> Date: Fri, 19 Nov 2021 21:56:19 +0100 (CET) > From: Mark Kettenis <mark.kette...@xs4all.nl> > > > Date: Fri, 19 Nov 2021 21:36:55 +0100 > > From: Christopher Zimmermann <chr...@openbsd.org> > > > > It is about MSI. After disabling MSI interrupts are delivered, albeit > > the performance seems to be a bit poor (reading at a constant 8 MB/s on > > a WD Red disk). > > > > Here's the dmesg with patched kernel - notice the intx instead msi: > > ahci0 at pci1 dev 0 function 0 "JMicron JMB58x AHCI" rev 0x00: intx, AHCI > > 1.3.1 > > > > Should I commit this quirk? Or might this break something on other > > platforms? > > Before we do this, we should establish whether this is a rkpcie(4) > problem or if that PCI device really has an issue with MSIs.
So MSIs still work fine on my firefly-rk3399: nvme0 at pci1 dev 0 function 0 "Silicon Motion SM2260 NVMe" rev 0x03: msix, NVMe 1.2 nvme0: ADATA SX8000NP, firmware C2.2.1, serial 2H1220011170 scsibus0 at nvme0: 2 targets, initiator 0 sd0 at scsibus0 targ 1 lun 0: <NVMe, ADATA SX8000NP, C2.2> sd0: 122104MB, 512 bytes/sector, 250069680 sectors Can you post a full dmesg for your RockPro64, and a pcidump -vxxx output of possible?
