linux_kernel/arch/powerpc/platforms/powernv/npu-dma.c

/*
 * This file implements the DMA operations for NVLink devices. The NPU
 * devices all point to the same iommu table as the parent PCI device.
 *
 * Copyright Alistair Popple, IBM Corporation 2015.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of version 2 of the GNU General Public
 * License as published by the Free Software Foundation.
 */

#include <linux/slab.h>
#include <linux/mmu_notifier.h>
#include <linux/mmu_context.h>
#include <linux/of.h>
#include <linux/export.h>
#include <linux/pci.h>
#include <linux/memblock.h>
#include <linux/iommu.h>

#include <asm/tlb.h>
#include <asm/powernv.h>
#include <asm/reg.h>
#include <asm/opal.h>
#include <asm/io.h>
#include <asm/iommu.h>
#include <asm/pnv-pci.h>
#include <asm/msi_bitmap.h>
#include <asm/opal.h>

#include "powernv.h"
#include "pci.h"

#define npu_to_phb(x) container_of(x, struct pnv_phb, npu)

/*
 * spinlock to protect initialisation of an npu_context for a particular
 * mm_struct.
 */
static DEFINE_SPINLOCK(npu_context_lock);

/*
 * When an address shootdown range exceeds this threshold we invalidate the
 * entire TLB on the GPU for the given PID rather than each specific address in
 * the range.
 */
#define ATSD_THRESHOLD (2*1024*1024)

/*
 * Other types of TCE cache invalidation are not functional in the
 * hardware.
 */
static struct pci_dev *get_pci_dev(struct device_node *dn)
{
	struct pci_dn *pdn = PCI_DN(dn);

	return pci_get_domain_bus_and_slot(pci_domain_nr(pdn->phb->bus),
					   pdn->busno, pdn->devfn);
}

/* Given a NPU device get the associated PCI device. */
struct pci_dev *pnv_pci_get_gpu_dev(struct pci_dev *npdev)
{
	struct device_node *dn;
	struct pci_dev *gpdev;

	if (WARN_ON(!npdev))
		return NULL;

	if (WARN_ON(!npdev->dev.of_node))
		return NULL;

	/* Get assoicated PCI device */
	dn = of_parse_phandle(npdev->dev.of_node, "ibm,gpu", 0);
	if (!dn)
		return NULL;

	gpdev = get_pci_dev(dn);
	of_node_put(dn);

	return gpdev;
}
EXPORT_SYMBOL(pnv_pci_get_gpu_dev);

/* Given the real PCI device get a linked NPU device. */
struct pci_dev *pnv_pci_get_npu_dev(struct pci_dev *gpdev, int index)
{
	struct device_node *dn;
	struct pci_dev *npdev;

	if (WARN_ON(!gpdev))
		return NULL;

	/* Not all PCI devices have device-tree nodes */
	if (!gpdev->dev.of_node)
		return NULL;

	/* Get assoicated PCI device */
	dn = of_parse_phandle(gpdev->dev.of_node, "ibm,npu", index);
	if (!dn)
		return NULL;

	npdev = get_pci_dev(dn);
	of_node_put(dn);

	return npdev;
}
EXPORT_SYMBOL(pnv_pci_get_npu_dev);

#define NPU_DMA_OP_UNSUPPORTED()					\
	dev_err_once(dev, "%s operation unsupported for NVLink devices\n", \
		__func__)

static void *dma_npu_alloc(struct device *dev, size_t size,
			   dma_addr_t *dma_handle, gfp_t flag,
			   unsigned long attrs)
{
	NPU_DMA_OP_UNSUPPORTED();
	return NULL;
}

static void dma_npu_free(struct device *dev, size_t size,
			 void *vaddr, dma_addr_t dma_handle,
			 unsigned long attrs)
{
	NPU_DMA_OP_UNSUPPORTED();
}

static dma_addr_t dma_npu_map_page(struct device *dev, struct page *page,
				   unsigned long offset, size_t size,
				   enum dma_data_direction direction,
				   unsigned long attrs)
{
	NPU_DMA_OP_UNSUPPORTED();
	return 0;
}

