<!DOCTYPE html><html><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body>
    <pre>
</pre>
    <div class="moz-cite-prefix">On 29-07-2025 13:47, Jani Nikula wrote:<br>
    </div>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">On Mon, 28 Jul 2025, Dibin Moolakadan Subrahmanian <a class="moz-txt-link-rfc2396E" href="mailto:dibin.moolakadan.subrahmanian@intel.com"><dibin.moolakadan.subrahmanian@intel.com></a> wrote:
</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre"> The current wait_panel_status() uses intel_de_wait() with a long timeout
 (5000ms), which is suboptimal on Xe platforms where the underlying
 xe_mmio_wait32() employs an exponential backoff strategy. This leads
 to unnecessary delays during resume or power-on  when the panel becomes
 ready earlier than the full timeout.
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
It's not about the timeout, it's about the exponentially increasing poll
delay.

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre"> This patch replaces intel_de_wait() with read_poll_timeout() +
 intel_de_read() to actively poll the register at given interval and exit
 early when panel is ready, improving resume latency
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
Please do not say "this patch" in commit messages. Just use the
imperative "Replace ...".

The commit messages is unnecessarily indented with a space.</pre>
    </blockquote>
    <pre wrap="" class="moz-quote-pre">will correct this.</pre>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre">Changes in v2:
 Replaced  two-phase intel_de_wait() with  read_poll_timeout()
 + intel_de_read()

Changes in v3:
 - Add poll_interval_ms argument  'wait_panel_status' function.
 - Modify 'wait_panel_status' callers with proper poll interval

Changes in v4:
 - Change 'wait_panel_off' poll interval to 10ms

Signed-off-by: Dibin Moolakadan Subrahmanian <a class="moz-txt-link-rfc2396E" href="mailto:dibin.moolakadan.subrahmanian@intel.com"><dibin.moolakadan.subrahmanian@intel.com></a>
---
 drivers/gpu/drm/i915/display/intel_pps.c | 41 +++++++++++++++++-------
 1 file changed, 30 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_pps.c b/drivers/gpu/drm/i915/display/intel_pps.c
index b64d0b30f5b1..56ef835fc2eb 100644
--- a/drivers/gpu/drm/i915/display/intel_pps.c
+++ b/drivers/gpu/drm/i915/display/intel_pps.c
@@ -22,6 +22,7 @@
 #include "intel_pps.h"
 #include "intel_pps_regs.h"
 #include "intel_quirks.h"
+#include <linux/iopoll.h>
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
Please look at how includes are ordered in every single file in i915.</pre>
    </blockquote>
    <pre wrap="" class="moz-quote-pre">will correct this.</pre>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre"> static void vlv_steal_power_sequencer(struct intel_display *display,
                                      enum pipe pipe);
@@ -600,14 +601,18 @@ void intel_pps_check_power_unlocked(struct intel_dp *intel_dp)
 #define IDLE_CYCLE_MASK                (PP_ON | PP_SEQUENCE_MASK | PP_CYCLE_DELAY_ACTIVE | PP_SEQUENCE_STATE_MASK)
 #define IDLE_CYCLE_VALUE       (0     | PP_SEQUENCE_NONE | 0                     | PP_SEQUENCE_STATE_OFF_IDLE)
 
+#define PANEL_MAXIMUM_ON_TIME_MS               (5000)
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
The name of the macro is misleading. For single-use things, maybe better
to just keep the value inline as it were.

Side note, the parenthesis are superfluous here.
</pre>
    </blockquote>
    <pre wrap="" class="moz-quote-pre">will correct this.</pre>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">
</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre">+
 static void intel_pps_verify_state(struct intel_dp *intel_dp);
 
 static void wait_panel_status(struct intel_dp *intel_dp,
-                             u32 mask, u32 value)
+                             u32 mask, u32 value, int poll_interval_ms)
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
Can we not add the extra parameter please? Can we have a meaningful
default instead? 10 ms? Is the 1 ms poll interval really required?</pre>
    </blockquote>
    <pre>Motive behind adding new parameter is to adjust the poll time based on case.
Currently each call is taking different time to complete as below 
for panel power off time   - 82 ms
for panel power cycle      - 0.074 ms
for panel power on         -  327 ms

Making default poll interval 10ms will  increase panel power cycle time  to 10ms


</pre>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre"> {
        struct intel_display *display = to_intel_display(intel_dp);
        struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
        i915_reg_t pp_stat_reg, pp_ctrl_reg;
+       int ret;
+       u32 reg_val;
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
Nitpick, usually just "val".</pre>
    </blockquote>
    will correct this.
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre">   lockdep_assert_held(&display->pps.mutex);
 
