Skip to contentSkip to conversation

2.6.29 acpi regression: acpi_ex_extract_from_field -- div by zero

DISCUSSION Jiri Slaby 4 participants 10 messages Mar 15, 2009 – Mar 20, 2009 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

Jiri Slaby

author
#0 lore raw

Hi,

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

This is the case:

        /* Mask off any extra bits in the last datum */

        buffer_tail_bits = obj_desc->common_field.bit_length %
            obj_desc->common_field.access_bit_width;

.L39:

        xorl    %edx, %edx      #
        movzbl  37(%rbx), %esi  # 

<variable>.common_field.access_bit_width, <variable>.common_field.access_bit_width
movl 24(%rbx), %eax # <variable>.common_field.bit_length,
<variable>.common_field.bit_length
--------------- here:

        divl    %esi    # <variable>.common_field.access_bit_width
        movl    %edx, %ecx      #, tmp121
        testl   %edx, %edx      # tmp121
        je      .L41    #,

[1] http://www.fi.muni.cz/~xslaby/sklad/panics/acpi_oops.png

#1 lore raw
Quoted 11 lines · Jiri Slaby

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

This is the case:

        /* Mask off any extra bits in the last datum */

        buffer_tail_bits = obj_desc->common_field.bit_length %
            obj_desc->common_field.access_bit_width;

Hmm. "bit_length" is zero too, according to the oops (it's in %eax).

Looks like the whole "obj_desc" is basically uninitialized. I have no idea why, and why it is apparently not reliable. ACPI is odd.

Does everything work if you just do something like the appended?

It's insane, but so is ACPI.

Linus

---
 drivers/acpi/acpica/exfldio.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
drivers/acpi/acpica/exfldio.c +3 -0
index ef58ac4..9297adb 100644
@@ -698,6 +698,9 @@ acpi_ex_extract_from_field(union acpi_operand_object *obj_desc,
 	}
 	ACPI_MEMSET(buffer, 0, buffer_length);
 
+	if (!obj_desc->common_field.bit_length)
+		return_ACPI_STATUS(AE_OK);
+
 	/* Compute the number of datums (access width data items) */
 
 	datum_count = ACPI_ROUND_UP_TO(obj_desc->common_field.bit_length,
#2 lore raw
In reply to Jiri Slaby

Can you please send or post the acpidump for the machine? Thanks.

Quoted 40 lines

-----Original Message-----
From: [email protected] [mailto:linux-acpi- [email protected]] On Behalf Of Jiri Slaby
Sent: Sunday, March 15, 2009 3:47 AM
To: Len Brown
Cc: [email protected]; [email protected]; Linux kernel mailing list; [email protected]
Subject: 2.6.29 acpi regression: acpi_ex_extract_from_field -- div by zero

Hi,

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

This is the case:

        /* Mask off any extra bits in the last datum */

        buffer_tail_bits = obj_desc->common_field.bit_length %
            obj_desc->common_field.access_bit_width;

.L39:

        xorl    %edx, %edx      #
        movzbl  37(%rbx), %esi  #

<variable>.common_field.access_bit_width,
<variable>.common_field.access_bit_width
movl 24(%rbx), %eax # <variable>.common_field.bit_length,
<variable>.common_field.bit_length
--------------- here:

        divl    %esi    # <variable>.common_field.access_bit_width
        movl    %edx, %ecx      #, tmp121
        testl   %edx, %edx      # tmp121
        je      .L41    #,

[1] http://www.fi.muni.cz/~xslaby/sklad/panics/acpi_oops.png
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#5 lore raw
In reply to a message not in the archive
Quoted 7 lines

Hi,

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

Would please try below debug patch to see which region filed is accessed?

drivers/acpi/acpica/exfldio.c +15 -0
index ef58ac4..10ac0c8 100644
@@ -683,6 +683,7 @@ acpi_ex_extract_from_field(union acpi_operand_object *obj_desc,
 	u32 datum_count;
 	u32 field_datum_count;
 	u32 i;
+	struct acpi_buffer name_buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 
 	ACPI_FUNCTION_TRACE(ex_extract_from_field);
 
@@ -765,6 +766,20 @@ acpi_ex_extract_from_field(union acpi_operand_object *obj_desc,
 		    raw_datum >> obj_desc->common_field.start_field_bit_offset;
 	}
 
