Created attachment 288221 [details] kernel .config (5.6.2, INLINE KASAN, PowerMac G4 DP) Was trying to do some testing with the PowerMac G4 DP again, running a 5.6.2 debug kernel w. KASAN INLINE. The G4 boots fine, but crashes shortly afterwards when using it, leaving no stack trace, but only this message on the screen: do_IRQ: stack overflow: 1760 CPU: 0 PID: 209 Comm: rsync Tained: G W 5.6.2-PowerMacG4+ #3 Call Trace: 120 seconds panic timer does not kick in. I have to manually switch off/switch on the G4.
So it hands in show_stack(). Does it also hang without CONFIG_DEBUG_STACKOVERFLOW ? If not, it means we have a problem with check_stack_overflow() Regardless of the result above, can you try increasing CONFIG_THREAD_SHIFT ? Can you maybe also do a test without CONFIG_VMAP_STACK ?
Created attachment 288229 [details] screenshot01.jpg Without CONFIG_DEBUG_STACKOVERFLOW things are better. The rsync completes, the G4 was building stuff for 2 hours or so until I got these errors and a hard freeze: [...] Oops: kernel stack overflow, sig: 11 [#1] BE PAGE_SIZE=4K MMU=Hash SMP NR_CPUS=2 PowerMac Modules linked in: ... CPU: 1 PID: 17105 Comm: kworker/u4:5 Tainted: G W 5.6.2-PowerMacG4+ #5 ------------[ cut here ]------------ kernel BUG at mm/usercopy.c:99! Oops: Exception in kernel mode, sig: 5 [#2] BE PAGE_SIZE=4K MMU=Hash SMP NR_CPUS=2 PowerMac Modules linked in: ... CPU: 1 PID: 17185 Comm: kworker/u4:5 Tainted: G W 5.6.2-PowerMacG4+ #5 usercopy: Kernel memory overwrite attempt detected to kernel text (offset 6336, size 4)! ------------[ cut here ]------------ kernel BUG at mm/usercopy.c:99! Oops: Exception in kernel mode, sig: 5 [#3] BE PAGE_SIZE=4K MMU=Hash SMP NR_CPUS=2 PowerMac Modules linked in: ... CPU: 1 PID: 17185 Comm: kworker/u4:5 Tainted: G W 5.6.2-PowerMacG4+ #5 usercopy: Kernel memory overwrite attempt detected to kernel text (offset 5336, size 4)! ------------[ cut here ]------------ kernel BUG at mm/usercopy.c:99! Oops: Exception in kernel mode, sig: 5 [#4] BE PAGE_SIZE=4K MMU=Hash SMP NR_CPUS=2 PowerMac Modules linked in: ... CPU: 1 PID: 17185 Comm: kworker/u4:5 Tainted: G W 5.6.2-PowerMacG4+ #5 usercopy: Kernel memory overwrite attempt detected to kernel text (offset 4336, size 4)! ------------[ cut here ]------------ kernel BUG at mm/usercopy.c:99! Oops: Exception in kernel mode, sig: 5 [#5] BE PAGE_SIZE=4K MMU=Hash SMP NR_CPUS=2 PowerMac Modules linked in: ... Unrecoverable FP Unavailable Exception 801 at 9b8 CPU: 1 PID: 17185 Comm: kworker/u4:5 Tainted: G W 5.6.2-PowerMacG4+ #5 usercopy: Kernel memory overwrite attempt detected to kernel text (offset 3336, size 4)! ------------[ cut here ]------------ Now running with CONFIG_THREAD_SHIFT=14 which runs fine so far... Did not try without CONFIG_VMAP_STACK yet.
Created attachment 288231 [details] screenshot02.jpg
Without CONFIG_VMAP_STACK I had one crash after 2-3 hours of building but the panic timer kicked in and rebooted the machine. Now it has been building packages for hours again without any anomalies.
