irqchip: GICv3: ITS: tables allocators
authorMarc Zyngier <marc.zyngier@arm.com>
Mon, 24 Nov 2014 14:35:14 +0000 (14:35 +0000)
committerJason Cooper <jason@lakedaemon.net>
Wed, 26 Nov 2014 15:55:13 +0000 (15:55 +0000)
The interrupt translation is driven by a set of tables (device,
ITT, and collection) to be in the end delivered to a CPU. Also,
the redistributors rely on a couple of tables (configuration, and
pending) to deliver the interrupts to the CPUs.

This patch adds the required allocators for these tables.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Link: https://lkml.kernel.org/r/1416839720-18400-8-git-send-email-marc.zyngier@arm.com
Signed-off-by: Jason Cooper <jason@lakedaemon.net>
drivers/irqchip/irq-gic-v3-its.c

index 4154a16..03f9831 100644 (file)
@@ -91,6 +91,14 @@ struct its_device {
        u32                     device_id;
 };
 
+static LIST_HEAD(its_nodes);
+static DEFINE_SPINLOCK(its_lock);
+static struct device_node *gic_root_node;
+static struct rdists *gic_rdists;
+
+#define gic_data_rdist()               (raw_cpu_ptr(gic_rdists->rdist))
+#define gic_data_rdist_rd_base()       (gic_data_rdist()->rd_base)
+
 /*
  * ITS command descriptors - parameters to be encoded in a command
  * block.
@@ -689,3 +697,287 @@ static void its_lpi_free(unsigned long *bitmap, int base, int nr_ids)
 
        kfree(bitmap);
 }
+
+/*
+ * We allocate 64kB for PROPBASE. That gives us at most 64K LPIs to
+ * deal with (one configuration byte per interrupt). PENDBASE has to
+ * be 64kB aligned (one bit per LPI, plus 8192 bits for SPI/PPI/SGI).
+ */
+#define LPI_PROPBASE_SZ                SZ_64K
+#define LPI_PENDBASE_SZ                (LPI_PROPBASE_SZ / 8 + SZ_1K)
+
+/*
+ * This is how many bits of ID we need, including the useless ones.
+ */
+#define LPI_NRBITS             ilog2(LPI_PROPBASE_SZ + SZ_8K)
+
+#define LPI_PROP_DEFAULT_PRIO  0xa0
+
+static int __init its_alloc_lpi_tables(void)
+{
+       phys_addr_t paddr;
+
+       gic_rdists->prop_page = alloc_pages(GFP_NOWAIT,
+                                          get_order(LPI_PROPBASE_SZ));
+       if (!gic_rdists->prop_page) {
+               pr_err("Failed to allocate PROPBASE\n");
+               return -ENOMEM;
+       }
+
+       paddr = page_to_phys(gic_rdists->prop_page);
+       pr_info("GIC: using LPI property table @%pa\n", &paddr);
+
+       /* Priority 0xa0, Group-1, disabled */
+       memset(page_address(gic_rdists->prop_page),
+              LPI_PROP_DEFAULT_PRIO | LPI_PROP_GROUP1,
+              LPI_PROPBASE_SZ);
+
+       /* Make sure the GIC will observe the written configuration */
+       __flush_dcache_area(page_address(gic_rdists->prop_page), LPI_PROPBASE_SZ);
+
+       return 0;
+}
+
+static const char *its_base_type_string[] = {
+       [GITS_BASER_TYPE_DEVICE]        = "Devices",
+       [GITS_BASER_TYPE_VCPU]          = "Virtual CPUs",
+       [GITS_BASER_TYPE_CPU]           = "Physical CPUs",
+       [GITS_BASER_TYPE_COLLECTION]    = "Interrupt Collections",
+       [GITS_BASER_TYPE_RESERVED5]     = "Reserved (5)",
+       [GITS_BASER_TYPE_RESERVED6]     = "Reserved (6)",
+       [GITS_BASER_TYPE_RESERVED7]     = "Reserved (7)",
+};
+
+static void its_free_tables(struct its_node *its)
+{
+       int i;
+
+       for (i = 0; i < GITS_BASER_NR_REGS; i++) {
+               if (its->tables[i]) {
+                       free_page((unsigned long)its->tables[i]);
+                       its->tables[i] = NULL;
+               }
+       }
+}
+
+static int its_alloc_tables(struct its_node *its)
+{
+       int err;
+       int i;
+       int psz = PAGE_SIZE;
+       u64 shr = GITS_BASER_InnerShareable;
+
+       for (i = 0; i < GITS_BASER_NR_REGS; i++) {
+               u64 val = readq_relaxed(its->base + GITS_BASER + i * 8);
+               u64 type = GITS_BASER_TYPE(val);
+               u64 entry_size = GITS_BASER_ENTRY_SIZE(val);
+               u64 tmp;
+               void *base;
+
+               if (type == GITS_BASER_TYPE_NONE)
+                       continue;
