The use of stack Variable Length Arrays needs to be avoided, as they
can be a vector for stack exhaustion, which can be both a runtime bug
(kernel Oops) or a security flaw (overwriting memory beyond the
stack). Also, in general, as code evolves it is easy to lose track of
how big a VLA can get. Thus, we can end up having runtime failures
that are hard to debug. As part of the directive[1] to remove all VLAs
from the kernel, and build with -Wvla.

Currently rsi code uses a VLA based on a function argument to
`rsi_sdio_load_data_master_write()`.  The function call chain is

Both these functions

	rsi_sdio_reinit_device()
	rsi_probe()

start the call chain:

	rsi_hal_device_init()
	rsi_load_fw()
	auto_fw_upgrade()
	ping_pong_write()
	rsi_sdio_load_data_master_write()

[Without familiarity with the code] it appears that none of the 4 locks

	mutex
	rx_mutex
	tx_mutex
	tx_bus_mutex

are held when `rsi_sdio_load_data_master_write()` is called.  It is therefore
safe to use kmalloc with GFP_KERNEL.

We can avoid using the VLA by using `kmalloc()` and free'ing the memory on all
exit paths.

Change buffer from 'u8 array' to 'u8 *'.  Call `kmalloc()` to allocate memory for
the buffer.  Using goto statement to call `kfree()` on all return paths.

It can be expected that this patch will result in a small increase in overhead
due to the use of `kmalloc()` however this code is only called on initialization
(and re-initialization) so this overhead should not degrade performance.

[1] https://lkml.org/lkml/2018/3/7/621Signed-off-by: Tobin C. Harding <me@tobin.cc>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>