linux-phytec-ti

ARM: OMAP2+: Extend DRA7 IPU1 MMU pdata quirks to DSP MDMA MMUs

The C66-based DSPs on DRA7xx SoCs do not support a Powerdown-RET
mode, and only supports a Powerdown-Grid OFF mode which requires
a boot from reset. The HW_AUTO setting and a target power domain
state of OFF implies that the DSPs are powered off as soon as
they are idled by executing an IDLE instruction. The DSPs lose
their context as a result and will be unable to resume operations
from any wakeup event.

The DSP power domains therefore need to be restricted to ON state
for the duration a DSP processor is actively running. This is
similar to the restriction required for DRA7 IPU1 processor (albeit
because of a different reason). The IPU1 behavior is handled in
commit 4e6bf3ca947b ("ARM: OMAP2+: Use separate IOMMU pdata to fix
DRA7 IPU1 boot") which adds a .set_pwrdm_constraint ops to the OMAP
IOMMU platform data to restrict the IPU1 power domain to ON state
during the active period of the IPU1 remote processor.

Extend the IPU1 iommu pdata quirks to the DRA7 MDMA MMUs as well
to restrict the DSP power domains to ON state. The MDMA MMU module
configuration will be the first and last steps in the boot and
shutdown sequences of the DSP processors. The existing IPU1 IOMMU
pdata variable has also been renamed appropriately to reflect the
common usage between the IPU1 and the DSPs.

NOTE:
1. The functional behavior is inconsistent between different DSPs on
   DRA74x, DRA72x and DRA71x SoCs and silicon revisions. DSP1 on
   DRA7 EVM rev.H (DRA752 ES2.0), AM57xx GP EVM (DRA752 ES2.0) and
   AM572x IDK (DRA752 ES2.0) boards is entering idle mode without any
   fix and are getting powered down and losing context, but none of
   the other boards (DRA72 EVM, DRA71 EVM, AM571x IDK) exhibit this
   behavior. The idling behavior is unchanged even after this patch
   is applied. DSP2 is not idled on any board. These behaviors are
   different from those observed on 4.14 kernel.
2. This patch is also needed to preserve the DSP contexts when proper
   clock-gating is achieved during inactive periods. DSP power domains
   on these platforms should not be hitting OFF at the moment (even
   with firmware images executing an IDLE instruction) because of the
   issue described in errata i879 ("DSP MStandby requires CD_EMU in
   SW_WKUP") affecting these SoCs, but the behavior is different between
   different DSPs and SoCs as explained in #1. The i879 errata fix in
   the following patch achieves the DSP clock-gating with HW_AUTO mode,
   and would result in a power domain sleep transition to OFF mode
   without any software context save mechanism for all the DSPs.

Signed-off-by: Suman Anna <s-anna@ti.com>