Jérôme had the idea checking if MSR bit is active before using
msr_read/msr_write in generic_x86.cpp

if (get_current_cpuid(&cpuInfo, 1) != B_OK
|| (cpuInfo.eax_1.features & IA32_FEATURE_MSR) == 0)
   return;

#define IA32_FEATURE_MSR (1UL << 5) was added to
haiku/trunk/headers/private/kernel/arch/x86/arch_cpu.h

First i separated completely, intel from
others, i did the following:
Index: generic_x86.cpp
===================================================================
--- generic_x86.cpp	(revision 19851)
+++ generic_x86.cpp	(working copy)
@@ -181,7 +181,7 @@
 
 module_info *modules[] = {
 	(module_info *)&gIntelModule,
-	(module_info *)&gAMDModule,
-	(module_info *)&gVIAModule,
+//	(module_info *)&gAMDModule,
+//	(module_info *)&gVIAModule,
 	NULL
 };


Index: Jamfile
===================================================================
--- Jamfile	(revision 19851)
+++ Jamfile	(working copy)
@@ -4,7 +4,7 @@
 
 KernelAddon generic_x86 :
 	generic_x86.cpp
-	amd.cpp
+#	amd.cpp
 	intel.cpp
-	via.cpp
+#	via.cpp
 	;

RESULT:
the same stacktrace as previous.

After some testings i find that function
generic_mtrr_compute_physical_mask() in intel.cpp is 
executing successfully.
So, next i try to be sure that generic_dump_mtrrs
is executing successfully in intel.cpp:
[...]
-	generic_dump_mtrrs(generic_count_mtrrs());
+	generic_dump_mtrrs(0);
[...]
If i understood correctly we want
generic_count_mtrrs() returns 0 in the case
of a pentium mmx without mttrs.
Surprise!! even with generic_dump_mtrrs(0);
a got the same stacktrace.
So i focused in generic_dump_mtrrs (in generic_x86.cpp)
and did the following: (svn diff generic_x86.cpp > generic_x86.diff)

Index: generic_x86.cpp
===================================================================
--- generic_x86.cpp	(revision 19851)
+++ generic_x86.cpp	(working copy)
@@ -28,6 +28,7 @@
 #define IA32_MTRR_ENABLE_FIXED			(1UL << 10)
 #define IA32_MTRR_VALID_RANGE			(1UL << 11)
 
+#define IA32_FEATURE_MSR (1UL << 5)
 
 struct mtrr_capabilities {
 	mtrr_capabilities(uint64 value) { *(uint64 *)this = value; }
@@ -48,7 +49,8 @@
 {
 	cpuid_info cpuInfo;
 	if (get_current_cpuid(&cpuInfo, 1) != B_OK
-		|| (cpuInfo.eax_1.features & IA32_FEATURE_MTRR) == 0)
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MTRR) == 0
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MSR) == 0)
 		return 0;
 
 	mtrr_capabilities 
capabilities(x86_read_msr(IA32_MSR_MTRR_CAPABILITIES));
@@ -160,6 +162,13 @@
 void
 generic_dump_mtrrs(uint32 count)
 {
+
+	cpuid_info cpuInfo;
+	if (get_current_cpuid(&cpuInfo, 1) != B_OK
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MTRR) == 0
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MSR) == 0)
+		return;
+
 	if (x86_read_msr(IA32_MSR_MTRR_DEFAULT_TYPE) & IA32_MTRR_ENABLE) {
 		TRACE(("MTRR enabled\n"));
 	} else {
@@ -181,7 +190,7 @@
 
 module_info *modules[] = {
 	(module_info *)&gIntelModule,
-	(module_info *)&gAMDModule,
-	(module_info *)&gVIAModule,
+//	(module_info *)&gAMDModule,
+//	(module_info *)&gVIAModule,
 	NULL
 };

RESULT: the module is loaded successfully.
I think indeed generic_count_mtrrs() returns 0
and generic_dump_mtrrs returns immediately after the if
statement. The problem was indeed the execution of
x86_read_msr in generic_dump_mtrrs.Jérôme had the idea checking if MSR bit is active before using
msr_read/msr_write in generic_x86.cpp

if (get_current_cpuid(&cpuInfo, 1) != B_OK
|| (cpuInfo.eax_1.features & IA32_FEATURE_MSR) == 0)
   return;

