linux-toradex.git/drivers/clocksource/sh_cmt.c, branch v6.12

clocksource/drivers/sh_cmt: Address race condition for clock events

2024-07-12T14:07:05+00:00

There is a race condition in the CMT interrupt handler. In the interrupt
handler the driver sets a driver private flag, FLAG_IRQCONTEXT. This
flag is used to indicate any call to set_next_event() should not be
directly propagated to the device, but instead cached. This is done as
the interrupt handler itself reprograms the device when needed before it
completes and this avoids this operation to take place twice.

It is unclear why this design was chosen, my suspicion is to allow the
struct clock_event_device.event_handler callback, which is called while
the FLAG_IRQCONTEXT is set, can update the next event without having to
write to the device twice.

Unfortunately there is a race between when the FLAG_IRQCONTEXT flag is
set and later cleared where the interrupt handler have already started to
write the next event to the device. If set_next_event() is called in
this window the value is only cached in the driver but not written. This
leads to the board to misbehave, or worse lockup and produce a splat.

   rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
   rcu:     0-...!: (0 ticks this GP) idle=f5e0/0/0x0 softirq=519/519 fqs=0 (false positive?)
   rcu:     (detected by 1, t=6502 jiffies, g=-595, q=77 ncpus=2)
   Sending NMI from CPU 1 to CPUs 0:
   NMI backtrace for cpu 0
   CPU: 0 PID: 0 Comm: swapper/0 Not tainted 6.10.0-rc5-arm64-renesas-00019-g74a6f86eaf1c-dirty #20
   Hardware name: Renesas Salvator-X 2nd version board based on r8a77965 (DT)
   pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
   pc : tick_check_broadcast_expired+0xc/0x40
   lr : cpu_idle_poll.isra.0+0x8c/0x168
   sp : ffff800081c63d70
   x29: ffff800081c63d70 x28: 00000000580000c8 x27: 00000000bfee5610
   x26: 0000000000000027 x25: 0000000000000000 x24: 0000000000000000
   x23: ffff00007fbb9100 x22: ffff8000818f1008 x21: ffff8000800ef07c
   x20: ffff800081c79ec0 x19: ffff800081c70c28 x18: 0000000000000000
   x17: 0000000000000000 x16: 0000000000000000 x15: 0000ffffc2c717d8
   x14: 0000000000000000 x13: ffff000009c18080 x12: ffff8000825f7fc0
   x11: 0000000000000000 x10: ffff8000818f3cd4 x9 : 0000000000000028
   x8 : ffff800081c79ec0 x7 : ffff800081c73000 x6 : 0000000000000000
   x5 : 0000000000000000 x4 : ffff7ffffe286000 x3 : 0000000000000000
   x2 : ffff7ffffe286000 x1 : ffff800082972900 x0 : ffff8000818f1008
   Call trace:
    tick_check_broadcast_expired+0xc/0x40
    do_idle+0x9c/0x280
    cpu_startup_entry+0x34/0x40
    kernel_init+0x0/0x11c
    do_one_initcall+0x0/0x260
    __primary_switched+0x80/0x88
   rcu: rcu_preempt kthread timer wakeup didn't happen for 6501 jiffies! g-595 f0x0 RCU_GP_WAIT_FQS(5) ->state=0x402
   rcu:     Possible timer handling issue on cpu=0 timer-softirq=262
   rcu: rcu_preempt kthread starved for 6502 jiffies! g-595 f0x0 RCU_GP_WAIT_FQS(5) ->state=0x402 ->cpu=0
   rcu:     Unless rcu_preempt kthread gets sufficient CPU time, OOM is now expected behavior.
   rcu: RCU grace-period kthread stack dump:
   task:rcu_preempt     state:I stack:0     pid:15    tgid:15    ppid:2      flags:0x00000008
   Call trace:
    __switch_to+0xbc/0x100
    __schedule+0x358/0xbe0
    schedule+0x48/0x148
    schedule_timeout+0xc4/0x138
    rcu_gp_fqs_loop+0x12c/0x764
    rcu_gp_kthread+0x208/0x298
    kthread+0x10c/0x110
    ret_from_fork+0x10/0x20

The design have been part of the driver since it was first merged in
early 2009. It becomes increasingly harder to trigger the issue the
older kernel version one tries. It only takes a few boots on v6.10-rc5,
while hundreds of boots are needed to trigger it on v5.10.

