On 02/17/2016 02:44 PM, Gavin Shan wrote:
This support releasing PEs dynamically. Firstly, this moves
pnv_pci_ioda2_release_dma_pe() around, which is called to
release DMA resource on releasing IODA2 PE.
imho move would only make sense if we could get rid of the forward
declarations but this is not the case.
Secondly, several
functions are implemented to release the consumed resources
on releasing the PE:
* pnv_pci_ioda1_unset_window() to unset TVEs for the PE.
* pnv_pci_ioda1_release_dma_pe() to unset TVEs for the PE and
destroy the IOMMU table.
* pnv_ioda_release_pe_seg() releases the consumed IO/M32/M64
segments by the PE.
Lastly, this adds a reference count of PE, representing the number
of PCI devices associated with the PE. The reference count is
increased when PCI device joins the PE. It's decreased when PCI
device leaves the PE in pnv_pci_release_device(). When the count
becomes zero, its consumed resources are released by functions
as mentioned above. Note that the count is accessed concurrently.
So a "counter" with "int" type is enough here.
Signed-off-by: Gavin Shan <gwshan@xxxxxxxxxxxxxxxxxx>
---
arch/powerpc/platforms/powernv/pci-ioda.c | 236 ++++++++++++++++++++++++++----
arch/powerpc/platforms/powernv/pci.h | 1 +
2 files changed, 209 insertions(+), 28 deletions(-)
diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
index 077f9db..fa428a8 100644
--- a/arch/powerpc/platforms/powernv/pci-ioda.c
+++ b/arch/powerpc/platforms/powernv/pci-ioda.c
@@ -119,6 +119,158 @@ static inline bool pnv_pci_is_mem_pref_64(unsigned long flags)
(IORESOURCE_MEM_64 | IORESOURCE_PREFETCH));
}
+static unsigned int pnv_pci_ioda_pe_dma_weight(struct pnv_ioda_pe *pe);
+static long pnv_pci_ioda1_unset_window(struct iommu_table_group *table_group,
+ int num);
+static void pnv_pci_ioda1_release_dma_pe(struct pnv_ioda_pe *pe)
+{
+ struct iommu_table *tbl;
+ unsigned int weight = pnv_pci_ioda_pe_dma_weight(pe);
+ int64_t rc;
+
+ if (!weight)
+ return;
+
+ tbl = pe->table_group.tables[0];
+ rc = pnv_pci_ioda1_unset_window(&pe->table_group, 0);
+ if (rc)
+ pe_warn(pe, "OPAL error %ld release DMA window\n", rc);
+
+ if (pe->table_group.group) {
+ iommu_group_put(pe->table_group.group);
+ WARN_ON(pe->table_group.group);
+ }
+
+ pnv_pci_ioda_table_free_pages(tbl);
+ iommu_free_table(tbl, "pnv");
+}
+
+static long pnv_pci_ioda2_unset_window(struct iommu_table_group *table_group,
+ int num);
+static void pnv_pci_ioda2_set_bypass(struct pnv_ioda_pe *pe, bool enable);
+static void pnv_pci_ioda2_release_dma_pe(struct pnv_ioda_pe *pe)
If you left this code in its old location, it would be lot more obvious
what you silently changed in this function (checking for weight). Please
either do not move the code (this is preferred as I am hacking same chunk
in "[PATCH kernel 0/2] powerpc/powernv: Fix crash on PF unbind when VF is
passed" and I'd like to reduce conflicts) or split it to a separate patch.
+{
+ struct iommu_table *tbl;
+ unsigned int weight = pnv_pci_ioda_pe_dma_weight(pe);
+ int64_t rc;
+
+ if (!weight)
+ return;
+
+ tbl = pe->table_group.tables[0];
+ rc = pnv_pci_ioda2_unset_window(&pe->table_group, 0);
+ if (rc)
+ pe_warn(pe, "OPAL error %ld release DMA window\n", rc);
+
+ pnv_pci_ioda2_set_bypass(pe, false);
+ if (pe->table_group.group) {
+ iommu_group_put(pe->table_group.group);
+ WARN_ON(pe->table_group.group);
+ }
+
+ pnv_pci_ioda_table_free_pages(tbl);
+ iommu_free_table(tbl, "pnv");
+}
+
+static void pnv_ioda_release_pe_seg(struct pnv_ioda_pe *pe)
+{
+ struct pnv_phb *phb = pe->phb;
+ int win, index, *segmap = NULL;
+ int64_t rc;
+
+ for (win = OPAL_M32_WINDOW_TYPE; win <= OPAL_IO_WINDOW_TYPE; win++) {
In "Re: [PATCH v7 27/50] powerpc/powernv: Dynamically release PEs" I
suggested shorter & cleaner pnv_ioda_release_window(), what was wrong with it?
