Re: [PATCH 2/8] vfio/igd: canonicalize memory size calculations

[Tomita Moeko]

On 12/2/24 17:10, Corvin Köhne wrote:
> On Sun, 2024-12-01 at 22:28 -0700, Alex Williamson wrote:
>> CAUTION: External Email!!
>> On Mon,  2 Dec 2024 00:09:32 +0800
>> Tomita Moeko <tomitamoeko@gmail.com> wrote:
>>
>>> Add helper functions igd_gtt_memory_size() and igd_stolen_size() for
>>> calculating GTT stolen memory and Data stolen memory size in bytes,
>>> and use macros to replace the hardware-related magic numbers for
>>> better readability.
>>>
>>> Signed-off-by: Tomita Moeko <tomitamoeko@gmail.com>
>>> ---
>>>  hw/vfio/igd.c | 101 ++++++++++++++++++++++++++++-------------------
>>> ---
>>>  1 file changed, 57 insertions(+), 44 deletions(-)
>>>
>>> diff --git a/hw/vfio/igd.c b/hw/vfio/igd.c
>>> index e40e601026..b449316ec0 100644
>>> --- a/hw/vfio/igd.c
>>> +++ b/hw/vfio/igd.c
>>> @@ -116,6 +116,54 @@ typedef struct VFIOIGDQuirk {
>>>  #define IGD_BDSM 0x5c /* Base Data of Stolen Memory */
>>>  #define IGD_BDSM_GEN11 0xc0 /* Base Data of Stolen Memory of gen 11
>>> and later */
>>>  
>>> +#define IGD_GMCH_GEN6_GMS_SHIFT     3
>>> +#define IGD_GMCH_GEN6_GMS_MASK      0x1f
>>> +#define IGD_GMCH_GEN6_GGMS_SHIFT    8
>>> +#define IGD_GMCH_GEN6_GGMS_MASK     0x3
>>> +#define IGD_GMCH_GEN8_GMS_SHIFT     8
>>> +#define IGD_GMCH_GEN8_GMS_MASK      0xff
>>> +#define IGD_GMCH_GEN8_GGMS_SHIFT    6
>>> +#define IGD_GMCH_GEN8_GGMS_MASK     0x3
>>> +
> 
> You should consider reusing the names used by Linux [1].
> 
> [1]
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/drm/intel/i915_drm.h?h=v6.12#n50

Original SNB/BDW in i915 is not easy to understand I think, probably
adding a comment here is better?

>>> +static uint64_t igd_gtt_memory_size(int gen, uint16_t gmch)
>>> +{
>>> +    uint64_t ggms;
>>> +
>>> +    if (gen < 8) {
>>> +        ggms = (gmch >> IGD_GMCH_GEN6_GGMS_SHIFT) &
>>> IGD_GMCH_GEN6_GGMS_MASK;
>>> +    } else {
>>> +        ggms = (gmch >> IGD_GMCH_GEN8_GGMS_SHIFT) &
>>> IGD_GMCH_GEN8_GGMS_MASK;
>>> +        ggms *= 2;
>>> +    }
>>> +
>>> +    return ggms * MiB;
>>> +}
>>> +
>>> +static uint64_t igd_stolen_memory_size(int gen, uint32_t gmch)
>>> +{
>>> +    uint64_t gms;
>>> +
>>> +    if (gen < 8) {
>>> +        gms = (gmch >> IGD_GMCH_GEN6_GMS_SHIFT) &
>>> IGD_GMCH_GEN6_GMS_MASK;
>>> +    } else {
>>> +        gms = (gmch >> IGD_GMCH_GEN8_GMS_SHIFT) &
>>> IGD_GMCH_GEN8_GMS_MASK;
>>> +    }
>>> +
>>> +    if (gen < 9) {
>>> +        if (gms < 0x10) {
>>> +            return gms * 32 * MiB;
>>> +        }
>>
>> igd_get_stolen_mb() used:
>>
>>         if (gms > 0x10) {
>>             error_report("Unsupported IGD GMS value 0x%x", gms);
>>             return 0;
>>         }
>>         return gms * 32;
>>
>> Therefore this should use <= to be equivalent.
> 
> I'd personally prefer moving the code in one commit and change the
> format in a second one.
>
> If you want to touch this, I'd align it with Linux, which multiplies gms
> by 32 MB unconditionally for gen 6 and 8 [1].
> 
> [1]
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/early-quirks.c?h=v6.12#n404
> 
>>> +    } else {
>>> +        if (gms < 0xf0) {
>>> +            return gms * 32 * MiB;
>>> +        } else {
>>> +            return (gms - 0xf0 + 1) * 4 * MiB;
>>
>> This does not match the line from igd_get_stolen_mb():
>>
>>             return (gms - 0xf0) * 4 + 4;
>>
>> Explain.
>>
> 
> I've chosen the current format because it aligns with Linux [1]. I would
> keep it as is to make it easier to sync with Linux.
> 
> [1]
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/early-quirks.c?h=v6.12#n460

