[PATCH 1/2] habanalabs: split the host MMU properties

From: Omer Shpigelman
Date: Sun Jan 05 2020 - 04:06:36 EST

Host memory may be alloacted with huge pages.
A different virtual range may be used for mapping in this case.
Add Huge PCI MMU (HPMMU) properties to support it.
This patch is a prerequisite for future ASICs support and has no effect on
Goya ASIC as currently a single virtual host range is used for all page
sizes.

Signed-off-by: Omer Shpigelman <oshpigelman@xxxxxxxxx>
---
drivers/misc/habanalabs/debugfs.c | 21 +-
drivers/misc/habanalabs/goya/goya.c | 25 +-
drivers/misc/habanalabs/goya/goya_coresight.c | 4 +-
drivers/misc/habanalabs/habanalabs.h | 31 ++-
drivers/misc/habanalabs/memory.c | 213 ++++++++++++------
drivers/misc/habanalabs/mmu.c | 68 +++---
6 files changed, 225 insertions(+), 137 deletions(-)

diff --git a/drivers/misc/habanalabs/debugfs.c b/drivers/misc/habanalabs/debugfs.c
index 20413e350343..599d17dfd542 100644
--- a/drivers/misc/habanalabs/debugfs.c
+++ b/drivers/misc/habanalabs/debugfs.c
@@ -393,9 +393,10 @@ static int mmu_show(struct seq_file *s, void *data)
}

is_dram_addr = hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
- prop->va_space_dram_start_address,
- prop->va_space_dram_end_address);
+ prop->dmmu.start_addr,
+ prop->dmmu.end_addr);

+ /* shifts and masks are the same in PMMU and HPMMU, use one of them */
mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;

mutex_lock(&ctx->mmu_lock);
@@ -547,12 +548,15 @@ static bool hl_is_device_va(struct hl_device *hdev, u64 addr)
goto out;

if (hdev->dram_supports_virtual_memory &&
- addr >= prop->va_space_dram_start_address &&
- addr < prop->va_space_dram_end_address)
+ (addr >= prop->dmmu.start_addr && addr < prop->dmmu.end_addr))
return true;

- if (addr >= prop->va_space_host_start_address &&
- addr < prop->va_space_host_end_address)
+ if (addr >= prop->pmmu.start_addr &&
+ addr < prop->pmmu.end_addr)
+ return true;
+
+ if (addr >= prop->pmmu_huge.start_addr &&
+ addr < prop->pmmu_huge.end_addr)
return true;
out:
return false;
@@ -575,9 +579,10 @@ static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr,
}

is_dram_addr = hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
- prop->va_space_dram_start_address,
- prop->va_space_dram_end_address);
+ prop->dmmu.start_addr,
+ prop->dmmu.end_addr);

+ /* shifts and masks are the same in PMMU and HPMMU, use one of them */
mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;

mutex_lock(&ctx->mmu_lock);
diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c
index 5750294400f4..382331cc00d3 100644
--- a/drivers/misc/habanalabs/goya/goya.c
+++ b/drivers/misc/habanalabs/goya/goya.c
@@ -393,19 +393,21 @@ void goya_get_fixed_properties(struct hl_device *hdev)
prop->dmmu.hop2_mask = HOP2_MASK;
prop->dmmu.hop3_mask = HOP3_MASK;
prop->dmmu.hop4_mask = HOP4_MASK;
- prop->dmmu.huge_page_size = PAGE_SIZE_2MB;
+ prop->dmmu.start_addr = VA_DDR_SPACE_START;
+ prop->dmmu.end_addr = VA_DDR_SPACE_END;
+ prop->dmmu.page_size = PAGE_SIZE_2MB;

- /* No difference between PMMU and DMMU except of page size */
+ /* shifts and masks are the same in PMMU and DMMU */
memcpy(&prop->pmmu, &prop->dmmu, sizeof(prop->dmmu));
- prop->dmmu.page_size = PAGE_SIZE_2MB;
+ prop->pmmu.start_addr = VA_HOST_SPACE_START;
+ prop->pmmu.end_addr = VA_HOST_SPACE_END;
prop->pmmu.page_size = PAGE_SIZE_4KB;

