2010 macbook pro kernel panic "supposed to be free page 0xffffffff83b778a0" has cache

[sunmacrosystems]

[here] (​https://0x0.st/XY4C.tar) are some pictures of the error message i got, and [here] (​https://0x0.st/XYJr.txt) is the output of lscpu on linux, ignore 59.jpg and 55.jpg in the first tarball cause they're bad quality. i tried the nightly and got the same kernel panic, unfortunately the keyboard doesn't work so i couldn't run bt, ints and co. the boot splash initially worked and highlighted all of the icons until the kernel panic message happened

[sunmacrosystems]

edit: this was from booting the installation media btw

[waddlesplash]

Please attach one good quality image to this ticket rather than linking a tarball.

[sunmacrosystems]

sorry about that

[waddlesplash]

That image is too small to be readable...

[sunmacrosystems]

​https://0x0.st/XYJc.jpg sorry for wasting your time, had trouble sending it from my phone to my computer

[waddlesplash]

Looks similar to #13189, but if this is reproducible on every boot, that's much more interesting.

I will see about adding some more debug assertions to this panic to see if we can get more information.

[sunmacrosystems]

i can't enter the bootloader menu by holding space, i tried chainloading the uefi partition of the usb from grub> but that didn't work either

[waddlesplash]

You may need to spam the spacebar, not just hold it.

[waddlesplash]

I've added a bit more debugging information with this KDL in hrev58176, please retest and get a new screenshot with the added "page" information.

[jckarter]

some anecdata: I was seeing a similar crash running beta 5 on a MacBookAir4,1. However, I installed the nightly hrev58293 in order to see if I could get more debug info, but the crashes seem to have gone away, at least in the situations where I had been able to reliably reproduce them under beta 5.

[waddlesplash]

Interesting. There were a number of VM changes on the nightlies since beta5 which could have affected this, or perhaps the SCSI DMA changes are somehow relevant. The page structures themselves used to be allocated at the beginning of the free memory, but that was recently changed so that they're usually allocated closer to the end, or at least above the 4GB boundary (if the machine has over 4GB of RAM), whichever is lower.

If something's corrupting memory below 4GB, then the recent changes may just hide this since it's no longer the page structures getting corrupted but something else less noticed. The change in question was hrev58243, I guess you can try a nightly from just before that and see if it's any different.

[jckarter]

So far under hrev58293, I've been able to run SoftwareUpdater, use HaikuDepot to install Falkon and a few other programs, and do a day or so's worth of web browsing under both WebPositive and Falkon without any apparent corruption or instability. That of course doesn't rule out the possibility of lurking corruption; I could try compiling something or some other more CPU- and memory-intensive tasks next. The MacBook Air I'm using only has 2GB of RAM so all RAM should be under the 4GB boundary.