[PATCH v11 6/6] iommu/arm-smmu: add support for PRR bit setup

Rob Clark robdclark at gmail.com
Wed Jun 26 11:46:39 PDT 2024


On Thu, Jun 13, 2024 at 1:11 AM Bibek Kumar Patro
<quic_bibekkum at quicinc.com> wrote:
>
>
>
> On 6/12/2024 12:32 AM, Rob Clark wrote:
> > On Wed, Jun 5, 2024 at 5:18 AM Bibek Kumar Patro
> > <quic_bibekkum at quicinc.com> wrote:
> >>
> >> Add an adreno-smmu-priv interface for drm/msm to call
> >> into arm-smmu-qcom and initiate the PRR bit setup or reset
> >> sequence as per request.
> >>
> >> This will be used by GPU side to setup the PRR bit and
> >> related configuration registers through adreno-smmu private
> >> interface instead of directly poking the smmu hardware.
> >>
> >> Signed-off-by: Bibek Kumar Patro <quic_bibekkum at quicinc.com>
> >> ---
> >>   drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c | 21 +++++++++++++++++++++
> >>   drivers/iommu/arm/arm-smmu/arm-smmu.h      |  2 ++
> >>   include/linux/adreno-smmu-priv.h           |  5 ++++-
> >>   3 files changed, 27 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
> >> index 8dabc26fa10e..2f4ee22f740a 100644
> >> --- a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
> >> +++ b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
> >> @@ -28,6 +28,7 @@
> >>   #define PREFETCH_SHALLOW       (1 << PREFETCH_SHIFT)
> >>   #define PREFETCH_MODERATE      (2 << PREFETCH_SHIFT)
> >>   #define PREFETCH_DEEP          (3 << PREFETCH_SHIFT)
> >> +#define GFX_ACTLR_PRR          (1 << 5)
> >>
> >>   static const struct actlr_config sc7280_apps_actlr_cfg[] = {
> >>          { 0x0800, 0x24e1, PREFETCH_DEFAULT | CMTLB },
> >> @@ -212,6 +213,25 @@ static void qcom_adreno_smmu_resume_translation(const void *cookie, bool termina
> >>          arm_smmu_cb_write(smmu, cfg->cbndx, ARM_SMMU_CB_RESUME, reg);
> >>   }
> >>
> >> +static void qcom_adreno_smmu_set_actlr_bit(const void *cookie, phys_addr_t page_addr, bool set)
> >> +{
> >
> > _set_actlr_bit() is a bit more of an implementation detail.  Maybe
> > _set_prr() would be a better name?
> >
>
> Yes set_prr sounds more explanatory. Infact Initially planned to name it
> as set_actlr_prr() but later changed it to set_actlr_bit() so to keep
> this interface extendible.
> Incase if gfx driver in future wants to control some other ACTLR bit as
> well along with PRR bit, then we can extend this same interface to
> handle other bits.
> So any additional adreno-smmu-priv interface would note be needed, and
> set_actlr_bit can itself be used for gfx handling of all required ACTLR
> bits.
> I think we can name it set_actlr_prr() for now, and later can change the
> name incase we extend it. What's your thought on this?

I think set_prr() or set_acltr_prr() is fine.. we can worry about
making it more generic if and when there are other ACLTR bits the gpu
wants to control

> > Also, the version of this patch that I typed up (but haven't sent to
> > list yet) took a `struct page *` instead of a phys_addr_t.. passing
> > NULL would disable PRR, so I didn't need the third arg
> >
> >> +       struct arm_smmu_domain *smmu_domain = (void *)cookie;
> >> +       struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
> >> +       struct arm_smmu_device *smmu = smmu_domain->smmu;
> >> +       u32 reg = 0;
> >> +
> >> +       writel_relaxed(lower_32_bits(page_addr),
> >> +                               (void *)smmu->ioaddr + ARM_SMMU_GFX_PRR_CFG_LADDR);
> >> +
> >> +       writel_relaxed(upper_32_bits(page_addr),
> >> +                               (void *)smmu->ioaddr + ARM_SMMU_GFX_PRR_CFG_UADDR);
> >
> > I know downstream writes it as upper+lower, but I'd guess we could
> > just writeq, couldn't we?
> >
>
> Actually ARM_SMMU_GFX_PRR_CFG_LADDR, ARM_SMMU_GFX_PRR_CFG_UADDR both are
> separate 32 bit registers. So I think writeq for 64bit write might not
> work, as these are not 64 bit registers nor these two are separated by
> 64 aligned address.

