From: Zhipeng Wang <zhipeng.wang_1@nxp.com>
This series makes the i.MX IRQSTEER driver buildable as a module.
v3 was a four-patch series: three fixes for pre-existing bugs that only
become reachable once the driver can be unbound/reloaded, followed by
the module conversion. On v3, Frank Li reviewed the unload-path fixes
and asked for three changes, all addressed here:
- The probe() error-path irq_domain leak was fixed by hand-rolling an
err_domain label. Frank suggested a devres-managed domain instead,
and to add a reusable devm_irq_domain_create_linear() helper so other
drivers benefit. Patch 1/5 adds that helper (the devres sibling of
irq_domain_create_linear()); patch 2/5 switches imx-irqsteer to it,
which fixes the leak structurally and lets remove() drop its explicit
irq_domain_remove().
- Frank questioned why remove() disposed of the mappings "twice". The
two loops disposed of two different sets: the parent output mappings
the driver created with irq_of_parse_and_map(), and the child input
mappings the domain handed out. The child mappings are freed by their
consumers and, with the domain now owned by devres, are torn down
automatically; only the parent mappings need explicit disposal. Patch
3/5 disposes of the parent mappings only, dropping the child loop.
- Frank pointed out that masking in remove() touches CHANMASK while the
device may be runtime-suspended with the clock gated. The remove()
masking is unnecessary: the next probe() quiesces the hardware before
it re-maps and unmasks the parent interrupts, which is the only window
in which a stale line could storm. Patch 4/5 therefore masks only in
probe().
Patch 5/5 then converts the driver to a module. As in v3, it lets devres
own the clock and runtime PM (devm_clk_get_enabled() +
devm_pm_runtime_set_active_enabled()).
Changes in v4:
- New patch 1/5: add devm_irq_domain_create_linear() (Frank Li).
- Patch 2/5 (was "Fix error handling path in probe()"): use the new
devm helper to manage the domain instead of hand-rolling an
err_domain label; remove() drops irq_domain_remove() (Frank Li).
- Patch 3/5 (was 2/4): dispose of the parent mappings only; the child
mapping disposal loop is removed (Frank Li).
- Patch 4/5 (was 3/4): mask interrupts in probe() only; the remove()
masking, which could touch registers while runtime-suspended, is
dropped (Frank Li).
- Patch 5/5 (was 4/4): no functional change; remove() and the probe()
error path now only dispose of the parent mappings.
v3: https://lore.kernel.org/r/20260807072346.1222389-1-Zhipeng.wang_1@oss.nxp.com
v2: https://lore.kernel.org/r/20260728092219.525449-1-Zhipeng.wang_1@oss.nxp.com
v1: https://lore.kernel.org/r/20260724090136.3595894-1-Zhipeng.wang_1@oss.nxp.com
Jindong Yue (1):
irqchip/imx-irqsteer: Allow building as module
Zhipeng Wang (4):
genirq/irqdomain: Add devm_irq_domain_create_linear()
irqchip/imx-irqsteer: Use devm to manage the IRQ domain
irqchip/imx-irqsteer: Dispose of parent IRQ mappings in remove()
irqchip/imx-irqsteer: Mask all interrupts in probe()
drivers/irqchip/Kconfig | 2 +-
drivers/irqchip/irq-imx-irqsteer.c | 52 +++++++++++++++++++-----------
include/linux/irqdomain.h | 30 +++++++++++++++++
3 files changed, 64 insertions(+), 20 deletions(-)
base-commit: bd5f485f3f026225b86573e559af0b7254ef4184
--
2.34.1