Skip to contentSkip to conversation

grub2 int15 hook for e820/e801 does not return CF correctly; makes recent Linux detect only 64MB (via BIOS-88)

PATCH Superseded Josh Triplett 2 participants 3 messages Jun 25, 2010 – Jun 28, 2010

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 grub2 int15 hook for e820/e801 does not return CF correctly; makes recent Linux detect only 64MB (via BIOS-88)

As I originally reported in http://bugs.debian.org/584846 , and later tracked down, grub2's int15 hook, used to augment functions e820/e801/88 for additional reserved memory (such as for drivemap), fails to return its error status correctly, causing recent Linux (and possibly other OSes) to detect only 64MB via BIOS-88, ignoring e820 and e801.

This occurs because the interrupt must return a status in CF, but after using clc or stc to clear or set CF, it calls iret, which restores the flags from the stack. Since recent Linux has CF set when it calls int15, CF remains set when the interrupt returns, indicating an error. Linux's detect_memory_e820 and detect_memory_e801 functions thus cannot retrieve any memory map, and Linux must fall back to function 88, which cannot return more than 64MB.

To reproduce this problem, just use the "drivemap" command and then boot a recent Linux kernel (git commit c549e71d073a6e9a4847497344db28a784061455 or newer, meaning v2.6.30-rc1 or newer). In my case, I booted a USB drive with grub2 on it, used "drivemap -s (hd1) (hd0)", then chainloaded the hard drive.

The following patch against latest bzr fixes this problem. (Thanks to H. Peter Anvin for noticing a bug in my original patch and suggesting a fix.) With this patch, Linux's memory detection can use e820 again, and detect all memory.

=== modified file 'mmap/i386/pc/mmap_helper.S'

i386/pc/mmap_helper.S +11 -4
@@ -59,7 +59,7 @@
 	movw %bx, %dx
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 
 LOCAL (h88):
 	popf
@@ -69,7 +69,7 @@
 	movw DS (LOCAL (kbin16mb)), %ax
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 
 LOCAL (e820):
 	popf
@@ -101,12 +101,19 @@
 	mov $0x534d4150, %eax
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 LOCAL (errexit):
 	mov $0x534d4150, %eax
 	pop %ds
+	xor %bx, %bx
 	stc
-	xor %bx, %bx
+	jmp LOCAL (iret_cf)
+
+LOCAL (iret_cf):
+	push %bp
+	mov %sp, %bp
+	setc 6(%bp)
+	pop %bp
 	iret
 
 VARIABLE(grub_machine_mmaphook_mmap_num)

ChangeLog entry for this patch:

2010-06-25 Josh Triplett <[email protected]>

* mmap/i386/pc/mmap_helper.S: Preserve CF by propagating it into the flags on the stack before popping them with iret. Add a new iret_cf label implementing this logic, and make the e820, e801, and 88 handlers jump to it after setting or clearing CF. This fixes memory detection with recent Linux (git commit c549e71d073a6e9a4847497344db28a784061455 or newer, meaning v2.6.30-rc1 or newer), which sets CF on entry to int15, and treats the preserved CF on exit as an error in e820 and e801; without this change, it can only successfully use BIOS-88 and detect only up to 64MB of RAM. Originally reported as Debian bug #584846.

- Josh Triplett

#1 lore raw

PATCH v2 1/1 grub2 int15 hook for e820/e801 does not return CF correctly; makes recent Linux detect only 64MB (via BIOS-88)

Quoted 20 lines · Josh Triplett

As I originally reported in http://bugs.debian.org/584846 , and later tracked down, grub2's int15 hook, used to augment functions e820/e801/88 for additional reserved memory (such as for drivemap), fails to return its error status correctly, causing recent Linux (and possibly other OSes) to detect only 64MB via BIOS-88, ignoring e820 and e801.

This occurs because the interrupt must return a status in CF, but after using clc or stc to clear or set CF, it calls iret, which restores the flags from the stack. Since recent Linux has CF set when it calls int15, CF remains set when the interrupt returns, indicating an error. Linux's detect_memory_e820 and detect_memory_e801 functions thus cannot retrieve any memory map, and Linux must fall back to function 88, which cannot return more than 64MB.

To reproduce this problem, just use the "drivemap" command and then boot a recent Linux kernel (git commit c549e71d073a6e9a4847497344db28a784061455 or newer, meaning v2.6.30-rc1 or newer). In my case, I booted a USB drive with grub2 on it, used "drivemap -s (hd1) (hd0)", then chainloaded the hard drive.

Quoted 4 lines

The following patch against latest bzr fixes this problem. (Thanks to H. Peter Anvin for noticing a bug in my original patch and suggesting a fix.) With this patch, Linux's memory detection can use e820 again, and detect all memory.

