I'd recommend making this the first change in the series. Make
'drm/amdkfd: force raven as "dgpu" path' the second patch. That way it
only needs to change one place.
A few more comments inline.
Am 2020-08-18 um 9:09 a.m. schrieb Huang Rui:
> It's better to use inline function to wrap the iommu checking.
>
> Signed-off-by: Huang Rui <ray.huang@xxxxxxx>
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 2 +-
> drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c | 4 ++--
> .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c | 2 +-
> drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 4 ++--
> drivers/gpu/drm/amd/amdkfd/kfd_iommu.c | 10 +++++-----
> drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 8 ++++++++
> drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 6 +++---
> 7 files changed, 22 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> index 20ef754dc62e..0598a1682854 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> @@ -1258,7 +1258,7 @@ bool kfd_dev_is_large_bar(struct kfd_dev *dev)
> return true;
> }
>
> - if (dev->device_info->needs_iommu_device && !ignore_crat)
> + if (kfd_go_iommu_v2(dev->device_info))
> return false;
>
> amdgpu_amdkfd_get_local_mem_info(dev->kgd, &mem_info);
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
> index 0c4161ac4102..04103273f52b 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
> @@ -309,7 +309,7 @@ static int dbgdev_address_watch_nodiq(struct kfd_dbgdev *dbgdev,
> for (i = 0; i < adw_info->num_watch_points; i++) {
> dbgdev_address_watch_set_registers(adw_info, &addrHi, &addrLo,
> &cntl, i, pdd->qpd.vmid,
> - dbgdev->dev->device_info->needs_iommu_device && !ignore_crat);
> + kfd_go_iommu_v2(dbgdev->dev->device_info));
>
> pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *");
> pr_debug("\t\t%20s %08x\n", "register index :", i);
> @@ -399,7 +399,7 @@ static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev,
> for (i = 0; i < adw_info->num_watch_points; i++) {
> dbgdev_address_watch_set_registers(adw_info, &addrHi, &addrLo,
> &cntl, i, vmid,
> - dbgdev->dev->device_info->needs_iommu_device && !ignore_crat);
> + kfd_go_iommu_v2(dbgdev->dev->device_info));
>
> pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *");
> pr_debug("\t\t%20s %08x\n", "register index :", i);
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
> index 48b86bdbb680..70cede08e555 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
> @@ -62,7 +62,7 @@ static int update_qpd_v9(struct device_queue_manager *dqm,
> SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
> SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
> if (amdgpu_noretry &&
> - !(dqm->dev->device_info->needs_iommu_device && !ignore_crat))
> + !(kfd_go_iommu_v2(dqm->dev->device_info)))
> qpd->sh_mem_config |=
> 1 << SH_MEM_CONFIG__RETRY_DISABLE__SHIFT;
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
> index 6044d36bceb6..fe7e31014289 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
> @@ -321,7 +321,7 @@ static void kfd_init_apertures_vi(struct kfd_process_device *pdd, uint8_t id)
> pdd->lds_base = MAKE_LDS_APP_BASE_VI();
> pdd->lds_limit = MAKE_LDS_APP_LIMIT(pdd->lds_base);
>
> - if (!(pdd->dev->device_info->needs_iommu_device && !ignore_crat)) {
> + if (!kfd_go_iommu_v2(pdd->dev->device_info)) {
> /* dGPUs: SVM aperture starting at 0
> * with small reserved space for kernel.
> * Set them to CANONICAL addresses.