Close the race condition by using the CMT channel lock for the two
competing sections. The channel lock was added to the driver after its
initial design.

Signed-off-by: Niklas Söderlund 
Link: https://lore.kernel.org/r/20240702190230.3825292-1-niklas.soderlund+renesas@ragnatech.se
Signed-off-by: Daniel Lezcano

clocksource: Explicitly include correct DT includes

2023-08-28T18:30:57+00:00

The DT of_device.h and of_platform.h date back to the separate
of_platform_bus_type before it was merged into the regular platform bus.
As part of that merge prepping Arm DT support 13 years ago, they
"temporarily" include each other. They also include platform_device.h
and of.h. As a result, there's a pretty much random mix of those include
files used throughout the tree. In order to detangle these headers and
replace the implicit includes with struct declarations, users need to
explicitly include the correct includes.

Link: https://lore.kernel.org/r/20230714174409.4053843-1-robh@kernel.org
Signed-off-by: Rob Herring

clocksource: remove MODULE_LICENSE in non-modules

2023-04-13T20:13:51+00:00

Since commit 8b41fc4454e ("kbuild: create modules.builtin without
Makefile.modbuiltin or tristate.conf"), MODULE_LICENSE declarations
are used to identify modules. As a consequence, uses of the macro
in non-modules will cause modprobe to misidentify their containing
object file as a module when it is not (false positives), and modprobe
might succeed rather than failing with a suitable error message.

So remove it in the files in this commit, none of which can be built as
modules.

Signed-off-by: Nick Alcock 
Suggested-by: Luis Chamberlain 
Cc: Luis Chamberlain 
Cc: linux-modules@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: Hitomi Hasegawa 
Cc: Daniel Lezcano 
Cc: Thomas Gleixner 
Signed-off-by: Luis Chamberlain

clocksource/drivers/sh_cmt: Mark driver as non-removable

2023-02-13T12:10:17+00:00

The comment in the remove callback suggests that the driver is not
supposed to be unbound. However returning an error code in the remove
callback doesn't accomplish that. Instead set the suppress_bind_attrs
property (which makes it impossible to unbind the driver via sysfs).
The only remaining way to unbind a sh_cmt device would be module
unloading, but that doesn't apply here, as the driver cannot be built as
a module.

Also drop the useless remove callback.

Signed-off-by: Uwe Kleine-König 
Link: https://lore.kernel.org/r/20230123220221.48164-1-u.kleine-koenig@pengutronix.de
Signed-off-by: Daniel Lezcano

clocksource/drivers/sh_cmt: Access registers according to spec

2022-12-01T10:56:25+00:00

Documentation for most CMTs say that it takes two input clocks before
changes propagate to the timer. This is especially relevant when the timer
is stopped to change further settings.

Implement the delays according to the spec. To avoid unnecessary delays in
atomic mode, also check if the to-be-written value actually differs.

CMCNT is a bit special because testing showed that it requires 3 cycles to
propagate, which affects all CMTs. Also, the WRFLAG needs to be checked
before writing. This fixes "cannot clear CMCNT" messages which occur often
on R-Car Gen4 SoCs, but only very rarely on older SoCs for some reason.

Fixes: 81b3b2711072 ("clocksource: sh_cmt: Add support for multiple channels per device")
Signed-off-by: Wolfram Sang 
Signed-off-by: Thomas Gleixner 
Link: https://lore.kernel.org/r/20221130210609.7718-1-wsa+renesas@sang-engineering.com

clocksource/drivers/sh_cmt: Add R-Car Gen4 support

2022-07-18T20:55:14+00:00

Add support for the R-Car Gen4 CMT types 0/1 which are the same as in
the previous two generations.

Signed-off-by: Wolfram Sang 
Link: https://lore.kernel.org/r/20220713100603.3391-4-wsa+renesas@sang-engineering.com
Signed-off-by: Daniel Lezcano

clocksource/drivers/sh_cmt: Fix wrong setting if don't request IRQ for clock source channel

2021-08-13T18:26:29+00:00

If CMT instance has at least two channels, one channel will be used
as a clock source and another one used as a clock event device.
In that case, IRQ is not requested for clock source channel so
sh_cmt_clock_event_program_verify() might work incorrectly.
Besides, when a channel is only used for clock source, don't need to
re-set the next match_value since it should be maximum timeout as
it still is.