static int dma_npu_map_sg(struct device *dev, struct scatterlist *sglist,
			  int nelems, enum dma_data_direction direction,
			  unsigned long attrs)
{
	NPU_DMA_OP_UNSUPPORTED();
	return 0;
}

static int dma_npu_dma_supported(struct device *dev, u64 mask)
{
	NPU_DMA_OP_UNSUPPORTED();
	return 0;
}

static u64 dma_npu_get_required_mask(struct device *dev)
{
	NPU_DMA_OP_UNSUPPORTED();
	return 0;
}

static const struct dma_map_ops dma_npu_ops = {
	.map_page		= dma_npu_map_page,
	.map_sg			= dma_npu_map_sg,
	.alloc			= dma_npu_alloc,
	.free			= dma_npu_free,
	.dma_supported		= dma_npu_dma_supported,
	.get_required_mask	= dma_npu_get_required_mask,
};

/*
 * Returns the PE assoicated with the PCI device of the given
 * NPU. Returns the linked pci device if pci_dev != NULL.
 */
static struct pnv_ioda_pe *get_gpu_pci_dev_and_pe(struct pnv_ioda_pe *npe,
						  struct pci_dev **gpdev)
{
	struct pnv_phb *phb;
	struct pci_controller *hose;
	struct pci_dev *pdev;
	struct pnv_ioda_pe *pe;
	struct pci_dn *pdn;

	pdev = pnv_pci_get_gpu_dev(npe->pdev);
	if (!pdev)
		return NULL;

	pdn = pci_get_pdn(pdev);
	if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
		return NULL;

	hose = pci_bus_to_host(pdev->bus);
	phb = hose->private_data;
	pe = &phb->ioda.pe_array[pdn->pe_number];

	if (gpdev)
		*gpdev = pdev;

	return pe;
}

long pnv_npu_set_window(struct pnv_ioda_pe *npe, int num,
		struct iommu_table *tbl)
{
	struct pnv_phb *phb = npe->phb;
	int64_t rc;
	const unsigned long size = tbl->it_indirect_levels ?
		tbl->it_level_size : tbl->it_size;
	const __u64 start_addr = tbl->it_offset << tbl->it_page_shift;
	const __u64 win_size = tbl->it_size << tbl->it_page_shift;

	pe_info(npe, "Setting up window %llx..%llx pg=%lx\n",
			start_addr, start_addr + win_size - 1,
			IOMMU_PAGE_SIZE(tbl));

	rc = opal_pci_map_pe_dma_window(phb->opal_id,
			npe->pe_number,
			npe->pe_number,
			tbl->it_indirect_levels + 1,
			__pa(tbl->it_base),
			size << 3,
			IOMMU_PAGE_SIZE(tbl));
	if (rc) {
		pe_err(npe, "Failed to configure TCE table, err %lld\n", rc);
		return rc;
	}
	pnv_pci_ioda2_tce_invalidate_entire(phb, false);

	/* Add the table to the list so its TCE cache will get invalidated */
	pnv_pci_link_table_and_group(phb->hose->node, num,
			tbl, &npe->table_group);

	return 0;
}

long pnv_npu_unset_window(struct pnv_ioda_pe *npe, int num)
{
	struct pnv_phb *phb = npe->phb;
	int64_t rc;

	pe_info(npe, "Removing DMA window\n");

	rc = opal_pci_map_pe_dma_window(phb->opal_id, npe->pe_number,
			npe->pe_number,
			0/* levels */, 0/* table address */,
			0/* table size */, 0/* page size */);
	if (rc) {
		pe_err(npe, "Unmapping failed, ret = %lld\n", rc);
		return rc;
	}
	pnv_pci_ioda2_tce_invalidate_entire(phb, false);

	pnv_pci_unlink_table_and_group(npe->table_group.tables[num],
			&npe->table_group);

	return 0;
}

/*
 * Enables 32 bit DMA on NPU.
 */
static void pnv_npu_dma_set_32(struct pnv_ioda_pe *npe)
{
	struct pci_dev *gpdev;
	struct pnv_ioda_pe *gpe;
	int64_t rc;

