On Thu, Mar 19, 2026 at 09:47:15AM +0100, Niklas Cassel wrote: > On Thu, Mar 19, 2026 at 10:28:41AM +0900, Koichiro Den wrote: > > On Wed, Mar 18, 2026 at 03:46:27PM +0100, Christian Bruel wrote: > > > Handle -ENOSPC error. If the number of available inbound ATU entries is > > > insufficient to map the subrange, skip the test instead of failing. > > > > > > Link: > > > https://lore.kernel.org/linux-pci/20260317152707.GA85951@bhelgaas/T/#m87e4c24173097a0ea70195b71aab294ad8d6c283 > > > Signed-off-by: Christian Bruel <[email protected]> > > > --- > > > tools/testing/selftests/pci_endpoint/pci_endpoint_test.c | 2 ++ > > > 1 file changed, 2 insertions(+) > > > > > > diff --git a/tools/testing/selftests/pci_endpoint/pci_endpoint_test.c > > > b/tools/testing/selftests/pci_endpoint/pci_endpoint_test.c > > > index > > > c417fb3a198b2d92c3060938c23807cc8bea5573..8ea2fda4539d11eb22b22800a7cb8bbaa99c91ba > > > 100644 > > > --- a/tools/testing/selftests/pci_endpoint/pci_endpoint_test.c > > > +++ b/tools/testing/selftests/pci_endpoint/pci_endpoint_test.c > > > @@ -88,6 +88,8 @@ TEST_F(pci_ep_bar, BAR_SUBRANGE_TEST) > > > SKIP(return, "Subrange map is not supported"); > > > if (ret == -ENOBUFS) > > > SKIP(return, "BAR is reserved"); > > > + if (ret == -ENOSPC) > > > + SKIP(return, "Not enough ATU entries to allocate subrange"); > > > > Thank you for handling this! > > > > pci_endpoint_test_bar_subrange() can also return -ENOSPC locally: > > https://github.com/torvalds/linux/blob/v7.0-rc4/drivers/misc/pci_endpoint_test.c#L538 > > While that is a different resource limit case, I think it would still be > > reasonable to treat it as SKIP as well. > > sub_size = bar_size / nsub; > if (sub_size < sizeof(u32)) { > ret = -ENOSPC; > goto out_clear; > } > > What is this test for anyway? > > If sub_size < 4 ? > return -ENOSPACE ? > > ENOSPC seem like a weird return for this. > > Wouldn't EINVAL be a better return code for this?
Thanks, good point. It is to ensure that each subrange is large enough for 32-bit accesses. I must have used -ENOSPC here thinking of it as "no space for subrange mapping". However, subranges smaller than 32-bit do not make sense for this test, and -EINVAL seems more appropriate here as you pointed out. Also, the PCIe spec defines the minimum Memory Space BAR size as 128 bytes, so subranges smaller than that would not normally exist in practice, which also aligns better with -EINVAL than -ENOSPC. That would keep -ENOSPC reserved for actual resource limitations (i.e. inbound regions limitation). Best regards, Koichiro > > > Kind regards, > Niklas