On the other hand, due to no IRQ, total_cycles is not counted up
when reaches compare match time (timer counter resets to zero),
so sh_cmt_clocksource_read() returns unexpected value.
Therefore, use 64-bit clocksoure's mask for 32-bit or 16-bit variants
will also lead to wrong delta calculation. Hence, this mask should
correspond to timer counter width, and above function just returns
the raw value of timer counter register.

Fixes: bfa76bb12f23 ("clocksource: sh_cmt: Request IRQ for clock event device only")
Fixes: 37e7742c55ba ("clocksource/drivers/sh_cmt: Fix clocksource width for 32-bit machines")
Signed-off-by: Phong Hoang 
Signed-off-by: Niklas Söderlund 
Signed-off-by: Daniel Lezcano 
Link: https://lore.kernel.org/r/20210422123443.73334-1-niklas.soderlund+renesas@ragnatech.se

clocksource/drivers/sh_cmt: Don't use CMTOUT_IE with R-Car Gen2/3

2021-04-08T11:24:16+00:00

CMTOUT_IE is only supported for older SoCs. Newer SoCs shall not set
this bit. So, add a version check.

Reported-by: Phong Hoang 
Reviewed-by: Geert Uytterhoeven 
Reviewed-by: Niklas Söderlund 
Signed-off-by: Wolfram Sang 
Signed-off-by: Daniel Lezcano 
Link: https://lore.kernel.org/r/20210309094448.31823-1-wsa+renesas@sang-engineering.com

clocksource/drivers/sh_cmt: Make sure channel clock supply is enabled

2021-01-18T15:20:37+00:00

The Renesas Compare Match Timer 0 and 1 (CMT0/1) variants have a
register to control the clock supply to the individual channels.
Currently the driver does not touch this register, and relies on the
documented initial value, which has the clock supply enabled for all
channels present.

However, when Linux starts on the APE6-EVM development board, only the
clock supply to the first CMT1 channel is enabled.  Hence the first
channel (used as a clockevent) works, while the second channel (used as
a clocksource) does not.  Note that the default system clocksource is
the Cortex-A15 architectured timer, and the user needs to manually
switch to the CMT1 clocksource to trigger the broken behavior.

Fix this by removing the fragile dependency on implicit reset and/or
boot loader state, and by enabling the clock supply explicitly for all
channels used instead.  This requires postponing the clk_disable() call,
else the timer's registers cannot be accessed in sh_cmt_setup_channel().

Signed-off-by: Geert Uytterhoeven 
Reviewed-by: Niklas Söderlund 
Signed-off-by: Daniel Lezcano 
Link: https://lore.kernel.org/r/20201210194648.2901899-1-geert+renesas@glider.be

Merge tag 'pm-5.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

2020-12-16T00:30:31+00:00

Pull power management updates from Rafael Wysocki:
 "These update cpufreq (core and drivers), cpuidle (polling state
  implementation and the PSCI driver), the OPP (operating performance
  points) framework, devfreq (core and drivers), the power capping RAPL
  (Running Average Power Limit) driver, the Energy Model support, the
  generic power domains (genpd) framework, the ACPI device power
  management, the core system-wide suspend code and power management
  utilities.

  Specifics:

   - Use local_clock() instead of jiffies in the cpufreq statistics to
     improve accuracy (Viresh Kumar).

   - Fix up OPP usage in the cpufreq-dt and qcom-cpufreq-nvmem cpufreq
     drivers (Viresh Kumar).

   - Clean up the cpufreq core, the intel_pstate driver and the
     schedutil cpufreq governor (Rafael Wysocki).

   - Fix up error code paths in the sti-cpufreq and mediatek cpufreq
     drivers (Yangtao Li, Qinglang Miao).

   - Fix cpufreq_online() to return error codes instead of success (0)
     in all cases when it fails (Wang ShaoBo).

   - Add mt8167 support to the mediatek cpufreq driver and blacklist
     mt8516 in the cpufreq-dt-platdev driver (Fabien Parent).

   - Modify the tegra194 cpufreq driver to always return values from the
     frequency table as the current frequency and clean up that driver
     (Sumit Gupta, Jon Hunter).

   - Modify the arm_scmi cpufreq driver to allow it to discover the
     power scale present in the performance protocol and provide this
     information to the Energy Model (Lukasz Luba).