- prop->va_space_host_start_address = VA_HOST_SPACE_START;
- prop->va_space_host_end_address = VA_HOST_SPACE_END;
- prop->va_space_dram_start_address = VA_DDR_SPACE_START;
- prop->va_space_dram_end_address = VA_DDR_SPACE_END;
- prop->dram_size_for_default_page_mapping =
- prop->va_space_dram_end_address;
+ /* PMMU and HPMMU are the same except of page size */
+ memcpy(&prop->pmmu_huge, &prop->pmmu, sizeof(prop->pmmu));
+ prop->pmmu_huge.page_size = PAGE_SIZE_2MB;
+
+ prop->dram_size_for_default_page_mapping = VA_DDR_SPACE_END;
prop->cfg_size = CFG_SIZE;
prop->max_asid = MAX_ASID;
prop->num_of_events = GOYA_ASYNC_EVENT_ID_SIZE;
@@ -3443,12 +3445,13 @@ static int goya_validate_dma_pkt_mmu(struct hl_device *hdev,
/*
* WA for HW-23.
* We can't allow user to read from Host using QMANs other than 1.
+ * PMMU and HPMMU addresses are equal, check only one of them.
*/
if (parser->hw_queue_id != GOYA_QUEUE_ID_DMA_1 &&
hl_mem_area_inside_range(le64_to_cpu(user_dma_pkt->src_addr),
le32_to_cpu(user_dma_pkt->tsize),
- hdev->asic_prop.va_space_host_start_address,
- hdev->asic_prop.va_space_host_end_address)) {
+ hdev->asic_prop.pmmu.start_addr,
+ hdev->asic_prop.pmmu.end_addr)) {
dev_err(hdev->dev,
"Can't DMA from host on queue other then 1\n");
return -EFAULT;
diff --git a/drivers/misc/habanalabs/goya/goya_coresight.c b/drivers/misc/habanalabs/goya/goya_coresight.c
index c1ee6e2b5dff..a1bc930d904f 100644
--- a/drivers/misc/habanalabs/goya/goya_coresight.c
+++ b/drivers/misc/habanalabs/goya/goya_coresight.c
@@ -364,8 +364,8 @@ static int goya_etr_validate_address(struct hl_device *hdev, u64 addr,
u64 range_start, range_end;

if (hdev->mmu_enable) {
- range_start = prop->va_space_dram_start_address;
- range_end = prop->va_space_dram_end_address;
+ range_start = prop->dmmu.start_addr;
+ range_end = prop->dmmu.end_addr;
} else {
range_start = prop->dram_user_base_address;
range_end = prop->dram_end_address;
diff --git a/drivers/misc/habanalabs/habanalabs.h b/drivers/misc/habanalabs/habanalabs.h
index df34227dea31..5c751b9517c0 100644
--- a/drivers/misc/habanalabs/habanalabs.h
+++ b/drivers/misc/habanalabs/habanalabs.h
@@ -132,6 +132,8 @@ enum hl_device_hw_state {

/**
* struct hl_mmu_properties - ASIC specific MMU address translation properties.
+ * @start_addr: virtual start address of the memory region.
+ * @end_addr: virtual end address of the memory region.
* @hop0_shift: shift of hop 0 mask.
* @hop1_shift: shift of hop 1 mask.
* @hop2_shift: shift of hop 2 mask.
@@ -143,9 +145,10 @@ enum hl_device_hw_state {
* @hop3_mask: mask to get the PTE address in hop 3.
* @hop4_mask: mask to get the PTE address in hop 4.
* @page_size: default page size used to allocate memory.
- * @huge_page_size: page size used to allocate memory with huge pages.
*/
struct hl_mmu_properties {
+ u64 start_addr;
+ u64 end_addr;
u64 hop0_shift;
u64 hop1_shift;
u64 hop2_shift;
@@ -157,7 +160,6 @@ struct hl_mmu_properties {
u64 hop3_mask;
u64 hop4_mask;
u32 page_size;
- u32 huge_page_size;
};

