Re: [PATCH v1] s390x/pci: fix interrupt blocking by returning only the device's summary bit

[Halil Pasic <pasic@linux.ibm.com>]

On Wed,  1 Oct 2025 10:39:38 -0500
Jaehoon Kim <jhkim@linux.ibm.com> wrote:

> Previously, set_ind_atomic() returned the entire byte containing
> multiple summary bits. This meant that if any other summary bit in the
> byte was set, interrupt injection could be incorrectly blocked, even
> when the current device's summary bit was not set. As a result, the
> guest could remain blocked after I/O completion during FIO tests.
> 
> This patch replaces set_ind_atomic() with set_ind_bit_atomic(), which
> returns true if the bit was set by this function, and false if it was
> already set or mapping failed. Interrupts are now blocked only when
> the device's own summary bit was not previously set, avoiding
> unintended blocking when multiple PCI summary bits exist within the
> same byte.
> 
> Signed-off-by: Jaehoon Kim <jhkim@linux.ibm.com>
> ---
>  hw/s390x/s390-pci-bus.c | 21 +++++++++++++++------
>  1 file changed, 15 insertions(+), 6 deletions(-)
> 
> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
> index f87d2748b6..e8e41c8a9a 100644
> --- a/hw/s390x/s390-pci-bus.c
> +++ b/hw/s390x/s390-pci-bus.c
> @@ -652,7 +652,16 @@ static const PCIIOMMUOps s390_iommu_ops = {
>      .get_address_space = s390_pci_dma_iommu,
>  };
>  
> -static uint8_t set_ind_atomic(uint64_t ind_loc, uint8_t to_be_set)
> +/**
> + * set_ind_bit_atomic - Atomically set a bit in an indicator
> + *
> + * @ind_loc:   Address of the indicator
> + * @to_be_set: Bit to set
> + *
> + * Returns true if the bit was set by this function, false if it was
> + * already set or mapping failed.
> + */
> +static bool set_ind_bit_atomic(uint64_t ind_loc, uint8_t to_be_set)

Unlike for example set_bit() in the Linux kernel which works with
a bit number, this one works with basically a mask in which only that
single bit (or possibly multiple bits) is set which needs to be set in
the target byte as well. I'm not sure if bit nr based would lead
to simplification or not, just wanted to point this out.

Reviewed-by: Halil Pasic <pasic@linux.ibm.com>