   - Add missing MODULE_DEVICE_TABLE to several cpufreq drivers (Pali
     Rohár).

   - Clean up the CPPC cpufreq driver (Ionela Voinescu).

   - Fix NVMEM_IMX_OCOTP dependency in the imx cpufreq driver (Arnd
     Bergmann).

   - Rework the poling interval selection for the polling state in
     cpuidle (Mel Gorman).

   - Enable suspend-to-idle for PSCI OSI mode in the PSCI cpuidle driver
     (Ulf Hansson).

   - Modify the OPP framework to support empty (node-less) OPP tables in
     DT for passing dependency information (Nicola Mazzucato).

   - Fix potential lockdep issue in the OPP core and clean up the OPP
     core (Viresh Kumar).

   - Modify dev_pm_opp_put_regulators() to accept a NULL argument and
     update its users accordingly (Viresh Kumar).

   - Add frequency changes tracepoint to devfreq (Matthias Kaehlcke).

   - Add support for governor feature flags to devfreq, make devfreq
     sysfs file permissions depend on the governor and clean up the
     devfreq core (Chanwoo Choi).

   - Clean up the tegra20 devfreq driver and deprecate it to allow
     another driver based on EMC_STAT to be used instead of it (Dmitry
     Osipenko).

   - Add interconnect support to the tegra30 devfreq driver, allow it to
     take the interconnect and OPP information from DT and clean it up
     (Dmitry Osipenko).

   - Add interconnect support to the exynos-bus devfreq driver along
     with interconnect properties documentation (Sylwester Nawrocki).

   - Add suport for AMD Fam17h and Fam19h processors to the RAPL power
     capping driver (Victor Ding, Kim Phillips).

   - Fix handling of overly long constraint names in the powercap
     framework (Lukasz Luba).

   - Fix the wakeup configuration handling for bridges in the ACPI
     device power management core (Rafael Wysocki).

   - Add support for using an abstract scale for power units in the
     Energy Model (EM) and document it (Lukasz Luba).

   - Add em_cpu_energy() micro-optimization to the EM (Pavankumar
     Kondeti).

   - Modify the generic power domains (genpd) framwework to support
     suspend-to-idle (Ulf Hansson).

   - Fix creation of debugfs nodes in genpd (Thierry Strudel).

   - Clean up genpd (Lina Iyer).

   - Clean up the core system-wide suspend code and make it print driver
     flags for devices with debug enabled (Alex Shi, Patrice Chotard,
     Chen Yu).

   - Modify the ACPI system reboot code to make it prepare for system
     power off to avoid confusing the platform firmware (Kai-Heng Feng).

   - Update the pm-graph (multiple changes, mostly usability-related)
     and cpupower (online and offline CPU information support) PM
     utilities (Todd Brandt, Brahadambal Srinivasan)"

* tag 'pm-5.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (86 commits)
  cpufreq: Fix cpufreq_online() return value on errors
  cpufreq: Fix up several kerneldoc comments
  cpufreq: stats: Use local_clock() instead of jiffies
  cpufreq: schedutil: Simplify sugov_update_next_freq()
  cpufreq: intel_pstate: Simplify intel_cpufreq_update_pstate()
  PM: domains: create debugfs nodes when adding power domains
  opp: of: Allow empty opp-table with opp-shared
  dt-bindings: opp: Allow empty OPP tables
  media: venus: dev_pm_opp_put_*() accepts NULL argument
  drm/panfrost: dev_pm_opp_put_*() accepts NULL argument
  drm/lima: dev_pm_opp_put_*() accepts NULL argument
  PM / devfreq: exynos: dev_pm_opp_put_*() accepts NULL argument
  cpufreq: qcom-cpufreq-nvmem: dev_pm_opp_put_*() accepts NULL argument
  cpufreq: dt: dev_pm_opp_put_regulators() accepts NULL argument
  opp: Allow dev_pm_opp_put_*() APIs to accept NULL opp_table
  opp: Don't create an OPP table from dev_pm_opp_get_opp_table()
  cpufreq: dt: Don't (ab)use dev_pm_opp_get_opp_table() to create OPP table
  opp: Reduce the size of critical section in _opp_kref_release()
  PM / EM: Micro optimization in em_cpu_energy
  cpufreq: arm_scmi: Discover the power scale in performance protocol
  ...