On 2018-09-25 11:33 a.m., Bart Van Assche wrote:

quoted

+	/*
+	 * p2pdma mappings are not compatible with devices that use
+	 * dma_virt_ops. If the upper layers do the right thing
+	 * this should never happen because it will be prevented
+	 * by the check in pci_p2pdma_add_client()
+	 */
+	if (WARN_ON_ONCE(IS_ENABLED(CONFIG_DMA_VIRT_OPS) &&
+			 dev->dma_ops == &dma_virt_ops))
+		return 0;

Are you assuming that the compiler will optimize out the dev->dma_ops
== &dma_virt_ops test if CONFIG_DMA_VIRT_OPS=n such that no reference to the
dma_virt_ops symbol appears in the object file? Are you sure all compilers
and compiler versions that are used to build the Linux kernel will do that?

Hmm, I can't say for certain but I would definitely expect any sane
compiler to do so. It's a fairly obvious optimization and the kbuild
robot is happy with it.

I just did a test compile on 4.9 (which is the earliest compiler version
I have readily available) and it worked fine. The minimum gcc the kernel
supports is 4.6 right now.

I've also done a similar test on godbolt.org[1] and can't find any
compiler version that doesn't work.

Logan


[1] https://godbolt.org/z/d5Mhee