Will use
        return ((gms - 0xf0) * 4 + 4) * MiB;
in v2

>>> +        }
>>> +    }
>>> +
>>> +    error_report("Unsupported IGD GMS value 0x%lx", gms);
>>> +    return 0;
>>> +}
>>>  
>>>  /*
>>>   * The rather short list of registers that we copy from the host
>>> devices.
>>> @@ -264,17 +312,10 @@ static int vfio_pci_igd_lpc_init(VFIOPCIDevice
>>> *vdev,
>>>  static int vfio_igd_gtt_max(VFIOPCIDevice *vdev)
>>>  {
>>>      uint32_t gmch = vfio_pci_read_config(&vdev->pdev, IGD_GMCH,
>>> sizeof(gmch));
>>> -    int ggms, gen = igd_gen(vdev);
>>> -
>>> -    gmch = vfio_pci_read_config(&vdev->pdev, IGD_GMCH,
>>> sizeof(gmch));
>>> -    ggms = (gmch >> (gen < 8 ? 8 : 6)) & 0x3;
>>> -    if (gen > 7) {
>>> -        ggms = 1 << ggms;
>>> -    }
>>> -
>>> -    ggms *= MiB;
>>> +    int gen = igd_gen(vdev);
>>> +    uint64_t ggms_size = igd_gtt_memory_size(gen, gmch);
>>>  
>>> -    return (ggms / (4 * KiB)) * (gen < 8 ? 4 : 8);
>>> +    return (ggms_size / (4 * KiB)) * (gen < 8 ? 4 : 8);
>>>  }
>>>  
>>>  /*
>>> @@ -481,30 +522,6 @@ void vfio_probe_igd_bar0_quirk(VFIOPCIDevice
>>> *vdev, int nr)
>>>      QLIST_INSERT_HEAD(&vdev->bars[nr].quirks, quirk, next);
>>>  }
>>>  
>>> -static int igd_get_stolen_mb(int gen, uint32_t gmch)
>>> -{
>>> -    int gms;
>>> -
>>> -    if (gen < 8) {
>>> -        gms = (gmch >> 3) & 0x1f;
>>> -    } else {
>>> -        gms = (gmch >> 8) & 0xff;
>>> -    }
>>> -
>>> -    if (gen < 9) {
>>> -        if (gms > 0x10) {
>>> -            error_report("Unsupported IGD GMS value 0x%x", gms);
>>> -            return 0;
>>> -        }
>>> -        return gms * 32;
>>> -    } else {
>>> -        if (gms < 0xf0)
>>> -            return gms * 32;
>>> -        else
>>> -            return (gms - 0xf0) * 4 + 4;
>>> -    }
>>> -}
>>> -
>>>  void vfio_probe_igd_bar4_quirk(VFIOPCIDevice *vdev, int nr)
>>>  {
>>>      g_autofree struct vfio_region_info *rom = NULL;
>>> @@ -514,7 +531,8 @@ void vfio_probe_igd_bar4_quirk(VFIOPCIDevice
>>> *vdev, int nr)
>>>      VFIOQuirk *quirk;
>>>      VFIOIGDQuirk *igd;
>>>      PCIDevice *lpc_bridge;
>>> -    int i, ret, ggms_mb, gms_mb = 0, gen;
>>> +    int i, ret, gen;
>>> +    uint64_t ggms_size, gms_size;
>>>      uint64_t *bdsm_size;
