On 11.06.2021 12:03, Borislav Petkov wrote:

On Wed, Jun 09, 2021 at 02:13:38PM -0700, Pawan Gupta wrote:

quoted

@@ -114,6 +127,26 @@ void __init tsx_init(void)
 			tsx_ctrl_state = TSX_CTRL_ENABLE;
 	}

+	/*
+	 * Hardware will always abort a TSX transaction if both CPUID bits
+	 * RTM_ALWAYS_ABORT and TSX_FORCE_ABORT are enumerated.  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)) {
+		tsx_ctrl_state = TSX_CTRL_RTM_ALWAYS_ABORT;
+		tsx_clear_cpuid();
+		setup_clear_cpu_cap(X86_FEATURE_RTM);
+		setup_clear_cpu_cap(X86_FEATURE_HLE);
+		return;
+	}

Why isn't this happening as the first thing on entry in that tsx_init()
function? IOW, there's no point to noodle through cmdline options etc if
the hardware will always abort transactions.

Because the next patch is adding tsx=fake cmdline, for that this code
needs to be executed after cmdline parsing. I wanted to avoid the churn
to move this code between patches.

But, I see your comment for 4/4 i.e. tsx=fake may not be needed. It was
added after an internal test failure.

tsx=fake is really for a corner case and if that patch goes away it
would make sense to move the code in question here before cmdline
parsing.

Thanks,
Pawan