On Mon, Jan 21, 2013 at 10:43 AM, Christoffer Dall <c.dall@xxxxxxxxxxxxxxxxxxxxxx> wrote: > On Mon, Jan 21, 2013 at 10:39 AM, Alexander Graf <agraf@xxxxxxx> wrote: >> On 01/19/2013 05:29 PM, Christoffer Dall wrote: >>> >>> On Sat, Jan 19, 2013 at 11:20 AM, Alexander Graf<agraf@xxxxxxx> wrote: >>>> >>>> On 19.01.2013, at 17:13, Christoffer Dall wrote: >>>> >>>>> On Sat, Jan 19, 2013 at 9:31 AM, Alexander Graf<agraf@xxxxxxx> wrote: >>>>>> >>>>>> Hey Christoffer, >>>>>> >>>>>> I'm trying to use KVM on ARM with your Arndale branch + the icache >>>>>> patch you posted recently. I am running into odd kernel oopses quite often >>>>>> though. Do any of these ring a bell? >>>>>> >>>>> Eh, not really. Arndale is quite stable for me (provided you compile >>>>> everything as ARM and not Thumb-2). Is the crash you're seeing on the >>>>> host or on the guest? >>>> >>>> These are all within the guest. The host has been completely stable for >>>> me so far. >>>> >>>>> Which icache patch do you mean? The dcache in Hyp mode? >>>> >>>> Yeah, the dcache in Hyp mode patch. I recompiled the kernel without and >>>> now I can't get the system up anymore, as it's locked in a room with lots of >>>> security measures I can't circumvent :). Monday I'll know more. >>>> >>>>> This branch should work at least: >>>>> >>>>> https://github.com/virtualopensystems/linux-kvm-arm/commits/lue_arndale-kvm-huge-vmexit >>>> >>>> Ah, I am basing off of >>>> https://github.com/columbia/linux-kvm-arm/tree/lue_arndale-kvm. I can try >>>> the other one again on monday. >>>> >>> I'll make sure my arndale branches are nicely up to date with recent >>> patch series and test things this weekend as well. >> >> >> Thanks a bunch :). >> >> Turns out this was only slightly related to KVM. I was running a guest with >> -m 1000, but the guest dtb said it had 1024 MB. So the OOPSes result from >> the kernel accessing memory from 1000-1024MB. Yay. >> >> > oh, how nice. So, on a somewhat unrelated note, if you want to play > with guests with "a lot" of memory, say around 1500 M, then you > probably want to configure your kernel as a 3/1 memory split using > high memory, as opposed to the 2/2 split which is in at least some of > the recent arm defconfigs. (Otherwise qemu will just crash miserably > on pthread_create). > by the way, I updated all the lue_arndale-kvm* branches to use the latest patch series if you want to play with that (should also be kvmtool compliant on the host side). _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm