Re: [PATCH 0/2] Xen: Use a dedicated pointer for IRQ data

[Thomas Gleixner <tglx@xxxxxxxxxxxxx>]

On Fri, Aug 21 2020 at 14:17, Jürgen Groß wrote:
> On 21.08.20 13:19, Sergei Temerkhanov wrote:
>>> Did you see any specific problem where handler_data is written by
>> another component?
>> 
>> I've posted this series in the thread
>> https://lists.xenproject.org/archives/html/xen-devel/2020-08/msg00957.html
>> where the problem is caused exactly by that behavior
>> 
>>> In case this is a real problem I don't think your approach will be accepted
>> Any comments/suggestions are welcome
>
> Not sure if the IRQ maintainers agree with me, but I would add
> a set_handler_data and get_handler_data function pointer to
> struct irq_chip. If those are set I'd call them for writing/reading
> handler_data instead doing it directly. Xen could then specify those
> and add a field to its own handler data struct for storing the data
> of the driver coming later.
>
> Xen would need another accessor function for its own primary data,
> of course.
>
> Adding the IRQ maintainer as he might have an opinion here. :-)

Without seeing the patches, and no I'm not going to grab them from a web
archive, I'd say they are wrong :)

Fiddling in irqchip is wrong to begin with.

int irq_set_handler_data(unsigned int irq, void *data);
static inline void *irq_get_handler_data(unsigned int irq)
static inline void *irq_data_get_irq_handler_data(struct irq_data *d)

are accessors to handler_data. Am I missing something?

Thanks,

        tglx