	/*
	 * Find the assoicated PCI devices and get the dma window
	 * information from there.
	 */
	if (!npe->pdev || !(npe->flags & PNV_IODA_PE_DEV))
		return;

	gpe = get_gpu_pci_dev_and_pe(npe, &gpdev);
	if (!gpe)
		return;

	rc = pnv_npu_set_window(npe, 0, gpe->table_group.tables[0]);

	/*
	 * We don't initialise npu_pe->tce32_table as we always use
	 * dma_npu_ops which are nops.
	 */
	set_dma_ops(&npe->pdev->dev, &dma_npu_ops);
}

/*
 * Enables bypass mode on the NPU. The NPU only supports one
 * window per link, so bypass needs to be explicitly enabled or
 * disabled. Unlike for a PHB3 bypass and non-bypass modes can't be
 * active at the same time.
 */
static int pnv_npu_dma_set_bypass(struct pnv_ioda_pe *npe)
{
	struct pnv_phb *phb = npe->phb;
	int64_t rc = 0;
	phys_addr_t top = memblock_end_of_DRAM();

	if (phb->type != PNV_PHB_NPU_NVLINK || !npe->pdev)
		return -EINVAL;

	rc = pnv_npu_unset_window(npe, 0);
	if (rc != OPAL_SUCCESS)
		return rc;

	/* Enable the bypass window */

	top = roundup_pow_of_two(top);
	dev_info(&npe->pdev->dev, "Enabling bypass for PE %x\n",
			npe->pe_number);
	rc = opal_pci_map_pe_dma_window_real(phb->opal_id,
			npe->pe_number, npe->pe_number,
			0 /* bypass base */, top);

	if (rc == OPAL_SUCCESS)
		pnv_pci_ioda2_tce_invalidate_entire(phb, false);

	return rc;
}

void pnv_npu_try_dma_set_bypass(struct pci_dev *gpdev, bool bypass)
{
	int i;
	struct pnv_phb *phb;
	struct pci_dn *pdn;
	struct pnv_ioda_pe *npe;
	struct pci_dev *npdev;

	for (i = 0; ; ++i) {
		npdev = pnv_pci_get_npu_dev(gpdev, i);

		if (!npdev)
			break;

		pdn = pci_get_pdn(npdev);
		if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
			return;

		phb = pci_bus_to_host(npdev->bus)->private_data;

		/* We only do bypass if it's enabled on the linked device */
		npe = &phb->ioda.pe_array[pdn->pe_number];

		if (bypass) {
			dev_info(&npdev->dev,
					"Using 64-bit DMA iommu bypass\n");
			pnv_npu_dma_set_bypass(npe);
		} else {
			dev_info(&npdev->dev, "Using 32-bit DMA via iommu\n");
			pnv_npu_dma_set_32(npe);
		}
	}
}

/* Switch ownership from platform code to external user (e.g. VFIO) */
void pnv_npu_take_ownership(struct pnv_ioda_pe *npe)
{
	struct pnv_phb *phb = npe->phb;
	int64_t rc;

	/*
	 * Note: NPU has just a single TVE in the hardware which means that
	 * while used by the kernel, it can have either 32bit window or
	 * DMA bypass but never both. So we deconfigure 32bit window only
	 * if it was enabled at the moment of ownership change.
	 */
	if (npe->table_group.tables[0]) {
		pnv_npu_unset_window(npe, 0);
		return;
	}

	/* Disable bypass */
	rc = opal_pci_map_pe_dma_window_real(phb->opal_id,
			npe->pe_number, npe->pe_number,
			0 /* bypass base */, 0);
	if (rc) {
		pe_err(npe, "Failed to disable bypass, err %lld\n", rc);
		return;
	}
	pnv_pci_ioda2_tce_invalidate_entire(npe->phb, false);
}

struct pnv_ioda_pe *pnv_pci_npu_setup_iommu(struct pnv_ioda_pe *npe)
{
	struct pnv_phb *phb = npe->phb;
	struct pci_bus *pbus = phb->hose->bus;
	struct pci_dev *npdev, *gpdev = NULL, *gptmp;
	struct pnv_ioda_pe *gpe = get_gpu_pci_dev_and_pe(npe, &gpdev);

