x86/tsx: Use MSR_TSX_CTRL to clear CPUID bits

tsx_clear_cpuid() uses MSR_TSX_FORCE_ABORT to clear CPUID.RTM and
CPUID.HLE. Not all CPUs support MSR_TSX_FORCE_ABORT, alternatively use
MSR_IA32_TSX_CTRL when supported.

  [ bp: Document how and why TSX gets disabled. ]

Fixes: 293649307e ("x86/tsx: Clear CPUID bits when TSX always force aborts")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Neelima Krishnan <neelima.krishnan@intel.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/5b323e77e251a9c8bcdda498c5cc0095be1e1d3c.1646943780.git.pawan.kumar.gupta@linux.intel.com
This commit is contained in:
Pawan Gupta 2022-03-10 14:00:59 -08:00 committed by Borislav Petkov
parent ce522ba9ef
commit 258f3b8c32
2 changed files with 48 additions and 7 deletions

View file

@ -722,6 +722,7 @@ static void init_intel(struct cpuinfo_x86 *c)
else if (tsx_ctrl_state == TSX_CTRL_DISABLE)
tsx_disable();
else if (tsx_ctrl_state == TSX_CTRL_RTM_ALWAYS_ABORT)
/* See comment over that function for more details. */
tsx_clear_cpuid();
split_lock_init();

View file

@ -58,7 +58,7 @@ void tsx_enable(void)
wrmsrl(MSR_IA32_TSX_CTRL, tsx);
}
static bool __init tsx_ctrl_is_supported(void)
static bool tsx_ctrl_is_supported(void)
{
u64 ia32_cap = x86_read_arch_cap_msr();
@ -84,6 +84,44 @@ static enum tsx_ctrl_states x86_get_tsx_auto_mode(void)
return TSX_CTRL_ENABLE;
}
/*
* Disabling TSX is not a trivial business.
*
* First of all, there's a CPUID bit: X86_FEATURE_RTM_ALWAYS_ABORT
* which says that TSX is practically disabled (all transactions are
* aborted by default). When that bit is set, the kernel unconditionally
* disables TSX.
*
* In order to do that, however, it needs to dance a bit:
*
* 1. The first method to disable it is through MSR_TSX_FORCE_ABORT and
* the MSR is present only when *two* CPUID bits are set:
*
* - X86_FEATURE_RTM_ALWAYS_ABORT
* - X86_FEATURE_TSX_FORCE_ABORT
*
* 2. The second method is for CPUs which do not have the above-mentioned
* MSR: those use a different MSR - MSR_IA32_TSX_CTRL and disable TSX
* through that one. Those CPUs can also have the initially mentioned
* CPUID bit X86_FEATURE_RTM_ALWAYS_ABORT set and for those the same strategy
* applies: TSX gets disabled unconditionally.
*
* When either of the two methods are present, the kernel disables TSX and
* clears the respective RTM and HLE feature flags.
*
* An additional twist in the whole thing presents late microcode loading
* which, when done, may cause for the X86_FEATURE_RTM_ALWAYS_ABORT CPUID
* bit to be set after the update.
*
* A subsequent hotplug operation on any logical CPU except the BSP will
* cause for the supported CPUID feature bits to get re-detected and, if
* RTM and HLE get cleared all of a sudden, but, userspace did consult
* them before the update, then funny explosions will happen. Long story
* short: the kernel doesn't modify CPUID feature bits after booting.
*
* That's why, this function's call in init_intel() doesn't clear the
* feature flags.
*/
void tsx_clear_cpuid(void)
{
u64 msr;
@ -97,6 +135,10 @@ void tsx_clear_cpuid(void)
rdmsrl(MSR_TSX_FORCE_ABORT, msr);
msr |= MSR_TFA_TSX_CPUID_CLEAR;
wrmsrl(MSR_TSX_FORCE_ABORT, msr);
} else if (tsx_ctrl_is_supported()) {
rdmsrl(MSR_IA32_TSX_CTRL, msr);
msr |= TSX_CTRL_CPUID_CLEAR;
wrmsrl(MSR_IA32_TSX_CTRL, msr);
}
}
@ -106,13 +148,11 @@ void __init tsx_init(void)
int ret;
/*
* Hardware will always abort a TSX transaction if both CPUID bits
* RTM_ALWAYS_ABORT and TSX_FORCE_ABORT are set. In this case, it is
* better not to enumerate CPUID.RTM and CPUID.HLE bits. Clear them
* here.
* Hardware will always abort a TSX transaction when the CPUID bit
* RTM_ALWAYS_ABORT is set. In this case, it is better not to enumerate
* CPUID.RTM and CPUID.HLE bits. Clear them here.
*/
if (boot_cpu_has(X86_FEATURE_RTM_ALWAYS_ABORT) &&
boot_cpu_has(X86_FEATURE_TSX_FORCE_ABORT)) {
if (boot_cpu_has(X86_FEATURE_RTM_ALWAYS_ABORT)) {
tsx_ctrl_state = TSX_CTRL_RTM_ALWAYS_ABORT;
tsx_clear_cpuid();
setup_clear_cpu_cap(X86_FEATURE_RTM);