Hello,
On Thu, Nov 15, 2018 at 11:02 AM Andrew Jones drjones@redhat.com wrote:
I think you should be able to pre-compile the selftests and save them, and then bisect the kernel further back than their introduction. Actually you may want to simplify the test case to just a mmap/madvise pair and see if that reproduces, since kvm selftests doesn't do much more than that. (Other than all the /dev/kvm ioctls - you have the right perms on /dev/kvm, or are running as root, right?)
yes I have /dev/kvm permissions, but it shouldn't matter at the moment.
well I stripped the kvm self tests from most stuff and ended up with simpler code (sorry for spaces, I am using web interface to send this email)
#include <stdio.h> #include <sys/mman.h> #include <string.h> #include <errno.h> static void *align(void *x, size_t size) { size_t mask = size - 1; return (void *) (((size_t) x + mask) & ~mask); }
int main(){
int ret; void *mmap_start; void *host_mem; size_t length = 6291456;
mmap_start = mmap(NULL, length, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | 0, -1, 0); host_mem = align(mmap_start, 1); ret = madvise(host_mem, length, MADV_NOHUGEPAGE); printf("madvise returned: %d\nerrno: %d %s\n", ret, errno, strerror(errno)); }
Now I couldn't find a single commit to bisect from, I checked 4.20, 4.19, 4.16, 4.10, 4.4 back to v3.18 tag and None of the builds I attempted did work. All of the kernels returned:
madvise returned: -1 errno: 22 Invalid argument
and the strace logs looks like this:
mmap(NULL, 6291456, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f46ea2a1000 madvise(0x7f46ea2a1000, 6291456, MADV_NOHUGEPAGE) = -1 EINVAL (Invalid argument)
For comprehension, this is done on intel core i7-4500U CPU @ 1.80GHz
As for now I will comment the madvise line and the assert when writing my own kvm self test. I think it wouldn't cause any trouble?, If it is not the case, please let me know.
Thanks,