	if (!gpe || !gpdev)
		return NULL;

	list_for_each_entry(npdev, &pbus->devices, bus_list) {
		gptmp = pnv_pci_get_gpu_dev(npdev);

		if (gptmp != gpdev)
			continue;

		pe_info(gpe, "Attached NPU %s\n", dev_name(&npdev->dev));
		iommu_group_add_device(gpe->table_group.group, &npdev->dev);
	}

	return gpe;
}

/* Maximum number of nvlinks per npu */
#define NV_MAX_LINKS 6

/* Maximum index of npu2 hosts in the system. Always < NV_MAX_NPUS */
static int max_npu2_index;

struct npu_context {
	struct mm_struct *mm;
	struct pci_dev *npdev[NV_MAX_NPUS][NV_MAX_LINKS];
	struct mmu_notifier mn;
	struct kref kref;
	bool nmmu_flush;

	/* Callback to stop translation requests on a given GPU */
	void (*release_cb)(struct npu_context *context, void *priv);

	/*
	 * Private pointer passed to the above callback for usage by
	 * device drivers.
	 */
	void *priv;
};

struct mmio_atsd_reg {
	struct npu *npu;
	int reg;
};

/*
 * Find a free MMIO ATSD register and mark it in use. Return -ENOSPC
 * if none are available.
 */
static int get_mmio_atsd_reg(struct npu *npu)
{
	int i;

	for (i = 0; i < npu->mmio_atsd_count; i++) {
		if (!test_and_set_bit_lock(i, &npu->mmio_atsd_usage))
			return i;
	}

	return -ENOSPC;
}

static void put_mmio_atsd_reg(struct npu *npu, int reg)
{
	clear_bit_unlock(reg, &npu->mmio_atsd_usage);
}

/* MMIO ATSD register offsets */
#define XTS_ATSD_AVA  1
#define XTS_ATSD_STAT 2

static void mmio_launch_invalidate(struct mmio_atsd_reg *mmio_atsd_reg,
				unsigned long launch, unsigned long va)
{
	struct npu *npu = mmio_atsd_reg->npu;
	int reg = mmio_atsd_reg->reg;

	__raw_writeq_be(va, npu->mmio_atsd_regs[reg] + XTS_ATSD_AVA);
	eieio();
	__raw_writeq_be(launch, npu->mmio_atsd_regs[reg]);
}

static void mmio_invalidate_pid(struct mmio_atsd_reg mmio_atsd_reg[NV_MAX_NPUS],
				unsigned long pid, bool flush)
{
	int i;
	unsigned long launch;

	for (i = 0; i <= max_npu2_index; i++) {
		if (mmio_atsd_reg[i].reg < 0)
			continue;

		/* IS set to invalidate matching PID */
		launch = PPC_BIT(12);

		/* PRS set to process-scoped */
		launch |= PPC_BIT(13);

		/* AP */
		launch |= (u64)
			mmu_get_ap(mmu_virtual_psize) << PPC_BITLSHIFT(17);

		/* PID */
		launch |= pid << PPC_BITLSHIFT(38);

		/* No flush */
		launch |= !flush << PPC_BITLSHIFT(39);

		/* Invalidating the entire process doesn't use a va */
		mmio_launch_invalidate(&mmio_atsd_reg[i], launch, 0);
	}
}

static void mmio_invalidate_va(struct mmio_atsd_reg mmio_atsd_reg[NV_MAX_NPUS],
			unsigned long va, unsigned long pid, bool flush)
{
	int i;
	unsigned long launch;

	for (i = 0; i <= max_npu2_index; i++) {
		if (mmio_atsd_reg[i].reg < 0)
			continue;

		/* IS set to invalidate target VA */
		launch = 0;

		/* PRS set to process scoped */
		launch |= PPC_BIT(13);

		/* AP */
		launch |= (u64)
			mmu_get_ap(mmu_virtual_psize) << PPC_BITLSHIFT(17);

		/* PID */
		launch |= pid << PPC_BITLSHIFT(38);