+	if (obj_desc->common_field.bit_length == 0 ||
+	    obj_desc->common_field.access_bit_width == 0) {
+		status = acpi_get_name(obj_desc->common_field.node,
+				ACPI_FULL_PATHNAME, &name_buffer);
+		if (ACPI_FAILURE(status)) {
+			printk(KERN_DEBUG "ACPI Debug: %s\n",
+				acpi_format_exception(status));
+		} else {
+			printk(KERN_DEBUG "ACPI Debug: field node path: %s\n",
+				(char *) name_buffer.pointer);
+			kfree(name_buffer.pointer);
+		}
+	}
+
 	/* Mask off any extra bits in the last datum */
 
 	buffer_tail_bits = obj_desc->common_field.bit_length %

---
Lin Ming

Quoted 22 lines

This is the case:

       /* Mask off any extra bits in the last datum */

       buffer_tail_bits = obj_desc->common_field.bit_length %
           obj_desc->common_field.access_bit_width;

.L39:

       xorl    %edx, %edx      #
       movzbl  37(%rbx), %esi  #

<variable>.common_field.access_bit_width,
<variable>.common_field.access_bit_width
movl 24(%rbx), %eax # <variable>.common_field.bit_length,
<variable>.common_field.bit_length
--------------- here:

       divl    %esi    # <variable>.common_field.access_bit_width
       movl    %edx, %ecx      #, tmp121
       testl   %edx, %edx      # tmp121
       je      .L41    #,

[1] http://www.fi.muni.cz/~xslaby/sklad/panics/acpi_oops.png
--

author
#6 lore raw
Quoted 7 lines · Lin Ming
Quoted 4 lines

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

Would please try below debug patch to see which region filed is accessed?

Yes, except the fact, we wouldn't see anything :) -- I'll add inside the if a return statement or something to disallow the oops to flood screen.

Also, it doesn't happen regularly. Mostly several first tries after cold start. I hope this is a regression and not a HW problem, but it emerged after switching to 2.6.29-*, I will keep you informed.

Thanks.

author
#7 lore raw
Quoted 9 lines · Jiri Slaby
Quoted 7 lines · Lin Ming
Quoted 4 lines

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

Would please try below debug patch to see which region filed is accessed?

Quoted 6 lines

Yes, except the fact, we wouldn't see anything :) -- I'll add inside the if a return statement or something to disallow the oops to flood screen.

Also, it doesn't happen regularly. Mostly several first tries after cold start. I hope this is a regression and not a HW problem, but it emerged after switching to 2.6.29-*, I will keep you informed.

Hmm, I didn't see it with rc8 and this patch applied so far (no 'ACPI Debug' in dmesg). Also I don't see anything changed within drivers/acpi/ et al. between rc7 and rc8. Maybe the timing changed somewhere else... I'm confused, but keep trying.

author
#8 lore raw
Quoted 11 lines · Jiri Slaby
Quoted 8 lines · Jiri Slaby
Quoted 7 lines · Lin Ming
Quoted 4 lines

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

Would please try below debug patch to see which region filed is accessed?

I'm confused, but keep trying.

Got it again. obj_desc->common_field.node is martian too (0x4000000), so the added acpi_get_name dies:
http://www.fi.muni.cz/~xslaby/sklad/panics/acpi_oops1.png

Whole common_field seems to be mangled. Ideas?

#9 lore raw
Quoted 11 lines · Jiri Slaby
Quoted 10 lines · Jiri Slaby
Quoted 8 lines · Jiri Slaby
Quoted 7 lines · Lin Ming
Quoted 4 lines

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

Would please try below debug patch to see which region filed is accessed?

Quoted 7 lines
Quoted 1 line

I'm confused, but keep trying.

Got it again. obj_desc->common_field.node is martian too (0x4000000), so the added acpi_get_name dies:
http://www.fi.muni.cz/~xslaby/sklad/panics/acpi_oops1.png

Whole common_field seems to be mangled. Ideas?

Thanks for the info. We are looking at this.

Lin Ming

#10 lore raw
In reply to Jiri Slaby
Quoted 11 lines · Jiri Slaby
Quoted 10 lines · Jiri Slaby
Quoted 8 lines · Jiri Slaby
Quoted 7 lines · Lin Ming
Quoted 4 lines

sometimes, when booting up/resuming from disk, I get an oops[1].

obj_desc->common_field.access_bit_width is zero, but even after the loop. Division before the loop is apparently OK.

Would please try below debug patch to see which region filed is accessed?

Quoted 7 lines
Quoted 1 line

I'm confused, but keep trying.

