[2/2] powerpc/powernv: Enable and setup PCI P2P

Implement the generic dma_map_resource callback on the PCI controller
for powernv. This will enable PCI P2P on POWER9 architecture. It will
allow catching a cross-PHB mmio mapping, which needs to be setup in
hardware by calling opal. Both the initiator and target PHBs need to be
configured, so we look for which PHB owns the mmio address being mapped.

Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
[maxg: added CONFIG_PCI_P2PDMA wrappers]
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
---
 arch/powerpc/include/asm/opal.h            |   5 +-
 arch/powerpc/platforms/powernv/opal-call.c |   1 +
 arch/powerpc/platforms/powernv/pci-ioda.c  | 164 +++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.h       |   9 ++
 4 files changed, 178 insertions(+), 1 deletion(-)

> +#ifdef CONFIG_PCI_P2PDMA
> +int64_t opal_pci_set_p2p(uint64_t phb_init, uint64_t phb_target,
> +			 uint64_t desc, uint16_t pe_number);
> +#endif

There should be no need for the ifdef here.

> +	/*
> +	 * Configuring the initiator's PHB requires to adjust its
> +	 * TVE#1 setting. Since the same device can be an initiator
> +	 * several times for different target devices, we need to keep
> +	 * a reference count to know when we can restore the default
> +	 * bypass setting on its TVE#1 when disabling. Opal is not
> +	 * tracking PE states, so we add a reference count on the PE
> +	 * in linux.
> +	 *
> +	 * For the target, the configuration is per PHB, so we keep a
> +	 * target reference count on the PHB.
> +	 */

This could be shortened a bit by using up the whole 80 lines available
in source files.

> +	mutex_lock(&p2p_mutex);
> +
> +	if (desc & OPAL_PCI_P2P_ENABLE) {
> +		/* always go to opal to validate the configuration */
> +		rc = opal_pci_set_p2p(phb_init->opal_id, phb_target->opal_id,
> +				      desc, pe_init->pe_number);
> +
> +		if (rc != OPAL_SUCCESS) {
> +			rc = -EIO;
> +			goto out;
> +		}
> +
> +		pe_init->p2p_initiator_count++;
> +		phb_target->p2p_target_count++;
> +	} else {
> +		if (!pe_init->p2p_initiator_count ||
> +		    !phb_target->p2p_target_count) {
> +			rc = -EINVAL;
> +			goto out;
> +		}
> +
> +		if (--pe_init->p2p_initiator_count == 0)
> +			pnv_pci_ioda2_set_bypass(pe_init, true);
> +
> +		if (--phb_target->p2p_target_count == 0) {
> +			rc = opal_pci_set_p2p(phb_init->opal_id,
> +					      phb_target->opal_id, desc,
> +					      pe_init->pe_number);
> +			if (rc != OPAL_SUCCESS) {
> +				rc = -EIO;
> +				goto out;
> +			}
> +		}
> +	}
> +	rc = 0;
> +out:
> +	mutex_unlock(&p2p_mutex);
> +	return rc;
> +}

The enable and disable path shares almost no code, why not split it into
two functions?

> +static bool pnv_pci_controller_owns_addr(struct pci_controller *hose,
> +					 phys_addr_t addr, size_t size)
> +{
> +	struct resource *r;
> +	int i;
> +
> +	/*
> +	 * it seems safe to assume the full range is under the same
> +	 * PHB, so we can ignore the size

Capitalize the first character in a multi-line comment, and use up the
whole 80 chars.

> +	 */
> +	for (i = 0; i < 3; i++) {

Replace the magic 3 with ARRAY_SIZE(hose->mem_resources) ?


> +		r = &hose->mem_resources[i];

Move the r declaration here and initialize it on the same line.

> +		if (r->flags && (addr >= r->start) && (addr < r->end))

No need for the inner braces.

> +			return true;
> +	}
> +	return false;
> +}
> +
> +/*
> + * find the phb owning a mmio address if not owned locally
> + */
> +static struct pnv_phb *pnv_pci_find_owning_phb(struct pci_dev *pdev,
> +					       phys_addr_t addr, size_t size)
> +{
> +	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
> +	struct pnv_phb *phb;
> +
> +	/* fast path */
> +	if (pnv_pci_controller_owns_addr(hose, addr, size))
> +		return NULL;

Maybe open code the pci_bus_to_host(pdev->bus) call here, as we just
overwrite host in the list iteration?

> +
> +	list_for_each_entry(hose, &hose_list, list_node) {
> +		phb = hose->private_data;

Move the ohb declaration here and initialize it on the same line.

> +		if (phb->type != PNV_PHB_NPU_NVLINK &&
> +		    phb->type != PNV_PHB_NPU_OCAPI) {
> +			if (pnv_pci_controller_owns_addr(hose, addr, size))
> +				return phb;
> +		}
> +	}


> +	return NULL;
> +}
> +
> +static int pnv_pci_dma_map_resource(struct pci_dev *pdev,
> +				    phys_addr_t phys_addr, size_t size,
> +				    enum dma_data_direction dir)
> +{
> +	struct pnv_phb *target_phb;
> +	int rc;
> +	u64 desc;
> +
> +	target_phb = pnv_pci_find_owning_phb(pdev, phys_addr, size);
> +	if (target_phb) {

Return early here for the !target_phb case?

