qemu/hw/ppc/spapr_pci_nvlink2.c

461 lines
16 KiB
C
Raw Normal View History

spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
/*
* QEMU sPAPR PCI for NVLink2 pass through
*
* Copyright (c) 2019 Alexey Kardashevskiy, IBM Corporation.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include "qemu/osdep.h"
#include "qapi/error.h"
#include "qemu-common.h"
#include "hw/pci/pci.h"
#include "hw/pci-host/spapr.h"
#include "qemu/error-report.h"
#include "hw/ppc/fdt.h"
#include "hw/pci/pci_bridge.h"
#define PHANDLE_PCIDEV(phb, pdev) (0x12000000 | \
(((phb)->index) << 16) | ((pdev)->devfn))
#define PHANDLE_GPURAM(phb, n) (0x110000FF | ((n) << 8) | \
(((phb)->index) << 16))
#define PHANDLE_NVLINK(phb, gn, nn) (0x00130000 | (((phb)->index) << 8) | \
((gn) << 4) | (nn))
#define SPAPR_GPU_NUMA_ID (cpu_to_be32(1))
typedef struct SpaprPhbPciNvGpuSlot {
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
uint64_t tgt;
uint64_t gpa;
unsigned numa_id;
PCIDevice *gpdev;
int linknum;
struct {
uint64_t atsd_gpa;
PCIDevice *npdev;
uint32_t link_speed;
} links[NVGPU_MAX_LINKS];
} SpaprPhbPciNvGpuSlot;
struct SpaprPhbPciNvGpuConfig {
uint64_t nv2_ram_current;
uint64_t nv2_atsd_current;
int num; /* number of non empty (i.e. tgt!=0) entries in slots[] */
SpaprPhbPciNvGpuSlot slots[NVGPU_MAX_NUM];
Error *err;
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
};
static SpaprPhbPciNvGpuSlot *
spapr_nvgpu_get_slot(SpaprPhbPciNvGpuConfig *nvgpus, uint64_t tgt)
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
{
int i;
/* Search for partially collected "slot" */
for (i = 0; i < nvgpus->num; ++i) {
if (nvgpus->slots[i].tgt == tgt) {
return &nvgpus->slots[i];
}
}
if (nvgpus->num == ARRAY_SIZE(nvgpus->slots)) {
return NULL;
}
i = nvgpus->num;
nvgpus->slots[i].tgt = tgt;
++nvgpus->num;
return &nvgpus->slots[i];
}
static void spapr_pci_collect_nvgpu(SpaprPhbPciNvGpuConfig *nvgpus,
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
PCIDevice *pdev, uint64_t tgt,
MemoryRegion *mr, Error **errp)
{
MachineState *machine = MACHINE(qdev_get_machine());
SpaprMachineState *spapr = SPAPR_MACHINE(machine);
SpaprPhbPciNvGpuSlot *nvslot = spapr_nvgpu_get_slot(nvgpus, tgt);
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
if (!nvslot) {
error_setg(errp, "Found too many GPUs per vPHB");
return;
}
g_assert(!nvslot->gpdev);
nvslot->gpdev = pdev;
nvslot->gpa = nvgpus->nv2_ram_current;
nvgpus->nv2_ram_current += memory_region_size(mr);
nvslot->numa_id = spapr->gpu_numa_id;
++spapr->gpu_numa_id;
}
static void spapr_pci_collect_nvnpu(SpaprPhbPciNvGpuConfig *nvgpus,
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
PCIDevice *pdev, uint64_t tgt,
MemoryRegion *mr, Error **errp)
{
SpaprPhbPciNvGpuSlot *nvslot = spapr_nvgpu_get_slot(nvgpus, tgt);
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
int j;
if (!nvslot) {
error_setg(errp, "Found too many NVLink bridges per vPHB");
return;
}
j = nvslot->linknum;
if (j == ARRAY_SIZE(nvslot->links)) {
error_setg(errp, "Found too many NVLink bridges per GPU");
return;
}
++nvslot->linknum;
g_assert(!nvslot->links[j].npdev);
nvslot->links[j].npdev = pdev;
nvslot->links[j].atsd_gpa = nvgpus->nv2_atsd_current;
nvgpus->nv2_atsd_current += memory_region_size(mr);
nvslot->links[j].link_speed =
object_property_get_uint(OBJECT(pdev), "nvlink2-link-speed", NULL);
}
static void spapr_phb_pci_collect_nvgpu(PCIBus *bus, PCIDevice *pdev,
void *opaque)
{
PCIBus *sec_bus;
Object *po = OBJECT(pdev);
uint64_t tgt = object_property_get_uint(po, "nvlink2-tgt", NULL);
if (tgt) {
Error *local_err = NULL;
SpaprPhbPciNvGpuConfig *nvgpus = opaque;
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
Object *mr_gpu = object_property_get_link(po, "nvlink2-mr[0]", NULL);
Object *mr_npu = object_property_get_link(po, "nvlink2-atsd-mr[0]",
NULL);
g_assert(mr_gpu || mr_npu);
if (mr_gpu) {
spapr_pci_collect_nvgpu(nvgpus, pdev, tgt, MEMORY_REGION(mr_gpu),
&local_err);
} else {
spapr_pci_collect_nvnpu(nvgpus, pdev, tgt, MEMORY_REGION(mr_npu),
