Re: [PATCH] hw/ppc: Improve build for PPC VFIO

["Philippe Mathieu-Daudé" <philmd@linaro.org>]

Hi Cédric,

On 23/11/23 08:33, Cédric Le Goater wrote:
> On 11/23/23 07:01, Zhenzhong Duan wrote:
>> VFIO is not a required subsystem for the pseries machine but it's
>> force enabled currently. When --without-default-devices is used
>> to drop some default devices including vfio-pci, vfio core code
>> is still kept which is unnecessary.
>>
>> Introduce a stub file to hold stub functions of VFIO EEH hooks,
>> then vfio core could be compiled out.
>>
>> Suggested-by: Cédric Le Goater <clg@redhat.com>
>> Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
> 
> 
> Nick,
> 
> I will take this patch through the vfio tree if that's OK for you.
> 
>> ---
>> Based on vfio-next/vfio-8.2
>>
>>   hw/ppc/spapr_pci_vfio_stub.c | 33 +++++++++++++++++++++++++++++++++
>>   hw/ppc/Kconfig               |  2 +-
>>   hw/ppc/meson.build           |  6 +++---
>>   3 files changed, 37 insertions(+), 4 deletions(-)
>>   create mode 100644 hw/ppc/spapr_pci_vfio_stub.c

We are trying to remove stubs: instead of checking late in the callee,
we shouldn't let the caller call functions depending on an unavailable
feature. So I'm a bit reluctant with this patch.

Can we add a simple 'bool pci_vfio_available(void);' helper? Or rework a
bit. For example looking quickly, we already have:

     #ifdef CONFIG_LINUX
     int spapr_phb_vfio_eeh_set_option(SpaprPhbState *sphb,
     #else
     static inline bool spapr_phb_eeh_available(SpaprPhbState *sphb)
     {
         return false;
     }
     #endif

This should be enough to protect the other calls.

Maybe we just need:

-- >8 --
--- a/include/hw/pci-host/spapr.h
+++ b/include/hw/pci-host/spapr.h
@@ -122,41 +122,20 @@ int spapr_pci_dt_populate(SpaprDrc *drc, 
SpaprMachineState *spapr,
                            void *fdt, int *fdt_start_offset, Error **errp);

  /* VFIO EEH hooks */
-#ifdef CONFIG_LINUX
+#if defined(CONFIG_LINUX) && defined(CONFIG_VFIO_PCI)
  bool spapr_phb_eeh_available(SpaprPhbState *sphb);
+#else
+static inline bool spapr_phb_eeh_available(SpaprPhbState *sphb)
+{
+    return false;
+}
+#endif
  int spapr_phb_vfio_eeh_set_option(SpaprPhbState *sphb,
                                    unsigned int addr, int option);
  int spapr_phb_vfio_eeh_get_state(SpaprPhbState *sphb, int *state);
  int spapr_phb_vfio_eeh_reset(SpaprPhbState *sphb, int option);
  int spapr_phb_vfio_eeh_configure(SpaprPhbState *sphb);
  void spapr_phb_vfio_reset(DeviceState *qdev);
-#else
-static inline bool spapr_phb_eeh_available(SpaprPhbState *sphb)
-{
-    return false;
-}
-static inline int spapr_phb_vfio_eeh_set_option(SpaprPhbState *sphb,
-                                                unsigned int addr, int 
option)
-{
-    return RTAS_OUT_HW_ERROR;
-}
-static inline int spapr_phb_vfio_eeh_get_state(SpaprPhbState *sphb,
-                                               int *state)
-{
-    return RTAS_OUT_HW_ERROR;
-}
-static inline int spapr_phb_vfio_eeh_reset(SpaprPhbState *sphb, int option)
-{
-    return RTAS_OUT_HW_ERROR;
-}
-static inline int spapr_phb_vfio_eeh_configure(SpaprPhbState *sphb)
-{
-    return RTAS_OUT_HW_ERROR;
-}
-static inline void spapr_phb_vfio_reset(DeviceState *qdev)
-{
-}
-#endif

  void spapr_phb_dma_reset(SpaprPhbState *sphb);
---

and massage a bit the calls not protected by spapr_phb_eeh_available().