> @@ -423,7 +423,7 @@ int kfd_init_apertures(struct kfd_process *process)
> return -EINVAL;
> }
>
> - if (!(dev->device_info->needs_iommu_device && !ignore_crat)) {
> + if (!kfd_go_iommu_v2(dev->device_info)) {
> /* dGPUs: the reserved space for kernel
> * before SVM
> */
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c b/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c
> index c767b524bea4..ea05446572e1 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c
> @@ -41,7 +41,7 @@ int kfd_iommu_check_device(struct kfd_dev *kfd)
> struct amd_iommu_device_info iommu_info;
> int err;
>
> - if (!(kfd->device_info->needs_iommu_device && !ignore_crat))
> + if (!kfd_go_iommu_v2(kfd->device_info))
> return -ENODEV;
>
> iommu_info.flags = 0;
> @@ -63,7 +63,7 @@ int kfd_iommu_device_init(struct kfd_dev *kfd)
> unsigned int pasid_limit;
> int err;
>
> - if (!(kfd->device_info->needs_iommu_device && !ignore_crat))
> + if (!kfd_go_iommu_v2(kfd->device_info))
> return 0;
>
> iommu_info.flags = 0;
> @@ -109,7 +109,7 @@ int kfd_iommu_bind_process_to_device(struct kfd_process_device *pdd)
> struct kfd_process *p = pdd->process;
> int err;
>
> - if (!(dev->device_info->needs_iommu_device && !ignore_crat) || pdd->bound == PDD_BOUND)
> + if (!kfd_go_iommu_v2(dev->device_info) || pdd->bound == PDD_BOUND)
> return 0;
>
> if (unlikely(pdd->bound == PDD_BOUND_SUSPENDED)) {
> @@ -284,7 +284,7 @@ static void kfd_unbind_processes_from_device(struct kfd_dev *kfd)
> */
> void kfd_iommu_suspend(struct kfd_dev *kfd)
> {
> - if (!(kfd->device_info->needs_iommu_device && !ignore_crat))
> + if (!kfd_go_iommu_v2(kfd->device_info))
> return;
>
> kfd_unbind_processes_from_device(kfd);
> @@ -304,7 +304,7 @@ int kfd_iommu_resume(struct kfd_dev *kfd)
> unsigned int pasid_limit;
> int err;
>
> - if (!(kfd->device_info->needs_iommu_device && !ignore_crat))
> + if (!kfd_go_iommu_v2(kfd->device_info))
> return 0;
>
> pasid_limit = kfd_get_pasid_limit();
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> index 60243798cce2..5087b23ccdb2 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> @@ -1232,6 +1232,14 @@ static inline int kfd_devcgroup_check_permission(struct kfd_dev *kfd)
> #endif
> }
>
> +static inline bool kfd_go_iommu_v2(const struct kfd_device_info *info)
I'd prefer a different function name: kfd_device_use_iommu_v2.
Then make the parameter const struct kfd_device *dev.
> +{
> + if (!info)
> + return false;
> +
> + return info->needs_iommu_device && !ignore_crat;
> +}
If you need a NULL check, you could still write it in a more compact way
(I like inline functions to be one-liners):
return !ignore_crat && dev && dev->device_info->needs_iommu_device;
If you make this the first patch in the series, remove the !ignore_crat
part. That would be added in the second change.
Regards,
Felix
> +
> /* Debugfs */
> #if defined(CONFIG_DEBUG_FS)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> index ad8139594b6f..e6f03867e0ed 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> @@ -545,7 +545,7 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
> * If the ASIC is APU except Kaveri, set local memory size
> * to 0 to disable local memory support
> */
> - if (!(dev->gpu->device_info->needs_iommu_device && !ignore_crat)
> + if (!kfd_go_iommu_v2(dev->gpu->device_info)
> || dev->gpu->device_info->asic_family == CHIP_KAVERI) {
> amdgpu_amdkfd_get_local_mem_info(dev->gpu->kgd,
> &local_mem_info);
> @@ -1198,7 +1198,7 @@ static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu)
> /* Discrete GPUs need their own topology device list
> * entries. Don't assign them to CPU/APU nodes.
> */
> - if (!(gpu->device_info->needs_iommu_device && !ignore_crat) &&
> + if (!kfd_go_iommu_v2(gpu->device_info) &&
> dev->node_props.cpu_cores_count)
> continue;
>
> @@ -1453,7 +1453,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
> * Overwrite ATS capability according to needs_iommu_device to fix
> * potential missing corresponding bit in CRAT of BIOS.
> */
> - if (dev->gpu->device_info->needs_iommu_device && !ignore_crat)
> + if (kfd_go_iommu_v2(dev->gpu->device_info))
> dev->node_props.capability |= HSA_CAP_ATS_PRESENT;
> else
> dev->node_props.capability &= ~HSA_CAP_ATS_PRESENT;
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
