Re: [PATCH] KVM: selftests: Speed up set_memory_region_test

[Paolo Bonzini]

On 23/06/21 11:03, Zenghui Yu wrote:
> On 2021/4/26 21:01, Vitaly Kuznetsov wrote:
> > After commit 4fc096a99e01 ("KVM: Raise the maximum number of user 
> > memslots")
> > set_memory_region_test may take too long, reports are that the default
> > timeout value we have (120s) may not be enough even on a physical host.
> >
> > Speed things up a bit by throwing away vm_userspace_mem_region_add() 
> > usage
> > from test_add_max_memory_regions(), we don't really need to do the 
> > majority
> > of the stuff it does for the sake of this test.
> >
> > On my AMD EPYC 7401P, # time ./set_memory_region_test
> > pre-patch:
> >  Testing KVM_RUN with zero added memory regions
> >  Allowed number of memory slots: 32764
> >  Adding slots 0..32763, each memory region with 2048K size
> >  Testing MOVE of in-use region, 10 loops
> >  Testing DELETE of in-use region, 10 loops
> >
> >  real    0m44.917s
> >  user    0m7.416s
> >  sys    0m34.601s
> >
> > post-patch:
> >  Testing KVM_RUN with zero added memory regions
> >  Allowed number of memory slots: 32764
> >  Adding slots 0..32763, each memory region with 2048K size
> >  Testing MOVE of in-use region, 10 loops
> >  Testing DELETE of in-use region, 10 loops
> >
> >  real    0m20.714s
> >  user    0m0.109s
> >  sys    0m18.359s
> >
> > Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
> > Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
>
> I've seen the failure on my arm64 server, # ./set_memory_region_test
>
> Allowed number of memory slots: 32767
> Adding slots 0..32766, each memory region with 2048K size
> ==== Test Assertion Failure ====
>    set_memory_region_test.c:391: ret == 0
>    pid=42696 tid=42696 errno=22 - Invalid argument
>       1    0x00000000004015a7: test_add_max_memory_regions at 
> set_memory_region_test.c:389
>       2     (inlined by) main at set_memory_region_test.c:426
>       3    0x0000ffffb7c63bdf: ?? ??:0
>       4    0x00000000004016db: _start at :?
>    KVM_SET_USER_MEMORY_REGION IOCTL failed,
>    rc: -1 errno: 22 slot: 2624
>
> > +    mem = mmap(NULL, MEM_REGION_SIZE * max_mem_slots + alignment,
>
> The problem is that max_mem_slots is declared as uint32_t, the result
> of (MEM_REGION_SIZE * max_mem_slots) is unexpectedly truncated to be
> 0xffe00000.
>
> > +           PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
> > +    TEST_ASSERT(mem != MAP_FAILED, "Failed to mmap() host");
> > +    mem_aligned = (void *)(((size_t) mem + alignment - 1) & 
> > ~(alignment - 1));
> > +
> >      for (slot = 0; slot < max_mem_slots; slot++) {
> > -        vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
> > -                        guest_addr, slot, mem_reg_npages,
> > -                        0);
> > -        guest_addr += MEM_REGION_SIZE;
> > +        ret = test_memory_region_add(vm, mem_aligned +
> > +                         ((uint64_t)slot * MEM_REGION_SIZE),
>
> These unmapped VAs got caught by access_ok() checker in
> __kvm_set_memory_region() as they happen to go beyond the task's
> address space on arm64. Casting max_mem_slots to size_t in both
> mmap() and munmap() fixes the issue for me.

Can you provide a patch for both?

Paolo