&local_err);
}
error_propagate(&nvgpus->err, local_err);
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
}
if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
PCI_HEADER_TYPE_BRIDGE)) {
return;
}
sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
if (!sec_bus) {
return;
}
pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
spapr_phb_pci_collect_nvgpu, opaque);
}
void spapr_phb_nvgpu_setup(SpaprPhbState *sphb, Error **errp)
{
int i, j, valid_gpu_num;
PCIBus *bus;
/* Search for GPUs and NPUs */
if (!sphb->nv2_gpa_win_addr || !sphb->nv2_atsd_win_addr) {
return;
}
sphb->nvgpus = g_new0(SpaprPhbPciNvGpuConfig, 1);
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
sphb->nvgpus->nv2_ram_current = sphb->nv2_gpa_win_addr;
sphb->nvgpus->nv2_atsd_current = sphb->nv2_atsd_win_addr;
bus = PCI_HOST_BRIDGE(sphb)->bus;
pci_for_each_device(bus, pci_bus_num(bus),
spapr_phb_pci_collect_nvgpu, sphb->nvgpus);
if (sphb->nvgpus->err) {
error_propagate(errp, sphb->nvgpus->err);
sphb->nvgpus->err = NULL;
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
goto cleanup_exit;
}
/* Add found GPU RAM and ATSD MRs if found */
for (i = 0, valid_gpu_num = 0; i < sphb->nvgpus->num; ++i) {
Object *nvmrobj;
SpaprPhbPciNvGpuSlot *nvslot = &sphb->nvgpus->slots[i];
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
if (!nvslot->gpdev) {
continue;
}
nvmrobj = object_property_get_link(OBJECT(nvslot->gpdev),
"nvlink2-mr[0]", NULL);
/* ATSD is pointless without GPU RAM MR so skip those */
if (!nvmrobj) {
continue;
}
++valid_gpu_num;
memory_region_add_subregion(get_system_memory(), nvslot->gpa,
MEMORY_REGION(nvmrobj));
for (j = 0; j < nvslot->linknum; ++j) {
Object *atsdmrobj;
atsdmrobj = object_property_get_link(OBJECT(nvslot->links[j].npdev),
"nvlink2-atsd-mr[0]", NULL);
if (!atsdmrobj) {
continue;
}
memory_region_add_subregion(get_system_memory(),
nvslot->links[j].atsd_gpa,
MEMORY_REGION(atsdmrobj));
}
}
if (valid_gpu_num) {
return;
}
/* We did not find any interesting GPU */
cleanup_exit:
g_free(sphb->nvgpus);
sphb->nvgpus = NULL;
}
void spapr_phb_nvgpu_free(SpaprPhbState *sphb)
{
int i, j;
if (!sphb->nvgpus) {
return;
}
for (i = 0; i < sphb->nvgpus->num; ++i) {
SpaprPhbPciNvGpuSlot *nvslot = &sphb->nvgpus->slots[i];
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
Object *nv_mrobj = object_property_get_link(OBJECT(nvslot->gpdev),
"nvlink2-mr[0]", NULL);
if (nv_mrobj) {
memory_region_del_subregion(get_system_memory(),
MEMORY_REGION(nv_mrobj));
}
for (j = 0; j < nvslot->linknum; ++j) {
PCIDevice *npdev = nvslot->links[j].npdev;
Object *atsd_mrobj;
atsd_mrobj = object_property_get_link(OBJECT(npdev),
"nvlink2-atsd-mr[0]", NULL);
if (atsd_mrobj) {
memory_region_del_subregion(get_system_memory(),
MEMORY_REGION(atsd_mrobj));
}
}
}
g_free(sphb->nvgpus);
sphb->nvgpus = NULL;
}
void spapr_phb_nvgpu_populate_dt(SpaprPhbState *sphb, void *fdt, int bus_off,
Error **errp)
{
int i, j, atsdnum = 0;
uint64_t atsd[8]; /* The existing limitation of known guests */
if (!sphb->nvgpus) {
return;
}
for (i = 0; (i < sphb->nvgpus->num) && (atsdnum < ARRAY_SIZE(atsd)); ++i) {
SpaprPhbPciNvGpuSlot *nvslot = &sphb->nvgpus->slots[i];
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
if (!nvslot->gpdev) {
continue;
}
for (j = 0; j < nvslot->linknum; ++j) {
if (!nvslot->links[j].atsd_gpa) {
continue;
}
if (atsdnum == ARRAY_SIZE(atsd)) {
error_report("Only %"PRIuPTR" ATSD registers supported",
ARRAY_SIZE(atsd));
break;
}
atsd[atsdnum] = cpu_to_be64(nvslot->links[j].atsd_gpa);
++atsdnum;
}
}
if (!atsdnum) {
error_setg(errp, "No ATSD registers found");
return;
}
if (!spapr_phb_eeh_available(sphb)) {
/*
* ibm,mmio-atsd contains ATSD registers; these belong to an NPU PHB
* which we do not emulate as a separate device. Instead we put
* ibm,mmio-atsd to the vPHB with GPU and make sure that we do not
* put GPUs from different IOMMU groups to the same vPHB to ensure
* that the guest will use ATSDs from the corresponding NPU.