/**
@@ -169,6 +171,8 @@ struct hl_mmu_properties {
* @preboot_ver: F/W Preboot version.
* @dmmu: DRAM MMU address translation properties.
* @pmmu: PCI (host) MMU address translation properties.
+ * @pmmu_huge: PCI (host) MMU address translation properties for memory
+ * allocated with huge pages.
* @sram_base_address: SRAM physical start address.
* @sram_end_address: SRAM physical end address.
* @sram_user_base_address - SRAM physical start address for user access.
@@ -178,14 +182,6 @@ struct hl_mmu_properties {
* @dram_size: DRAM total size.
* @dram_pci_bar_size: size of PCI bar towards DRAM.
* @max_power_default: max power of the device after reset
- * @va_space_host_start_address: base address of virtual memory range for
- * mapping host memory.
- * @va_space_host_end_address: end address of virtual memory range for
- * mapping host memory.
- * @va_space_dram_start_address: base address of virtual memory range for
- * mapping DRAM memory.
- * @va_space_dram_end_address: end address of virtual memory range for
- * mapping DRAM memory.
* @dram_size_for_default_page_mapping: DRAM size needed to map to avoid page
* fault.
* @pcie_dbi_base_address: Base address of the PCIE_DBI block.
@@ -218,6 +214,7 @@ struct asic_fixed_properties {
char preboot_ver[VERSION_MAX_LEN];
struct hl_mmu_properties dmmu;
struct hl_mmu_properties pmmu;
+ struct hl_mmu_properties pmmu_huge;
u64 sram_base_address;
u64 sram_end_address;
u64 sram_user_base_address;
@@ -227,10 +224,6 @@ struct asic_fixed_properties {
u64 dram_size;
u64 dram_pci_bar_size;
u64 max_power_default;
- u64 va_space_host_start_address;
- u64 va_space_host_end_address;
- u64 va_space_dram_start_address;
- u64 va_space_dram_end_address;
u64 dram_size_for_default_page_mapping;
u64 pcie_dbi_base_address;
u64 pcie_aux_dbi_reg_addr;
@@ -658,6 +651,8 @@ struct hl_va_range {
* this hits 0l. It is incremented on CS and CS_WAIT.
* @cs_pending: array of DMA fence objects representing pending CS.
* @host_va_range: holds available virtual addresses for host mappings.
+ * @host_huge_va_range: holds available virtual addresses for host mappings
+ * with huge pages.
* @dram_va_range: holds available virtual addresses for DRAM mappings.
* @mem_hash_lock: protects the mem_hash.
* @mmu_lock: protects the MMU page tables. Any change to the PGT, modifing the
@@ -688,8 +683,9 @@ struct hl_ctx {
struct hl_device *hdev;
struct kref refcount;
struct dma_fence *cs_pending[HL_MAX_PENDING_CS];
- struct hl_va_range host_va_range;
- struct hl_va_range dram_va_range;
+ struct hl_va_range *host_va_range;
+ struct hl_va_range *host_huge_va_range;
+ struct hl_va_range *dram_va_range;
struct mutex mem_hash_lock;
struct mutex mmu_lock;
struct list_head debugfs_list;
@@ -1291,6 +1287,8 @@ struct hl_device_idle_busy_ts {
* otherwise.
* @dram_supports_virtual_memory: is MMU enabled towards DRAM.
* @dram_default_page_mapping: is DRAM default page mapping enabled.
+ * @pmmu_huge_range: is a different virtual addresses range used for PMMU with
+ * huge pages.
* @init_done: is the initialization of the device done.
* @mmu_enable: is MMU enabled.
* @device_cpu_disabled: is the device CPU disabled (due to timeouts)
@@ -1372,6 +1370,7 @@ struct hl_device {
u8 reset_on_lockup;
u8 dram_supports_virtual_memory;
u8 dram_default_page_mapping;
+ u8 pmmu_huge_range;
u8 init_done;
u8 device_cpu_disabled;
u8 dma_mask;
diff --git a/drivers/misc/habanalabs/memory.c b/drivers/misc/habanalabs/memory.c
index b612b1ad0aac..a72f766ca470 100644
--- a/drivers/misc/habanalabs/memory.c
+++ b/drivers/misc/habanalabs/memory.c
@@ -530,7 +530,7 @@ static u64 get_va_block(struct hl_device *hdev,
* or not, hence we continue with the biggest possible
* granularity.
*/
- page_size = hdev->asic_prop.pmmu.huge_page_size;
+ page_size = hdev->asic_prop.pmmu_huge.page_size;
else
page_size = hdev->asic_prop.dmmu.page_size;

