes that impossible to do, so
I'm going to re-enable it in rawhide, because the timeout that's being
discussed in that bugzilla thread is already there, and so this firmware
bugs people saw before shouldn't make things completely unusable.
If you *do* see debilitating stalls on AMD EPYC hardware starting after
the next rawhide build, make some noise so we can disable it again.
(I've also posted this to kernel@l.fp.o: https://lists.fedoraproject.org/archives/list/kernel@lists.fedoraproject.org/thread/22XCMRU4R75M4D5OSXE7YMWYVOBQIU2M/ )
Signed-off-by: Peter Jones <pjones@redhat.com>
The shipped .config files all have CONFIG_TICK_CPU_ACCOUNTING and
CONFIG_VIRT_CPU_ACCOUNTING_NATIVE set. The x86_64 .config files
additionaly have CONFIG_VIRT_CPU_ACCOUNTING_GEN set.
This is wrong because each .config should have just one of these three
set. As a result the build generates
[...]⚠️ override: VIRT_CPU_ACCOUNTING_NATIVE changes choice state
24 times and
[...]⚠️ override: VIRT_CPU_ACCOUNTING_GEN changes choice state
4 times. So let's make sure the shipped .config files only have one of
these three Kconfig macros set (ie, the one that's relevant for that
.config target).
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
The shipped .config files all have CONFIG_FLATMEM_MANUAL and
CONFIG_SPARSEMEM_MANUAL set.
This is wrong because each .config should have just one of these two set
set. (There's also CONFIG_DISCONTIGMEM_MANUAL but Fedora doesn't use
that.) As a result the build generates
[...]⚠️ override: SPARSEMEM_MANUAL changes choice state
24 times.
If the configuration generation system drops CONFIG_FLATMEM_MANUAL the
warnings go away. And that doesn't change the final .config files that
the kernel's build system actually uses.
(Two details that very few people will care about:
1. x86/i686 still needs to unset CONFIG_FLATMEM_MANUAL to keep
process_configs.sh from killing the build;
2. the kernel build of armv7hl will silently drop
CONFIG_SPARSEMEM_MANUAL. Because of that I didn't bother with an
explicit override for arm/armv7.)
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
The shipped .config files for i686 have both CONFIG_UNWINDER_ORC and
CONFIG_UNWINDER_FRAME_POINTER set. This is wrong because those .config
files should only have CONFIG_UNWINDER_FRAME_POINTER set. As a result
the build generates
[...]⚠️ override: UNWINDER_ORC changes choice state
4 times.
Move CONFIG_UNWINDER_ORC to x86/x86_64 to correct this. Note that this
doesn't change the final .config files that the kernel's build system
actually uses.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
This script detects unknown Kconfig symbols by checking whether the
Kconfig macros used in the various *.config files have a corresponding
Kconfig symbol in the tree. Its core test is done with grep().
It turns out that match pattern used in grep() is too broad. For
instance, it doesn't report CONFIG_ACPI_PROCFS as unknown because there
is a Kconfig entry for ACPI_PROCFS_POWER in the tree. Make the pattern
match Kconfig symbols exactly to correct this.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Five files that are used in the configuration generation have typos in
their names. Those typos have no effect on the build, but should still
be fixed since they could be confusing (to people and scripts).
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
With the introduction of the new method of configuration generation the
names of the shipped kernel configuration files changed too. Adjust the
glob pattern in the check-configs.pl script to those changed names.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>