regmap: clean up and improve __regmap_init()
Get series with b4
b4 am [email protected] change-id 20261109-regmap-improve-915eee5a80ecVersions newest first
- v2 Latest10 patches 2 repliesIn review
- v19 patches 1 replySuperseded
Changes in v2 from the cover letter
- Rebased onto next-20261112.
- New patch 1: regmap: fix reinit cache failure handling
- Patch 3: regmap-i2c: fix write limit clamping for adapter quirks Correct fixes tag
Rename the patch subject.
Fix write limit clamping.
- Patch 5: regmap: factor out regmap_set_lock():
Move ret = -EINVAL below the successful regmap_set_lock() call so it cannot be clobbered with 0 on success.
Patches in v2 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/10 | regmap: fix reinit cache failure handlingdrivers/ drivers/ | +15 | 0 replies | |
| 2/10 | regmap-i2c: keep bus raw write limit when applying read quirkdrivers/ | +8 | 0 replies | |
| 3/10 | regmap-i2c: fix write limit clamping for adapter quirksdrivers/ | +17 | 0 replies | |
| 4/10 | regmap: deduplicate fast_io checksdrivers/ | +5 | 0 replies | |
| 5/10 | regmap: factor out regmap_set_lock()drivers/ | +79 | 0 replies | |
| 6/10 | regmap: factor out regmap_init_ranges()drivers/ | +101 | 0 replies | |
| 7/10 | regmap: factor out format initialisation into helpersdrivers/ | +200 | 0 replies | |
| 8/10 | regmap: drop regmap_bus::free_on_exitdrivers/ drivers/ drivers/ +1 more | +75 | 0 replies | |
| 9/10 | regmap: use cleanup API in __regmap_init()drivers/ | +33 | 0 replies | |
| 10/10 | regmap: kunit: add test for regmap_bus::init_limitsdrivers/ | +89 | 0 replies |
Reviews trailers per patch
No Reviewed-by, Acked-by or Tested-by trailers for v2 yet.
Cover letter [PATCH v2 00/10] regmap: clean up and improve __regmap_init()
This series includes three fixes for pre-existing regmap issues noticed during review, improves the __regmap_init() error handling and removes the last user of regmap_bus::free_on_exit. Patches 1-3 are fixes suitable for stable; patches 4-9 are cleanups with no functional change intended; patch 10 adds KUnit coverage for the new bus callback.
Patch 1 fixes the cache reinitialisation failure paths. regcache_exit() freed reg_defaults_raw without clearing cache_free, and a failed regmap_reinit_cache() left map->cache_type set, so a later regmap_exit() could free the raw defaults a second time. It also drops the old cache_ops after tearing down the cache, so a successful reinitialisation to REGCACHE_NONE cannot leave stale cache ops behind. Clear the ownership state and leave the map without a cache if reinitialisation fails.
Patch 2 fixes regmap_get_i2c_bus() so that applying an adapter read quirk no longer drops the raw write limit advertised by the selected bus (and vice versa). It currently starts from zero-initialised locals and overwrites both limits of the bus copy.
Patch 3 fixes the raw write limit calculation when the adapter's max_write_len quirk is applied. It compares the bus limit against the quirk payload size rather than the whole-message size, avoids the underflow when the register address alone exceeds the quirk limit, and keeps the existing limit when the quirk cannot fit a single value, which would otherwise lead to a division by zero in the raw write chunking. The same minimum-size guard is applied to the read quirk.
Patches 4-7 are mechanical: evaluate fast_io once, and factor out regmap_set_lock(), regmap_init_ranges() and the format selection into helpers. The forward goto and the label-based unwinding are reduced to a single error path.
Patch 8 drops regmap_bus::free_on_exit. Instead of kmemdup()ing the static I2C/SPI bus to store per-device raw I/O limits, add an optional regmap_bus::init_limits() callback. __regmap_init() invokes it before regcache_init(), which is required because regcache_hw_init() may read raw register defaults from the device during initialization and must already see the effective max_raw_read limit. The static bus structures can then be used directly and the per-device allocations disappear.
Patch 9 converts the error unwinding to the cleanup API and drops all err_* labels. map->cache_type is now assigned just before regcache_init() and reset to REGCACHE_NONE on failure, so the release helper can safely call regcache_exit() on a partially initialised map.
Patch 10 adds a KUnit test which verifies that init_limits() is invoked before regcache_init() by checking that the raw default readback is split according to the limit installed by the callback.
Testing:
- Built drivers/base/regmap/ with W=1 on x86_64.
- checkpatch.pl clean for all patches.
- regmap KUnit suite pass.
---
Changes in v2:
- Rebased onto next-20261112.
- New patch 1: regmap: fix reinit cache failure handling
- Patch 3: regmap-i2c: fix write limit clamping for adapter quirks Correct fixes tag
Rename the patch subject.
Fix write limit clamping.
- Patch 5: regmap: factor out regmap_set_lock():
Move ret = -EINVAL below the successful regmap_set_lock() call so it cannot be clobbered with 0 on success.
---
Peng Fan (10):
regmap: fix reinit cache failure handling
regmap-i2c: keep bus raw write limit when applying read quirk
regmap-i2c: fix write limit clamping for adapter quirks
regmap: deduplicate fast_io checks
regmap: factor out regmap_set_lock()
regmap: factor out regmap_init_ranges()
regmap: factor out format initialisation into helpers
regmap: drop regmap_bus::free_on_exit
regmap: use cleanup API in __regmap_init()
regmap: kunit: add test for regmap_bus::init_limits
drivers/base/regmap/regcache.c | 8 +- drivers/base/regmap/regmap-i2c.c | 68 ++-- drivers/base/regmap/regmap-kunit.c | 89 +++++ drivers/base/regmap/regmap-spi.c | 64 ++-- drivers/base/regmap/regmap.c | 714 +++++++++++++++++++------------------ include/linux/regmap.h | 7 +- 6 files changed, 541 insertions(+), 409 deletions(-)
---
base-commit: 6d582d78fad9d828e575ca0cdcff8ea47a77220b
change-id: 20261109-regmap-improve-915eee5a80ec
Best regards,
Signature · 1 line
Peng Fan <[email protected]>