RE: [LinuxBIOS] LNXI Merge: lnxi-patch-11/16

[Lu, Yinghai]

Before cache as ram working, I used fixed size to replace unsigned to
make sure romcc can compile my in_coherent.c

Do you have own internal version ROMCC that make use unsigned well....?

YH

-----Original Message-----
From: [e-mail]
[mailto:[e-mail]] On Behalf Of jason schildt
Sent: Friday, September 02, 2005 3:03 PM
To: [e-mail]
Subject: [LinuxBIOS] LNXI Merge: lnxi-patch-11/16

DESCRIPTION:
------------------------------------------------

## lnxi-patch-11 ##
src/northbridge/amd/amdk8/incoherent_ht.c
    Removed K8_HT_FREQ_1G test as we can now compile this with
romcc.
    Bug fix added to ht_optimize_link()
    	To ignore status bits when we read the link freq.	
    Comment added that suggests s/uint8_t/unsigned/g that could save
    code generated by romcc.

DIFFSTAT:
------------------------------------------------

incoherent_ht.c |   22 ++++++++++++----------
  1 files changed, 12 insertions(+), 10 deletions(-)

PATCH:
------------------------------------------------

Index: northbridge/amd/amdk8/incoherent_ht.c
===================================================================
--- northbridge/amd/amdk8/incoherent_ht.c	(revision 1105)
+++ northbridge/amd/amdk8/incoherent_ht.c	(working copy)
@@ -1,15 +1,16 @@
  /*
    This should be done by Eric
    2004.12 yhlu add multi ht chain dynamically support
+
  */
  #include <device/pci_def.h>
  #include <device/pci_ids.h>
  #include <device/hypertransport_def.h>

-#ifndef K8_HT_FREQ_1G_SUPPORT
-        #define K8_HT_FREQ_1G_SUPPORT 0
-#endif
-
+/* We can reduce the size of code generated by romcc by
+ * changing all of the fixed size types that live in registers
+ * into simple unsigned variables. (ie s/uint8_t/unsigned/g)
+ */
  #ifndef K8_SCAN_PCI_BUS
    #define K8_SCAN_PCI_BUS 0
  #endif
@@ -130,21 +131,20 @@
    /* AMD 8131 Errata 48 */
    if (id == (PCI_VENDOR_ID_AMD | (PCI_DEVICE_ID_AMD_8131_PCIX <<
16))) {
    	freq_cap &= ~(1 << HT_FREQ_800Mhz);
-		return freq_cap;
    }

/* AMD 8151 Errata 23 */
    if (id == (PCI_VENDOR_ID_AMD | (PCI_DEVICE_ID_AMD_8151_SYSCTRL
<< 16))) {
    	freq_cap &= ~(1 << HT_FREQ_800Mhz);
-		return freq_cap;
    }
    
    /* AMD K8 Unsupported 1Ghz? */
    if (id == (PCI_VENDOR_ID_AMD | (0x1100 << 16))) {
-	#if K8_HT_FREQ_1G_SUPPORT == 1
-	        if (is_cpu_pre_e0())  // CK804 support 1G?
-	#endif	
-                        freq_cap &= ~(1 << HT_FREQ_1000Mhz);
+	    	/* Supported starting with E0 */
+		device_t dev_2 = PCI_DEV(0,0x18,2);
+              	if(pci_read_config32(dev_2,0x9c) < 0x20f00) {
+			freq_cap &= ~(1 << HT_FREQ_1000Mhz);
+		}
    }

return freq_cap;
@@ -199,8 +199,10 @@

/* See if I am changing the link freqency */
    old_freq = pci_read_config8(dev1, pos1 + LINK_FREQ(offs1));
+	old_freq &= 0x0f;
    needs_reset |= old_freq != freq;
    old_freq = pci_read_config8(dev2, pos2 + LINK_FREQ(offs2));
+	old_freq &= 0x0f;
    needs_reset |= old_freq != freq;

/* Set the Calulcated link frequency */

[LinuxBIOS] LNXI Merge: lnxi-patch-11/16

"Lu, Yinghai"  writes:
> Before cache as ram working, I used fixed size to replace unsigned to
> make sure romcc can compile my in_coherent.c

To the best of my knowledge this is based on a myth of how romcc works.
Explanation already sent.
> Do you have own internal version ROMCC that make use unsigned well....?

No special internal version.  Occasionally minor bug fixes that haven't
propagated up, but I don't think there are any more outstanding romcc
patches right now.
> I remember that if changed that to unsigned, for four ways system it
> will not be compiled by ROMCC.

Without seeing the output I cannot comment on why it failed for you.  I
recall making and testing that change without problems.

It is fascinating that a comment about how to reduce code size has generated
more discussion than most of the actual code changes.

Eric