@@ -638,13 +638,12 @@ static int init_phys_pg_pack_from_userptr(struct hl_ctx *ctx,
struct hl_userptr *userptr,
struct hl_vm_phys_pg_pack **pphys_pg_pack)
{
- struct hl_mmu_properties *mmu_prop = &ctx->hdev->asic_prop.pmmu;
struct hl_vm_phys_pg_pack *phys_pg_pack;
struct scatterlist *sg;
dma_addr_t dma_addr;
u64 page_mask, total_npages;
u32 npages, page_size = PAGE_SIZE,
- huge_page_size = mmu_prop->huge_page_size;
+ huge_page_size = ctx->hdev->asic_prop.pmmu_huge.page_size;
bool first = true, is_huge_page_opt = true;
int rc, i, j;
u32 pgs_in_huge_page = huge_page_size >> __ffs(page_size);
@@ -856,6 +855,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
struct hl_vm_phys_pg_pack *phys_pg_pack;
struct hl_userptr *userptr = NULL;
struct hl_vm_hash_node *hnode;
+ struct hl_va_range *va_range;
enum vm_type_t *vm_type;
u64 ret_vaddr, hint_addr;
u32 handle = 0;
@@ -927,9 +927,16 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
goto hnode_err;
}

- ret_vaddr = get_va_block(hdev,
- is_userptr ? &ctx->host_va_range : &ctx->dram_va_range,
- phys_pg_pack->total_size, hint_addr, is_userptr);
+ if (is_userptr)
+ if (phys_pg_pack->page_size == hdev->asic_prop.pmmu.page_size)
+ va_range = ctx->host_va_range;
+ else
+ va_range = ctx->host_huge_va_range;
+ else
+ va_range = ctx->dram_va_range;
+
+ ret_vaddr = get_va_block(hdev, va_range, phys_pg_pack->total_size,
+ hint_addr, is_userptr);
if (!ret_vaddr) {
dev_err(hdev->dev, "no available va block for handle %u\n",
handle);
@@ -968,10 +975,8 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
return 0;

map_err:
- if (add_va_block(hdev,
- is_userptr ? &ctx->host_va_range : &ctx->dram_va_range,
- ret_vaddr,
- ret_vaddr + phys_pg_pack->total_size - 1))
+ if (add_va_block(hdev, va_range, ret_vaddr,
+ ret_vaddr + phys_pg_pack->total_size - 1))
dev_warn(hdev->dev,
"release va block failed for handle 0x%x, vaddr: 0x%llx\n",
handle, ret_vaddr);
@@ -1033,7 +1038,6 @@ static int unmap_device_va(struct hl_ctx *ctx, u64 vaddr, bool ctx_free)

if (*vm_type == VM_TYPE_USERPTR) {
is_userptr = true;
- va_range = &ctx->host_va_range;
userptr = hnode->ptr;
rc = init_phys_pg_pack_from_userptr(ctx, userptr,
&phys_pg_pack);
@@ -1043,9 +1047,15 @@ static int unmap_device_va(struct hl_ctx *ctx, u64 vaddr, bool ctx_free)
vaddr);
goto vm_type_err;
}
+
+ if (phys_pg_pack->page_size ==
+ hdev->asic_prop.pmmu.page_size)
+ va_range = ctx->host_va_range;
+ else
+ va_range = ctx->host_huge_va_range;
} else if (*vm_type == VM_TYPE_PHYS_PACK) {
is_userptr = false;
- va_range = &ctx->dram_va_range;
+ va_range = ctx->dram_va_range;
phys_pg_pack = hnode->ptr;
} else {
dev_warn(hdev->dev,
@@ -1441,19 +1451,18 @@ bool hl_userptr_is_pinned(struct hl_device *hdev, u64 addr,
}

