Skip to contentSkip to conversation

makedumpfile: fix for xen-3.1.2

RFC PATCH In review Akio Takebe 2 participants 3 messages Apr 4, 2008

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 makedumpfile: fix for xen-3.1.2

Hi,

This patch fixes makedumpfile for xen-3.1. xen-3.1.2+ hypervisor doesn't have idle_pg_table_l3, we need to find a symbole of idle_pg_table.

Best Regards,

Akio Takebe

makedumpfile.c +5 -0
@@ -5298,7 +5298,12 @@ get_symbol_info_xen()
 	 */
 	SYMBOL_INIT(pgd_l2, "idle_pg_table_l2");	/* x86 */
 	SYMBOL_INIT(pgd_l3, "idle_pg_table_l3");	/* x86-PAE */
+	if (SYMBOL(pgd_l3) == NOT_FOUND_SYMBOL)
+		SYMBOL_INIT(pgd_l3, "idle_pg_table");	/* x86-PAE */
 	SYMBOL_INIT(pgd_l4, "idle_pg_table_4");		/* x86_64 */
+	if (SYMBOL(pgd_l4) == NOT_FOUND_SYMBOL)
+		SYMBOL_INIT(pgd_l4, "idle_pg_table");		/* x86_64 */
+
 	SYMBOL_INIT(xen_heap_start, "xen_heap_start");	/* ia64 */
 	SYMBOL_INIT(xen_pstart, "xen_pstart");		/* ia64 */
 	SYMBOL_INIT(frametable_pg_dir, "frametable_pg_dir");	/* ia64 */
#1 lore raw

Hi Takebe-san,

Quoted 3 lines · Akio Takebe

This patch fixes makedumpfile for xen-3.1. xen-3.1.2+ hypervisor doesn't have idle_pg_table_l3, we need to find a symbole of idle_pg_table.

Thank you for the patch !
I tested your patch with RHEL5.2-beta, and it worked fine on i386 machine. But it doesn't work on x86_64 machine. makedumpfile went to the loop in kvtop_xen_x86_64(), and it failed with a segmentation fault:

makedumpfile/x86_64.c
196 if ((dirp = kvtop_xen_x86_64(SYMBOL(pgd_l4))) == NOT_PADDR) <- *HERE*
197 return NOT_PADDR;

If you know, please teach me. Should this patch fix the problem for not only the upstream but also RHEL5.2 ?
Unfortunately I don't know the detail of xen, and I haven't used the upstream xen yet.

Thanks
Ken'ichi Ohmichi

Quoted 15 lines
makedumpfile.c +5 -0
@@ -5298,7 +5298,12 @@ get_symbol_info_xen()
 	 */
 	SYMBOL_INIT(pgd_l2, "idle_pg_table_l2");	/* x86 */
 	SYMBOL_INIT(pgd_l3, "idle_pg_table_l3");	/* x86-PAE */
+	if (SYMBOL(pgd_l3) == NOT_FOUND_SYMBOL)
+		SYMBOL_INIT(pgd_l3, "idle_pg_table");	/* x86-PAE */
 	SYMBOL_INIT(pgd_l4, "idle_pg_table_4");		/* x86_64 */
+	if (SYMBOL(pgd_l4) == NOT_FOUND_SYMBOL)
+		SYMBOL_INIT(pgd_l4, "idle_pg_table");		/* x86_64 */
+
 	SYMBOL_INIT(xen_heap_start, "xen_heap_start");	/* ia64 */
 	SYMBOL_INIT(xen_pstart, "xen_pstart");		/* ia64 */
 	SYMBOL_INIT(frametable_pg_dir, "frametable_pg_dir");	/* ia64 */
#2 lore raw

Hi, Ohmichi-san

Thank you very much for you report.

Quoted 16 lines
Quoted 3 lines · Akio Takebe

This patch fixes makedumpfile for xen-3.1. xen-3.1.2+ hypervisor doesn't have idle_pg_table_l3, we need to find a symbole of idle_pg_table.

Thank you for the patch !
I tested your patch with RHEL5.2-beta, and it worked fine on i386 machine. But it doesn't work on x86_64 machine. makedumpfile went to the loop in kvtop_xen_x86_64(), and it failed with a segmentation fault:

makedumpfile/x86_64.c
196 if ((dirp = kvtop_xen_x86_64(SYMBOL(pgd_l4))) == NOT_PADDR) <-
*HERE*
197 return NOT_PADDR;

Quoted 4 lines

If you know, please teach me. Should this patch fix the problem for not only the upstream but also RHEL5.2 ?
Unfortunately I don't know the detail of xen, and I haven't used the upstream xen yet.

Sorry for that.
Yes, this patch will fix also RHEL5.2 because the base is 3.1.2.
https://www.redhat.com/archives/rhelv5-list/2008-March/msg00111.html

The bug may be occurred by changing __XEN_VIRT_START.
http://xenbits.xensource.com/xen-3.1-testing.hg?rev/a06a28ebad5d

So we may be able to fix it by changing DIRECTMAP_VIRT_START.

I didn't have x86_64 system.
I'll setup x86_64 system, and try to investigate it next week.

Best Regards,

Akio Takebe