Revised patch, incorporating suggestion from phcoder to eliminate the unnecessary jmp right before the iret_cf label:

=== modified file 'mmap/i386/pc/mmap_helper.S'

i386/pc/mmap_helper.S +10 -4
@@ -59,7 +59,7 @@
 	movw %bx, %dx
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 
 LOCAL (h88):
 	popf
@@ -69,7 +69,7 @@
 	movw DS (LOCAL (kbin16mb)), %ax
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 
 LOCAL (e820):
 	popf
@@ -101,12 +101,18 @@
 	mov $0x534d4150, %eax
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 LOCAL (errexit):
 	mov $0x534d4150, %eax
 	pop %ds
+	xor %bx, %bx
 	stc
-	xor %bx, %bx
+
+LOCAL (iret_cf):
+	push %bp
+	mov %sp, %bp
+	setc 6(%bp)
+	pop %bp
 	iret
 
 VARIABLE(grub_machine_mmaphook_mmap_num)

ChangeLog entry for this patch:

2010-06-25 Josh Triplett <[email protected]>

* mmap/i386/pc/mmap_helper.S: Preserve CF by propagating it into the flags on the stack before popping them with iret. Add a new iret_cf label implementing this logic, and make the e820, e801, and 88 handlers jump to it after setting or clearing CF. This fixes memory detection with recent Linux (git commit c549e71d073a6e9a4847497344db28a784061455 or newer, meaning v2.6.30-rc1 or newer), which sets CF on entry to int15, and treats the preserved CF on exit as an error in e820 and e801; without this change, it can only successfully use BIOS-88 and detect only up to 64MB of RAM. Originally reported as Debian bug #584846.

- Josh Triplett

#2 lore raw
Quoted 22 lines · Josh Triplett
Quoted 20 lines · Josh Triplett

As I originally reported in http://bugs.debian.org/584846 , and later tracked down, grub2's int15 hook, used to augment functions e820/e801/88 for additional reserved memory (such as for drivemap), fails to return its error status correctly, causing recent Linux (and possibly other OSes) to detect only 64MB via BIOS-88, ignoring e820 and e801.

This occurs because the interrupt must return a status in CF, but after using clc or stc to clear or set CF, it calls iret, which restores the flags from the stack. Since recent Linux has CF set when it calls int15, CF remains set when the interrupt returns, indicating an error. Linux's detect_memory_e820 and detect_memory_e801 functions thus cannot retrieve any memory map, and Linux must fall back to function 88, which cannot return more than 64MB.

To reproduce this problem, just use the "drivemap" command and then boot a recent Linux kernel (git commit c549e71d073a6e9a4847497344db28a784061455 or newer, meaning v2.6.30-rc1 or newer). In my case, I booted a USB drive with grub2 on it, used "drivemap -s (hd1) (hd0)", then chainloaded the hard drive.

Quoted 9 lines
Quoted 5 lines

The following patch against latest bzr fixes this problem. (Thanks to H. Peter Anvin for noticing a bug in my original patch and suggesting a fix.) With this patch, Linux's memory detection can use e820 again, and detect all memory.

Revised patch, incorporating suggestion from phcoder to eliminate the unnecessary jmp right before the iret_cf label:

Comitted. Thanks

Quoted 69 lines

=== modified file 'mmap/i386/pc/mmap_helper.S'

i386/pc/mmap_helper.S +10 -4
@@ -59,7 +59,7 @@
 	movw %bx, %dx
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 
 LOCAL (h88):
 	popf
@@ -69,7 +69,7 @@
 	movw DS (LOCAL (kbin16mb)), %ax
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 
 LOCAL (e820):
 	popf
@@ -101,12 +101,18 @@
 	mov $0x534d4150, %eax
 	pop %ds
 	clc
-	iret
+	jmp LOCAL (iret_cf)
 LOCAL (errexit):
 	mov $0x534d4150, %eax
 	pop %ds
+	xor %bx, %bx
 	stc
-	xor %bx, %bx
+
+LOCAL (iret_cf):
+	push %bp
+	mov %sp, %bp
+	setc 6(%bp)
+	pop %bp
 	iret
 
 VARIABLE(grub_machine_mmaphook_mmap_num)

ChangeLog entry for this patch:

2010-06-25 Josh Triplett <[email protected]>

* mmap/i386/pc/mmap_helper.S: Preserve CF by propagating it into the flags on the stack before popping them with iret. Add a new iret_cf label implementing this logic, and make the e820, e801, and 88 handlers jump to it after setting or clearing CF. This fixes memory detection with recent Linux (git commit c549e71d073a6e9a4847497344db28a784061455 or newer, meaning v2.6.30-rc1 or newer), which sets CF on entry to int15, and treats the preserved CF on exit as an error in e820 and e801; without this change, it can only successfully use BIOS-88 and detect only up to 64MB of RAM. Originally reported as Debian bug #584846.

- Josh Triplett

_______________________________________________
Grub-devel mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/grub-devel

Signature · 2 lines
Regards
Vladimir 'φ-coder/phcoder' Serbinenko