On Tue, Feb 1, 2022 at 2:20 PM Ben Widawsky <ben.widawsky@xxxxxxxxx> wrote: > > On 22-02-01 14:15:22, Dan Williams wrote: > > On Tue, Feb 1, 2022 at 2:11 PM Ben Widawsky <ben.widawsky@xxxxxxxxx> wrote: > > > > > > On 22-02-01 13:41:50, Dan Williams wrote: > > > > On Tue, Feb 1, 2022 at 7:24 AM Ben Widawsky <ben.widawsky@xxxxxxxxx> wrote: > > > > > > > > > > On 22-01-31 18:19:24, Jonathan Cameron wrote: > > > > > > On Sun, 23 Jan 2022 16:31:02 -0800 > > > > > > Dan Williams <dan.j.williams@xxxxxxxxx> wrote: > > > > > > > > > > > > > From: Ben Widawsky <ben.widawsky@xxxxxxxxx> > > > > > > > > > > > > > > The PCIe device DVSEC, defined in the CXL 2.0 spec, 8.1.3 is required to > > > > > > > be implemented by CXL 2.0 endpoint devices. Since the information > > > > > > > contained within this DVSEC will be critically important, it makes sense > > > > > > > to find the value early, and error out if it cannot be found. > > > > > > > > > > > > > > Signed-off-by: Ben Widawsky <ben.widawsky@xxxxxxxxx> > > > > > > > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx> > > > > > > Guess the logic makes sense about checking this early though my cynical > > > > > > mind says, that if someone is putting in devices that claim to be > > > > > > CXL ones and this isn't there it is there own problem if they > > > > > > kernel wastes effort bringing the driver up only to find later > > > > > > it can't finish doing so... > > > > > > > > > > I don't remember if Dan and I discussed actually failing to bind this early if > > > > > the DVSEC isn't there. > > > > > > > > On second look, the error message does not make sense because there is > > > > "no functionality" not "limited functionality" as a result of this > > > > failure because the cxl_pci driver just gives up. This failure should > > > > be limited to cxl_mem, not cxl_pci as there might still be value in > > > > accessing the mailbox on this device. > > > > > > > > > I think the concern is less about wasted effort and more > > > > > about the inability to determine if the device is actively decoding something > > > > > and then having the kernel driver tear that out when it takes over the decoder > > > > > resources. This was specifically targeted toward the DVSEC range registers > > > > > (obviously things would fail later if we couldn't find the MMIO). > > > > > > > > If there is no CXL DVSEC then cxl_mem should fail, that's it. > > > > > > > > > > If there is no CXL DVSEC we have no way to find the device's MMIO. You need the > > > register locator dvsec. Not sure how you intend to do anything with the device > > > at that point, but if you see something I don't, then by all means, change it. > > > > I see: > > > > pci_find_dvsec_capability(pdev, PCI_DVSEC_VENDOR_ID_CXL, CXL_DVSEC_PCIE_DEVICE); > > > > ...and: > > > > pci_find_dvsec_capability(pdev, PCI_DVSEC_VENDOR_ID_CXL, CXL_DVSEC_REG_LOCATOR); > > > > ...aren't they independent? > > My mistake. I was thinking of a different patch, "cxl/pci: Retrieve CXL DVSEC > memory info". You're correct, they are independent (both mandatory for type 3 > devices). > > However, Jonathan was the one who originally suggested it. I had it as a warn > originally. > https://lore.kernel.org/linux-cxl/20211122223430.gvkwj3yeckriffes@xxxxxxxxx/ At least to the concern of "nothing" working without the base CXL DVSEC the cxl_mem driver failing to attach catches that case. Otherwise a device that only implements the mailbox seems not outside the realm of possibility. Jonathan?
