Re: Linux 5.14-rc3

[Guenter Roeck <linux@roeck-us.net>]

On Sun, Jul 25, 2021 at 03:54:56PM -0700, Linus Torvalds wrote:
> Here we are, a week later. After a relatively big rc2, things seem to
> have calmed down and rc3 looks pretty normal. Most of the fixes here
> are small, and the diffstat looks largely flat. And there's not an
> undue amount of stuff.
> 
> The fixes are spread fairly evenly all over - driver changes do
> dominate, but it all seems commensurate with code size in general, so
> not anything odd or unusual. There's two bigger commits, but one of
> them is a revert, and the other is just making the amdgpu codec info
> structure initializations much clearer.
> 
> There are still various discussions about a few pending issues still,
> but on the whole things are looking pretty good, and I know of nothing
> so scary as to keep people from getting their toes wet and testing
> this all out.
> 
> Please test,

Here we are:

Build results:
	total: 154 pass: 153 fail: 1
Failed builds:
	riscv32:allmodconfig
Qemu test results:
	total: 471 pass: 470 fail: 1
Failed tests:
	arm:z2:pxa_defconfig:nodebug:nocd:nofs:nonvme:noscsi:notests:novirt:nofdt:flash8,384k,2:rootfs

riscv32:allmodconfig still reports

cc1: error: '6152' is not a valid offset in '-mstack-protector-guard-offset='

The arm image doesn't boot because of a regression in flash code.
The fix is in the pending-fixes branch and should hopefully find its
way into mainline soon.

There is also still the following warning all over the place.

alg: self-tests for drbg_nopr_hmac_sha512 (stdrng) failed (rc=-22)
WARNING: CPU: 0 PID: 57 at crypto/testmgr.c:5652 alg_test.part.0+0x12c/0x430

The riscv32 boot problem has been fixed.

The riscv compile problem is seen randomly (roughly every other build)
if both CONFIG_STACKPROTECTOR_PER_TASK and CONFIG_GCC_PLUGIN_RANDSTRUCT
are enabled. It looks like the problem may be difficult to reproduce
for others, and I am not sure if it will be fixed anytime soon.
Since the problem may hide other compile problems, I am going to
disable CONFIG_GCC_PLUGIN_RANDSTRUCT in future riscv test builds.
The problem won't be gone, but I will no longer report it.

Similar, the warning about drbg_nopr_hmac_sha512 is only seen if crypto
self tests are enabled and CONFIG_CRYPTO_SHA512 is not enabled. The fix
is queued in -next but not in pending-fixes, meaning it might not make it
into v5.14. Since the warning is distracting and may hide other problems,
I'll force-enable CONFIG_CRYPTO_SHA512 in future test builds and no longer
report this warning.

Guenter