*** This bug is a duplicate of bug 1967130 ***
https://bugs.launchpad.net/bugs/1967130
Public bug reported:
[ Impact ]
The kernel crashes when trying to access the EFI runtime
services/variables if the kernel is configured with CONFIG_VMAP_STACK.
Indeed, to access EFI variables/services, the kernel switches to the EFI
page table which is created during the boot process *but* is never
updated during the life of the kernel: as CONFIG_VMAP_STACK allocates
the kernel stack in the vmalloc region and maps it in the *kernel page
table*, the EFI page table never has this new mapping and then segfaults
when trying to access the kernel stack of the current process.
This is easier to reproduce on the Unmatched than on QEMU (never
reproduced on this platform):
ubuntu@ubuntu:~$ cat /sys/firmware/efi/efivars/*
Which gives rise to a kernel crash, gdb traces below:
(gdb) bt
#0 handle_exception () at /home/alex/work/linux/arch/riscv/kernel/entry.S:27
#1 0xffffffff8080741a in efi_virtmap_load () at
/home/alex/work/linux/arch/riscv/include/asm/current.h:31
#2 0x0000000000000000 in ?? ()
(gdb) p/x $stval
$1 = 0xffffffc8946cbc38
(gdb) display /i $sepc
1: x/i $sepc
0xffffffff8000392c <handle_exception+44>: sd ra,8(sp)
(gdb) p/x $sp
$6 = 0xffffffc8946cbc30
A workaround can be to disable CONFIG_VMAP_STACK but actually any
mapping in the kernel page table that is not also added to the EFI page
table can cause such crash: but I'm not sure this can happen and this
workaround was enough in the previous release to fix this bug and to be
able to live install a system.
I advise to implement this workaround for now as I'm implementing the
proper patch which consists in updating the EFI page table before
switching to it, just like x86 does.
[ Test Plan ]
Use a kernel with CONFIG_VMAP_STACK disabled and simply enter the
following command, preferably on the Unmatched:
ubuntu@ubuntu:~$ cat /sys/firmware/efi/efivars/*
This should not trigger a crash.
[ Where problems could occur ]
As said above, just disabling CONFIG_VMAP_STACK may not be enough.
** Affects: linux-riscv (Ubuntu)
Importance: Undecided
Status: New
** This bug has been marked a duplicate of bug 1967130
rcu_sched detected stalls on CPUs/tasks
--
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux-riscv in Ubuntu.
https://bugs.launchpad.net/bugs/1992458
Title:
Disable VMAP_STACK as it causes crash in EFI
Status in linux-riscv package in Ubuntu:
New
Bug description:
[ Impact ]
The kernel crashes when trying to access the EFI runtime
services/variables if the kernel is configured with CONFIG_VMAP_STACK.
Indeed, to access EFI variables/services, the kernel switches to the
EFI page table which is created during the boot process *but* is never
updated during the life of the kernel: as CONFIG_VMAP_STACK allocates
the kernel stack in the vmalloc region and maps it in the *kernel page
table*, the EFI page table never has this new mapping and then
segfaults when trying to access the kernel stack of the current
process.
This is easier to reproduce on the Unmatched than on QEMU (never
reproduced on this platform):
ubuntu@ubuntu:~$ cat /sys/firmware/efi/efivars/*
Which gives rise to a kernel crash, gdb traces below:
(gdb) bt
#0 handle_exception () at /home/alex/work/linux/arch/riscv/kernel/entry.S:27
#1 0xffffffff8080741a in efi_virtmap_load () at
/home/alex/work/linux/arch/riscv/include/asm/current.h:31
#2 0x0000000000000000 in ?? ()
(gdb) p/x $stval
$1 = 0xffffffc8946cbc38
(gdb) display /i $sepc
1: x/i $sepc
0xffffffff8000392c <handle_exception+44>: sd ra,8(sp)
(gdb) p/x $sp
$6 = 0xffffffc8946cbc30
A workaround can be to disable CONFIG_VMAP_STACK but actually any
mapping in the kernel page table that is not also added to the EFI
page table can cause such crash: but I'm not sure this can happen and
this workaround was enough in the previous release to fix this bug and
to be able to live install a system.
I advise to implement this workaround for now as I'm implementing the
proper patch which consists in updating the EFI page table before
switching to it, just like x86 does.
[ Test Plan ]
Use a kernel with CONFIG_VMAP_STACK disabled and simply enter the
following command, preferably on the Unmatched:
ubuntu@ubuntu:~$ cat /sys/firmware/efi/efivars/*
This should not trigger a crash.
[ Where problems could occur ]
As said above, just disabling CONFIG_VMAP_STACK may not be enough.
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-riscv/+bug/1992458/+subscriptions
--
Mailing list: https://launchpad.net/~kernel-packages
Post to : kernel-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~kernel-packages
More help : https://help.launchpad.net/ListHelp