/*
- * hl_va_range_init - initialize virtual addresses range
- *
- * @hdev : pointer to the habanalabs device structure
- * @va_range : pointer to the range to initialize
- * @start : range start address
- * @end : range end address
+ * va_range_init - initialize virtual addresses range
+ * @hdev: pointer to the habanalabs device structure
+ * @va_range: pointer to the range to initialize
+ * @start: range start address
+ * @end: range end address
*
* This function does the following:
* - Initializes the virtual addresses list of the given range with the given
* addresses.
*/
-static int hl_va_range_init(struct hl_device *hdev,
- struct hl_va_range *va_range, u64 start, u64 end)
+static int va_range_init(struct hl_device *hdev, struct hl_va_range *va_range,
+ u64 start, u64 end)
{
int rc;

@@ -1488,47 +1497,105 @@ static int hl_va_range_init(struct hl_device *hdev,
}

/*
- * hl_vm_ctx_init_with_ranges - initialize virtual memory for context
+ * va_range_fini() - clear a virtual addresses range
+ * @hdev: pointer to the habanalabs structure
+ * va_range: pointer to virtual addresses range
*
- * @ctx : pointer to the habanalabs context structure
- * @host_range_start : host virtual addresses range start
- * @host_range_end : host virtual addresses range end
- * @dram_range_start : dram virtual addresses range start
- * @dram_range_end : dram virtual addresses range end
+ * This function does the following:
+ * - Frees the virtual addresses block list and its lock
+ */
+static void va_range_fini(struct hl_device *hdev,
+ struct hl_va_range *va_range)
+{
+ mutex_lock(&va_range->lock);
+ clear_va_list_locked(hdev, &va_range->list);
+ mutex_unlock(&va_range->lock);
+
+ mutex_destroy(&va_range->lock);
+ kfree(va_range);
+}
+
+/*
+ * vm_ctx_init_with_ranges() - initialize virtual memory for context
+ * @ctx: pointer to the habanalabs context structure
+ * @host_range_start: host virtual addresses range start.
+ * @host_range_end: host virtual addresses range end.
+ * @host_huge_range_start: host virtual addresses range start for memory
+ * allocated with huge pages.
+ * @host_huge_range_end: host virtual addresses range end for memory allocated
+ * with huge pages.
+ * @dram_range_start: dram virtual addresses range start.
+ * @dram_range_end: dram virtual addresses range end.
*
* This function initializes the following:
* - MMU for context
* - Virtual address to area descriptor hashtable
* - Virtual block list of available virtual memory
*/
-static int hl_vm_ctx_init_with_ranges(struct hl_ctx *ctx, u64 host_range_start,
- u64 host_range_end, u64 dram_range_start,
- u64 dram_range_end)
+static int vm_ctx_init_with_ranges(struct hl_ctx *ctx,
+ u64 host_range_start,
+ u64 host_range_end,
+ u64 host_huge_range_start,
+ u64 host_huge_range_end,
+ u64 dram_range_start,
+ u64 dram_range_end)
{
struct hl_device *hdev = ctx->hdev;
int rc;

