Virtualization - Apr 2007 - [PATCH 2/3] Pnp bios gdt fix

Zachary Amsden wrote:
>PnP BIOS for x86 is part of drivers, so I missed it in the initial
>GDT page alignment patch.  Kudos to Andrew for fixing that.
>Unfortunately, fixing the build introduced a kernel panic when
>trying to setup the as of yet unallocated GDTs for the APs.
>This fixes the problem by setting only the BSP's GDT, then copying
>the PnP segments back to the cpu_gdt_table template.
>
>  
>
Bogus patch!  I was just getting lucky and not running PnP BIOS code on 
the AP's -- until now.  Turns out the fix and the bug were much 
simpler.  Please apply this instead.


-------------- next part --------------
PnP BIOS for x86 is part of drivers, so I missed it in the initial
GDT page alignment patch.  Kudos to Andrew for fixing that.
Unfortunately, fixing the build introduced a kernel panic when
trying to setup the as of yet unallocated GDTs for the APs.

I totally misdiagnosed this.  The problem happens only
when NR_CPUS > physical CPUs.  Fix is to ignore missing GDT's.

Signed-off-by: Zachary Amsden <zach@vmware.com>
Index: linux-2.6.14-rc2/drivers/pnp/pnpbios/bioscalls.c
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
--- linux-2.6.14-rc2.orig/drivers/pnp/pnpbios/bioscalls.c	2005-09-28
15:57:42.000000000 -0700
+++ linux-2.6.14-rc2/drivers/pnp/pnpbios/bioscalls.c	2005-09-28
15:59:37.000000000 -0700
@@ -537,6 +537,8 @@ void pnpbios_calls_init(union pnp_bios_i
 	_set_limit((char *)&bad_bios_desc, 4095 - (0x40 << 4));
 	for(i=3D0; i < NR_CPUS; i++)
 	{
+		if (!get_cpu_gdt_table(i))
+			continue;
 		Q2_SET_SEL(i, PNP_CS32, &pnp_bios_callfunc, 64 * 1024);
 		Q_SET_SEL(i, PNP_CS16, header->fields.pm16cseg, 64 * 1024);
 		Q_SET_SEL(i, PNP_DS, header->fields.pm16dseg, 64 * 1024);

PnP BIOS for x86 is part of drivers, so I missed it in the initial
GDT page alignment patch.  Kudos to Andrew for fixing that.
Unfortunately, fixing the build introduced a kernel panic when
trying to setup the as of yet unallocated GDTs for the APs.
This fixes the problem by setting only the BSP's GDT, then copying
the PnP segments back to the cpu_gdt_table template.

Signed-off-by: Zachary Amsden <zach@vmware.com>
Index: linux-2.6.14-rc2/drivers/pnp/pnpbios/bioscalls.c
==================================================================---
linux-2.6.14-rc2.orig/drivers/pnp/pnpbios/bioscalls.c	2005-09-28
14:16:34.000000000 -0700
+++ linux-2.6.14-rc2/drivers/pnp/pnpbios/bioscalls.c	2005-09-28
14:23:57.000000000 -0700
@@ -528,17 +528,24 @@ static int pnp_bios_write_escd(char *dat
 
 void pnpbios_calls_init(union pnp_bios_install_struct *header)
 {
-	int i;
 	spin_lock_init(&pnp_bios_lock);
 	pnp_bios_callpoint.offset = header->fields.pm16offset;
 	pnp_bios_callpoint.segment = PNP_CS16;
 
 	set_base(bad_bios_desc, __va((unsigned long)0x40 << 4));
 	_set_limit((char *)&bad_bios_desc, 4095 - (0x40 << 4));
-	for(i=0; i < NR_CPUS; i++)
-	{
-		Q2_SET_SEL(i, PNP_CS32, &pnp_bios_callfunc, 64 * 1024);
-		Q_SET_SEL(i, PNP_CS16, header->fields.pm16cseg, 64 * 1024);
-		Q_SET_SEL(i, PNP_DS, header->fields.pm16dseg, 64 * 1024);
-	}
+
+	/*
+	 * This is awkward; GDT entries needed for this driver must
+	 * be set during init on the BSP, but also copied into the
+	 * cpu_gdt_table template for the APs to acquire.  This is
+	 * because APs will not have allocated GDTs until later in
+	 * the boot process.
+	 */
+	Q2_SET_SEL(0, PNP_CS32, &pnp_bios_callfunc, 64 * 1024);
+	Q_SET_SEL(0, PNP_CS16, header->fields.pm16cseg, 64 * 1024);
+	Q_SET_SEL(0, PNP_DS, header->fields.pm16dseg, 64 * 1024);
+	memcpy(&cpu_gdt_table[GDT_ENTRY_PNPBIOS_BASE],
+		&get_cpu_gdt_table(0)[GDT_ENTRY_PNPBIOS_BASE],
+		3 * sizeof(struct desc_struct));
 }