*/
error_setg(errp, "ATSD requires separate vPHB per GPU IOMMU group");
return;
}
_FDT((fdt_setprop(fdt, bus_off, "ibm,mmio-atsd", atsd,
atsdnum * sizeof(atsd[0]))));
}
void spapr_phb_nvgpu_ram_populate_dt(SpaprPhbState *sphb, void *fdt)
{
int i, j, linkidx, npuoff;
char *npuname;
if (!sphb->nvgpus) {
return;
}
npuname = g_strdup_printf("npuphb%d", sphb->index);
npuoff = fdt_add_subnode(fdt, 0, npuname);
_FDT(npuoff);
_FDT(fdt_setprop_cell(fdt, npuoff, "#address-cells", 1));
_FDT(fdt_setprop_cell(fdt, npuoff, "#size-cells", 0));
/* Advertise NPU as POWER9 so the guest can enable NPU2 contexts */
_FDT((fdt_setprop_string(fdt, npuoff, "compatible", "ibm,power9-npu")));
g_free(npuname);
for (i = 0, linkidx = 0; i < sphb->nvgpus->num; ++i) {
for (j = 0; j < sphb->nvgpus->slots[i].linknum; ++j) {
char *linkname = g_strdup_printf("link@%d", linkidx);
int off = fdt_add_subnode(fdt, npuoff, linkname);
_FDT(off);
/* _FDT((fdt_setprop_cell(fdt, off, "reg", linkidx))); */
_FDT((fdt_setprop_string(fdt, off, "compatible",
"ibm,npu-link")));
_FDT((fdt_setprop_cell(fdt, off, "phandle",
PHANDLE_NVLINK(sphb, i, j))));
_FDT((fdt_setprop_cell(fdt, off, "ibm,npu-link-index", linkidx)));
g_free(linkname);
++linkidx;
}
}
/* Add memory nodes for GPU RAM and mark them unusable */
for (i = 0; i < sphb->nvgpus->num; ++i) {
SpaprPhbPciNvGpuSlot *nvslot = &sphb->nvgpus->slots[i];
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
Object *nv_mrobj = object_property_get_link(OBJECT(nvslot->gpdev),
"nvlink2-mr[0]",
&error_abort);
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
uint32_t associativity[] = {
cpu_to_be32(0x4),
cpu_to_be32(nvslot->numa_id),
cpu_to_be32(nvslot->numa_id),
cpu_to_be32(nvslot->numa_id),
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
cpu_to_be32(nvslot->numa_id)
};
uint64_t size = object_property_get_uint(nv_mrobj, "size", NULL);
uint64_t mem_reg[2] = { cpu_to_be64(nvslot->gpa), cpu_to_be64(size) };
char *mem_name = g_strdup_printf("memory@%"PRIx64, nvslot->gpa);
int off = fdt_add_subnode(fdt, 0, mem_name);
_FDT(off);
_FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
_FDT((fdt_setprop(fdt, off, "reg", mem_reg, sizeof(mem_reg))));
if (sphb->pre_5_1_assoc) {
associativity[1] = SPAPR_GPU_NUMA_ID;
associativity[2] = SPAPR_GPU_NUMA_ID;
associativity[3] = SPAPR_GPU_NUMA_ID;
}
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
_FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
sizeof(associativity))));
_FDT((fdt_setprop_string(fdt, off, "compatible",
"ibm,coherent-device-memory")));
mem_reg[1] = cpu_to_be64(0);
_FDT((fdt_setprop(fdt, off, "linux,usable-memory", mem_reg,
sizeof(mem_reg))));
_FDT((fdt_setprop_cell(fdt, off, "phandle",
PHANDLE_GPURAM(sphb, i))));
g_free(mem_name);
}
}
void spapr_phb_nvgpu_populate_pcidev_dt(PCIDevice *dev, void *fdt, int offset,
SpaprPhbState *sphb)
{
int i, j;
if (!sphb->nvgpus) {
return;
}
for (i = 0; i < sphb->nvgpus->num; ++i) {
SpaprPhbPciNvGpuSlot *nvslot = &sphb->nvgpus->slots[i];