ahh, ok

> >> +
> >> +       reg =  arm_smmu_cb_read(smmu, cfg->cbndx, ARM_SMMU_CB_ACTLR);
> >> +       reg |= FIELD_PREP(GFX_ACTLR_PRR, set ? 1 : 0);
> >
> > this won't clear the ENABLE_PRR bit if we try to disable it after
> > enabling (unless this bit is read-as-zero
> >
>
> Ah okay right, got it. Thanks for pointing this out.
> Will take care of this in next version.
> will use set/reset in side if() instead with different
> bitwise ops for set and reset.
>
>
> > Also, can we give a name to PRR?  I'm guessing it is something like
> > physical-range-remap?
>
> Yes sure. I checked on this, PRR here actually stands for
> partially-resident-region.

ahh, ok, that makes sense

> So would be better If we add expansion for PRR in a comment ?
> Because if we expand PRR in variables/defines then wouldn't
> the names become too long?

Yeah, I was just thinking to mention it in the comment for set_prr()

BR,
-R

>
> Thanks & regards,
> Bibek
>
> >
> > BR,
> > -R
> >
> >> +       arm_smmu_cb_write(smmu, cfg->cbndx, ARM_SMMU_CB_ACTLR, reg);
> >> +
> >> +}
> >> +
> >>   #define QCOM_ADRENO_SMMU_GPU_SID 0
> >>
> >>   static bool qcom_adreno_smmu_is_gpu_device(struct device *dev)
> >> @@ -384,6 +404,7 @@ static int qcom_adreno_smmu_init_context(struct arm_smmu_domain *smmu_domain,
> >>          priv->get_fault_info = qcom_adreno_smmu_get_fault_info;
> >>          priv->set_stall = qcom_adreno_smmu_set_stall;
> >>          priv->resume_translation = qcom_adreno_smmu_resume_translation;
> >> +       priv->set_actlr_bit = qcom_adreno_smmu_set_actlr_bit;
> >>
> >>          actlrvar = qsmmu->data->actlrvar;
> >>          if (!actlrvar)
> >> diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu.h b/drivers/iommu/arm/arm-smmu/arm-smmu.h
> >> index d9c2ef8c1653..3076bef49e20 100644
> >> --- a/drivers/iommu/arm/arm-smmu/arm-smmu.h
> >> +++ b/drivers/iommu/arm/arm-smmu/arm-smmu.h
> >> @@ -154,6 +154,8 @@ enum arm_smmu_cbar_type {
> >>   #define ARM_SMMU_SCTLR_M               BIT(0)
> >>
> >>   #define ARM_SMMU_CB_ACTLR              0x4
> >> +#define ARM_SMMU_GFX_PRR_CFG_LADDR     0x6008
> >> +#define ARM_SMMU_GFX_PRR_CFG_UADDR     0x600C
> >>
> >>   #define ARM_SMMU_CB_RESUME             0x8
> >>   #define ARM_SMMU_RESUME_TERMINATE      BIT(0)
> >> diff --git a/include/linux/adreno-smmu-priv.h b/include/linux/adreno-smmu-priv.h
> >> index c637e0997f6d..448e191eeb52 100644
> >> --- a/include/linux/adreno-smmu-priv.h
> >> +++ b/include/linux/adreno-smmu-priv.h
> >> @@ -49,7 +49,9 @@ struct adreno_smmu_fault_info {
> >>    *                 before set_ttbr0_cfg().  If stalling on fault is enabled,
> >>    *                 the GPU driver must call resume_translation()
> >>    * @resume_translation: Resume translation after a fault
> >> - *
> >> + * @set_actlr_bits: Extendible interface to be used  by GPU to modify the
> >> + *                 ACTLR bits, currently used to intitate PRR bit setup or
> >> + *                 reset sequence for ACTLR registers as requested.
> >>    *
> >>    * The GPU driver (drm/msm) and adreno-smmu work together for controlling
> >>    * the GPU's SMMU instance.  This is by necessity, as the GPU is directly
> >> @@ -67,6 +69,7 @@ struct adreno_smmu_priv {
> >>       void (*get_fault_info)(const void *cookie, struct adreno_smmu_fault_info *info);
> >>       void (*set_stall)(const void *cookie, bool enabled);
> >>       void (*resume_translation)(const void *cookie, bool terminate);
> >> +    void (*set_actlr_bit)(const void *cookie, phys_addr_t page_addr, bool set);
> >>   };
> >>
> >>   #endif /* __ADRENO_SMMU_PRIV_H */
> >> --
> >> 2.34.1
> >>



More information about the linux-arm-kernel mailing list