[Why]
A hang was observed where a read-modify-write access occurred due to the
register for idle state being shared between DMCUB and driver.
dmcub read - idle allow / no commit
driver read - idle allow / no commit
driver write - idle disallow / no commit
dmcub write - idle allow / commit
Resulting in DMCUB re-entering IPS after a disable and keeping the allow
high.
[How]
Long term we need to split commit/allow into two registers or use shared
DRAM state, but short term we can reduce the repro rate by ensuring that
the disallow went through by bounding the expected worst case scenario.
Tested-by: Daniel Wheeler <[email protected]>
Reviewed-by: Hansen Dsouza <[email protected]>
Reviewed-by: Ovidiu Bunea <[email protected]>
Acked-by: Rodrigo Siqueira <[email protected]>
Signed-off-by: Nicholas Kazlauskas <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
if (!(allow_state & DMUB_IPS2_ALLOW_MASK)) {
// Wait for evaluation time
- udelay(dc->debug.ips2_eval_delay_us);
- commit_state = dc->hwss.get_idle_state(dc);
+ for (;;) {
+ udelay(dc->debug.ips2_eval_delay_us);
+ commit_state = dc->hwss.get_idle_state(dc);
+ if (commit_state & DMUB_IPS2_ALLOW_MASK)
+ break;
+
+ /* allow was still set, retry eval delay */
+ dc->hwss.set_idle_state(dc, false);
+ }
+
if (!(commit_state & DMUB_IPS2_COMMIT_MASK)) {
// Tell PMFW to exit low power state
dc->clk_mgr->funcs->exit_low_power_state(dc->clk_mgr);