hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4V388
CVE: NA

--------------------------------

KFENCE is designed to be enabled in production kernels, but it can
be also useful in some debug situations. For machines with limited
memory and CPU resources, KASAN is really hard to run. Fortunately,
KFENCE can be a suitable candidate. For KFENCE running on a single
machine, the possibility of discovering existed bugs will increase
as the increasing of KFENCE objects, but this will cost more memory.
In order to balance the possibility of discovering existed bugs and
memory cost, KFENCE objects need to be adjusted according to memory
resources for a compiled kernel Image. Add a module parameter to
adjust KFENCE objects will make kfence to use in different machines
with the same kernel Image.

In short, the following reasons motivate us to add this parameter.
1) In some debug situations, this will make kfence flexible.
2) For some production machines with different memory and CPU size,
this will reduce the kernel-Image-version burden.

The main change is just using kfence_num_objects to replace config
CONFIG_KFENCE_NUM_OBJECTS for dynamic configuration convenient. To
make compatible, kfence_metadata and alloc_covered are alloced by
memblock_alloc. Considering "cat /sys/kernel/debug/kfence/objects"
will read kfence_metadata, the initialization of this fs should
check whether kfence_metadata is successfully allocated.

Unfortunately, dynamic allocation require the KFENCE pool size to
be a configurable variable, which lead to additional instructions
(eg, load) added to the fast path of the memory allocation. As a
result, the performance will degrade. To avoid bad performance in
production machine, an ugly macro is used to isolate the changes.
Signed-off-by: NPeng Liu <liupeng256@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: NCheng Jian <cj.chengjian@huawei.com>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>