spapr: Support NVIDIA V100 GPU with NVLink2 NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver implements special regions for such GPUs and emulates an NVLink bridge. NVLink2-enabled POWER9 CPUs also provide address translation services which includes an ATS shootdown (ATSD) register exported via the NVLink bridge device. This adds a quirk to VFIO to map the GPU memory and create an MR; the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses this to get the MR and map it to the system address space. Another quirk does the same for ATSD. This adds additional steps to sPAPR PHB setup: 1. Search for specific GPUs and NPUs, collect findings in sPAPRPHBState::nvgpus, manage system address space mappings; 2. Add device-specific properties such as "ibm,npu", "ibm,gpu", "memory-block", "link-speed" to advertise the NVLink2 function to the guest; 3. Add "mmio-atsd" to vPHB to advertise the ATSD capability; 4. Add new memory blocks (with extra "linux,memory-usable" to prevent the guest OS from accessing the new memory until it is onlined) and npuphb# nodes representing an NPU unit for every vPHB as the GPU driver uses it for link discovery. This allocates space for GPU RAM and ATSD like we do for MMIOs by adding 2 new parameters to the phb_placement() hook. Older machine types set these to zero. This puts new memory nodes in a separate NUMA node to as the GPU RAM needs to be configured equally distant from any other node in the system. Unlike the host setup which assigns numa ids from 255 downwards, this adds new NUMA nodes after the user configures nodes or from 1 if none were configured. This adds requirement similar to EEH - one IOMMU group per vPHB. The reason for this is that ATSD registers belong to a physical NPU so they cannot invalidate translations on GPUs attached to another NPU. It is guaranteed by the host platform as it does not mix NVLink bridges or GPUs from different NPU in the same IOMMU group. If more than one IOMMU group is detected on a vPHB, this disables ATSD support for that vPHB and prints a warning. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> [aw: for vfio portions] Acked-by: Alex Williamson <alex.williamson@redhat.com> Message-Id: <20190312082103.130561-1-aik@ozlabs.ru> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-03-12 11:21:03 +03:00
/* Skip "slot" without attached GPU */
if (!nvslot->gpdev) {
continue;
}
if (dev == nvslot->gpdev) {
uint32_t npus[nvslot->linknum];
for (j = 0; j < nvslot->linknum; ++j) {
PCIDevice *npdev = nvslot->links[j].npdev;
npus[j] = cpu_to_be32(PHANDLE_PCIDEV(sphb, npdev));
}
_FDT(fdt_setprop(fdt, offset, "ibm,npu", npus,
j * sizeof(npus[0])));
_FDT((fdt_setprop_cell(fdt, offset, "phandle",
PHANDLE_PCIDEV(sphb, dev))));
continue;
}
for (j = 0; j < nvslot->linknum; ++j) {
if (dev != nvslot->links[j].npdev) {
continue;
}
_FDT((fdt_setprop_cell(fdt, offset, "phandle",
PHANDLE_PCIDEV(sphb, dev))));
_FDT(fdt_setprop_cell(fdt, offset, "ibm,gpu",
PHANDLE_PCIDEV(sphb, nvslot->gpdev)));
_FDT((fdt_setprop_cell(fdt, offset, "ibm,nvlink",
PHANDLE_NVLINK(sphb, i, j))));
/*
* If we ever want to emulate GPU RAM at the same location as on
* the host - here is the encoding GPA->TGT:
*
* gta = ((sphb->nv2_gpa >> 42) & 0x1) << 42;
* gta |= ((sphb->nv2_gpa >> 45) & 0x3) << 43;
* gta |= ((sphb->nv2_gpa >> 49) & 0x3) << 45;
* gta |= sphb->nv2_gpa & ((1UL << 43) - 1);
*/
_FDT(fdt_setprop_cell(fdt, offset, "memory-region",
PHANDLE_GPURAM(sphb, i)));
_FDT(fdt_setprop_u64(fdt, offset, "ibm,device-tgt-addr",
nvslot->tgt));
_FDT(fdt_setprop_cell(fdt, offset, "ibm,nvlink-speed",
nvslot->links[j].link_speed));
}
}
}