* Ignore CRAT table during KFD initialization. By default, KFD uses the ACPI CRAT
* table to get information about AMD APUs. This option can serve as a workaround on
* systems with a broken CRAT table.
+ *
+ * Default is auto (according to asic type, iommu_v2, and crat table, to decide
+ * whehter use CRAT)
*/
int ignore_crat;
module_param(ignore_crat, int, 0444);
MODULE_PARM_DESC(ignore_crat,
- "Ignore CRAT table during KFD initialization (0 = use CRAT (default), 1 = ignore CRAT)");
+ "Ignore CRAT table during KFD initialization (0 = auto (default), 1 = ignore CRAT)");
/**
* DOC: halt_if_hws_hang (int)
return true;
}
- if (dev->device_info->needs_iommu_device)
+ if (dev->use_iommu_v2)
return false;
amdgpu_amdkfd_get_local_mem_info(dev->kgd, &mem_info);
return 0;
}
+static bool kfd_ignore_crat(void)
+{
+ bool ret;
+
+ if (ignore_crat)
+ return true;
+
+#ifndef KFD_SUPPORT_IOMMU_V2
+ ret = true;
+#else
+ ret = false;
+#endif
+
+ return ret;
+}
+
/*
* kfd_create_crat_image_acpi - Allocates memory for CRAT image and
* copies CRAT from ACPI (if available).
return -EINVAL;
}
- if (ignore_crat) {
+ if (kfd_ignore_crat()) {
pr_info("CRAT table disabled by module option\n");
return -ENODATA;
}
.num_xgmi_sdma_engines = 0,
.num_sdma_queues_per_engine = 2,
};
+#endif
static const struct kfd_device_info raven_device_info = {
.asic_family = CHIP_RAVEN,
.num_xgmi_sdma_engines = 0,
.num_sdma_queues_per_engine = 2,
};
-#endif
static const struct kfd_device_info hawaii_device_info = {
.asic_family = CHIP_HAWAII,
goto gws_error;
}
+ /* If CRAT is broken, won't set iommu enabled */
+ kfd_double_confirm_iommu_support(kfd);
+
if (kfd_iommu_device_init(kfd)) {
dev_err(kfd_device, "Error initializing iommuv2\n");
goto device_iommu_error;
SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
if (amdgpu_noretry &&
- !dqm->dev->device_info->needs_iommu_device)
+ !dqm->dev->use_iommu_v2)
qpd->sh_mem_config |=
1 << SH_MEM_CONFIG__RETRY_DISABLE__SHIFT;
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) {
+ if (!pdd->dev->use_iommu_v2) {
/* dGPUs: SVM aperture starting at 0
* with small reserved space for kernel.
* Set them to CANONICAL addresses.
return -EINVAL;
}
- if (!dev->device_info->needs_iommu_device) {
+ if (!dev->use_iommu_v2) {
/* dGPUs: the reserved space for kernel
* before SVM
*/
struct amd_iommu_device_info iommu_info;
int err;
- if (!kfd->device_info->needs_iommu_device)
+ if (!kfd->use_iommu_v2)
return -ENODEV;
iommu_info.flags = 0;
unsigned int pasid_limit;
int err;
- if (!kfd->device_info->needs_iommu_device)
+ if (!kfd->use_iommu_v2)
return 0;
iommu_info.flags = 0;
struct kfd_process *p = pdd->process;
int err;
- if (!dev->device_info->needs_iommu_device || pdd->bound == PDD_BOUND)
+ if (!dev->use_iommu_v2 || pdd->bound == PDD_BOUND)
return 0;
if (unlikely(pdd->bound == PDD_BOUND_SUSPENDED)) {
*/
void kfd_iommu_suspend(struct kfd_dev *kfd)
{
- if (!kfd->device_info->needs_iommu_device)
+ if (!kfd->use_iommu_v2)
return;
kfd_unbind_processes_from_device(kfd);
unsigned int pasid_limit;
int err;
- if (!kfd->device_info->needs_iommu_device)
+ if (!kfd->use_iommu_v2)
return 0;
pasid_limit = kfd_get_pasid_limit();
bool pci_atomic_requested;
+ /* Use IOMMU v2 flag */
+ bool use_iommu_v2;
+
/* SRAM ECC flag */
atomic_t sram_ecc_flag;
struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd);
int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev);
int kfd_numa_node_to_apic_id(int numa_node_id);
+void kfd_double_confirm_iommu_support(struct kfd_dev *gpu);
/* Interrupts */
int kfd_interrupt_init(struct kfd_dev *dev);
sysfs_show_32bit_prop(buffer, offs, "cpu_cores_count",
dev->node_props.cpu_cores_count);
sysfs_show_32bit_prop(buffer, offs, "simd_count",
- dev->node_props.simd_count);
+ dev->gpu ? dev->node_props.simd_count : 0);
sysfs_show_32bit_prop(buffer, offs, "mem_banks_count",
dev->node_props.mem_banks_count);
sysfs_show_32bit_prop(buffer, offs, "caches_count",
/* Discrete GPUs need their own topology device list
* entries. Don't assign them to CPU/APU nodes.
*/
- if (!gpu->device_info->needs_iommu_device &&
+ if (!gpu->use_iommu_v2 &&
dev->node_props.cpu_cores_count)
continue;
* 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)
+ if (dev->gpu->use_iommu_v2)
dev->node_props.capability |= HSA_CAP_ATS_PRESENT;
else
dev->node_props.capability &= ~HSA_CAP_ATS_PRESENT;
return kfd_cpumask_to_apic_id(cpumask_of_node(numa_node_id));
}
+void kfd_double_confirm_iommu_support(struct kfd_dev *gpu)
+{
+ struct kfd_topology_device *dev;
+
+ gpu->use_iommu_v2 = false;
+
+ if (!gpu->device_info->needs_iommu_device)
+ return;
+
+ down_read(&topology_lock);
+
+ /* Only use IOMMUv2 if there is an APU topology node with no GPU
+ * assigned yet. This GPU will be assigned to it.
+ */
+ list_for_each_entry(dev, &topology_device_list, list)
+ if (dev->node_props.cpu_cores_count &&
+ dev->node_props.simd_count &&
+ !dev->gpu)
+ gpu->use_iommu_v2 = true;
+
+ up_read(&topology_lock);
+}
+
#if defined(CONFIG_DEBUG_FS)
int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data)