>> diff --git a/hw/ppc/spapr_pci_vfio_stub.c b/hw/ppc/spapr_pci_vfio_stub.c
>> new file mode 100644
>> index 0000000000..adb3fb5e35
>> --- /dev/null
>> +++ b/hw/ppc/spapr_pci_vfio_stub.c
>> @@ -0,0 +1,33 @@
>> +#include "qemu/osdep.h"
>> +#include "hw/pci-host/spapr.h"
>> +
>> +bool spapr_phb_eeh_available(SpaprPhbState *sphb)
>> +{
>> +    return false;
>> +}
>> +
>> +void spapr_phb_vfio_reset(DeviceState *qdev)
>> +{
>> +}
>> +
>> +int spapr_phb_vfio_eeh_set_option(SpaprPhbState *sphb,
>> +                                  unsigned int addr, int option)
>> +{
>> +    return RTAS_OUT_NOT_SUPPORTED;
>> +}
>> +
>> +int spapr_phb_vfio_eeh_get_state(SpaprPhbState *sphb, int *state)
>> +{
>> +    return RTAS_OUT_NOT_SUPPORTED;
>> +}
>> +
>> +int spapr_phb_vfio_eeh_reset(SpaprPhbState *sphb, int option)
>> +{
>> +    return RTAS_OUT_NOT_SUPPORTED;
>> +}
>> +
>> +int spapr_phb_vfio_eeh_configure(SpaprPhbState *sphb)
>> +{
>> +    return RTAS_OUT_NOT_SUPPORTED;
>> +}
>> +
>> diff --git a/hw/ppc/Kconfig b/hw/ppc/Kconfig
>> index edc6d2d139..b8dabdbfbe 100644
>> --- a/hw/ppc/Kconfig
>> +++ b/hw/ppc/Kconfig
>> @@ -3,11 +3,11 @@ config PSERIES
>>       imply PCI_DEVICES
>>       imply TEST_DEVICES
>>       imply VIRTIO_VGA
>> +    imply VFIO if LINUX   # needed by spapr_pci_vfio.c
> 
> Zhenzhong,
> 
> I changed VFIO to VFIO_PCI because PPC only supports this type
> of passthrough devices.
> 
> With that,
> 
> Reviewed-by: Cédric Le Goater <clg@redhat.com>
> 
> Thanks,
> 
> C.
> 
> 
> 
>>       select NVDIMM
>>       select DIMM
>>       select PCI
>>       select SPAPR_VSCSI
>> -    select VFIO_PCI if LINUX   # needed by spapr_pci_vfio.c
>>       select XICS
>>       select XIVE
>>       select MSI_NONBROKEN
>> diff --git a/hw/ppc/meson.build b/hw/ppc/meson.build
>> index ea44856d43..2df5db2eef 100644
>> --- a/hw/ppc/meson.build
>> +++ b/hw/ppc/meson.build
>> @@ -34,9 +34,9 @@ ppc_ss.add(when: ['CONFIG_PSERIES', 'CONFIG_TCG'], 
>> if_true: files(
>>     'spapr_softmmu.c',
>>   ))
>>   ppc_ss.add(when: 'CONFIG_SPAPR_RNG', if_true: files('spapr_rng.c'))
>> -ppc_ss.add(when: ['CONFIG_PSERIES', 'CONFIG_LINUX'], if_true: files(
>> -  'spapr_pci_vfio.c',
>> -))
>> +ppc_ss.add(when: [ 'CONFIG_VFIO_PCI', 'CONFIG_PSERIES', 'CONFIG_LINUX'],
>> +           if_true: files('spapr_pci_vfio.c'),
>> +           if_false: files('spapr_pci_vfio_stub.c'))
>>   # IBM PowerNV
>>   ppc_ss.add(when: 'CONFIG_POWERNV', if_true: files(
> 
>