> +		desc = OPAL_PCI_P2P_ENABLE;
> +		if (dir == DMA_TO_DEVICE)
> +			desc |= OPAL_PCI_P2P_STORE;
> +		else if (dir == DMA_FROM_DEVICE)
> +			desc |= OPAL_PCI_P2P_LOAD;
> +		else if (dir == DMA_BIDIRECTIONAL)
> +			desc |= OPAL_PCI_P2P_LOAD | OPAL_PCI_P2P_STORE;

Seems like this could be split into a little helper.

> +		rc = pnv_pci_ioda_set_p2p(pdev, target_phb, desc);
> +		if (rc) {
> +			dev_err(&pdev->dev, "Failed to setup PCI peer-to-peer for address %llx: %d\n",
> +				phys_addr, rc);
> +			return rc;
> +		}

No need for this printk, the callers should already deal with mapping
failures.

On 4/17/2020 10:02 AM, Christoph Hellwig wrote:
>> +#ifdef CONFIG_PCI_P2PDMA
>> +int64_t opal_pci_set_p2p(uint64_t phb_init, uint64_t phb_target,
>> +			 uint64_t desc, uint16_t pe_number);
>> +#endif
> There should be no need for the ifdef here.
>
>> +	/*
>> +	 * Configuring the initiator's PHB requires to adjust its
>> +	 * TVE#1 setting. Since the same device can be an initiator
>> +	 * several times for different target devices, we need to keep
>> +	 * a reference count to know when we can restore the default
>> +	 * bypass setting on its TVE#1 when disabling. Opal is not
>> +	 * tracking PE states, so we add a reference count on the PE
>> +	 * in linux.
>> +	 *
>> +	 * For the target, the configuration is per PHB, so we keep a
>> +	 * target reference count on the PHB.
>> +	 */
> This could be shortened a bit by using up the whole 80 lines available
> in source files.
>
>> +	mutex_lock(&p2p_mutex);
>> +
>> +	if (desc & OPAL_PCI_P2P_ENABLE) {
>> +		/* always go to opal to validate the configuration */
>> +		rc = opal_pci_set_p2p(phb_init->opal_id, phb_target->opal_id,
>> +				      desc, pe_init->pe_number);
>> +
>> +		if (rc != OPAL_SUCCESS) {
>> +			rc = -EIO;
>> +			goto out;
>> +		}
>> +
>> +		pe_init->p2p_initiator_count++;
>> +		phb_target->p2p_target_count++;
>> +	} else {
>> +		if (!pe_init->p2p_initiator_count ||
>> +		    !phb_target->p2p_target_count) {
>> +			rc = -EINVAL;
>> +			goto out;
>> +		}
>> +
>> +		if (--pe_init->p2p_initiator_count == 0)
>> +			pnv_pci_ioda2_set_bypass(pe_init, true);
>> +
>> +		if (--phb_target->p2p_target_count == 0) {
>> +			rc = opal_pci_set_p2p(phb_init->opal_id,
>> +					      phb_target->opal_id, desc,
>> +					      pe_init->pe_number);
>> +			if (rc != OPAL_SUCCESS) {
>> +				rc = -EIO;
>> +				goto out;
>> +			}
>> +		}
>> +	}
>> +	rc = 0;
>> +out:
>> +	mutex_unlock(&p2p_mutex);
>> +	return rc;
>> +}
> The enable and disable path shares almost no code, why not split it into
> two functions?

how about also changing the defines OPAL_PCI_P2P_* to an enum ?

/* PCI p2p operation descriptors */
enum opal_pci_p2p {

         OPAL_PCI_P2P_DISABLE    = 0,

         OPAL_PCI_P2P_ENABLE     = (1 << 0),
         OPAL_PCI_P2P_LOAD       = (1 << 1),
         OPAL_PCI_P2P_STORE      = (1 << 2),
};

Fred ?


