On Thursday, June 15, 2017 11:46 PM, Krzysztof Kozlowski wrote:
After commit 8b1bd11c1f8f ("pinctrl: samsung: Add the support the
multiple IORESOURCE_MEM for one pin-bank"), the S3C24xx (and probably
S3C64xx as well) fails:

        Unable to handle kernel NULL pointer dereference at virtual address 
000000a8
        ...
        (s3c24xx_demux_eint4_7) from [<c004469c>] 
(__handle_domain_irq+0x6c/0xcc)
        (__handle_domain_irq) from [<c0009444>] (s3c24xx_handle_irq+0x6c/0x12c)
        (s3c24xx_handle_irq) from [<c000e5fc>] (__irq_svc+0x5c/0x78)

Mentioned commit moved the pointer to controller's base IO memory address
from each controller's driver data (samsung_pinctrl_drv_data) to per-bank
structure (samsung_pin_bank).  The external interrupt demux
handlers (s3c24xx_demux_eint()) tried to get this base address from opaque
pointer stored under irq_chip data:

        struct irq_data *irqd = irq_desc_get_irq_data(desc);
        struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(irqd);
        ...
        pend = readl(bank->eint_base + EINTPEND_REG);

which is wrong because this is hardware irq and it bank was never set
for this irq_chip.

For S3C24xx and S3C64xx, this partially reverts mentioned commit by
bringing back the virt_base stored under each controller's driver data
(samsung_pinctrl_drv_data).  This virt_base address will be now
duplicated:
  - samsung_pinctrl_drv_data->virt_base: used on S3C24xx and S3C64xx,
  - samsung_pin_bank->pctl_base: used on Exynos.

Fixes: 8b1bd11c1f8f ("pinctrl: samsung: Add the support the multiple IORESOURCE_MEM 
for one pin-bank")
Cc: <sta...@vger.kernel.org>
Cc: Sergio Prado <sergio.pr...@e-labworks.com>
Reported-by: Sergio Prado <sergio.pr...@e-labworks.com>
Signed-off-by: Krzysztof Kozlowski <k...@kernel.org>

---

Tested on Odroid XU3.
Please kindly test on S3C24xx and S3C64xx. Other tests, including
Exynos5433 are also welcomed.

Changes since v1:
1. Fix NULL pointer exception when setting virt_base because bdata
    pointer is invalid at this point (incremented by loop). Spotted
    by Yao Lihua.
---
Works well for me. Tested on TINY6410 board.

Tested-by: Lihua Yao <ylhua...@163.com>

Thanks!
Lihua

Reply via email to