+
+               /* We're lazy and only allocate a single page for now */
+               base = (void *)get_zeroed_page(GFP_KERNEL);
+               if (!base) {
+                       err = -ENOMEM;
+                       goto out_free;
+               }
+
+               its->tables[i] = base;
+
+retry_baser:
+               val = (virt_to_phys(base)                                |
+                      (type << GITS_BASER_TYPE_SHIFT)                   |
+                      ((entry_size - 1) << GITS_BASER_ENTRY_SIZE_SHIFT) |
+                      GITS_BASER_WaWb                                   |
+                      shr                                               |
+                      GITS_BASER_VALID);
+
+               switch (psz) {
+               case SZ_4K:
+                       val |= GITS_BASER_PAGE_SIZE_4K;
+                       break;
+               case SZ_16K:
+                       val |= GITS_BASER_PAGE_SIZE_16K;
+                       break;
+               case SZ_64K:
+                       val |= GITS_BASER_PAGE_SIZE_64K;
+                       break;
+               }
+
+               val |= (PAGE_SIZE / psz) - 1;
+
+               writeq_relaxed(val, its->base + GITS_BASER + i * 8);
+               tmp = readq_relaxed(its->base + GITS_BASER + i * 8);
+
+               if ((val ^ tmp) & GITS_BASER_SHAREABILITY_MASK) {
+                       /*
+                        * Shareability didn't stick. Just use
+                        * whatever the read reported, which is likely
+                        * to be the only thing this redistributor
+                        * supports.
+                        */
+                       shr = tmp & GITS_BASER_SHAREABILITY_MASK;
+                       goto retry_baser;
+               }
+
+               if ((val ^ tmp) & GITS_BASER_PAGE_SIZE_MASK) {
+                       /*
+                        * Page size didn't stick. Let's try a smaller
+                        * size and retry. If we reach 4K, then
+                        * something is horribly wrong...
+                        */
+                       switch (psz) {
+                       case SZ_16K:
+                               psz = SZ_4K;
+                               goto retry_baser;
+                       case SZ_64K:
+                               psz = SZ_16K;
+                               goto retry_baser;
+                       }
+               }
+
+               if (val != tmp) {
+                       pr_err("ITS: %s: GITS_BASER%d doesn't stick: %lx %lx\n",
+                              its->msi_chip.of_node->full_name, i,
+                              (unsigned long) val, (unsigned long) tmp);
+                       err = -ENXIO;
+                       goto out_free;
+               }
+
+               pr_info("ITS: allocated %d %s @%lx (psz %dK, shr %d)\n",
+                       (int)(PAGE_SIZE / entry_size),
+                       its_base_type_string[type],
+                       (unsigned long)virt_to_phys(base),
+                       psz / SZ_1K, (int)shr >> GITS_BASER_SHAREABILITY_SHIFT);
+       }
+
+       return 0;
+
+out_free:
+       its_free_tables(its);
+
+       return err;
+}
+
+static int its_alloc_collections(struct its_node *its)
+{
+       its->collections = kzalloc(nr_cpu_ids * sizeof(*its->collections),
+                                  GFP_KERNEL);
+       if (!its->collections)
+               return -ENOMEM;
+
+       return 0;
+}
+
+static void its_cpu_init_lpis(void)
+{
+       void __iomem *rbase = gic_data_rdist_rd_base();
+       struct page *pend_page;
+       u64 val, tmp;
+
+       /* If we didn't allocate the pending table yet, do it now */
+       pend_page = gic_data_rdist()->pend_page;
+       if (!pend_page) {
+               phys_addr_t paddr;
+               /*
+                * The pending pages have to be at least 64kB aligned,
+                * hence the 'max(LPI_PENDBASE_SZ, SZ_64K)' below.