Ok, so as a summary: - With CONFIG_THREAD_SHIFT = 13 and CONFIG_DEBUG_STACKOVERFLOW, the system gets stuck - With CONFIG_THREAD_SHIFT = 13 and without CONFIG_DEBUG_STACKOVERFLOW, stack overflow is not really detected until it gets into kernel text !!! - With CONFIG_THREAD_SHIFT = 14 it runs fine - With CONFIG_VMAP_STACK, the automatic restart doesn't work - Without CONFIG_VMAP_STACK, the automatic restart works So I'll send a patch to set CONFIG_THREAD_SHIFT to 14 when CONFIG_KASAN is selected. x86 and arm64 already do that. And I'll try to investigate the other points when I have time.
Yes, precisely summarized! Thanks for your efforts! CONFIG_KASAN though only is x86_64 not x86 AFAIK.
I think this was resolved by increasing the stack size for KASAN builds. ie. edbadaf06710 ("powerpc/kasan: Fix stack overflow by increasing THREAD_SHIFT") and later 3e8635fb2e07 ("powerpc/kasan: Force thread size increase with KASAN") If not feel free to reopen.
Currently (as of v6.8-rc6) and since quite a while my G4 does not boot at all with CONFIG_KASAN_INLINE=y. When I try booting an KASAN_INLINE enabled kernel it fails with an invalid memory access and I get dropped out to the OpenFirmaware console. A small difference is when I boot the 56M vmlinux-6.8.0-rc6-PMacG4: Please wait, loading kernel... Elf32 kernel loaded... Invalid memory access at %SRR0: 00000013 %SRR1: 00001300 Apple PowerMac3,6 4.6.0f1 BootROM built on 02/20/03 at 13:52:27 [...] vs. booting the 16M arch/powerpc/boot/zImage: Please wait, loading kernel... Elf32 kernel loaded... zImage starting: loaded at 0x00400000 (sp: 0x012eefb0) OF version = 'OpenFirmware 3' Allocating 0x2c337e0 bytes for kernel... Trying to claim from 0x400000 to 0x12ef5d8 (0xeef5d8) got ffffffff Decompressing (0x01414000 <- 0x00410000:0x12ea9c9)... Done! Decompressed 0x2bcc80c bytes Linux/PowerPC load: ro root=/dev/sda5 slub_debug=FZP page_poison=1 netconsole=6666@192.168.2.8/eth0,6666@192.168.2.3/A8:A1:59:16:4F:EA debug Finalizing device tree... using OF tree (promptr=ff847240) Invalid memory access at %SRR0: 40000000 %SRR1: 00000000 Apple PowerMac3,6 4.6.0f1 BootROM built on 02/20/03 at 13:52:27 [...] Same kernel with CONFIG_KASAN_OUTLINE=y instead of KASAN_INLINE boots and runs ok.
Created attachment 305910 [details] kernel .config (6.8-rc6, KASAN_INLINE=y, PowerMac G4 DP)
I built a kernel with your .config, the problem is a size problem. PPC32 kernels are not designed to be that big. Extract from generated System.map: c2394000 D _etext c2800000 T _sinittext c2bf5000 B _end You need to keep the size of the kernel below 32Mbytes, or a deep work is required to enable the kernel to perform far jumps before the kernel is relocated.
You were correct! I forgot about that... I shrunk the size by using -Os and disabling some debugging stuff and changing some statically built-in stuff to 'M' without sacrificing debugging capabilities too much until it fit < 32 MiB: KASAN_OUTLINE vs. # size vmlinux-6.8.0-rc6-PMacG4 text data bss dec hex filename 12367737 6652440 426336 19446513 128baf1 vmlinux-6.8.0-rc6-PMacG4 KASAN_INLINE # size vmlinux-6.8.0-rc6-PMacG4 text data bss dec hex filename 24660169 6652440 426336 31738945 1e44c41 vmlinux-6.8.0-rc6-PMacG4 Apart from that I can confirm inline KASAN runs fine now and I really no longer get this stack overflow when using it.