On 17.05.24 16:53, Kirill A. Shutemov wrote:
On Fri, May 17, 2024 at 04:37:16PM +0200, Juergen Gross wrote:
On 17.05.24 16:32, Kirill A. Shutemov wrote:
On Mon, Feb 26, 2024 at 12:25:41AM -0800, isaku.yamahata@xxxxxxxxx wrote:
@@ -725,6 +967,17 @@ static int __init tdx_module_setup(void)
tdx_info->nr_tdcs_pages = tdcs_base_size / PAGE_SIZE;
+ /*
+ * Make TDH.VP.ENTER preserve RBP so that the stack unwinder
+ * always work around it. Query the feature.
+ */
+ if (!(tdx_info->features0 & MD_FIELD_ID_FEATURES0_NO_RBP_MOD) &&
+ !IS_ENABLED(CONFIG_FRAME_POINTER)) {
I think it supposed to be IS_ENABLED(CONFIG_FRAME_POINTER). "!" shouldn't
be here.
No, I don't think so.
With CONFIG_FRAME_POINTER %rbp is being saved and restored, so there is no
problem in case the seamcall is clobbering it.
Could you check setup_tdparams() in your tree?
Commit
[SEAM-WORKAROUND] KVM: TDX: Don't use NO_RBP_MOD for backward compatibility
in my tree comments out the setting TDX_CONTROL_FLAG_NO_RBP_MOD.
I now remember there was problem in EDK2 using RBP. So the patch is
temporary until EDK2 is fixed.
I have the following line in setup_tdparams() (not commented out):
td_params->exec_controls = TDX_CONTROL_FLAG_NO_RBP_MOD;
Juergen
