You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
afs: fix tracepoint string placement with built-in AFS
I was adding custom tracepoint to the kernel, grabbed full F34 kernel
.config, disabled modules and booted whole shebang as VM kernel.
Then did
perf record -a -e ...
It crashed:
general protection fault, probably for non-canonical address 0x435f5346592e4243: 0000 [#1] SMP PTI
CPU: 1 PID: 842 Comm: cat Not tainted 5.12.6+ #26
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-1.fc33 04/01/2014
RIP: 0010:t_show+0x22/0xd0
Then reproducer was narrowed to
# cat /sys/kernel/tracing/printk_formats
Original F34 kernel with modules didn't crash.
So I started to disable options and after disabling AFS everything started
working again.
The root cause is that AFS was placing char arrays content into a section
full of _pointers_ to strings with predictable consequences.
Non canonical address 435f5346592e4243 is "CB.YFS_" which came from
CM_NAME macro.
The fix is to create char array and pointer to it separatedly.
Steps to reproduce:
CONFIG_AFS=y
CONFIG_TRACING=y
# cat /sys/kernel/tracing/printk_formats
Link: https://lkml.kernel.org/r/YLAXfvZ+rObEOdc/@localhost.localdomain
Fixes: 8e8d7f1 ("afs: Add some tracepoints")
Signed-off-by: Alexey Dobriyan (SK hynix) <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: David Howells <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Stephen Rothwell <[email protected]>
0 commit comments