@@ -624,14 +629,27 @@ static void wait_panel_status(struct intel_dp *intel_dp,
                    intel_de_read(display, pp_stat_reg),
                    intel_de_read(display, pp_ctrl_reg));
 
-       if (intel_de_wait(display, pp_stat_reg, mask, value, 5000))
-               drm_err(display->drm,
-                       "[ENCODER:%d:%s] %s panel status timeout: PP_STATUS: 0x%08x PP_CONTROL: 0x%08x\n",
-                       dig_port->base.base.base.id, dig_port->base.base.name,
-                       pps_name(intel_dp),
-                       intel_de_read(display, pp_stat_reg),
-                       intel_de_read(display, pp_ctrl_reg));
+       if (poll_interval_ms <= 0)
+               poll_interval_ms = 1; //if <0 is passed go with 1ms
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
Without the parameter, we could get rid of checks like this.

The comment just duplicates what the code already says.

Also, we don't use // comments.</pre>
    </blockquote>
    will correct this.
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre">+
+       ret = read_poll_timeout(intel_de_read, reg_val,
+                               ((reg_val & mask) == value),
+                               (poll_interval_ms * 1000),  // poll intervell
+                               (PANEL_MAXIMUM_ON_TIME_MS * 1000),  // total timeout (us)
+                               true,
+                               display, pp_stat_reg);
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
The outer parenthesis in the parameters are superfluous.

The comments are useless (and have a typo too).

</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre">+
+       if (ret == 0)
+               goto panel_wait_complete;
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
We do use goto in kernel, but primarily for error handling. Please use

        if (ret)

here, and the whole drm_err() thing remains unchanged, and doesn't
become part of the patch...

</pre>
    </blockquote>
    <pre wrap="" class="moz-quote-pre">will correct this.</pre>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre"> 
+       drm_err(display->drm,
+               "dibin [ENCODER:%d:%s] %s panel status timeout: PP_STATUS: 0x%08x PP_CONTROL: 0x%08x\n",
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
...and it'll be easier to notice you've left your name in the debug
logs. Oops.
</pre>
    </blockquote>
    <blockquote type="cite" cite="mid:018afd66168bd9a55abe4e14c24e9b11a89ab15c@intel.com">
      <pre wrap="" class="moz-quote-pre">
</pre>
      <blockquote type="cite">
        <pre wrap="" class="moz-quote-pre">+          dig_port->base.base.base.id, dig_port->base.base.name,
+               pps_name(intel_dp),
+               intel_de_read(display, pp_stat_reg),
+               intel_de_read(display, pp_ctrl_reg));
+
+panel_wait_complete:
        drm_dbg_kms(display->drm, "Wait complete\n");
 }
 
@@ -644,7 +662,8 @@ static void wait_panel_on(struct intel_dp *intel_dp)
                    "[ENCODER:%d:%s] %s wait for panel power on\n",
                    dig_port->base.base.base.id, dig_port->base.base.name,
                    pps_name(intel_dp));
-       wait_panel_status(intel_dp, IDLE_ON_MASK, IDLE_ON_VALUE);
+
+       wait_panel_status(intel_dp, IDLE_ON_MASK, IDLE_ON_VALUE, 20);
 }
 
 static void wait_panel_off(struct intel_dp *intel_dp)
@@ -656,7 +675,7 @@ static void wait_panel_off(struct intel_dp *intel_dp)
                    "[ENCODER:%d:%s] %s wait for panel power off time\n",
                    dig_port->base.base.base.id, dig_port->base.base.name,
                    pps_name(intel_dp));
-       wait_panel_status(intel_dp, IDLE_OFF_MASK, IDLE_OFF_VALUE);
+       wait_panel_status(intel_dp, IDLE_OFF_MASK, IDLE_OFF_VALUE, 10);
 }
 
 static void wait_panel_power_cycle(struct intel_dp *intel_dp)
@@ -683,7 +702,7 @@ static void wait_panel_power_cycle(struct intel_dp *intel_dp)
        if (remaining)
                wait_remaining_ms_from_jiffies(jiffies, remaining);
 
-       wait_panel_status(intel_dp, IDLE_CYCLE_MASK, IDLE_CYCLE_VALUE);
+       wait_panel_status(intel_dp, IDLE_CYCLE_MASK, IDLE_CYCLE_VALUE, 1);
 }
 
 void intel_pps_wait_power_cycle(struct intel_dp *intel_dp)
</pre>
      </blockquote>
      <pre wrap="" class="moz-quote-pre">
</pre>
    </blockquote>
  </body>
</html>