+ ctx->host_va_range = kzalloc(sizeof(*ctx->host_va_range), GFP_KERNEL);
+ if (!ctx->host_va_range)
+ return -ENOMEM;
+
+ ctx->host_huge_va_range = kzalloc(sizeof(*ctx->host_huge_va_range),
+ GFP_KERNEL);
+ if (!ctx->host_huge_va_range) {
+ rc = -ENOMEM;
+ goto host_huge_va_range_err;
+ }
+
+ ctx->dram_va_range = kzalloc(sizeof(*ctx->dram_va_range), GFP_KERNEL);
+ if (!ctx->dram_va_range) {
+ rc = -ENOMEM;
+ goto dram_va_range_err;
+ }
+
rc = hl_mmu_ctx_init(ctx);
if (rc) {
dev_err(hdev->dev, "failed to init context %d\n", ctx->asid);
- return rc;
+ goto mmu_ctx_err;
}

mutex_init(&ctx->mem_hash_lock);
hash_init(ctx->mem_hash);

- mutex_init(&ctx->host_va_range.lock);
+ mutex_init(&ctx->host_va_range->lock);

- rc = hl_va_range_init(hdev, &ctx->host_va_range, host_range_start,
- host_range_end);
+ rc = va_range_init(hdev, ctx->host_va_range, host_range_start,
+ host_range_end);
if (rc) {
dev_err(hdev->dev, "failed to init host vm range\n");
- goto host_vm_err;
+ goto host_page_range_err;
+ }
+
+ if (hdev->pmmu_huge_range) {
+ mutex_init(&ctx->host_huge_va_range->lock);
+
+ rc = va_range_init(hdev, ctx->host_huge_va_range,
+ host_huge_range_start,
+ host_huge_range_end);
+ if (rc) {
+ dev_err(hdev->dev,
+ "failed to init host huge vm range\n");
+ goto host_hpage_range_err;
+ }
+ } else {
+ ctx->host_huge_va_range = ctx->host_va_range;
}

- mutex_init(&ctx->dram_va_range.lock);
+ mutex_init(&ctx->dram_va_range->lock);

- rc = hl_va_range_init(hdev, &ctx->dram_va_range, dram_range_start,
+ rc = va_range_init(hdev, ctx->dram_va_range, dram_range_start,
dram_range_end);
if (rc) {
dev_err(hdev->dev, "failed to init dram vm range\n");
@@ -1540,15 +1607,29 @@ static int hl_vm_ctx_init_with_ranges(struct hl_ctx *ctx, u64 host_range_start,
return 0;

dram_vm_err:
- mutex_destroy(&ctx->dram_va_range.lock);
+ mutex_destroy(&ctx->dram_va_range->lock);

- mutex_lock(&ctx->host_va_range.lock);
- clear_va_list_locked(hdev, &ctx->host_va_range.list);
- mutex_unlock(&ctx->host_va_range.lock);
-host_vm_err:
- mutex_destroy(&ctx->host_va_range.lock);
+ if (hdev->pmmu_huge_range) {
+ mutex_lock(&ctx->host_huge_va_range->lock);
+ clear_va_list_locked(hdev, &ctx->host_huge_va_range->list);
+ mutex_unlock(&ctx->host_huge_va_range->lock);
+ }
+host_hpage_range_err:
+ if (hdev->pmmu_huge_range)
+ mutex_destroy(&ctx->host_huge_va_range->lock);
+ mutex_lock(&ctx->host_va_range->lock);
+ clear_va_list_locked(hdev, &ctx->host_va_range->list);
+ mutex_unlock(&ctx->host_va_range->lock);
+host_page_range_err:
+ mutex_destroy(&ctx->host_va_range->lock);
mutex_destroy(&ctx->mem_hash_lock);
hl_mmu_ctx_fini(ctx);
+mmu_ctx_err:
+ kfree(ctx->dram_va_range);
+dram_va_range_err:
+ kfree(ctx->host_huge_va_range);
+host_huge_va_range_err:
+ kfree(ctx->host_va_range);

return rc;
}
@@ -1556,8 +1637,8 @@ static int hl_vm_ctx_init_with_ranges(struct hl_ctx *ctx, u64 host_range_start,
int hl_vm_ctx_init(struct hl_ctx *ctx)
{
struct asic_fixed_properties *prop = &ctx->hdev->asic_prop;
- u64 host_range_start, host_range_end, dram_range_start,
- dram_range_end;
+ u64 host_range_start, host_range_end, host_huge_range_start,
+ host_huge_range_end, dram_range_start, dram_range_end;

atomic64_set(&ctx->dram_phys_mem, 0);

