Re: [Qemu-devel] Re: hpet emulation problems

[Beth Kon <eak@us.ibm.com>]

Andriy Gapon wrote:
> I obtained qemu code and looked through hw/hpet.c, below are some observations.
>
>
> on 21/07/2009 15:29 Andriy Gapon said the following:
>   
>> I observe the following problems with qemu-emulated HPET:
>>
>> 1. setting lower 32bits of a 64-bit register clears the higher 32 bits;
>> At least this happens with TIMn_CONF register - I set some bits at offset 0x100
>> and all bits at 0x104 become cleared. The problem is aggravated by the fact that
>> those bits are supposed to be RO - they specify interrupt routing capabilities.
>>     
>
> This probably happens because of the following.
> New value is set using a filter function, e.g.:
> timer->config = hpet_fixup_reg(new_val, old_val,
>                                HPET_TN_CFG_WRITE_MASK);
> But old_val was set to:
> old_val = hpet_ram_readl(opaque, addr);
>
> Apparently hpet_ram_readl returns value in the lower 32 bits and thus higher 32
> bits are lost.
> timer->config is a 64-bit variable that is supposed to hold all bits of TIMn_CONF
> (judging from hpet_ram_readl).
>
>   
>> 2. Setting interrupt type to level-triggered has no effect in the sense that
>> interrupt status bits are not set in GINTR_STA when interrupts are generated.
>>     
>
> From the code I see that level-triggered interrupts are not supposed to be
> supported at all:
> if (new_val & HPET_TIMER_TYPE_LEVEL) {
>     printf("qemu: level-triggered hpet not supported\n");
>     exit (-1);
> }
>
> The code is quite harsh in calling exit(), but it is incorrect too.
> This how HPET_TIMER_TYPE_LEVEL is defined:
> #define HPET_TIMER_TYPE_LEVEL 1
> #define HPET_TIMER_TYPE_EDGE 0
>
> But Interrupt Type is bit #1 in TIMn_CONF, bit #0 is reserved and is typically
> zero. The check should be:
> if (new_val & (HPET_TIMER_TYPE_LEVEL << 1))
> or something like that.
>
> But maybe level-triggered HPET interrupts could be supported after all.
>   
Thanks for catching these bugs! I'll submit corrections shortly.

As far as supporting level-triggered interrupts, all the guests I tested 
used edge-triggered interrupts, so this is what I implemented as a first 
pass. I don't think there is any reason not to implement 
level-triggered, though I am not planning to work on that at this point.