		/* No flush */
		launch |= !flush << PPC_BITLSHIFT(39);

		mmio_launch_invalidate(&mmio_atsd_reg[i], launch, va);
	}
}

#define mn_to_npu_context(x) container_of(x, struct npu_context, mn)

static void mmio_invalidate_wait(
	struct mmio_atsd_reg mmio_atsd_reg[NV_MAX_NPUS])
{
	struct npu *npu;
	int i, reg;

	/* Wait for all invalidations to complete */
	for (i = 0; i <= max_npu2_index; i++) {
		if (mmio_atsd_reg[i].reg < 0)
			continue;

		/* Wait for completion */
		npu = mmio_atsd_reg[i].npu;
		reg = mmio_atsd_reg[i].reg;
		while (__raw_readq(npu->mmio_atsd_regs[reg] + XTS_ATSD_STAT))
			cpu_relax();
	}
}

/*
 * Acquires all the address translation shootdown (ATSD) registers required to
 * launch an ATSD on all links this npu_context is active on.
 */
static void acquire_atsd_reg(struct npu_context *npu_context,
			struct mmio_atsd_reg mmio_atsd_reg[NV_MAX_NPUS])
{
	int i, j;
	struct npu *npu;
	struct pci_dev *npdev;
	struct pnv_phb *nphb;

	for (i = 0; i <= max_npu2_index; i++) {
		mmio_atsd_reg[i].reg = -1;
		for (j = 0; j < NV_MAX_LINKS; j++) {
			/*
			 * There are no ordering requirements with respect to
			 * the setup of struct npu_context, but to ensure
			 * consistent behaviour we need to ensure npdev[][] is
			 * only read once.
			 */
			npdev = READ_ONCE(npu_context->npdev[i][j]);
			if (!npdev)
				continue;

			nphb = pci_bus_to_host(npdev->bus)->private_data;
			npu = &nphb->npu;
			mmio_atsd_reg[i].npu = npu;
			mmio_atsd_reg[i].reg = get_mmio_atsd_reg(npu);
			while (mmio_atsd_reg[i].reg < 0) {
				mmio_atsd_reg[i].reg = get_mmio_atsd_reg(npu);
				cpu_relax();
			}
			break;
		}
	}
}

/*
 * Release previously acquired ATSD registers. To avoid deadlocks the registers
 * must be released in the same order they were acquired above in
 * acquire_atsd_reg.
 */
static void release_atsd_reg(struct mmio_atsd_reg mmio_atsd_reg[NV_MAX_NPUS])
{
	int i;

	for (i = 0; i <= max_npu2_index; i++) {
		/*
		 * We can't rely on npu_context->npdev[][] being the same here
		 * as when acquire_atsd_reg() was called, hence we use the
		 * values stored in mmio_atsd_reg during the acquire phase
		 * rather than re-reading npdev[][].
		 */
		if (mmio_atsd_reg[i].reg < 0)
			continue;

		put_mmio_atsd_reg(mmio_atsd_reg[i].npu, mmio_atsd_reg[i].reg);
	}
}

/*
 * Invalidate either a single address or an entire PID depending on
 * the value of va.
 */
static void mmio_invalidate(struct npu_context *npu_context, int va,
			unsigned long address, bool flush)
{
	struct mmio_atsd_reg mmio_atsd_reg[NV_MAX_NPUS];
	unsigned long pid = npu_context->mm->context.id;

	if (npu_context->nmmu_flush)
		/*
		 * Unfortunately the nest mmu does not support flushing specific
		 * addresses so we have to flush the whole mm once before
		 * shooting down the GPU translation.
		 */
		flush_all_mm(npu_context->mm);

