Re: [PATCH v4 17/36] arm64/mm: Handle GCS data aborts

On Fri, Aug 11, 2023 at 04:09:33PM +0100, Catalin Marinas wrote:

On Mon, Aug 07, 2023 at 11:00:22PM +0100, Mark Brown wrote:

quoted

+		if (is_write_abort(esr) &&
+		    !(is_gcs_fault(esr) || is_el1_data_abort(esr)))
+			return VM_FAULT_BADACCESS;

Related to my PIE permissions comment: when do we have a valid EL1 data
write abort that's not a GCS fault? Does a faulting GCSSTTR set the
ESR_ELx_GCS bit?

Yes, it should do.  The GCS instructions have access descriptors created
with CreateAccDescGCS() which results in the access being flagged as a
GCS access.

quoted

+	} else {
+		/*
+		 * GCS faults should never happen for pages that are
+		 * not part of a GCS and the operation being attempted
+		 * can never succeed.
+		 */
+		if (is_gcs_fault(esr))
+			return VM_FAULT_BADACCESS;

If one does a GCS push/store to a non-GCS page, do we get a GCS fault or
something else? I couldn't figure out from the engineering spec. If the
hardware doesn't generate such exceptions, we might as well remove this
'else' branch. But maybe it does generate a GCS-specific fault as you
added a similar check in is_invalid_el0_gcs_access().

Yes, see AddGCSRecord() and LoadCheckGCSRecord() - all GCS initiated
accesses need to be AccDescGCS so appropriate permissions enforcement
can happen and that's what causes the fault to be flagged as GCS.

quoted

@@ -595,6 +644,19 @@ static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
 	if (!vma)
 		goto lock_mmap;
 
+	/*
+	 * We get legitimate write faults for GCS pages from GCS
+	 * operations and from EL1 writes to EL0 pages but just plain

What are the EL1 writes to the shadow stack? Would it not use
copy_to_user_gcs()?

They should, yes - I'll reword the comment.