Hello,

On 29/11/17 06:15, Yisheng Xie wrote:

Hi Jean,

On 2017/10/6 21:31, Jean-Philippe Brucker wrote:

quoted

-	if (domain->ext_handler) {
+	if (domain->handler_flags & IOMMU_FAULT_HANDLER_ATOMIC) {
+		fault->flags |= IOMMU_FAULT_ATOMIC;

Why remove the condition of domain->ext_handler? should it be much better like:
  if ((domain->handler_flags & IOMMU_FAULT_HANDLER_ATOMIC) && domain->ext_handler)

If domain->ext_handler is NULL, and (domain->handler_flags & IOMMU_FAULT_HANDLER_ATOMIC)
is true. It will oops, right?

I removed the check because ext_handler shouldn't be NULL if handler_flags
has a bit set (as per iommu_set_ext_fault_handler). But you're right that
this is fragile, and I overlooked the case where users could call
set_ext_fault_handler to clear the fault handler.

(Note that this ext_handler will most likely be replaced by the fault
infrastructure that Jacob is working on:
https://patchwork.kernel.org/patch/10063385/ to which we should add the
atomic/blocking flags)

Thanks,
Jean