[PATCH kernel] powerpc/xive: Ignore kmemleak false positives
Michael Ellerman
mpe at ellerman.id.au
Fri Jun 12 22:43:22 AEST 2020
Alexey Kardashevskiy <aik at ozlabs.ru> writes:
> xive_native_provision_pages() allocates memory and passes the pointer to
> OPAL so kmemleak cannot find the pointer usage in the kernel memory and
> produces a false positive report (below) (even if the kernel did scan
> OPAL memory, it is unable to deal with __pa() addresses anyway).
>
> This silences the warning.
>
> unreferenced object 0xc000200350c40000 (size 65536):
> comm "qemu-system-ppc", pid 2725, jiffies 4294946414 (age 70776.530s)
> hex dump (first 32 bytes):
> 02 00 00 00 50 00 00 00 00 00 00 00 00 00 00 00 ....P...........
> 01 00 08 07 00 00 00 00 00 00 00 00 00 00 00 00 ................
> backtrace:
> [<0000000081ff046c>] xive_native_alloc_vp_block+0x120/0x250
> [<00000000d555d524>] kvmppc_xive_compute_vp_id+0x248/0x350 [kvm]
> [<00000000d69b9c9f>] kvmppc_xive_connect_vcpu+0xc0/0x520 [kvm]
> [<000000006acbc81c>] kvm_arch_vcpu_ioctl+0x308/0x580 [kvm]
> [<0000000089c69580>] kvm_vcpu_ioctl+0x19c/0xae0 [kvm]
> [<00000000902ae91e>] ksys_ioctl+0x184/0x1b0
> [<00000000f3e68bd7>] sys_ioctl+0x48/0xb0
> [<0000000001b2c127>] system_call_exception+0x124/0x1f0
> [<00000000d2b2ee40>] system_call_common+0xe8/0x214
>
> Signed-off-by: Alexey Kardashevskiy <aik at ozlabs.ru>
> ---
>
> Does kmemleak actually check the OPAL memory?
No it shouldn't.
The memory used by OPAL should all be reserved in the device tree. That
means we never give it to any of the Linux memory allocators, and
therefore kmemleak will never see an allocation from those areas and add
that area to its list of areas to scan.
At least that's my understanding of how kmemleak works.
> Because if it did, we would still have a warning as kmemleak does not
> trace __pa() addresses anyway.
Right.
I think this patch is an OK solution.
It's kind of odd that we donate pages and don't keep track of them. But
they're used by xive until it's reset, and we don't do that until we
kexec, at which point we don't need to know about them anyway.
cheers
More information about the Linuxppc-dev
mailing list