On Fri, Sep 25, 2020 at 05:47:14PM -0600, Shuah Khan wrote:
> This patch series is a result of discussion at the refcount_t BOF
> the Linux Plumbers Conference. In this discussion, we identified
> a need for looking closely and investigating atomic_t usages in
> the kernel when it is used strictly as a counter without it
> controlling object lifetimes and state changes.
>
> There are a number of atomic_t usages in the kernel where atomic_t api
> is used strictly for counting and not for managing object lifetime. In
> some cases, atomic_t might not even be needed.
>
> The purpose of these counters is twofold: 1. clearly differentiate
> atomic_t counters from atomic_t usages that guard object lifetimes,
> hence prone to overflow and underflow errors. It allows tools that scan
> for underflow and overflow on atomic_t usages to detect overflow and
> underflows to scan just the cases that are prone to errors. 2. provides
> non-atomic counters for cases where atomic isn't necessary.
Nice series :)
It appears there is no user of counter_simple in this series other than the
selftest. Would you be planning to add any conversions in the series itself,
for illustration of use? Sorry if I missed a usage.
Also how do we guard against atomicity of counter_simple RMW operations? Is
the implication that it should be guarded using other synchronization to
prevent lost-update problem?
Some more comments:
1. atomic RMW operations that have a return value are fully ordered. Would
you be adding support to counter_simple for such ordering as well, for
consistency?
2. I felt counter_atomic and counter_atomic64 would be nice equivalents to
the atomic and atomic64 naming currently used (i.e. dropping the '32').
However that is just my opinion and I am ok with either naming.
thanks!
- Joel
>
> Simple atomic and non-atomic counters api provides interfaces for simple
> atomic and non-atomic counters that just count, and don't guard resource
> lifetimes. Counters will wrap around to 0 when it overflows and should
> not be used to guard resource lifetimes, device usage and open counts
> that control state changes, and pm states.
>
> Using counter_atomic to guard lifetimes could lead to use-after free
> when it overflows and undefined behavior when used to manage state
> changes and device usage/open states.
>
> This patch series introduces Simple atomic and non-atomic counters.
> Counter atomic ops leverage atomic_t and provide a sub-set of atomic_t
> ops.
>
> In addition this patch series converts a few drivers to use the new api.
> The following criteria is used for select variables for conversion:
>
> 1. Variable doesn't guard object lifetimes, manage state changes e.g:
> device usage counts, device open counts, and pm states.
> 2. Variable is used for stats and counters.
> 3. The conversion doesn't change the overflow behavior.
>
> Changes since RFC:
> -- Thanks for reviews and reviewed-by, and Acked-by tags. Updated
> the patches with the tags.
> -- Addressed Kees's comments:
> 1. Non-atomic counters renamed to counter_simple32 and counter_simple64
> to clearly indicate size.
> 2. Added warning for counter_simple* usage and it should be used only
> when there is no need for atomicity.
> 3. Renamed counter_atomic to counter_atomic32 to clearly indicate size.
> 4. Renamed counter_atomic_long to counter_atomic64 and it now uses
> atomic64_t ops and indicates size.
> 5. Test updated for the API renames.
> 6. Added helper functions for test results printing
> 7. Verified that the test module compiles in kunit env. and test
> module can be loaded to run the test.
> 8. Updated Documentation to reflect the intent to make the API
> restricted so it can never be used to guard object lifetimes
> and state management. I left _return ops for now, inc_return
> is necessary for now as per the discussion we had on this topic.
> -- Updated driver patches with API name changes.
> -- We discussed if binder counters can be non-atomic. For now I left
> them the same as the RFC patch - using counter_atomic32
> -- Unrelated to this patch series:
> The patch series review uncovered improvements could be made to
> test_async_driver_probe and vmw_vmci/vmci_guest. I will track
> these for fixing later.
>
> Shuah Khan (11):
> counters: Introduce counter_simple* and counter_atomic* counters
> selftests:lib:test_counters: add new test for counters
> drivers/base: convert deferred_trigger_count and probe_count to
> counter_atomic32
> drivers/base/devcoredump: convert devcd_count to counter_atomic32
> drivers/acpi: convert seqno counter_atomic32
> drivers/acpi/apei: convert seqno counter_atomic32
> drivers/android/binder: convert stats, transaction_log to
> counter_atomic32
> drivers/base/test/test_async_driver_probe: convert to use
> counter_atomic32
> drivers/char/ipmi: convert stats to use counter_atomic32
> drivers/misc/vmw_vmci: convert num guest devices counter to
> counter_atomic32
> drivers/edac: convert pci counters to counter_atomic32
>
> Documentation/core-api/counters.rst | 174 +++++++++
> MAINTAINERS | 8 +
> drivers/acpi/acpi_extlog.c | 5 +-
> drivers/acpi/apei/ghes.c | 5 +-
> drivers/android/binder.c | 41 +--
> drivers/android/binder_internal.h | 3 +-
> drivers/base/dd.c | 19 +-
> drivers/base/devcoredump.c | 5 +-
> drivers/base/test/test_async_driver_probe.c | 23 +-
> drivers/char/ipmi/ipmi_msghandler.c | 9 +-
> drivers/char/ipmi/ipmi_si_intf.c | 9 +-
> drivers/edac/edac_pci.h | 5 +-
> drivers/edac/edac_pci_sysfs.c | 28 +-
> drivers/misc/vmw_vmci/vmci_guest.c | 9 +-
> include/linux/counters.h | 350 +++++++++++++++++++
> lib/Kconfig | 10 +
> lib/Makefile | 1 +
> lib/test_counters.c | 276 +++++++++++++++
> tools/testing/selftests/lib/Makefile | 1 +
> tools/testing/selftests/lib/config | 1 +
> tools/testing/selftests/lib/test_counters.sh | 5 +
> 21 files changed, 913 insertions(+), 74 deletions(-)
> create mode 100644 Documentation/core-api/counters.rst
> create mode 100644 include/linux/counters.h
> create mode 100644 lib/test_counters.c
> create mode 100755 tools/testing/selftests/lib/test_counters.sh
>
> --
> 2.25.1
>
