Re: [Qemu-devel] [PULL 00/20] s390x/kvm: Features and fixes for 2.3

[Christian Borntraeger <borntraeger@de.ibm.com>]

Am 11.03.2015 um 10:00 schrieb Thomas Huth:
> On Wed, 11 Mar 2015 09:09:21 +0100
> Christian Borntraeger <borntraeger@de.ibm.com> wrote:
> 
>> Am 11.03.2015 um 08:57 schrieb Markus Armbruster:
>>> Christian Borntraeger <borntraeger@de.ibm.com> writes:
>>>
>>>> Am 10.03.2015 um 19:01 schrieb Peter Maydell:
>>>>> On 10 March 2015 at 08:37, Christian Borntraeger
>>>>> <borntraeger@de.ibm.com> wrote:
>>>>>> Peter,
>>>>>>
>>>>>> this pull reuqest contains a header sync to 4.0-rc3. This should cause
>>>>>> no merge conflicts with other header syncs against 4.0-rc2. So here
>>>>>> is the s390 pull request.
>>>>>>
>>>>>> The following changes since commit 277263e1b320d759a760ba6c5ea75ec268f929e5:
>>>>>>
>>>>>>   Merge remote-tracking branch
>>>>>> remotes/agraf/tags/signed-ppc-for-upstream' into staging
>>>>>> (2015-03-09 14:04:14 +0000)
>>>>>>
>>>>>> are available in the git repository at:
>>>>>>
>>>>>>   git://github.com/borntraeger/qemu.git tags/s390x-20150310
>>>>>>
>>>>>> for you to fetch changes up to 6342cb1443b37d249a0e2005128ade07a80a97f5:
>>>>>>
>>>>>>   s390-ccw: rebuild BIOS (2015-03-10 09:26:33 +0100)
>>>>>
>>>>> Unfortunately I see build failures on 32 bit hosts:
>>>>>
>>>>> /root/qemu/hw/s390x/ipl.c: In function ‘bios_translate_addr’:
>>>>> /root/qemu/hw/s390x/ipl.c:101:24: error: cast from pointer to integer
>>>>> of different size [-Werror=pointer-to-int-cast]
>>>>> /root/qemu/hw/s390x/ipl.c: In function ‘s390_ipl_init’:
>>>>> /root/qemu/hw/s390x/ipl.c:135:30: error: cast to pointer from integer
>>>>> of different size [-Werror=int-to-pointer-cast]
>>>>>
>>>>
>>>> O dear. Seems that every pull request after your vacation is somewhat broken.
>>>> I fixed up patch 19 (" s390/bios: Make the s390-ccw.img relocatable") as
>>>> such
>>>>
>>>> diff --git a/hw/s390x/ipl.c b/hw/s390x/ipl.c
>>>> index ab7fd8a..6ed18d3 100644
>>>> --- a/hw/s390x/ipl.c
>>>> +++ b/hw/s390x/ipl.c
>>>> @@ -98,7 +98,7 @@ static const VMStateDescription vmstate_ipl = {
>>>>  
>>>>  static uint64_t bios_translate_addr(void *opaque, uint64_t srcaddr)
>>>>  {
>>>> -    uint64_t dstaddr = (uint64_t)opaque;
>>>> +    uint64_t dstaddr = (uint64_t) (unsigned long) opaque;
>>>>      /*
>>>>       * Assuming that our s390-ccw.img was linked for starting at address 0,
>>>>       * we can simply add the destination address for the final location
>>>
>>> Won't this shift the warning from hosts where sizeof(void *) !=
>>> sizeof(uint64_t) to hosts where it's != sizeof(unsigned long)?
>>
>> Indeed. I was only thinking about Linux, where sizeof(unsigned long) == sizeof(void*).
>> But Windows has a different model. We have to cover all kind of lp64, llp64 and friends.
>>
>>> What about using uintptr_t instead?
>>
>> Should work. Will have a try on Linux 32bit/64bit..no way to test on windows, though :-(
> 
> Thinking about this again, I think the really proper solution is rather
> this:
> 
> diff --git a/hw/s390x/ipl.c b/hw/s390x/ipl.c
> index 437c3ea..2eb55ca 100644
> --- a/hw/s390x/ipl.c
> +++ b/hw/s390x/ipl.c
> @@ -98,7 +98,7 @@ static const VMStateDescription vmstate_ipl = {
>  
>  static uint64_t bios_translate_addr(void *opaque, uint64_t srcaddr)
>  {
> -    uint64_t dstaddr = (uint64_t)opaque;
> +    uint64_t dstaddr = *(uint64_t*)opaque;
>      /*
>       * Assuming that our s390-ccw.img was linked for starting at address 0,
>       * we can simply add the destination address for the final location
> @@ -132,7 +132,7 @@ static int s390_ipl_init(SysBusDevice *dev)
>          }
>  
>          bios_size = load_elf(bios_filename, bios_translate_addr,
> -                             (void *)fwbase, &ipl->bios_start_addr,
> +                             &fwbase, &ipl->bios_start_addr,
>                               NULL, NULL, 1, ELF_MACHINE, 0);
>  
>          if (bios_size > 0) {
> 
> Instead of passing the uint64 value in the pointer, let's rather pass a
> pointer to that value instead. Then we can also be sure to get the full
> 64-bit value in bios_translate_addr().

Yes, thats even better. Will fixup and resend.