+ if (phb->type == PNV_PHB_IODA2 &&
+ (win == OPAL_IO_WINDOW_TYPE || win == OPAL_M64_WINDOW_TYPE))
+ continue;
+
+ switch (win) {
+ case OPAL_IO_WINDOW_TYPE:
+ segmap = phb->ioda.io_segmap;
+ break;
+ case OPAL_M32_WINDOW_TYPE:
+ segmap = phb->ioda.m32_segmap;
+ break;
+ case OPAL_M64_WINDOW_TYPE:
+ segmap = phb->ioda.m64_segmap;
+ break;
+ }
+
+ for (index = 0; index < phb->ioda.total_pe_num; index++) {
+ if (segmap[index] != pe->pe_number)
+ continue;
+
+ if (win == OPAL_M64_WINDOW_TYPE)
+ rc = opal_pci_map_pe_mmio_window(phb->opal_id,
+ phb->ioda.reserved_pe_idx, win,
+ index / PNV_IODA1_M64_SEGS,
+ index % PNV_IODA1_M64_SEGS);
+ else
+ rc = opal_pci_map_pe_mmio_window(phb->opal_id,
+ phb->ioda.reserved_pe_idx, win,
+ 0, index);
+ if (rc != OPAL_SUCCESS)
+ pe_warn(pe, "Error %ld unmapping (%d) segment#%d\n",
+ rc, win, index);
+
+ segmap[index] = IODA_INVALID_PE;
+ }
+ }
+}
+
+static int pnv_ioda_deconfigure_pe(struct pnv_phb *phb,
+ struct pnv_ioda_pe *pe);
+static void pnv_ioda_free_pe(struct pnv_ioda_pe *pe);
+static void pnv_ioda_release_pe(struct pnv_ioda_pe *pe)
+{
+ struct pnv_phb *phb = pe->phb;
+ struct pnv_ioda_pe *tmp, *slave;
+
+ /* Release slave PEs in compound PE */
+ if (pe->flags & PNV_IODA_PE_MASTER) {
+ list_for_each_entry_safe(slave, tmp, &pe->slaves, list)
+ pnv_ioda_release_pe(slave);
+ }
+
+ /* Remove the PE from the list */
+ list_del(&pe->list);
+
+ /* Release DMA segments */
+ switch (phb->type) {
+ case PNV_PHB_IODA1:
+ pnv_pci_ioda1_release_dma_pe(pe);
+ break;
+ case PNV_PHB_IODA2:
+ pnv_pci_ioda2_release_dma_pe(pe);
+ break;
+ default:
+ WARN_ON(1);
+ }
+
+ pnv_ioda_release_pe_seg(pe);
+ pnv_ioda_deconfigure_pe(pe->phb, pe);
+
+ pnv_ioda_free_pe(pe);
+}
+
+static void pnv_pci_release_device(struct pci_dev *pdev)
+{
+ struct pci_controller *hose = pci_bus_to_host(pdev->bus);
+ struct pnv_phb *phb = hose->private_data;
+ struct pci_dn *pdn = pci_get_pdn(pdev);
+ struct pnv_ioda_pe *pe;
+
+ if (pdev->is_virtfn)
+ return;
+
+ if (!pdn || pdn->pe_number == IODA_INVALID_PE)
+ return;
+
+ pe = &phb->ioda.pe_array[pdn->pe_number];
+ WARN_ON(--pe->device_count < 0);
+ if (pe->device_count == 0)
+ pnv_ioda_release_pe(pe);
+}
+
static struct pnv_ioda_pe *pnv_ioda_init_pe(struct pnv_phb *phb, int pe_no)
{
phb->ioda.pe_array[pe_no].phb = phb;
@@ -715,7 +867,6 @@ static int pnv_ioda_set_peltv(struct pnv_phb *phb,
return 0;
}
-#ifdef CONFIG_PCI_IOV
static int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
{
struct pci_dev *parent;
@@ -750,9 +901,11 @@ static int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
}
rid_end = pe->rid + (count << 8);
} else {
+#ifdef CONFIG_PCI_IOV
if (pe->flags & PNV_IODA_PE_VF)
parent = pe->parent_dev;
else
+#endif
parent = pe->pdev->bus->self;
bcomp = OpalPciBusAll;
dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
@@ -790,11 +943,12 @@ static int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
pe->pbus = NULL;
pe->pdev = NULL;
+#ifdef CONFIG_PCI_IOV
pe->parent_dev = NULL;
+#endif
return 0;
}
-#endif /* CONFIG_PCI_IOV */
static int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
{
@@ -1031,6 +1185,7 @@ static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe)
if (pdn->pe_number != IODA_INVALID_PE)
continue;
+ pe->device_count++;
pdn->pcidev = dev;
pdn->pe_number = pe->pe_number;
if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate)
@@ -1095,9 +1250,8 @@ static struct pnv_ioda_pe *pnv_ioda_setup_bus_PE(struct pci_bus *bus, bool all)
bus->busn_res.start, pe->pe_number);
if (pnv_ioda_configure_pe(phb, pe)) {
- /* XXX What do we do here ? */
- pnv_ioda_free_pe(pe);
pe->pbus = NULL;