>> +static bool pnv_pci_controller_owns_addr(struct pci_controller *hose,
>> +					 phys_addr_t addr, size_t size)
>> +{
>> +	struct resource *r;
>> +	int i;
>> +
>> +	/*
>> +	 * it seems safe to assume the full range is under the same
>> +	 * PHB, so we can ignore the size
> Capitalize the first character in a multi-line comment, and use up the
> whole 80 chars.
>
>> +	 */
>> +	for (i = 0; i < 3; i++) {
> Replace the magic 3 with ARRAY_SIZE(hose->mem_resources) ?
>
>
>> +		r = &hose->mem_resources[i];
> Move the r declaration here and initialize it on the same line.
>
>> +		if (r->flags && (addr >= r->start) && (addr < r->end))
> No need for the inner braces.
>
>> +			return true;
>> +	}
>> +	return false;
>> +}
>> +
>> +/*
>> + * find the phb owning a mmio address if not owned locally
>> + */
>> +static struct pnv_phb *pnv_pci_find_owning_phb(struct pci_dev *pdev,
>> +					       phys_addr_t addr, size_t size)
>> +{
>> +	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
>> +	struct pnv_phb *phb;
>> +
>> +	/* fast path */
>> +	if (pnv_pci_controller_owns_addr(hose, addr, size))
>> +		return NULL;
> Maybe open code the pci_bus_to_host(pdev->bus) call here, as we just
> overwrite host in the list iteration?

if this is more readable so no problem:

if (pnv_pci_controller_owns_addr(pdev->bus->sysdata, addr, size))


>
>> +
>> +	list_for_each_entry(hose, &hose_list, list_node) {
>> +		phb = hose->private_data;
> Move the ohb declaration here and initialize it on the same line.
>
>> +		if (phb->type != PNV_PHB_NPU_NVLINK &&
>> +		    phb->type != PNV_PHB_NPU_OCAPI) {
>> +			if (pnv_pci_controller_owns_addr(hose, addr, size))
>> +				return phb;
>> +		}
>> +	}
>
>> +	return NULL;
>> +}
>> +
>> +static int pnv_pci_dma_map_resource(struct pci_dev *pdev,
>> +				    phys_addr_t phys_addr, size_t size,
>> +				    enum dma_data_direction dir)
>> +{
>> +	struct pnv_phb *target_phb;
>> +	int rc;
>> +	u64 desc;
>> +
>> +	target_phb = pnv_pci_find_owning_phb(pdev, phys_addr, size);
>> +	if (target_phb) {
> Return early here for the !target_phb case?
>
>> +		desc = OPAL_PCI_P2P_ENABLE;
>> +		if (dir == DMA_TO_DEVICE)
>> +			desc |= OPAL_PCI_P2P_STORE;
>> +		else if (dir == DMA_FROM_DEVICE)
>> +			desc |= OPAL_PCI_P2P_LOAD;
>> +		else if (dir == DMA_BIDIRECTIONAL)
>> +			desc |= OPAL_PCI_P2P_LOAD | OPAL_PCI_P2P_STORE;
> Seems like this could be split into a little helper.

sure.


>
>> +		rc = pnv_pci_ioda_set_p2p(pdev, target_phb, desc);
>> +		if (rc) {
>> +			dev_err(&pdev->dev, "Failed to setup PCI peer-to-peer for address %llx: %d\n",
>> +				phys_addr, rc);
>> +			return rc;
>> +		}
> No need for this printk, the callers should already deal with mapping
> failures.
>

On Fri, Apr 17, 2020 at 9:17 PM Max Gurtovoy <maxg@mellanox.com> wrote:
>
> > The enable and disable path shares almost no code, why not split it into
> > two functions?
>
> how about also changing the defines OPAL_PCI_P2P_* to an enum ?
>
> /* PCI p2p operation descriptors */
> enum opal_pci_p2p {
>
>          OPAL_PCI_P2P_DISABLE    = 0,
>
>          OPAL_PCI_P2P_ENABLE     = (1 << 0),
>          OPAL_PCI_P2P_LOAD       = (1 << 1),
>          OPAL_PCI_P2P_STORE      = (1 << 2),
> };
>
> Fred ?

I'd rather you didn't. We try to keep the definitions in opal-api.h
the same as the skiboot's opal-api.h since the skiboot version is
canonical.

Also, generally patches to anything PowerNV related go through the
powerpc tree rather than the pci tree even if they're PCI related. Can
you make sure you have linuxppc-dev CCed when you post v2.

Oliver

On 4/17/2020 5:04 PM, Oliver O'Halloran wrote:
> On Fri, Apr 17, 2020 at 9:17 PM Max Gurtovoy <maxg@mellanox.com> wrote:
>>> The enable and disable path shares almost no code, why not split it into
>>> two functions?
>> how about also changing the defines OPAL_PCI_P2P_* to an enum ?
>>
>> /* PCI p2p operation descriptors */
>> enum opal_pci_p2p {
>>
>>           OPAL_PCI_P2P_DISABLE    = 0,
>>
>>           OPAL_PCI_P2P_ENABLE     = (1 << 0),
>>           OPAL_PCI_P2P_LOAD       = (1 << 1),
>>           OPAL_PCI_P2P_STORE      = (1 << 2),
>> };
>>
>> Fred ?
> I'd rather you didn't. We try to keep the definitions in opal-api.h
> the same as the skiboot's opal-api.h since the skiboot version is
> canonical.
>
> Also, generally patches to anything PowerNV related go through the
> powerpc tree rather than the pci tree even if they're PCI related. Can
> you make sure you have linuxppc-dev CCed when you post v2.

Sure, no problem.


>
> Oliver