Got it again. obj_desc->common_field.node is martian too (0x4000000), so the added acpi_get_name dies:
http://www.fi.muni.cz/~xslaby/sklad/panics/acpi_oops1.png

Whole common_field seems to be mangled. Ideas?

I should save the field object and node path name at the entry of acpi_ex_extract_from_field.
Would you please try below debug patch?
I also add a return statement before the fault to disallow the oops to flood screen.

Thanks.

drivers/acpi/acpica/exfldio.c +66 -0
index ef58ac4..5109f5b 100644
@@ -670,6 +670,34 @@ acpi_ex_write_with_update_rule(union acpi_operand_object *obj_desc,
  *
  ******************************************************************************/
 
+void debug_dump_field(union acpi_operand_object *obj_desc)
+{
+	struct acpi_object_field_common *obj_field;
+
+	if (!obj_desc) {
+		printk(KERN_DEBUG "ACPI Debug: NULL field object\n");
+		return;
+	}
+
+	printk(KERN_DEBUG "next_object: %p\n", obj_desc->common.next_object);
+	printk(KERN_DEBUG "descriptor_type: %x\n", obj_desc->common.descriptor_type);
+	printk(KERN_DEBUG "type: %x\n", obj_desc->common.type);
+	printk(KERN_DEBUG "reference_count: %x\n", obj_desc->common.reference_count);
+	printk(KERN_DEBUG "flags: %x\n", obj_desc->common.flags);
+
+	obj_field = &obj_desc->common_field;
+
+	printk(KERN_DEBUG "field_flags: %x\n", obj_field->field_flags);
+	printk(KERN_DEBUG "attribute: %x\n", obj_field->attribute);
+	printk(KERN_DEBUG "access_byte_width: %x\n", obj_field->access_byte_width);
+	printk(KERN_DEBUG "node: %p\n", obj_field->node);
+	printk(KERN_DEBUG "bit_length: %x\n", obj_field->bit_length);
+	printk(KERN_DEBUG "base_byte_offset: %x\n", obj_field->base_byte_offset);
+	printk(KERN_DEBUG "value: %x\n", obj_field->value);
+	printk(KERN_DEBUG "start_field_bit_offset: %x\n", obj_field->start_field_bit_offset);
+	printk(KERN_DEBUG "access_bit_width: %x\n", obj_field->access_bit_width);
+}
+
 acpi_status
 acpi_ex_extract_from_field(union acpi_operand_object *obj_desc,
 			   void *buffer, u32 buffer_length)
@@ -683,9 +711,24 @@ acpi_ex_extract_from_field(union acpi_operand_object *obj_desc,
 	u32 datum_count;
 	u32 field_datum_count;
 	u32 i;
+	struct acpi_buffer name_buffer = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_operand_object backup_obj;
 
 	ACPI_FUNCTION_TRACE(ex_extract_from_field);
 
+	/* backup the object */
+	memcpy(&backup_obj, obj_desc, sizeof(*obj_desc));
+
+	name_buffer.pointer = NULL;
+	if (obj_desc->common_field.node) {
+		status = acpi_get_name(obj_desc->common_field.node,
+				ACPI_FULL_PATHNAME, &name_buffer);
+		if (ACPI_FAILURE(status)) {
+			printk(KERN_DEBUG "ACPI Debug: %s\n",
+				acpi_format_exception(status));
+		}
+	}
+
 	/* Validate target buffer and clear it */
 
 	if (buffer_length <
@@ -765,6 +808,29 @@ acpi_ex_extract_from_field(union acpi_operand_object *obj_desc,
 		    raw_datum >> obj_desc->common_field.start_field_bit_offset;
 	}
 
+	if (obj_desc->common_field.bit_length == 0 ||
+	    obj_desc->common_field.access_bit_width == 0) {
+
+		/* oops, dump the object */
+		if (name_buffer.pointer) {
+			printk(KERN_DEBUG "ACPI Debug: field node path: %s\n",
+				(char *) name_buffer.pointer);
+		}
+
+		printk(KERN_DEBUG "ACPI Debug: The original object:\n");
+		debug_dump_field(&backup_obj);
+
+		printk(KERN_DEBUG "ACPI Debug: The bad object:\n");
+		debug_dump_field(obj_desc);
+
+		kfree(name_buffer.pointer);
+
+		/* we return here to disallow the oops to flood screen */
+		return_ACPI_STATUS(AE_ERROR);
+	}
+
+	kfree(name_buffer.pointer);
+
 	/* Mask off any extra bits in the last datum */
 
 	buffer_tail_bits = obj_desc->common_field.bit_length %