Re: [Qemu-devel] [PATCH 04/10] x86,MIPS: make vmware_vga optional

[Blue Swirl]

On Sat, Feb 12, 2011 at 6:48 PM, Markus Armbruster <address@hidden> wrote:
> Blue Swirl <address@hidden> writes:
>
>> Allow failure with vmware_vga device creation and use standard
>> VGA instead.
>>
>> Signed-off-by: Blue Swirl <address@hidden>
>> ---
>>  hw/mips_malta.c |    6 +++++-
>>  hw/pc.c         |   11 ++++++++---
>>  hw/vmware_vga.h |   11 +++++++++--
>>  3 files changed, 22 insertions(+), 6 deletions(-)
>>
>> diff --git a/hw/mips_malta.c b/hw/mips_malta.c
>> index 2d3f242..930c51c 100644
>> --- a/hw/mips_malta.c
>> +++ b/hw/mips_malta.c
>> @@ -957,7 +957,11 @@ void mips_malta_init (ram_addr_t ram_size,
>>      if (cirrus_vga_enabled) {
>>          pci_cirrus_vga_init(pci_bus);
>>      } else if (vmsvga_enabled) {
>> -        pci_vmsvga_init(pci_bus);
>> +        if (!pci_vmsvga_init(pci_bus)) {
>> +            fprintf(stderr, "Warning: vmware_vga not available,"
>> +                    " using standard VGA instead\n");
>> +            pci_vga_init(pci_bus);
>> +        }
>>      } else if (std_vga_enabled) {
>>          pci_vga_init(pci_bus);
>>      }
>> diff --git a/hw/pc.c b/hw/pc.c
>> index 4dfdc0b..fcee09a 100644
>> --- a/hw/pc.c
>> +++ b/hw/pc.c
>> @@ -1053,10 +1053,15 @@ void pc_vga_init(PCIBus *pci_bus)
>>              isa_cirrus_vga_init();
>>          }
>>      } else if (vmsvga_enabled) {
>> -        if (pci_bus)
>> -            pci_vmsvga_init(pci_bus);
>> -        else
>> +        if (pci_bus) {
>> +            if (!pci_vmsvga_init(pci_bus)) {
>> +                fprintf(stderr, "Warning: vmware_vga not available,"
>> +                        " using standard VGA instead\n");
>
> I don't like "vmware_vga" here.  The command line option that makes us
> go here is spelled "-vga vmware", and the qdev is called "vmware-svga".
> What about "-vga vmware is not available"?

Maybe. Patches welcome.

>
>> +                pci_vga_init(pci_bus);
>> +            }
>> +        } else {
>>              fprintf(stderr, "%s: vmware_vga: no PCI bus\n", __FUNCTION__);
>> +        }
>>  #ifdef CONFIG_SPICE
>>      } else if (qxl_enabled) {
>>          if (pci_bus)
>> diff --git a/hw/vmware_vga.h b/hw/vmware_vga.h
>> index e7bcb22..5132573 100644
>> --- a/hw/vmware_vga.h
>> +++ b/hw/vmware_vga.h
>> @@ -4,9 +4,16 @@
>>  #include "qemu-common.h"
>>
>>  /* vmware_vga.c */
>> -static inline void pci_vmsvga_init(PCIBus *bus)
>> +static inline bool pci_vmsvga_init(PCIBus *bus)
>>  {
>> -    pci_create_simple(bus, -1, "vmware-svga");
>> +    PCIDevice *dev;
>> +
>> +    dev = pci_try_create(bus, -1, "vmware-svga");
>> +    if (!dev || qdev_init(&dev->qdev) < 0) {
>> +        return false;
>> +    } else {
>> +        return true;
>> +    }
>>  }
>>
>>  #endif
>
> Two failure modes:
>
> * pci_try_create() fails, which can happen only if no such qdev
>  "vmware-svga" exists.
>
> * qdev_init() fails.
>
> The caller can't distinguish between the two, and assumes any failure
> must be the former.
>
> The assumption is actually correct, because pci_vmsvga_initfn() never
> fails, and thus qdev_init() never fails.  Brittle.

Instead of bool, the function could return int with various error
values like -ENOENT etc. Do we care enough?

> pci_vmsvga_init() is a convenience function for board setup code.  Why
> not make it more convenient and concentrate the error handling there
> rather than duplicating it in each caller?
>
> Nice side effect: no need to conflate the failure modes, no need for the
> brittle assumption.

Nice idea, how about a patch?