[openfirmware] r1394 - cpu/x86/pc/olpc/via

[svn at openfirmware.info]

Author: wmb
Date: 2009-10-06 05:26:47 +0200 (Tue, 06 Oct 2009)
New Revision: 1394

Modified:
   cpu/x86/pc/olpc/via/romreset.bth
   cpu/x86/pc/olpc/via/startcominit.fth
Log:
OLPC trac 9451 - eliminated the startup delay before accessing CMOS RAM
in the wakeup-from-S3 case, thus speeding-up resume.


Modified: cpu/x86/pc/olpc/via/romreset.bth
===================================================================
--- cpu/x86/pc/olpc/via/romreset.bth	2009-10-06 03:18:14 UTC (rev 1393)
+++ cpu/x86/pc/olpc/via/romreset.bth	2009-10-06 03:26:47 UTC (rev 1394)
@@ -93,10 +93,6 @@
    2 92 port-wb  \ Enable A20
    \ External A20GATE is disabled automatically because the internal KBC is on by default (cfg reg 8851 bit 0)
 
-   acpi-io-base 1 + 8888 config-ww   \ Set ACPI base address
-\   88 8881 config-wb   \ Enable ACPI regs, 32-bit PM timer, guard RTC against power glitches
-   8c 8881 config-wb   \ Enable ACPI regs, 32-bit PM timer, guard RTC against power glitches
-
    fload ${BP}/cpu/x86/pc/olpc/via/startcacheasram.fth
 
 \ cache_as_ram_auto.c: amd64_main() 
@@ -122,10 +118,14 @@
 
    fload ${BP}/cpu/x86/pc/olpc/via/starthostctl.fth
 
+   acpi-io-base 1 + 8888 config-ww   \ Set ACPI base address
+
 [ifdef] debug-startup
    cominit #) call
 [then]
 
+   8c 8881 config-wb   \ Enable ACPI regs, 32-bit PM timer, guard RTC against power glitches
+
    char + report
    long-offsets on
 
@@ -138,8 +138,7 @@
          char X report
          begin  hlt  again
       then
-      carret report   linefeed report
-      char F report
+      carret report   linefeed report   char F report
       fload ${BP}/cpu/x86/pc/olpc/via/demodram.fth
    then
 
@@ -182,7 +181,7 @@
 
 0 [if]
 long-offsets on
-acpi-io-base 48 + port-rl  h# 1000.0000 # ax and  0<>  if  \ Memory ID1 bit
+acpi-io-base 48 + port-rl  h# 1000.0000 # ax and  0<>  if  \ Memory ID1 bit - set for 32bit memory width
 
    mmxcr
    39 3c4 port-wb  ax ax xor  3c5 port-rb  ax mmxdot  

Modified: cpu/x86/pc/olpc/via/startcominit.fth
===================================================================
--- cpu/x86/pc/olpc/via/startcominit.fth	2009-10-06 03:18:14 UTC (rev 1393)
+++ cpu/x86/pc/olpc/via/startcominit.fth	2009-10-06 03:26:47 UTC (rev 1394)
@@ -29,9 +29,7 @@
    \ First we check for a cached board ID in CMOS RAM, to avoid the
    \ possibly time-consuming operation of asking the EC.
 
-   \ To read the high half of CMOS RAM we must enable it
    d# 17 0 devfunc
-   4e 18 18 mreg  \ Enable ports 74/75 for CMOS RAM access  - 10 res be like Phx
    end-table
 
    \ Configure the I/O decoding to enable access to the EC
@@ -46,24 +44,27 @@
    64 0f 07 mreg  \ PCS0 size is 8 bytes - to include 68 and 6c
    66 01 01 mreg  \ PCS0 Enable
    67 10 10 mreg  \ PCS0 to LPC Bus
+   \ When the RTC is guarded against power glitches, there is a 50 mS delay before
+   \ reads return the correct data upon wakeup from S3.  On power-up, the delay is
+   \ present even with the RTC unguarded.  Unguarding it here speeds up S3 startup.
+   81 ff 88 mreg  \ Enable ACPI regs, 32-bit PM timer, disable RTC power glitch guard
+   4e 08 08 mreg  \ Enable ports 74/75 for high-bank CMOS RAM access
    end-table
 
-   d# 17 0 devfunc
+   d# 17 7 devfunc
    70 fb 82 mreg  \ CPU to PCI flow control - CPU to PCI posted write, Enable Delay Transaction
    end-table
 
-h# 83 # al mov  al h# 74 # out  h# 75 # al in    \ check byte - should be ~board-id
-ax bx mov
+   acpi-io-base 4 + port-rw           \ Get APCI Status register
+   d# 10 # ax shr  7 # ax and  1 # ax cmp  <>  if  \ Type 1 is wakeup from S3
+      \ The following applies only to power-up; no delay is needed for wakeup from S3
+      \ This delay is empirically necessary before reading CMOS - minimum is 36000 - about 50 ms
+      \ Before the delay has elapsed, the CMOS RAM returns 0 instead of the stored value.
+      d# 40000 wait-us
+   then
 
-   \ This delay is empirically necessary before reading CMOS - minimum is 36000 - about 50 ms
-   \ Before the delay has elapsed, the CMOS RAM returns 0 instead of the stored value.
-   d# 40000 wait-us
-
-h# 88 # al mov  al h# 74 # out  bx ax mov  al h# 75 # out    \ Test for CMOS read timing
-
    \ As an optimization to avoid long waits for the EC to respond, read the board ID
-   \ that is cached in CMOS RAM.  This might not in fact be an optimization in light
-   \ of the above delay ...
+   \ that is cached in CMOS RAM.
    h# 83 # al mov  al h# 74 # out  h# 75 # al in    \ check byte - should be ~board-id
    al ah mov   ah not                               \ ~check byte in AH
    h# 82 # al mov  al h# 74 # out  h# 75 # al in    \ board-id in AL