@@ -1569,38 +1650,26 @@ int hl_vm_ctx_init(struct hl_ctx *ctx)
* address of the memory related to the given handle.
*/
if (ctx->hdev->mmu_enable) {
- dram_range_start = prop->va_space_dram_start_address;
- dram_range_end = prop->va_space_dram_end_address;
- host_range_start = prop->va_space_host_start_address;
- host_range_end = prop->va_space_host_end_address;
+ dram_range_start = prop->dmmu.start_addr;
+ dram_range_end = prop->dmmu.end_addr;
+ host_range_start = prop->pmmu.start_addr;
+ host_range_end = prop->pmmu.end_addr;
+ host_huge_range_start = prop->pmmu_huge.start_addr;
+ host_huge_range_end = prop->pmmu_huge.end_addr;
} else {
dram_range_start = prop->dram_user_base_address;
dram_range_end = prop->dram_end_address;
host_range_start = prop->dram_user_base_address;
host_range_end = prop->dram_end_address;
+ host_huge_range_start = prop->dram_user_base_address;
+ host_huge_range_end = prop->dram_end_address;
}

- return hl_vm_ctx_init_with_ranges(ctx, host_range_start, host_range_end,
- dram_range_start, dram_range_end);
-}
-
-/*
- * hl_va_range_fini - clear a virtual addresses range
- *
- * @hdev : pointer to the habanalabs structure
- * va_range : pointer to virtual addresses range
- *
- * This function does the following:
- * - Frees the virtual addresses block list and its lock
- */
-static void hl_va_range_fini(struct hl_device *hdev,
- struct hl_va_range *va_range)
-{
- mutex_lock(&va_range->lock);
- clear_va_list_locked(hdev, &va_range->list);
- mutex_unlock(&va_range->lock);
-
- mutex_destroy(&va_range->lock);
+ return vm_ctx_init_with_ranges(ctx, host_range_start, host_range_end,
+ host_huge_range_start,
+ host_huge_range_end,
+ dram_range_start,
+ dram_range_end);
}

/*
@@ -1667,8 +1736,10 @@ void hl_vm_ctx_fini(struct hl_ctx *ctx)
}
spin_unlock(&vm->idr_lock);

- hl_va_range_fini(hdev, &ctx->dram_va_range);
- hl_va_range_fini(hdev, &ctx->host_va_range);
+ va_range_fini(hdev, ctx->dram_va_range);
+ if (hdev->pmmu_huge_range)
+ va_range_fini(hdev, ctx->host_huge_va_range);
+ va_range_fini(hdev, ctx->host_va_range);

mutex_destroy(&ctx->mem_hash_lock);
hl_mmu_ctx_fini(ctx);
diff --git a/drivers/misc/habanalabs/mmu.c b/drivers/misc/habanalabs/mmu.c
index 006eee47909d..a290d6b49d78 100644
--- a/drivers/misc/habanalabs/mmu.c
+++ b/drivers/misc/habanalabs/mmu.c
@@ -254,6 +254,15 @@ static inline u64 get_phys_addr(struct hl_ctx *ctx, u64 shadow_addr)
return phys_hop_addr + pte_offset;
}

+static bool is_dram_va(struct hl_device *hdev, u64 virt_addr)
+{
+ struct asic_fixed_properties *prop = &hdev->asic_prop;
+
+ return hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
+ prop->dmmu.start_addr,
+ prop->dmmu.end_addr);
+}
+
static int dram_default_mapping_init(struct hl_ctx *ctx)
{
struct hl_device *hdev = ctx->hdev;
@@ -548,6 +557,7 @@ static int _hl_mmu_unmap(struct hl_ctx *ctx, u64 virt_addr, bool is_dram_addr)
curr_pte;
bool is_huge, clear_hop3 = true;