	/*
	 * Loop over all the NPUs this process is active on and launch
	 * an invalidate.
	 */
	acquire_atsd_reg(npu_context, mmio_atsd_reg);
	if (va)
		mmio_invalidate_va(mmio_atsd_reg, address, pid, flush);
	else
		mmio_invalidate_pid(mmio_atsd_reg, pid, flush);

	mmio_invalidate_wait(mmio_atsd_reg);
	if (flush) {
		/*
		 * The GPU requires two flush ATSDs to ensure all entries have
		 * been flushed. We use PID 0 as it will never be used for a
		 * process on the GPU.
		 */
		mmio_invalidate_pid(mmio_atsd_reg, 0, true);
		mmio_invalidate_wait(mmio_atsd_reg);
		mmio_invalidate_pid(mmio_atsd_reg, 0, true);
		mmio_invalidate_wait(mmio_atsd_reg);
	}
	release_atsd_reg(mmio_atsd_reg);
}

static void pnv_npu2_mn_release(struct mmu_notifier *mn,
				struct mm_struct *mm)
{
	struct npu_context *npu_context = mn_to_npu_context(mn);

	/* Call into device driver to stop requests to the NMMU */
	if (npu_context->release_cb)
		npu_context->release_cb(npu_context, npu_context->priv);

	/*
	 * There should be no more translation requests for this PID, but we
	 * need to ensure any entries for it are removed from the TLB.
	 */
	mmio_invalidate(npu_context, 0, 0, true);
}

static void pnv_npu2_mn_change_pte(struct mmu_notifier *mn,
				struct mm_struct *mm,
				unsigned long address,
				pte_t pte)
{
	struct npu_context *npu_context = mn_to_npu_context(mn);

	mmio_invalidate(npu_context, 1, address, true);
}

static void pnv_npu2_mn_invalidate_range(struct mmu_notifier *mn,
					struct mm_struct *mm,
					unsigned long start, unsigned long end)
{
	struct npu_context *npu_context = mn_to_npu_context(mn);
	unsigned long address;

	if (end - start > ATSD_THRESHOLD) {
		/*
		 * Just invalidate the entire PID if the address range is too
		 * large.
		 */
		mmio_invalidate(npu_context, 0, 0, true);
	} else {
		for (address = start; address < end; address += PAGE_SIZE)
			mmio_invalidate(npu_context, 1, address, false);

		/* Do the flush only on the final addess == end */
		mmio_invalidate(npu_context, 1, address, true);
	}
}

static const struct mmu_notifier_ops nv_nmmu_notifier_ops = {
	.release = pnv_npu2_mn_release,
	.change_pte = pnv_npu2_mn_change_pte,
	.invalidate_range = pnv_npu2_mn_invalidate_range,
};

/*
 * Call into OPAL to setup the nmmu context for the current task in
 * the NPU. This must be called to setup the context tables before the
 * GPU issues ATRs. pdev should be a pointed to PCIe GPU device.
 *
 * A release callback should be registered to allow a device driver to
 * be notified that it should not launch any new translation requests
 * as the final TLB invalidate is about to occur.
 *
 * Returns an error if there no contexts are currently available or a
 * npu_context which should be passed to pnv_npu2_handle_fault().
 *
 * mmap_sem must be held in write mode and must not be called from interrupt
 * context.
 */
struct npu_context *pnv_npu2_init_context(struct pci_dev *gpdev,
			unsigned long flags,
			void (*cb)(struct npu_context *, void *),
			void *priv)
{
	int rc;
	u32 nvlink_index;
	struct device_node *nvlink_dn;
	struct mm_struct *mm = current->mm;
	struct pnv_phb *nphb;
	struct npu *npu;
	struct npu_context *npu_context;

	/*
	 * At present we don't support GPUs connected to multiple NPUs and I'm
	 * not sure the hardware does either.
	 */
	struct pci_dev *npdev = pnv_pci_get_npu_dev(gpdev, 0);

	if (!firmware_has_feature(FW_FEATURE_OPAL))
		return ERR_PTR(-ENODEV);

	if (!npdev)
		/* No nvlink associated with this GPU device */
		return ERR_PTR(-ENODEV);

	nvlink_dn = of_parse_phandle(npdev->dev.of_node, "ibm,nvlink", 0);
	if (WARN_ON(of_property_read_u32(nvlink_dn, "ibm,npu-link-index",
							&nvlink_index)))
		return ERR_PTR(-ENODEV);

	if (!mm || mm->context.id == 0) {
		/*
		 * Kernel thread contexts are not supported and context id 0 is
		 * reserved on the GPU.
		 */
		return ERR_PTR(-EINVAL);
	}

	nphb = pci_bus_to_host(npdev->bus)->private_data;
	npu = &nphb->npu;

