swsusp console change/userspace hang
Hi!
I'm currently working on an entirely userspace bootsplash program. It works quite happily, except in the case of resuming from hibernation. The splash program is launched at the start of initramfs, and at the end of initramfs (after the disk modules have been loaded) we attempt to trigger resume from userspace.
The code registers a signal that's fired on VT change. If a console change is requested when it's currently drawing, it disables drawing and schedules an alarm to fire 0.1 seconds later in order to avoid switching the console when part-way through a framebuffer operation.
The problem seems to be that swsusp tries to change the console and then immediately freezes userspace. For reasons I don't entirely understand, this freezes the machine. If I remove the pm_prepare_console call from pm_prepare_processes, resume functions correctly.
For now I'll probably just work around this by removing the console change from our kernels (we can do that in userspace scripting instead), but this still seems to be a less than ideal situation - I'm guessing that the same would happen if we were displaying the splash on suspend. Any ideas what might be causing this, and how to rectify it?
Signature · 1 line
Matthew Garrett | [email protected]
Quoted 13 lines · Matthew Garrett
Hi!
I'm currently working on an entirely userspace bootsplash program. It works quite happily, except in the case of resuming from hibernation. The splash program is launched at the start of initramfs, and at the end of initramfs (after the disk modules have been loaded) we attempt to trigger resume from userspace.
The code registers a signal that's fired on VT change. If a console change is requested when it's currently drawing, it disables drawing and schedules an alarm to fire 0.1 seconds later in order to avoid switching the console when part-way through a framebuffer operation.
Quoted 4 lines
The problem seems to be that swsusp tries to change the console and then immediately freezes userspace. For reasons I don't entirely understand, this freezes the machine. If I remove the
pm_prepare_consolecall frompm_prepare_processes, resume functions correctly.
Hrm.. it uses pm_prepare_console(), which should wait for the console to become active... However, it calls vt_waitactive without dealing with -EINTR. You might be getting a signal or something, can you check what's going on inside pm_prepare_console ?
Quoted 5 lines
For now I'll probably just work around this by removing the console change from our kernels (we can do that in userspace scripting instead), but this still seems to be a less than ideal situation - I'm guessing that the same would happen if we were displaying the splash on suspend. Any ideas what might be causing this, and how to rectify it?