Hi all,
While comparing register values read from debugfs files under
/sys/kernel/debug/ccree/, I noticed some oddities.
Apparently there is no guarantee these registers are read from the
device while it is resumed. This may lead to bogus values, or crashes
and lock-ups.
This patch series:
1. Allows debugfs_create_regset32() to be used for devices whose
registers must be accessed when resumed,
2. Fixes the CCREE driver to make use of this.
I have identified several other drivers that may be affected (i.e.
using debugfs_create_regset32() and pm_runtime_*()):
- drivers/gpu/drm/msm/disp/dpu1
- drivers/usb/dwc3
- drivers/usb/host/ehci-omap.c
- drivers/usb/host/ehci-tegra.c
- drivers/usb/host/ohci-platform.c
- drivers/usb/host/xhci.c
- drivers/usb/host/xhci-dbgcap.c
- drivers/usb/host/xhci-histb.c
- drivers/usb/host/xhci-hub.c
- drivers/usb/host/xhci-mtk.c
- drivers/usb/host/xhci-pci.c
- drivers/usb/host/xhci-plat.c
- drivers/usb/host/xhci-tegra.c
- drivers/usb/mtu3
- drivers/usb/musb
Some of these call pm_runtime_forbid(), but given the comment "users
should enable runtime pm using power/control in sysfs", this can be
overridden from userspace, so these are unsafe, too?
Thanks for your comments!
Geert Uytterhoeven (2):
debugfs: regset32: Add Runtime PM support
crypto: ccree - fix debugfs register access while suspended
drivers/crypto/ccree/cc_debugfs.c | 2 ++
fs/debugfs/file.c | 8 ++++++++
include/linux/debugfs.h | 1 +
3 files changed, 11 insertions(+)
--
2.17.1
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
