@@ -14,6 +14,19 @@
#include "iommu-pasid.h"
+/*
+ * Linear: when less than 1024 SSIDs are supported
+ * 2lvl: at most 1024 L1 entrie,
+ * 1024 lazy entries per table.
+ */
+#define CTXDESC_SPLIT 10
+#define CTXDESC_NUM_L2_ENTRIES (1 << CTXDESC_SPLIT)
+
+#define CTXDESC_L1_DESC_DWORD 1
+#define CTXDESC_L1_DESC_VALID 1
+#define CTXDESC_L1_DESC_L2PTR_SHIFT 12
+#define CTXDESC_L1_DESC_L2PTR_MASK 0xfffffffffUL
+
#define CTXDESC_CD_DWORDS 8
#define CTXDESC_CD_0_TCR_T0SZ_SHIFT 0
#define ARM64_TCR_T0SZ_SHIFT 0
@@ -86,7 +99,17 @@ struct arm_smmu_cd_table {
struct arm_smmu_cd_tables {
struct iommu_pasid_table pasid;
- struct arm_smmu_cd_table table;
+ bool linear;
+ union {
+ struct arm_smmu_cd_table table;
+ struct {
+ __le64 *ptr;
+ dma_addr_t ptr_dma;
+ size_t num_entries;
+
+ struct arm_smmu_cd_table *tables;
+ } l1;
+ };
};
#define pasid_to_cd_tables(pasid_table) \
@@ -122,9 +145,44 @@ static void arm_smmu_free_cd_leaf_table(struct device *dev,
dmam_free_coherent(dev, size, desc->ptr, desc->ptr_dma);
}
+static void arm_smmu_write_cd_l1_desc(__le64 *dst,
+ struct arm_smmu_cd_table *desc)
+{
+ u64 val = (desc->ptr_dma & CTXDESC_L1_DESC_L2PTR_MASK <<
+ CTXDESC_L1_DESC_L2PTR_SHIFT) | CTXDESC_L1_DESC_VALID;
+
+ *dst = cpu_to_le64(val);
+}
+
static __le64 *arm_smmu_get_cd_ptr(struct arm_smmu_cd_tables *tbl, u32 ssid)
{
- return tbl->table.ptr + ssid * CTXDESC_CD_DWORDS;
+ unsigned long idx;
+ struct arm_smmu_cd_table *l1_desc;
+ struct iommu_pasid_table_cfg *cfg = &tbl->pasid.cfg;
+
+ if (tbl->linear)
+ return tbl->table.ptr + ssid * CTXDESC_CD_DWORDS;
+
+ idx = ssid >> CTXDESC_SPLIT;
+ if (idx >= tbl->l1.num_entries)
+ return NULL;
+
+ l1_desc = &tbl->l1.tables[idx];
+ if (!l1_desc->ptr) {
+ __le64 *l1ptr = tbl->l1.ptr + idx * CTXDESC_L1_DESC_DWORD;
+
+ if (arm_smmu_alloc_cd_leaf_table(cfg->iommu_dev, l1_desc,
+ CTXDESC_NUM_L2_ENTRIES))
+ return NULL;
+
+ arm_smmu_write_cd_l1_desc(l1ptr, l1_desc);
+ /* An invalid L1 entry is allowed to be cached */
+ iommu_pasid_flush(&tbl->pasid, idx << CTXDESC_SPLIT, false);
+ }
+
+ idx = ssid & (CTXDESC_NUM_L2_ENTRIES - 1);
+
+ return l1_desc->ptr + idx * CTXDESC_CD_DWORDS;
}
static u64 arm_smmu_cpu_tcr_to_cd(u64 tcr)
@@ -307,16 +365,51 @@ static struct iommu_pasid_table *
arm_smmu_alloc_cd_tables(struct iommu_pasid_table_cfg *cfg, void *cookie)
{
int ret;
+ size_t size = 0;
struct arm_smmu_cd_tables *tbl;
struct device *dev = cfg->iommu_dev;
+ struct arm_smmu_cd_table *leaf_table;
+ size_t num_contexts, num_leaf_entries;
tbl = devm_kzalloc(dev, sizeof(*tbl), GFP_KERNEL);
if (!tbl)
return NULL;
- ret = arm_smmu_alloc_cd_leaf_table(dev, &tbl->table, 1 << cfg->order);
+ num_contexts = 1 << cfg->order;
+ if (num_contexts <= CTXDESC_NUM_L2_ENTRIES) {
+ /* Fits in a single table */
+ tbl->linear = true;
+ num_leaf_entries = num_contexts;
+ leaf_table = &tbl->table;