	/*
	 * Setup the NPU context table for a particular GPU. These need to be
	 * per-GPU as we need the tables to filter ATSDs when there are no
	 * active contexts on a particular GPU. It is safe for these to be
	 * called concurrently with destroy as the OPAL call takes appropriate
	 * locks and refcounts on init/destroy.
	 */
	rc = opal_npu_init_context(nphb->opal_id, mm->context.id, flags,
				PCI_DEVID(gpdev->bus->number, gpdev->devfn));
	if (rc < 0)
		return ERR_PTR(-ENOSPC);

	/*
	 * We store the npu pci device so we can more easily get at the
	 * associated npus.
	 */
	spin_lock(&npu_context_lock);
	npu_context = mm->context.npu_context;
	if (npu_context) {
		if (npu_context->release_cb != cb ||
			npu_context->priv != priv) {
			spin_unlock(&npu_context_lock);
			opal_npu_destroy_context(nphb->opal_id, mm->context.id,
						PCI_DEVID(gpdev->bus->number,
							gpdev->devfn));
			return ERR_PTR(-EINVAL);
		}

		WARN_ON(!kref_get_unless_zero(&npu_context->kref));
	}
	spin_unlock(&npu_context_lock);

	if (!npu_context) {
		/*
		 * We can set up these fields without holding the
		 * npu_context_lock as the npu_context hasn't been returned to
		 * the caller meaning it can't be destroyed. Parallel allocation
		 * is protected against by mmap_sem.
		 */
		rc = -ENOMEM;
		npu_context = kzalloc(sizeof(struct npu_context), GFP_KERNEL);
		if (npu_context) {
			kref_init(&npu_context->kref);
			npu_context->mm = mm;
			npu_context->mn.ops = &nv_nmmu_notifier_ops;
			rc = __mmu_notifier_register(&npu_context->mn, mm);
		}

		if (rc) {
			kfree(npu_context);
			opal_npu_destroy_context(nphb->opal_id, mm->context.id,
					PCI_DEVID(gpdev->bus->number,
						gpdev->devfn));
			return ERR_PTR(rc);
		}

		mm->context.npu_context = npu_context;
	}

	npu_context->release_cb = cb;
	npu_context->priv = priv;

	/*
	 * npdev is a pci_dev pointer setup by the PCI code. We assign it to
	 * npdev[][] to indicate to the mmu notifiers that an invalidation
	 * should also be sent over this nvlink. The notifiers don't use any
	 * other fields in npu_context, so we just need to ensure that when they
	 * deference npu_context->npdev[][] it is either a valid pointer or
	 * NULL.
	 */
	WRITE_ONCE(npu_context->npdev[npu->index][nvlink_index], npdev);

	if (!nphb->npu.nmmu_flush) {
		/*
		 * If we're not explicitly flushing ourselves we need to mark
		 * the thread for global flushes
		 */
		npu_context->nmmu_flush = false;
		mm_context_add_copro(mm);
	} else
		npu_context->nmmu_flush = true;

	return npu_context;
}
EXPORT_SYMBOL(pnv_npu2_init_context);

static void pnv_npu2_release_context(struct kref *kref)
{
	struct npu_context *npu_context =
		container_of(kref, struct npu_context, kref);

	if (!npu_context->nmmu_flush)
		mm_context_remove_copro(npu_context->mm);

	npu_context->mm->context.npu_context = NULL;
}

/*
 * Destroy a context on the given GPU. May free the npu_context if it is no
 * longer active on any GPUs. Must not be called from interrupt context.
 */
void pnv_npu2_destroy_context(struct npu_context *npu_context,
			struct pci_dev *gpdev)
{
	int removed;
	struct pnv_phb *nphb;
	struct npu *npu;
	struct pci_dev *npdev = pnv_pci_get_npu_dev(gpdev, 0);
	struct device_node *nvlink_dn;
	u32 nvlink_index;