>>>      uint32_t gmch;
>>>      uint16_t cmd_orig, cmd;
>>> @@ -676,13 +694,8 @@ void vfio_probe_igd_bar4_quirk(VFIOPCIDevice
>>> *vdev, int nr)
>>>  
>>>      QLIST_INSERT_HEAD(&vdev->bars[nr].quirks, quirk, next);
>>>  
>>> -    /* Determine the size of stolen memory needed for GTT */
>>> -    ggms_mb = (gmch >> (gen < 8 ? 8 : 6)) & 0x3;
>>> -    if (gen > 7) {
>>> -        ggms_mb = 1 << ggms_mb;
>>> -    }
>>> -
>>> -    gms_mb = igd_get_stolen_mb(gen, gmch);
>>> +    ggms_size = igd_gtt_memory_size(gen, gmch);
>>> +    gms_size = igd_stolen_memory_size(gen, gmch);
>>>  
>>>      /*
>>>       * Request reserved memory for stolen memory via fw_cfg.  VM
>>> firmware
>>> @@ -693,7 +706,7 @@ void vfio_probe_igd_bar4_quirk(VFIOPCIDevice
>>> *vdev, int nr)
>>>       * config offset 0x5C.
>>>       */
>>>      bdsm_size = g_malloc(sizeof(*bdsm_size));
>>> -    *bdsm_size = cpu_to_le64((ggms_mb + gms_mb) * MiB);
>>> +    *bdsm_size = cpu_to_le64(ggms_size + gms_size);
>>>      fw_cfg_add_file(fw_cfg_find(), "etc/igd-bdsm-size",
>>>                      bdsm_size, sizeof(*bdsm_size));
>>>  
>>> @@ -744,5 +757,5 @@ void vfio_probe_igd_bar4_quirk(VFIOPCIDevice
>>> *vdev, int nr)
>>>                       vdev-
>>>> https://nospamproxywebp.beckhoff.com/enQsig/link?id=BAgAAAD3YuKN4_e
>>> If2wAAADsWKVxcuZtvheRgRqIemuzgT-OABVjIqlMqKN-
>>> EU30bxV28qXZ_Y5_SMs0YfH8q_KtHOPLp1QG1mJuzWERjIVCEePSftEKzXou_Qr3jboY
>>> Ii_VUdvptqZTIPyfY-GJ9iOrnUrKoc_oifOB9wY1 );
>>>      }
>>>  
>>> -    trace_vfio_pci_igd_bdsm_enabled(vdev-
>>>> https://nospamproxywebp.beckhoff.com/enQsig/link?id=BAgAAAD3YuKN4_e
>>> If2wAAADsWKVxcuZtvheRgRqIemuzgT-OABVjIqlMqKN-
>>> EU30bxV28qXZ_Y5_SMs0YfH8q_KtHOPLp1QG1mJuzWERjIVCEePSftEKzXou_Qr3jboY
>>> Ii_VUdvptqZTIPyfY-GJ9iOrnUrKoc_oifOB9wY1 , ggms_mb + gms_mb);
>>> +    trace_vfio_pci_igd_bdsm_enabled(vdev-
>>>> https://nospamproxywebp.beckhoff.com/enQsig/link?id=BAgAAAD3YuKN4_e
>>> If2wAAADsWKVxcuZtvheRgRqIemuzgT-OABVjIqlMqKN-
>>> EU30bxV28qXZ_Y5_SMs0YfH8q_KtHOPLp1QG1mJuzWERjIVCEePSftEKzXou_Qr3jboY
>>> Ii_VUdvptqZTIPyfY-GJ9iOrnUrKoc_oifOB9wY1 , *bdsm_size / MiB);
>>
>> Technically this should use le64_to_cpu(*bdsm_size), or retain the
>> addition logic.  Thanks,
>>
>> Alex
>