+ } else {
+ /*
+ * SSID[S1CDmax-1:10] indexes 1st-level table, SSID[9:0] indexes
+ * 2nd-level
+ */
+ tbl->l1.num_entries = num_contexts / CTXDESC_NUM_L2_ENTRIES;
+
+ tbl->l1.tables = devm_kzalloc(dev,
+ sizeof(struct arm_smmu_cd_table) *
+ tbl->l1.num_entries, GFP_KERNEL);
+ if (!tbl->l1.tables)
+ goto err_free_tbl;
+
+ size = tbl->l1.num_entries * (CTXDESC_L1_DESC_DWORD << 3);
+ tbl->l1.ptr = dmam_alloc_coherent(dev, size, &tbl->l1.ptr_dma,
+ GFP_KERNEL | __GFP_ZERO);
+ if (!tbl->l1.ptr) {
+ dev_warn(dev, "failed to allocate L1 context table\n");
+ devm_kfree(dev, tbl->l1.tables);
+ goto err_free_tbl;
+ }
+
+ num_leaf_entries = CTXDESC_NUM_L2_ENTRIES;
+ leaf_table = tbl->l1.tables;
+ }
+
+ ret = arm_smmu_alloc_cd_leaf_table(dev, leaf_table, num_leaf_entries);
if (ret)
- goto err_free_tbl;
+ goto err_free_l1;
tbl->pasid.ops = (struct iommu_pasid_table_ops) {
.alloc_priv_entry = arm_smmu_alloc_priv_cd,
@@ -326,11 +419,22 @@ arm_smmu_alloc_cd_tables(struct iommu_pasid_table_cfg *cfg, void *cookie)
.clear_entry = arm_smmu_clear_cd,
};
- cfg->base = tbl->table.ptr_dma;
- cfg->arm_smmu.s1fmt = ARM_SMMU_S1FMT_LINEAR;
+ if (tbl->linear) {
+ cfg->base = leaf_table->ptr_dma;
+ cfg->arm_smmu.s1fmt = ARM_SMMU_S1FMT_LINEAR;
+ } else {
+ cfg->base = tbl->l1.ptr_dma;
+ cfg->arm_smmu.s1fmt = ARM_SMMU_S1FMT_64K_L2;
+ arm_smmu_write_cd_l1_desc(tbl->l1.ptr, leaf_table);
+ }
return &tbl->pasid;
+err_free_l1:
+ if (!tbl->linear) {
+ dmam_free_coherent(dev, size, tbl->l1.ptr, tbl->l1.ptr_dma);
+ devm_kfree(dev, tbl->l1.tables);
+ }
err_free_tbl:
devm_kfree(dev, tbl);
@@ -343,7 +447,26 @@ static void arm_smmu_free_cd_tables(struct iommu_pasid_table *pasid_table)
struct device *dev = cfg->iommu_dev;
struct arm_smmu_cd_tables *tbl = pasid_to_cd_tables(pasid_table);
- arm_smmu_free_cd_leaf_table(dev, &tbl->table, 1 << cfg->order);
+ if (tbl->linear) {
+ arm_smmu_free_cd_leaf_table(dev, &tbl->table, 1 << cfg->order);
+ } else {
+ size_t i, size;
+
+ for (i = 0; i < tbl->l1.num_entries; i++) {
+ struct arm_smmu_cd_table *table = &tbl->l1.tables[i];
+
+ if (!table->ptr)
+ continue;
+
+ arm_smmu_free_cd_leaf_table(dev, table,
+ CTXDESC_NUM_L2_ENTRIES);
+ }
+
+ size = tbl->l1.num_entries * (CTXDESC_L1_DESC_DWORD << 3);
+ dmam_free_coherent(dev, size, tbl->l1.ptr, tbl->l1.ptr_dma);
+ devm_kfree(dev, tbl->l1.tables);
+ }
+
devm_kfree(dev, tbl);
}
The SMMU can support up to 20 bits of SSID. Add a second level of page tables to accommodate this. Devices that support more than 1024 SSIDs now have a table of 1024 L1 entries (8kB), pointing to tables of 1024 context descriptors (64kB), allocated on demand. Signed-off-by: Jean-Philippe Brucker <jean-philippe.brucker@arm.com> --- drivers/iommu/arm-smmu-v3-context.c | 137 ++++++++++++++++++++++++++++++++++-- 1 file changed, 130 insertions(+), 7 deletions(-)