Date: Fri, 1 May 2020 04:24:49 +0300 From: Yuri Pankov <ypankov@fastmail.com> To: current@freebsd.org Subject: Re: nvme on 2019 macbooks Message-ID: <79fec490-d3b1-b91b-652c-e9bf0d121d7c@fastmail.com> In-Reply-To: <cb02bcb1-482b-0920-507e-79213c1ba1ff@fastmail.com> References: <cb02bcb1-482b-0920-507e-79213c1ba1ff@fastmail.com>
index | next in thread | previous in thread | raw e-mail
[-- Attachment #1 --] Yuri Pankov wrote: > I have tried booting latest -current snapshot on 2019 macbook pro 16, > and noticed that nvme driver didn't attach, as class reported is > 0x018002 and not 0x010802 that we expect in nvme_pci.c. > > The following linux changes seem to be related: > https://github.com/torvalds/linux/commit/66341331ba0d2de4ff421cdc401a1e34de50502a > > https://github.com/torvalds/linux/commit/d38e9f04ebf667d9cb8185b45bff747485f1d3e9 > > > I have tried adding the exact PCI IDs, but that also fails as number of > msix vectors seems to be 0 in nvme_ctrlr_setup_interrupts() and > nvme_ctrlr_configure_intx() fails with "unable to allocate shared IRQ". > > Any hints on how to proceed here? So it looks like we need to fallback to MSI if we failed to enable MSI-X. With the attached patch we still fail to attach the target device (as below) most likely due to the quirks needed as seen in linux driver, but it's definitely a start: nvme0: CREATE IO CQ (05) sqid:0 cid:15 nsid:0 cdw10:00800001 cdw11:00010003 nvme0: INVALID_FIELD (00/02) sqid:0 cid:15 cdw0:0 nvme0: nvme_create_io_cq failed! [-- Attachment #2 --] diff --git a/sys/dev/nvme/nvme_pci.c b/sys/dev/nvme/nvme_pci.c index 448bfda6a718..e609967b53fe 100644 --- a/sys/dev/nvme/nvme_pci.c +++ b/sys/dev/nvme/nvme_pci.c @@ -90,6 +90,7 @@ static struct _pcsid { 0x05401c5f, 0, 0, "Memblaze Pblaze4", QUIRK_DELAY_B4_CHK_RDY }, { 0xa821144d, 0, 0, "Samsung PM1725", QUIRK_DELAY_B4_CHK_RDY }, { 0xa822144d, 0, 0, "Samsung PM1725a", QUIRK_DELAY_B4_CHK_RDY }, + { 0x2005106b, 0, 0, "ANS2 NVMe Controller" }, { 0x00000000, 0, 0, NULL } }; @@ -267,7 +268,7 @@ nvme_ctrlr_setup_interrupts(struct nvme_controller *ctrlr) force_intx = 0; TUNABLE_INT_FETCH("hw.nvme.force_intx", &force_intx); - if (force_intx || pci_msix_count(dev) < 2) { + if (force_intx) { nvme_ctrlr_configure_intx(ctrlr); return; } @@ -297,9 +298,14 @@ nvme_ctrlr_setup_interrupts(struct nvme_controller *ctrlr) /* One vector for per core I/O queue, plus one vector for admin queue. */ num_vectors_requested = num_io_queues + 1; num_vectors_allocated = num_vectors_requested; + + /* Try MSI-X */ if (pci_alloc_msix(dev, &num_vectors_allocated) != 0) { - nvme_ctrlr_configure_intx(ctrlr); - return; + /* MSI-X failed, try MSI */ + if (pci_alloc_msi(dev, &num_vectors_allocated) != 0) { + nvme_ctrlr_configure_intx(ctrlr); + return; + } } if (num_vectors_allocated < 2) { pci_release_msi(dev);help
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?79fec490-d3b1-b91b-652c-e9bf0d121d7c>