suspend/resume improved?

[Nate Lawson]

Ulrich Spoerlein wrote:
> On Fri, 10.12.2004 at 10:43:26 -0800, Nate Lawson wrote:
> 
>>>Line 444 is the culprit:
>>>   ACPI_FLUSH_CPU_CACHE ();
>>>
>>>   Status = AcpiHwRegisterWrite (ACPI_MTX_DO_NOT_LOCK, 
>>>   ACPI_REGISTER_PM1A_CONTROL, PM1AControl);
>>>   if (ACPI_FAILURE (Status))
>>>   {
>>>       return_ACPI_STATUS (Status);
>>>   }
>>>
>>>Putting the loop before AcpiHwRegisterWrite will enter infinite loop,
>>>putting it after it -> Reset
>>
>>Interesting.  If you add a "DELAY(10000);" before the register write, 
>>does this help?  A more likely issue is that we need to write pm1a/b at 
>>the same time.  Let me think about this and get you more info later.
> 
> 
> It "looks" like the delay does nothing, however I noticed that all three
> LEDs (num/scroll/caps lock) were lighting up for 1-2s.
> 
> I then swapped pm1a and pm1b, but nothing changed.

That's not quite what I meant.  The acpi-ca code splits the write of 
SLP_TYP and SLP_EN into two separate writes.  I suspect some BIOSes 
don't like this.  Try the attached patch that combines them.

-- 
Nate
-------------- next part --------------
Index: sys/contrib/dev/acpica/hwsleep.c
===================================================================
RCS file: /home/ncvs/src/sys/contrib/dev/acpica/hwsleep.c,v
retrieving revision 1.18
diff -u -r1.18 hwsleep.c
--- sys/contrib/dev/acpica/hwsleep.c	1 Dec 2004 23:40:48 -0000	1.18
+++ sys/contrib/dev/acpica/hwsleep.c	11 Dec 2004 02:23:18 -0000
@@ -420,6 +420,7 @@
 
     /* Write #1: fill in SLP_TYP data */
 
+#if 0
     Status = AcpiHwRegisterWrite (ACPI_MTX_DO_NOT_LOCK, ACPI_REGISTER_PM1A_CONTROL, PM1AControl);
     if (ACPI_FAILURE (Status))
     {
@@ -431,6 +432,7 @@
     {
         return_ACPI_STATUS (Status);
     }
+#endif
 
     /* Insert SLP_ENABLE bit */