[PATCH] powerpc/kexec_file: Fix division by zero in extra size estimation

Michael Ellerman mpe at ellerman.id.au
Wed Feb 1 12:55:27 AEDT 2023


Sourabh Jain <sourabhjain at linux.ibm.com> writes:
> On 30/01/23 07:17, Michael Ellerman wrote:
>> In kexec_extra_fdt_size_ppc64() there's logic to estimate how much
>> extra space will be needed in the device tree for some memory related
>> properties.
>>
>> That logic uses the size of RAM divided by drmem_lmb_size() to do the
>> estimation. However drmem_lmb_size() can be zero if the machine has no
>> hotpluggable memory configured, which is the case when booting with qemu
>> and no maxmem=x parameter is passed (the default).
>>
>> The division by zero is reported by UBSAN, and can also lead to an
>> overflow and a warning from kvmalloc, and kdump kernel loading fails:
>>
>>    WARNING: CPU: 0 PID: 133 at mm/util.c:596 kvmalloc_node+0x15c/0x160
>>    Modules linked in:
>>    CPU: 0 PID: 133 Comm: kexec Not tainted 6.2.0-rc5-03455-g07358bd97810 #223
>>    Hardware name: IBM pSeries (emulated by qemu) POWER9 (raw) 0x4e1200 0xf000005 of:SLOF,git-dd0dca pSeries
>>    NIP:  c00000000041ff4c LR: c00000000041fe58 CTR: 0000000000000000
>>    REGS: c0000000096ef750 TRAP: 0700   Not tainted  (6.2.0-rc5-03455-g07358bd97810)
>>    MSR:  800000000282b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE>  CR: 24248242  XER: 2004011e
>>    CFAR: c00000000041fed0 IRQMASK: 0
>>    ...
>>    NIP kvmalloc_node+0x15c/0x160
>>    LR  kvmalloc_node+0x68/0x160
>>    Call Trace:
>>      kvmalloc_node+0x68/0x160 (unreliable)
>>      of_kexec_alloc_and_setup_fdt+0xb8/0x7d0
>>      elf64_load+0x25c/0x4a0
>>      kexec_image_load_default+0x58/0x80
>>      sys_kexec_file_load+0x5c0/0x920
>>      system_call_exception+0x128/0x330
>>      system_call_vectored_common+0x15c/0x2ec
>>
>> To fix it, skip the calculation if drmem_lmb_size() is zero.
>>
>> Fixes: 2377c92e37fe ("powerpc/kexec_file: fix FDT size estimation for kdump kernel")
>> Cc: stable at vger.kernel.org # v5.12+
>> Signed-off-by: Michael Ellerman <mpe at ellerman.id.au>
>> ---
>>   arch/powerpc/kexec/file_load_64.c | 10 ++++++----
>>   1 file changed, 6 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/powerpc/kexec/file_load_64.c b/arch/powerpc/kexec/file_load_64.c
>> index af8854f9eae3..3caee570e79b 100644
>> --- a/arch/powerpc/kexec/file_load_64.c
>> +++ b/arch/powerpc/kexec/file_load_64.c
>> @@ -989,10 +989,12 @@ unsigned int kexec_extra_fdt_size_ppc64(struct kimage *image)
>>   	 * linux,drconf-usable-memory properties. Get an approximate on the
>>   	 * number of usable memory entries and use for FDT size estimation.
>>   	 */
>> -	usm_entries = ((memblock_end_of_DRAM() / drmem_lmb_size()) +
>> -		       (2 * (resource_size(&crashk_res) / drmem_lmb_size())));
>> -
>> -	extra_size = (unsigned int)(usm_entries * sizeof(u64));
>> +	if (drmem_lmb_size()) {
>> +		usm_entries = ((memblock_end_of_DRAM() / drmem_lmb_size()) +
>> +			       (2 * (resource_size(&crashk_res) / drmem_lmb_size())));
>> +		extra_size = (unsigned int)(usm_entries * sizeof(u64));
>> +	} else
>> +		extra_size = 0;
>>   
>>   	/*
>>   	 * Get the number of CPU nodes in the current DT. This allows to
>
> I failed to replicate this issue.
>
> Qemu command used:
> $ qemu-system-ppc64 -enable-kvm -smp 4,cores=2 -drive 
> file=my-image.qcow2 -nographic -m 2G
>
>
> lsmem (inside guest):
> RANGE                                 SIZE  STATE REMOVABLE BLOCK
> 0x0000000000000000-0x000000007fffffff   2G online       yes 0-127
>
> Memory block size:        16M
> Total online memory:       2G
> Total offline memory:      0B
>
> Not sure what I am missing, but changes looks good to me.

Hmm, interesting.

Do you have /proc/device-tree/ibm,dynamic-reconfiguration-memory in the VM?

What version of qemu are you using? I think I tested with 6.2 and 7.1.

cheers


More information about the Linuxppc-dev mailing list