To make the interrupt driver operable with SWAP_IO_SPACE
config enabled, replace all instances of readl/writel with
their __raw variants. Otherwise readl/writel will do a byte
swap which this driver does not intend to do.

Tested-by: Carlo Szelinsky <github@szelinsky.de>
Signed-off-by: Rustam Adilov <adilov@disroot.org>
---
 drivers/irqchip/irq-realtek-rtl.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/drivers/irqchip/irq-realtek-rtl.c b/drivers/irqchip/irq-realtek-rtl.c
index 4b59e0cd86bf..3077f4efb4e2 100644
--- a/drivers/irqchip/irq-realtek-rtl.c
+++ b/drivers/irqchip/irq-realtek-rtl.c
@@ -50,18 +50,18 @@ static inline void enable_gimr(unsigned int cpu, unsigned int hw_irq)
 {
 	u32 gimr;
 
-	gimr = readl(REG(cpu, RTL_ICTL_GIMR));
+	gimr = __raw_readl(REG(cpu, RTL_ICTL_GIMR));
 	gimr |= BIT(hw_irq);
-	writel(gimr, REG(cpu, RTL_ICTL_GIMR));
+	__raw_writel(gimr, REG(cpu, RTL_ICTL_GIMR));
 }
 
 static inline void disable_gimr(unsigned int cpu, unsigned int hw_irq)
 {
 	u32 gimr;
 
-	gimr = readl(REG(cpu, RTL_ICTL_GIMR));
+	gimr = __raw_readl(REG(cpu, RTL_ICTL_GIMR));
 	gimr &= ~BIT(hw_irq);
-	writel(gimr, REG(cpu, RTL_ICTL_GIMR));
+	__raw_writel(gimr, REG(cpu, RTL_ICTL_GIMR));
 }
 
 static void write_irr(unsigned int cpu, int hw_irq, u32 value)
@@ -71,9 +71,9 @@ static void write_irr(unsigned int cpu, int hw_irq, u32 value)
 	unsigned int shift = IRR_SHIFT(hw_irq);
 	u32 irr;
 
-	irr = readl(irr0 + offset) & ~(0xf << shift);
+	irr = __raw_readl(irr0 + offset) & ~(0xf << shift);
 	irr |= (value & 0xf) << shift;
-	writel(irr, irr0 + offset);
+	__raw_writel(irr, irr0 + offset);
 }
 
 static void realtek_ictl_unmask_irq(struct irq_data *i)
@@ -159,7 +159,8 @@ static void realtek_irq_dispatch(struct irq_desc *desc)
 	unsigned int hw_irq;
 
 	chained_irq_enter(chip, desc);
-	pending = readl(REG(cpu, RTL_ICTL_GIMR)) & readl(REG(cpu, RTL_ICTL_GISR)) & output->mask;
+	pending = __raw_readl(REG(cpu, RTL_ICTL_GIMR)) &
+		  __raw_readl(REG(cpu, RTL_ICTL_GISR)) & output->mask;
 
 	if (unlikely(!pending)) {
 		spurious_interrupt();
-- 
2.55.0

On Sat, Jul 11 2026 at 13:23, Rustam Adilov wrote:

> To make the interrupt driver operable with SWAP_IO_SPACE
> config enabled, replace all instances of readl/writel with
> their __raw variants. Otherwise readl/writel will do a byte
> swap which this driver does not intend to do.

Sorry, but this word salad does not qualify as a change log. See

  https://docs.kernel.org/process/maintainer-tip.html#changelog

First you want to explain what the context:

  When CONFIG_SWAP_IO_SPACE is enabled readl() and writel() imply a byte
  swap.

Then you want to explain why this is not correct for this
driver/hardware. Just saying 'does not intend to do' contains zero
information:

  This is incorrect for the RTL driver, because <Insert proper technical explanation>

Then you tell how you cure it:

  Fix this by converting the MMIO accesses to __raw_readl() and
  __raw_writel(), which do not byte swap the data.

Thanks,

        tglx

On 2026-07-15 07:10, Thomas Gleixner wrote:
> On Sat, Jul 11 2026 at 13:23, Rustam Adilov wrote:
> 
>> To make the interrupt driver operable with SWAP_IO_SPACE
>> config enabled, replace all instances of readl/writel with
>> their __raw variants. Otherwise readl/writel will do a byte
>> swap which this driver does not intend to do.
> 
> Sorry, but this word salad does not qualify as a change log. See
> 
>   https://docs.kernel.org/process/maintainer-tip.html#changelog
> 
> First you want to explain what the context:
> 
>   When CONFIG_SWAP_IO_SPACE is enabled readl() and writel() imply a byte
>   swap.
> 
> Then you want to explain why this is not correct for this
> driver/hardware. Just saying 'does not intend to do' contains zero
> information:
> 
>   This is incorrect for the RTL driver, because <Insert proper technical explanation>
> 
> Then you tell how you cure it:
> 
>   Fix this by converting the MMIO accesses to __raw_readl() and
>   __raw_writel(), which do not byte swap the data.

Thanks for the review,

Honestly my bad for not thinking a bit more on the commit message.
That does mean i will have to change commit message in another patch
as well as it is a essentially an almost copy of this one...

Either way, would this commit message work?

   When CONFIG_SWAP_IO_SPACE is enabled, readl() and writel() are
   changed to perform a byte swap to little endian type. This is
   incorrect because the devices that use the irq-realtek-rtl driver
   are all big endian MIPS chips.

   Fix this by converting the MMIO accesses to __raw_readl() and
   __raw_writel(), which do not byte swap the data.

I think just saying "big endian MIPS" should be enough info to tell
why byte swapping to "little endian type" (dunno how know else to
describe the __le type) is an issue.

> Thanks,
> 
>         tglx

Best,
Rustam