+                */
+               pend_page = alloc_pages(GFP_NOWAIT | __GFP_ZERO,
+                                       get_order(max(LPI_PENDBASE_SZ, SZ_64K)));
+               if (!pend_page) {
+                       pr_err("Failed to allocate PENDBASE for CPU%d\n",
+                              smp_processor_id());
+                       return;
+               }
+
+               /* Make sure the GIC will observe the zero-ed page */
+               __flush_dcache_area(page_address(pend_page), LPI_PENDBASE_SZ);
+
+               paddr = page_to_phys(pend_page);
+               pr_info("CPU%d: using LPI pending table @%pa\n",
+                       smp_processor_id(), &paddr);
+               gic_data_rdist()->pend_page = pend_page;
+       }
+
+       /* Disable LPIs */
+       val = readl_relaxed(rbase + GICR_CTLR);
+       val &= ~GICR_CTLR_ENABLE_LPIS;
+       writel_relaxed(val, rbase + GICR_CTLR);
+
+       /*
+        * Make sure any change to the table is observable by the GIC.
+        */
+       dsb(sy);
+
+       /* set PROPBASE */
+       val = (page_to_phys(gic_rdists->prop_page) |
+              GICR_PROPBASER_InnerShareable |
+              GICR_PROPBASER_WaWb |
+              ((LPI_NRBITS - 1) & GICR_PROPBASER_IDBITS_MASK));
+
+       writeq_relaxed(val, rbase + GICR_PROPBASER);
+       tmp = readq_relaxed(rbase + GICR_PROPBASER);
+
+       if ((tmp ^ val) & GICR_PROPBASER_SHAREABILITY_MASK) {
+               pr_info_once("GIC: using cache flushing for LPI property table\n");
+               gic_rdists->flags |= RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING;
+       }
+
+       /* set PENDBASE */
+       val = (page_to_phys(pend_page) |
+              GICR_PROPBASER_InnerShareable |
+              GICR_PROPBASER_WaWb);
+
+       writeq_relaxed(val, rbase + GICR_PENDBASER);
+
+       /* Enable LPIs */
+       val = readl_relaxed(rbase + GICR_CTLR);
+       val |= GICR_CTLR_ENABLE_LPIS;
+       writel_relaxed(val, rbase + GICR_CTLR);
+
+       /* Make sure the GIC has seen the above */
+       dsb(sy);
+}
+
+static void its_cpu_init_collection(void)
+{
+       struct its_node *its;
+       int cpu;
+
+       spin_lock(&its_lock);
+       cpu = smp_processor_id();
+
+       list_for_each_entry(its, &its_nodes, entry) {
+               u64 target;
+
+               /*
+                * We now have to bind each collection to its target
+                * redistributor.
+                */
+               if (readq_relaxed(its->base + GITS_TYPER) & GITS_TYPER_PTA) {
+                       /*
+                        * This ITS wants the physical address of the
+                        * redistributor.
+                        */
+                       target = gic_data_rdist()->phys_base;
+               } else {
+                       /*
+                        * This ITS wants a linear CPU number.
+                        */
+                       target = readq_relaxed(gic_data_rdist_rd_base() + GICR_TYPER);
+                       target = GICR_TYPER_CPU_NUMBER(target);
+               }
+
+               /* Perform collection mapping */
+               its->collections[cpu].target_address = target;
+               its->collections[cpu].col_id = cpu;
+
+               its_send_mapc(its, &its->collections[cpu], 1);
+               its_send_invall(its, &its->collections[cpu]);
+       }
+
+       spin_unlock(&its_lock);
+}