On Thu, May 14, 2020 at 03:32:51PM +0200, Jaromír Doleček wrote: > Le jeu. 14 mai 2020 à 15:19, <[email protected]> a écrit : > > > Bumped ATA_DELAY to 30000 (was 10000), and the VM stayed up overnight, > > > only logging the one correctable soft error: > > > > > > May 7 04:19:29 qemu /netbsd: [ 16290.3345912] autoconfiguration error: > > > piixide0:0:0: lost interrupt > > > May 7 04:19:29 qemu /netbsd: [ 16290.3345912] type: ata tc_bcount: 512 > > > tc_skip: 0 > > > May 7 04:19:29 qemu /netbsd: [ 16290.3345912] autoconfiguration error: > > > piixide0:0:0: bus-master DMA error: missing interrupt, status=0x21 > > > May 7 04:19:29 qemu /netbsd: [ 16290.6088515] wd0a: DMA error writing > > > fsbn 1801813 (wd0 bn 1801876; cn 879 tn 52 sn 20), xfer 38, retry 0 > > > May 7 04:19:29 qemu /netbsd: [ 16292.6053372] wd0: soft error > > > (corrected) xfer 38 > > > > > > Would making ATA_DELAY configurable via options(4) be worth it? > > > > QEMU emulation isn't a niche setup, we should aim to have it work out of > > the box without adjusting sysctls, IMO. > > I agree. Let's fix QEMU IDE emulation?
I don't know that there's anything specifically wrong with the qemu emulation - perhaps the fact that slow host I/O triggers PCI timeouts and missed interrupts in the VM? The reason I'm seeing issues is mostly due to the fact my machine has a mirrored pair of slow WD Green disks, and runs a bunch of junk. At times (esp overnight), I/O bogs down badly, and I hit the current 10s timeouts. > Seriously though I think that it wouldn't hurt to just bump ATA_DELAY > to 30 seconds by default. > > Also as another general suggestion - you'll generally get more > performance by using the PV drivers i.e. virtio, or ahcisata > emulation. > > Jaromir -- Paul Ripke "Great minds discuss ideas, average minds discuss events, small minds discuss people." -- Disputed: Often attributed to Eleanor Roosevelt. 1948.
