Most recent kernel where this bug did not occur: 2.6.16.24 Distribution: vanilla kernel on Gentoo Hardware Environment: Dell Inspiron 9100, Intel Pentium 4 with w/HT stepping 9 Software Environment: gcc (GCC) 3.4.6 (Gentoo 3.4.6-r1, ssp-3.4.5-1.0, pie-8.7.9) Problem Description: Failure to resume. With radeonfb, "Linu" is displayed in probably 80x25 vga mode (as normal) but the system then hangs. With vgacon, a call stack is produced, as will be attached. Steps to reproduce: hibernate-ram, which performs echo mem > /sys/power/state. press power button to resume. (Both radeonfb and vgacon resumed fine without acpi_sleep= for 2.6.16.)
Created attachment 8605 [details] call stack screenshot I don't know whether this is the first message displayed and I can't think how to get previous messages. By trying various vga= options I could only get the previous 3 lines: divide error: 0000 [#1] PREEMPT SMP Modules linked in: usbcore ext2 (The patch in bug #6492 is applied, but the problem occurred without this patch.) gdb vmlinux (gdb) l *0xc0222ee9 0xc0222ee9 is in acpi_pm_enter (main.c:109). 104 105 default: 106 return -EINVAL; 107 } 108 local_irq_restore(flags); 109 printk(KERN_DEBUG "Back to C!\n"); 110 111 /* restore processor state 112 * We should only be here if we're coming back from STR or STD. 113 * And, in the case of the latter, the memory image should have already (gdb) x/100i acpi_pm_enter ... 0xc0222ebc <acpi_pm_enter+78>: call 0xc0112150 <do_suspend_lowlevel> 0xc0222ec1 <acpi_pm_enter+83>: jmp 0xc0222ee5 <acpi_pm_enter+119> ... 0xc0222ee5 <acpi_pm_enter+119>: pushl 0xfffffff0(%ebp) 0xc0222ee8 <acpi_pm_enter+122>: popf 0xc0222ee9 <acpi_pm_enter+123>: push $0xc0314844 0xc0222eee <acpi_pm_enter+128>: call 0xc0122aac <printk> 0xc0222ef3 <acpi_pm_enter+133>: dec %esi 0xc0222ef4 <acpi_pm_enter+134>: pop %eax 0xc0222ef5 <acpi_pm_enter+135>: jle 0xc0222efc <acpi_pm_enter+142> 0xc0222ef7 <acpi_pm_enter+137>: call 0xc011027e <acpi_restore_state_mem>
Created attachment 8606 [details] config
do you see similar issue without fb module?
Yes, I saw the same issue without CONFIG_FB. I added CONFIG_FB for FB_VESA in an unsuccessful attempt to get more text on the display. The call stack attached here was produced with fbcon=map:1 and/or vga=ask command line options to revert to using vgacon, but the output is very similar to that without CONFIG_FB.
Thanks! How about a UP kernel? And do you ever try a latest kernel like 2.6.18- rc2?
I don't usually try out the rc's, but no point trying to track down a bug if it has already been fixed. suspend2 to ram seems to be working with 2.6.18-rc2 SMP (radeonfb and vgacon), but the kernel config I used is quite different and some other things are not working properly: e.g. it is now 2032, but I haven't yet worked out whether this is related to suspend. I suggest leaving this bug open until I do some comparisons with more similar configs, but it may be a couple of days before I get a chance to look at this. Thanks for your help.
2.6.18-rc2 works fine when configured very similarly to 2.6.17.6 which demonstrated this bug. (And I should have read the documentation before enabling CONFIG_PM_TRACE.) Marking this resolved. Not sure if PATCH_ALREADY_AVAILABLE is quite right as this large patch may not have been consciously designed to resolve this bug: http://www.kernel.org/pub/linux/kernel/v2.6/testing/patch-2.6.18-rc2.bz2