[REQUEST] Backport 8250_dw BUSY deassert series to 6.12.y stable

[REQUEST] Backport 8250_dw BUSY deassert series to 6.12.y stable

[Ionut Nechita (Wind River)]

From: Ionut Nechita <ionut.nechita@windriver.com>

Hi Ilpo, Greg,

We're running kernel 6.12 (LTS) on production systems with DesignWare
8250 UARTs and are hitting the BUSY assertion issue that your recent
series addresses in 6.18:

  Ilpo Järvinen (7):
    serial: 8250: Protect LCR write in shutdown
    serial: 8250_dw: Avoid unnecessary LCR writes
    serial: 8250: Add serial8250_handle_irq_locked()
    serial: 8250_dw: Rework dw8250_handle_irq() locking and IIR handling
    serial: 8250_dw: Rework IIR_NO_INT handling to stop interrupt storm
    serial: 8250: Add late synchronize_irq() to shutdown to handle DW UART BUSY
    serial: 8250_dw: Ensure BUSY is deasserted

Patch 7/7 has Cc: stable, but it depends on patches 1-6 for the new
infrastructure (serial8250_handle_irq_locked(), the reworked locking in
dw8250_handle_irq(), etc.).

Could the full series be nominated for 6.12.y stable backport?  Or if
that's too invasive for stable, could you advise on the minimal subset
that would allow patch 7 to apply cleanly on 6.12?

We've attempted a standalone backport of patch 7 (squashing the
necessary helpers from 8250_port.c), but the IRQ handler refactoring
in patches 3-5 changes the locking model significantly, and we'd
rather carry the maintainer-blessed version than risk introducing
subtle races.

Our environment:
  - Kernel: 6.12.57-rt/6.12.87-rt (PREEMPT_RT)
  - Hardware: Intel platforms with DW APB UART (snps,dw-apb-uart)
  - Symptom: LCR writes silently ignored under Rx load, causing
    baud rate / framing mismatches after set_termios

We are available to test any backport candidates on our hardware.

Thanks,
Ionut
---

Re: [REQUEST] Backport 8250_dw BUSY deassert series to 6.12.y stable

[Greg KH]

On Fri, May 08, 2026 at 06:16:14PM +0300, Ionut Nechita (Wind River) wrote:
> From: Ionut Nechita <ionut.nechita@windriver.com>
> 
> Hi Ilpo, Greg,
> 
> We're running kernel 6.12 (LTS) on production systems with DesignWare
> 8250 UARTs and are hitting the BUSY assertion issue that your recent
> series addresses in 6.18:
> 
>   Ilpo Järvinen (7):
>     serial: 8250: Protect LCR write in shutdown
>     serial: 8250_dw: Avoid unnecessary LCR writes
>     serial: 8250: Add serial8250_handle_irq_locked()
>     serial: 8250_dw: Rework dw8250_handle_irq() locking and IIR handling
>     serial: 8250_dw: Rework IIR_NO_INT handling to stop interrupt storm
>     serial: 8250: Add late synchronize_irq() to shutdown to handle DW UART BUSY
>     serial: 8250_dw: Ensure BUSY is deasserted
> 
> Patch 7/7 has Cc: stable, but it depends on patches 1-6 for the new
> infrastructure (serial8250_handle_irq_locked(), the reworked locking in
> dw8250_handle_irq(), etc.).
> 
> Could the full series be nominated for 6.12.y stable backport?  Or if
> that's too invasive for stable, could you advise on the minimal subset
> that would allow patch 7 to apply cleanly on 6.12?

Why not just move to 6.18.y instead if you have this issue?

thanks,

greg k-h

Re: [REQUEST] Backport 8250_dw BUSY deassert series to 6.12.y stable

[Ionut Nechita (Wind River)]

From: Ionut Nechita <ionut.nechita@windriver.com>

Hi Greg,

Fair question. We're shipping 6.12 as part of a certified platform
(StarlingX/Yocto) that is already deployed in production at customer
sites. A major kernel version bump requires a full re-qualification
cycle across the entire platform stack (RT latency validation,
out-of-tree driver compat, storage/networking regression testing),
which is a multi-month effort we cannot justify for a single
subsystem fix.

We are planning the move to a newer LTS for our next major release,
but for the current production branch 6.12 is what we're committed to.

Since patch 7/7 already carries Cc: stable, would it be possible to
pull in just the minimal dependencies needed to make it apply on
6.12.y? We're happy to do the backport work ourselves and submit it
for review if that helps — we just wanted to check with Ilpo first
on what the correct minimal subset would be before sending something
that might be wrong.

Thanks,
Ionut
---

Re: [REQUEST] Backport 8250_dw BUSY deassert series to 6.12.y stable

[Ilpo Järvinen]

[-- Attachment #1: Type: text/plain, Size: 2333 bytes --]

On Fri, 8 May 2026, Ionut Nechita (Wind River) wrote:

> From: Ionut Nechita <ionut.nechita@windriver.com>
> 
> Hi Greg,
> 
> Fair question. We're shipping 6.12 as part of a certified platform
> (StarlingX/Yocto) that is already deployed in production at customer
> sites. A major kernel version bump requires a full re-qualification
> cycle across the entire platform stack (RT latency validation,
> out-of-tree driver compat, storage/networking regression testing),
> which is a multi-month effort we cannot justify for a single
> subsystem fix.
> 
> We are planning the move to a newer LTS for our next major release,
> but for the current production branch 6.12 is what we're committed to.
> 
> Since patch 7/7 already carries Cc: stable, would it be possible to
> pull in just the minimal dependencies needed to make it apply on
> 6.12.y? We're happy to do the backport work ourselves and submit it
> for review if that helps — we just wanted to check with Ilpo first
> on what the correct minimal subset would be before sending something
> that might be wrong.

Ah, this didn't get backported that far likely due to some syntax related 
changes causing conflicts with the series.

To backport the entire series my suggestion is to take parts of the 
commits:

dbd26a886e94 ("serial: 8250: use serial_port_in/out() helpers") (8250_dw.c only or just all if it goes cleanly)
b339809edda1 ("serial: 8250: use guard()s") (8250_port.c shutdown hunks)

And this completely (just syntax changes):

bd8cad85561b ("serial: 8250_dw: Comment possible corner cases in serial_out() implementation")

Those above are effectively just syntax changes but result in conflicts if 
not applied prior to the BUSY fix series.

This leaves only minor conflicts in:

a7b9ce39fbe4 ("serial: 8250_dw: Ensure BUSY is deasserted")
with type change in
fc9ceb501e38 ("serial: 8250: sanitize uart_port::serial_{in,out}() types")
(trivial resolution as the change was in the code removed by the BUSY 
change)

73a4ed8f9efa ("serial: 8250_dw: Rework IIR_NO_INT handling to stop interrupt storm")
with probe function changes (handle_irq assignment changes and unrelated 
dw8250_setup_dma_filter() change) in
c213375e3283 ("serial: 8250_dw: Call dw8250_quirks() conditionally")

Hope this helps.

-- 
 i.