	if (WARN_ON(!npdev))
		return;

	if (!firmware_has_feature(FW_FEATURE_OPAL))
		return;

	nphb = pci_bus_to_host(npdev->bus)->private_data;
	npu = &nphb->npu;
	nvlink_dn = of_parse_phandle(npdev->dev.of_node, "ibm,nvlink", 0);
	if (WARN_ON(of_property_read_u32(nvlink_dn, "ibm,npu-link-index",
							&nvlink_index)))
		return;
	WRITE_ONCE(npu_context->npdev[npu->index][nvlink_index], NULL);
	opal_npu_destroy_context(nphb->opal_id, npu_context->mm->context.id,
				PCI_DEVID(gpdev->bus->number, gpdev->devfn));
	spin_lock(&npu_context_lock);
	removed = kref_put(&npu_context->kref, pnv_npu2_release_context);
	spin_unlock(&npu_context_lock);

	/*
	 * We need to do this outside of pnv_npu2_release_context so that it is
	 * outside the spinlock as mmu_notifier_destroy uses SRCU.
	 */
	if (removed) {
		mmu_notifier_unregister(&npu_context->mn,
					npu_context->mm);

		kfree(npu_context);
	}

}
EXPORT_SYMBOL(pnv_npu2_destroy_context);

/*
 * Assumes mmap_sem is held for the contexts associated mm.
 */
int pnv_npu2_handle_fault(struct npu_context *context, uintptr_t *ea,
			unsigned long *flags, unsigned long *status, int count)
{
	u64 rc = 0, result = 0;
	int i, is_write;
	struct page *page[1];

	/* mmap_sem should be held so the struct_mm must be present */
	struct mm_struct *mm = context->mm;

	if (!firmware_has_feature(FW_FEATURE_OPAL))
		return -ENODEV;

	WARN_ON(!rwsem_is_locked(&mm->mmap_sem));

	for (i = 0; i < count; i++) {
		is_write = flags[i] & NPU2_WRITE;
		rc = get_user_pages_remote(NULL, mm, ea[i], 1,
					is_write ? FOLL_WRITE : 0,
					page, NULL, NULL);

		/*
		 * To support virtualised environments we will have to do an
		 * access to the page to ensure it gets faulted into the
		 * hypervisor. For the moment virtualisation is not supported in
		 * other areas so leave the access out.
		 */
		if (rc != 1) {
			status[i] = rc;
			result = -EFAULT;
			continue;
		}

		status[i] = 0;
		put_page(page[0]);
	}

	return result;
}
EXPORT_SYMBOL(pnv_npu2_handle_fault);

int pnv_npu2_init(struct pnv_phb *phb)
{
	unsigned int i;
	u64 mmio_atsd;
	struct device_node *dn;
	struct pci_dev *gpdev;
	static int npu_index;
	uint64_t rc = 0;

	phb->npu.nmmu_flush =
		of_property_read_bool(phb->hose->dn, "ibm,nmmu-flush");
	for_each_child_of_node(phb->hose->dn, dn) {
		gpdev = pnv_pci_get_gpu_dev(get_pci_dev(dn));
		if (gpdev) {
			rc = opal_npu_map_lpar(phb->opal_id,
				PCI_DEVID(gpdev->bus->number, gpdev->devfn),
				0, 0);
			if (rc)
				dev_err(&gpdev->dev,
					"Error %lld mapping device to LPAR\n",
					rc);
		}
	}

	for (i = 0; !of_property_read_u64_index(phb->hose->dn, "ibm,mmio-atsd",
							i, &mmio_atsd); i++)
		phb->npu.mmio_atsd_regs[i] = ioremap(mmio_atsd, 32);

	pr_info("NPU%lld: Found %d MMIO ATSD registers", phb->opal_id, i);
	phb->npu.mmio_atsd_count = i;
	phb->npu.mmio_atsd_usage = 0;
	npu_index++;
	if (WARN_ON(npu_index >= NV_MAX_NPUS))
		return -ENOSPC;
	max_npu2_index = npu_index;
	phb->npu.index = npu_index;

	return 0;
}