#define IA32_FEATURE_MSR (1UL << 5) was added to
haiku/trunk/headers/private/kernel/arch/x86/arch_cpu.h

First i separated completely, intel from
others, i did the following:
Index: generic_x86.cpp
===================================================================
--- generic_x86.cpp	(revision 19851)
+++ generic_x86.cpp	(working copy)
@@ -181,7 +181,7 @@
 
 module_info *modules[] = {
 	(module_info *)&gIntelModule,
-	(module_info *)&gAMDModule,
-	(module_info *)&gVIAModule,
+//	(module_info *)&gAMDModule,
+//	(module_info *)&gVIAModule,
 	NULL
 };


Index: Jamfile
===================================================================
--- Jamfile	(revision 19851)
+++ Jamfile	(working copy)
@@ -4,7 +4,7 @@
 
 KernelAddon generic_x86 :
 	generic_x86.cpp
-	amd.cpp
+#	amd.cpp
 	intel.cpp
-	via.cpp
+#	via.cpp
 	;

RESULT:
the same stacktrace as previous.

After some testings i find that function
generic_mtrr_compute_physical_mask() in intel.cpp is 
executing successfully.
So, next i try to be sure that generic_dump_mtrrs
is executing successfully in intel.cpp:
[...]
-	generic_dump_mtrrs(generic_count_mtrrs());
+	generic_dump_mtrrs(0);
[...]
If i understood correctly we want
generic_count_mtrrs() returns 0 in the case
of a pentium mmx without mttrs.
Surprise!! even with generic_dump_mtrrs(0);
a got the same stacktrace.
So i focused in generic_dump_mtrrs (in generic_x86.cpp)
and did the following: (svn diff generic_x86.cpp > generic_x86.diff)

Index: generic_x86.cpp
===================================================================
--- generic_x86.cpp	(revision 19851)
+++ generic_x86.cpp	(working copy)
@@ -28,6 +28,7 @@
 #define IA32_MTRR_ENABLE_FIXED			(1UL << 10)
 #define IA32_MTRR_VALID_RANGE			(1UL << 11)
 
+#define IA32_FEATURE_MSR (1UL << 5)
 
 struct mtrr_capabilities {
 	mtrr_capabilities(uint64 value) { *(uint64 *)this = value; }
@@ -48,7 +49,8 @@
 {
 	cpuid_info cpuInfo;
 	if (get_current_cpuid(&cpuInfo, 1) != B_OK
-		|| (cpuInfo.eax_1.features & IA32_FEATURE_MTRR) == 0)
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MTRR) == 0
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MSR) == 0)
 		return 0;
 
 	mtrr_capabilities 
capabilities(x86_read_msr(IA32_MSR_MTRR_CAPABILITIES));
@@ -160,6 +162,13 @@
 void
 generic_dump_mtrrs(uint32 count)
 {
+
+	cpuid_info cpuInfo;
+	if (get_current_cpuid(&cpuInfo, 1) != B_OK
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MTRR) == 0
+		|| (cpuInfo.eax_1.features & IA32_FEATURE_MSR) == 0)
+		return;
+
 	if (x86_read_msr(IA32_MSR_MTRR_DEFAULT_TYPE) & IA32_MTRR_ENABLE) {
 		TRACE(("MTRR enabled\n"));
 	} else {
@@ -181,7 +190,7 @@
 
 module_info *modules[] = {
 	(module_info *)&gIntelModule,
-	(module_info *)&gAMDModule,
-	(module_info *)&gVIAModule,
+//	(module_info *)&gAMDModule,
+//	(module_info *)&gVIAModule,
 	NULL
 };

RESULT: the module is loaded successfully.
I think indeed generic_count_mtrrs() returns 0
and generic_dump_mtrrs returns immediately after the if
statement. The problem was indeed the execution of
x86_read_msr in generic_dump_mtrrs.