powerpc: Disable HFSCR[TM] if TM is not supported
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>
Mon, 20 Mar 2017 06:49:03 +0000 (17:49 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 12 Apr 2017 10:41:13 +0000 (12:41 +0200)
commit6fbf84b5da23e7436b710660d2fd5d136e1bcfd3
tree9468245aa934e5f0b5239564d16c86bc6f672fe0
parenta1db9b2c1b120c7779cba2587f0cb3af4c3f8042
powerpc: Disable HFSCR[TM] if TM is not supported

commit 7ed23e1bae8bf7e37fd555066550a00b95a3a98b upstream.

On Power8 & Power9 the early CPU inititialisation in __init_HFSCR()
turns on HFSCR[TM] (Hypervisor Facility Status and Control Register
[Transactional Memory]), but that doesn't take into account that TM
might be disabled by CPU features, or disabled by the kernel being built
with CONFIG_PPC_TRANSACTIONAL_MEM=n.

So later in boot, when we have setup the CPU features, clear HSCR[TM] if
the TM CPU feature has been disabled. We use CPU_FTR_TM_COMP to account
for the CONFIG_PPC_TRANSACTIONAL_MEM=n case.

Without this a KVM guest might try use TM, even if told not to, and
cause an oops in the host kernel. Typically the oops is seen in
__kvmppc_vcore_entry() and may or may not be fatal to the host, but is
always bad news.

In practice all shipping CPU revisions do support TM, and all host
kernels we are aware of build with TM support enabled, so no one should
actually be able to hit this in the wild.

Fixes: 2a3563b023e5 ("powerpc: Setup in HFSCR for POWER8")
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Tested-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
[mpe: Rewrite change log with input from Sam, add Fixes/stable]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/powerpc/kernel/setup_64.c