Re: [PATCH v2 00/21] Runtime TDX Module update support

[Chao Gao]

On Wed, Nov 19, 2025 at 04:44:50PM -0600, Sagi Shahar wrote:
>On Thu, Oct 30, 2025 at 9:53 PM Chao Gao <chao.gao@xxxxxxxxx> wrote:
>>
>> >A reference patch we tested for "Avoid updates during update-sensitive
>> >times" and one caveat was that
>> >/sys/devices/virtual/tdx/tdx_tsm/version was not available post update
>> >failure until a subsequent successful update:
>>
>> I also tested this. It works well to prevent updates during TD build, so,
>>
>>   Tested-by: Chao Gao <chao.gao@xxxxxxxxx>
>>
>> And I can integrate this change into my next version if you don't object.
>>
>> Regarding the caveat, could you check if the diff [*] I posted earlier this
>> week can fix it?
>>
>> [1]: https://lore.kernel.org/linux-coco/aQAwRrvYMcaMsu02@xxxxxxxxx/
>
>[Now in plaintext]
>
>I tried testing it with the 1.5.24 TDX module and it sometimes fails,
>but the failure does not appear consistent.
>
>I added a local change to add the
>TDX_SYS_SHUTDOWN_AVOID_COMPAT_SENSITIVE flag when calling
>TDH_SYS_SHUTDOWN and TDH_SYS_SHUTDOWN fails as expected if a VM is
>under build:
>[ 1224.571177] virt/tdx: SEAMCALL (52) failed: 0x8000051200010000
>
>But then sometimes trying to finalize the VM fail with the following error:
>[ 1230.915145] kvm_intel: SEAMCALL TDH_MR_FINALIZE failed: 0x8000ff00ffff0000
>[ 1230.948264] kvm_intel: tdh_mng_vpflushdone() failed. HKID 3 is leaked.
>
>At this point the module seems to be in a broken state and trying to
>create more TDs will fail:
>[ 1543.745606] kvm_intel: SEAMCALL TDH_MNG_CREATE failed: 0x8000ff00ffff0000
>
>Trying to update the module will fail shutdown with -ENODEV

Can you apply this incremental change to see if the issue gets fixed?

diff --git a/arch/x86/virt/vmx/tdx/seamldr.c b/arch/x86/virt/vmx/tdx/seamldr.c
index e525bbd16610..f0bea1fecc52 100644
--- a/arch/x86/virt/vmx/tdx/seamldr.c
+++ b/arch/x86/virt/vmx/tdx/seamldr.c
@@ -317,8 +317,9 @@ static int do_seamldr_install_module(void *params)
					tdx_module_set_error();
					print_update_failure_message();
				}
+			} else {
+				ack_state();
			}
-			ack_state();
		} else {
			touch_nmi_watchdog();
			rcu_momentary_eqs();


The problem is if the failing CPU is the last one to ack the TDP_SHUTDOWN
state, the state will move to TDP_CPU_INSTALL state. Other CPUs may proceed to
install the new module before seeing tdp_data.failed. This disables TDX ISA, so
any subsequent SEAMCALLs get 0x8000ff00ffff0000.