+ /* shifts and masks are the same in PMMU and HPMMU, use one of them */
mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;

hop0_addr = get_hop0_addr(ctx);
@@ -702,26 +712,25 @@ int hl_mmu_unmap(struct hl_ctx *ctx, u64 virt_addr, u32 page_size,
if (!hdev->mmu_enable)
return 0;

- is_dram_addr = hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
- prop->va_space_dram_start_address,
- prop->va_space_dram_end_address);
+ is_dram_addr = is_dram_va(hdev, virt_addr);

- mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;
+ if (is_dram_addr)
+ mmu_prop = &prop->dmmu;
+ else if ((page_size % prop->pmmu_huge.page_size) == 0)
+ mmu_prop = &prop->pmmu_huge;
+ else
+ mmu_prop = &prop->pmmu;

/*
* The H/W handles mapping of specific page sizes. Hence if the page
* size is bigger, we break it to sub-pages and unmap them separately.
*/
- if ((page_size % mmu_prop->huge_page_size) == 0) {
- real_page_size = mmu_prop->huge_page_size;
- } else if ((page_size % mmu_prop->page_size) == 0) {
+ if ((page_size % mmu_prop->page_size) == 0) {
real_page_size = mmu_prop->page_size;
} else {
dev_err(hdev->dev,
- "page size of %u is not %uKB nor %uMB aligned, can't unmap\n",
- page_size,
- mmu_prop->page_size >> 10,
- mmu_prop->huge_page_size >> 20);
+ "page size of %u is not %uKB aligned, can't unmap\n",
+ page_size, mmu_prop->page_size >> 10);

return -EFAULT;
}
@@ -759,8 +768,6 @@ static int _hl_mmu_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr,
hop4_new = false, is_huge;
int rc = -ENOMEM;

- mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;
-
/*
* This mapping function can map a page or a huge page. For huge page
* there are only 3 hops rather than 4. Currently the DRAM allocation
@@ -768,11 +775,15 @@ static int _hl_mmu_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr,
* one of the two page sizes. Since this is a common code for all the
* three cases, we need this hugs page check.
*/
- is_huge = page_size == mmu_prop->huge_page_size;
-
- if (is_dram_addr && !is_huge) {
- dev_err(hdev->dev, "DRAM mapping should use huge pages only\n");
- return -EFAULT;
+ if (is_dram_addr) {
+ mmu_prop = &prop->dmmu;
+ is_huge = true;
+ } else if (page_size == prop->pmmu_huge.page_size) {
+ mmu_prop = &prop->pmmu_huge;
+ is_huge = true;
+ } else {
+ mmu_prop = &prop->pmmu;
+ is_huge = false;
}

hop0_addr = get_hop0_addr(ctx);
@@ -942,26 +953,25 @@ int hl_mmu_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr, u32 page_size,
if (!hdev->mmu_enable)
return 0;

- is_dram_addr = hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
- prop->va_space_dram_start_address,
- prop->va_space_dram_end_address);
+ is_dram_addr = is_dram_va(hdev, virt_addr);

- mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;
+ if (is_dram_addr)
+ mmu_prop = &prop->dmmu;
+ else if ((page_size % prop->pmmu_huge.page_size) == 0)
+ mmu_prop = &prop->pmmu_huge;
+ else
+ mmu_prop = &prop->pmmu;

/*
* The H/W handles mapping of specific page sizes. Hence if the page
* size is bigger, we break it to sub-pages and map them separately.
*/
- if ((page_size % mmu_prop->huge_page_size) == 0) {
- real_page_size = mmu_prop->huge_page_size;
- } else if ((page_size % mmu_prop->page_size) == 0) {
+ if ((page_size % mmu_prop->page_size) == 0) {
real_page_size = mmu_prop->page_size;
} else {
dev_err(hdev->dev,
- "page size of %u is not %dKB nor %dMB aligned, can't unmap\n",
- page_size,
- mmu_prop->page_size >> 10,
- mmu_prop->huge_page_size >> 20);
+ "page size of %u is not %uKB aligned, can't unmap\n",
+ page_size, mmu_prop->page_size >> 10);

return -EFAULT;
}
--
2.17.1