Driver does both wmb() and writel(). The latter already has a barrier
on some architectures like arm64. This ends up with CPU observing two
barriers back to back before executing the register write.

Drivers should generally assume that the barrier implied by writel() is
sufficient for ordering DMA. Remove the extraneous wmb() before it.

[mkp: Squashed Arnd's and Sinan's patches]
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reported-by: Sinan Kaya <okaya@codeaurora.org>
Acked-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>