+ pnv_ioda_release_pe(pe);
return NULL;
}
@@ -1333,29 +1487,6 @@ m64_failed:
return -EBUSY;
}
-static long pnv_pci_ioda2_unset_window(struct iommu_table_group *table_group,
- int num);
-static void pnv_pci_ioda2_set_bypass(struct pnv_ioda_pe *pe, bool enable);
-
-static void pnv_pci_ioda2_release_dma_pe(struct pci_dev *dev, struct pnv_ioda_pe *pe)
-{
- struct iommu_table *tbl;
- int64_t rc;
-
- tbl = pe->table_group.tables[0];
- rc = pnv_pci_ioda2_unset_window(&pe->table_group, 0);
- if (rc)
- pe_warn(pe, "OPAL error %ld release DMA window\n", rc);
-
- pnv_pci_ioda2_set_bypass(pe, false);
- if (pe->table_group.group) {
- iommu_group_put(pe->table_group.group);
- BUG_ON(pe->table_group.group);
- }
- pnv_pci_ioda_table_free_pages(tbl);
- iommu_free_table(tbl, of_node_full_name(dev->dev.of_node));
-}
-
static void pnv_ioda_release_vf_PE(struct pci_dev *pdev)
{
struct pci_bus *bus;
@@ -1376,7 +1507,7 @@ static void pnv_ioda_release_vf_PE(struct pci_dev *pdev)
if (pe->parent_dev != pdev)
continue;
- pnv_pci_ioda2_release_dma_pe(pdev, pe);
+ pnv_pci_ioda2_release_dma_pe(pe);
/* Remove from list */
mutex_lock(&phb->ioda.pe_list_mutex);
@@ -1780,6 +1911,16 @@ static void pnv_pci_ioda1_tce_invalidate(struct iommu_table *tbl,
*/
}
+static void pnv_pci_ioda1_tce_invalidate_entire(struct pnv_ioda_pe *pe)
+{
+ struct iommu_table *tbl = pe->table_group.tables[0];
+
+ if (!tbl)
+ return;
+
+ pnv_pci_ioda1_tce_invalidate(tbl, tbl->it_offset, tbl->it_size, false);
+}
+
static int pnv_ioda1_tce_build(struct iommu_table *tbl, long index,
long npages, unsigned long uaddr,
enum dma_data_direction direction,
@@ -2144,6 +2285,44 @@ static void pnv_pci_ioda1_setup_dma_pe(struct pnv_phb *phb,
}
}
+static long pnv_pci_ioda1_unset_window(struct iommu_table_group *table_group,
+ int num)
+{
+ struct pnv_ioda_pe *pe = container_of(table_group, struct pnv_ioda_pe,
+ table_group);
+ struct pnv_phb *phb = pe->phb;
+ int start, count, i;
+ long rc = OPAL_SUCCESS;
+
+ pe_info(pe, "Removing DMA window #%d\n", num);
+
+ /* Search the used DMA32 segments */
+ start = -1;
+ count = 0;
+ for (i = 0; i < phb->ioda.dma32_count; i++) {
+ if (phb->ioda.dma32_segmap[i] != pe->pe_number)
+ continue;
+
+ if (count++ == 0)
+ start = i;
+ }
+
+ if (!count)
+ return OPAL_SUCCESS;
+
+ for (i = start; i < start + count; i++)
+ rc |= opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number,
+ i, 0, 0ul, 0ul, 0ul);
If there are 2 different bad rc, they will make up absolutely irrelevant
error code.
+ if (rc)
+ pe_warn(pe, "Failure %ld unmapping TVEs\n");
+ else
+ pnv_pci_ioda1_tce_invalidate_entire(pe);
+
+ pnv_pci_unlink_table_and_group(table_group->tables[num], table_group);
+
+ return rc;
+}
+
static long pnv_pci_ioda2_set_window(struct iommu_table_group *table_group,
int num, struct iommu_table *tbl)
{
@@ -3318,6 +3497,7 @@ static const struct pci_controller_ops pnv_pci_ioda_controller_ops = {
.teardown_msi_irqs = pnv_teardown_msi_irqs,
#endif
.enable_device_hook = pnv_pci_enable_device_hook,
+ .release_device = pnv_pci_release_device,
.window_alignment = pnv_pci_window_alignment,
.setup_bridge = pnv_pci_setup_bridge,
.reset_secondary_bus = pnv_pci_reset_secondary_bus,
diff --git a/arch/powerpc/platforms/powernv/pci.h b/arch/powerpc/platforms/powernv/pci.h
index 01f2428..0cddde3 100644
--- a/arch/powerpc/platforms/powernv/pci.h
+++ b/arch/powerpc/platforms/powernv/pci.h
@@ -31,6 +31,7 @@ struct pnv_phb;
struct pnv_ioda_pe {
unsigned long flags;
struct pnv_phb *phb;
+ int device_count;
#define PNV_IODA_MAX_PEER_PES 8
struct pnv_ioda_pe *peers[PNV_IODA_MAX_PEER_PES];
--
Alexey
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
