Re: [PATCH 4.18 00/79] 4.18.1-stable review

From: Byron Stanoszek
Date: Thu Aug 16 2018 - 00:39:38 EST

On Wed, 15 Aug 2018, Greg Kroah-Hartman wrote:

On Wed, Aug 15, 2018 at 01:24:25PM -0400, Byron Stanoszek wrote:
Hi Greg & Thomas,

I'd like to report a regression in Linux 4.18.1 regarding the L1TF patches.

The kernel no longer thinks I have SMT enabled in the BIOS. This works fine in

Not sure if this matters, but in my particular 4-core system, my third core is
broken (core #2). So I must boot using "maxcpus=2" and then online the other
cores & SMT threads at startup using:

echo 1 > /sys/devices/system/cpu/cpu3/online
echo 1 > /sys/devices/system/cpu/cpu4/online
echo 1 > /sys/devices/system/cpu/cpu5/online
echo 1 > /sys/devices/system/cpu/cpu7/online

In 4.18.0, dmesg shows:

smpboot: Booting Node 0 Processor 3 APIC 0x6
smpboot: Booting Node 0 Processor 4 APIC 0x1
smpboot: Booting Node 0 Processor 5 APIC 0x3
smpboot: Booting Node 0 Processor 7 APIC 0x7

In 4.18.1, dmesg shows:

smpboot: Booting Node 0 Processor 3 APIC 0x6
smpboot: Booting Node 0 Processor 4 APIC 0x1
smpboot: CPU 4 is now offline
smpboot: Booting Node 0 Processor 5 APIC 0x3
smpboot: CPU 5 is now offline
smpboot: Booting Node 0 Processor 7 APIC 0x7
smpboot: CPU 7 is now offline

and I get an "Operation cancelled" error in the shell when trying to online 4,
5, and 7.

In 4.18.1, /sys/devices/system/cpu/smt/control says "notsupported".

- - -

A possible second regression is the following:

My CPU normally runs at 3600 MHz. I usually run my CPU at 2800 MHz to keep from
overheating under full load (it is a fanless system). I do this by running
"echo 1 > /sys/class/thermal/cooling_device5/cur_state", and confirm with "cat
/proc/cpuinfo" (shows 2800).

This works in 4.18.0 but not in 4.18.1. I get no error from the "echo" command
(and the state reads back as "1"), but the CPU remains running at 3600 MHz.

How about Linus's tree at the moment, is it ok there?


greg k-h

It also fails in Linus's tree. Seems like this logic is to blame:

* If SMT was disabled by BIOS, detect it here, after the CPUs have been
* brought online. This ensures the smt/l1tf sysfs entries are consistent
* with reality. cpu_smt_available is set to true during the bringup of non
* boot CPUs when a SMT sibling is detected. Note, this may overwrite
* cpu_smt_control's previous setting.
void __init cpu_smt_check_topology(void)
if (!cpu_smt_available)
cpu_smt_control = CPU_SMT_NOT_SUPPORTED;

SMT is enabled in my BIOS, but because I booted with maxcpus=2, the init code
never officially booted any SMT thread yet--just primary threads. I suspect the
line 'cpu_smt_available = true;' in kernel/cpu.c function cpu_smt_allowed() is
never being reached.

It is then impossible to boot any SMT thread after init is done, since
'cpu_smt_available' is false.

The following test patch makes everything work for me on both mainline and
4.18.1, but we might as well throw out 'cpu_smt_available' altogether then (or
find another way to set it appropriately). Just because we didn't boot any SMT
threads at init shouldn't mean that SMT is disabled by the BIOS.


x86/l1tf: Fix booting with low maxcpus=# causes SMT to be disabled

If maxcpus=# is given on the kernel command line where # is too low for
any SMT CPU (thread) to be booted during init, then the kernel thinks
SMT is disabled by the BIOS. SMT threads are then unable to be manually
brought online later after init.

Set 'cpu_smt_available' early on in init instead, if
topology_smt_supported() returns true.

Fixes: 958f338e96f8 ("Merge branch 'l1tf-final' of git://")
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Signed-off-by: Byron Stanoszek <gandalf@xxxxxxxxx>
kernel/cpu.c | 10 ++--------
1 file changed, 2 insertions(+), 8 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index ed44d7d34c2d..bf9be8b8c0a0 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -378,6 +378,8 @@ void __init cpu_smt_check_topology_early(void)
if (!topology_smt_supported())
cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
+ else
+ cpu_smt_available = true;

@@ -405,14 +407,6 @@ static inline bool cpu_smt_allowed(unsigned int cpu)
if (topology_is_primary_thread(cpu))
return true;

- /*
- * If the CPU is not a 'primary' thread and the booted_once bit is
- * set then the processor has SMT support. Store this information
- * for the late check of SMT support in cpu_smt_check_topology().
- */
- if (per_cpu(cpuhp_state, cpu).booted_once)
- cpu_smt_available = true;
if (cpu_smt_control == CPU_SMT_ENABLED)
return true;