gpu: nvgpu: split vidmem_is_vidmem

As the vidmem_is_vidmem flag has got two separate meanings in one bit,
split it in two bits into the enabled() API:

Add NVGPU_MM_HONORS_APERTURE bit, which is the same as vidmem_is_vidmem
with its original meaning, and use it to test which aperture bits to
write to hardware.

Add NVGPU_MM_UNIFIED_MEMORY bit, which has the opposite meaning: that
the GPU shares the SoC memory. When this flag is false, the GPU has its
own local video memory.

Jira NVGPU-86

Change-Id: I2d0bed3b1ede5a712be99323d3035b154bb23c3a
Signed-off-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-on: http://git-master/r/1496080
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

11 files changed, 33 insertions, 25 deletions

diff --git a/drivers/gpu/nvgpu/common/linux/dma.c b/drivers/gpu/nvgpu/common/linux/dma.c
index 53f3a35e..5fee878a 100644
--- a/drivers/gpu/nvgpu/common/linux/dma.c
+++ b/drivers/gpu/nvgpu/common/linux/dma.c
@@ -21,6 +21,7 @@
 #include <nvgpu/lock.h>
 #include <nvgpu/bug.h>
 #include <nvgpu/gmmu.h>
+#include <nvgpu/enabled.h>
 
 #include <nvgpu/linux/dma.h>
 
@@ -69,7 +70,7 @@ int nvgpu_dma_alloc(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
 int nvgpu_dma_alloc_flags(struct gk20a *g, unsigned long flags, size_t size,
                 struct nvgpu_mem *mem)
 {
-        if (g->mm.vidmem_is_vidmem) {
+        if (!nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY)) {
                 /*
                  * Force the no-kernel-mapping flag on because we don't support
                  * the lack of it for vidmem - the user should not care when
@@ -251,7 +252,7 @@ int nvgpu_dma_alloc_map(struct vm_gk20a *vm, size_t size,
 int nvgpu_dma_alloc_map_flags(struct vm_gk20a *vm, unsigned long flags,
                 size_t size, struct nvgpu_mem *mem)
 {
-        if (vm->mm->vidmem_is_vidmem) {
+        if (!nvgpu_is_enabled(gk20a_from_vm(vm), NVGPU_MM_UNIFIED_MEMORY)) {
                 /*
                  * Force the no-kernel-mapping flag on because we don't support
                  * the lack of it for vidmem - the user should not care when
diff --git a/drivers/gpu/nvgpu/common/linux/driver_common.c b/drivers/gpu/nvgpu/common/linux/driver_common.c
index f85016d4..bd9a4e77 100644
--- a/drivers/gpu/nvgpu/common/linux/driver_common.c
+++ b/drivers/gpu/nvgpu/common/linux/driver_common.c
@@ -133,8 +133,10 @@ static void nvgpu_init_mm_vars(struct gk20a *g)
 
         g->mm.bypass_smmu = platform->bypass_smmu;
         g->mm.disable_bigpage = platform->disable_bigpage;
-        g->mm.vidmem_is_vidmem = platform->vidmem_is_vidmem;
-
+        __nvgpu_set_enabled(g, NVGPU_MM_HONORS_APERTURE,
+                            platform->honors_aperture);
+        __nvgpu_set_enabled(g, NVGPU_MM_UNIFIED_MEMORY,
+                            platform->unified_memory);
         __nvgpu_set_enabled(g, NVGPU_MM_UNIFY_ADDRESS_SPACES,
                             platform->unify_address_spaces);
 
diff --git a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
index 58488067..34fd6626 100644
--- a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
+++ b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
@@ -19,6 +19,7 @@
 #include <nvgpu/page_allocator.h>
 #include <nvgpu/log.h>
 #include <nvgpu/bug.h>
+#include <nvgpu/enabled.h>
 
 #include <nvgpu/linux/dma.h>
 
@@ -30,8 +31,9 @@ u32 __nvgpu_aperture_mask(struct gk20a *g, enum nvgpu_aperture aperture,
 {
         switch (aperture) {
         case APERTURE_SYSMEM:
-                /* sysmem for dgpus; some igpus consider system memory vidmem */
-                return g->mm.vidmem_is_vidmem ? sysmem_mask : vidmem_mask;
+                /* some igpus consider system memory vidmem */
+                return nvgpu_is_enabled(g, NVGPU_MM_HONORS_APERTURE)
+                        ? sysmem_mask : vidmem_mask;
         case APERTURE_VIDMEM:
                 /* for dgpus only */
                 return vidmem_mask;
diff --git a/drivers/gpu/nvgpu/common/linux/pci.c b/drivers/gpu/nvgpu/common/linux/pci.c
index 0a5095fe..b66a6658 100644
--- a/drivers/gpu/nvgpu/common/linux/pci.c
+++ b/drivers/gpu/nvgpu/common/linux/pci.c
@@ -86,7 +86,7 @@ static struct gk20a_platform nvgpu_pci_device[] = {
 
         .ch_wdt_timeout_ms = 7000,
 
-        .vidmem_is_vidmem = true,
+        .honors_aperture = true,
         .vbios_min_version = 0x86063000,
         .hardcode_sw_threshold = true,
         .ina3221_dcb_index = 0,
@@ -121,7 +121,7 @@ static struct gk20a_platform nvgpu_pci_device[] = {
 
         .ch_wdt_timeout_ms = 7000,
 
-        .vidmem_is_vidmem = true,
+        .honors_aperture = true,
         .vbios_min_version = 0x86062d00,
         .hardcode_sw_threshold = true,
         .ina3221_dcb_index = 0,
@@ -156,7 +156,7 @@ static struct gk20a_platform nvgpu_pci_device[] = {
 
         .ch_wdt_timeout_ms = 7000,
 
-        .vidmem_is_vidmem = true,
+        .honors_aperture = true,
         .vbios_min_version = 0x86063000,
         .hardcode_sw_threshold = true,
         .ina3221_dcb_index = 0,
@@ -191,7 +191,7 @@ static struct gk20a_platform nvgpu_pci_device[] = {
 
         .ch_wdt_timeout_ms = 7000,
 
-        .vidmem_is_vidmem = true,
+        .honors_aperture = true,
         .vbios_min_version = 0x86065600,
         .hardcode_sw_threshold = false,
         .ina3221_dcb_index = 1,
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
index 08e2e9cc..ec020d5f 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
@@ -665,7 +665,8 @@ int gk20a_init_mm_setup_sw(struct gk20a *g)
          * this requires fixed allocations in vidmem which must be
          * allocated before all other buffers
          */
-        if (g->ops.pmu.alloc_blob_space && g->mm.vidmem_is_vidmem) {
+        if (g->ops.pmu.alloc_blob_space
+                        && !nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY)) {
                 err = g->ops.pmu.alloc_blob_space(g, 0, &g->acr.ucode_blob);
                 if (err)
                         return err;
@@ -1234,10 +1235,12 @@ enum nvgpu_aperture gk20a_dmabuf_aperture(struct gk20a *g,
                                           struct dma_buf *dmabuf)
 {
         struct gk20a *buf_owner = gk20a_vidmem_buf_owner(dmabuf);
+        bool unified_memory = nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY);
+
         if (buf_owner == NULL) {
                 /* Not nvgpu-allocated, assume system memory */
                 return APERTURE_SYSMEM;
-        } else if (WARN_ON(buf_owner == g && !g->mm.vidmem_is_vidmem)) {
+        } else if (WARN_ON(buf_owner == g && unified_memory)) {
                 /* Looks like our video memory, but this gpu doesn't support
                  * it. Warn about a bug and bail out */
                 nvgpu_warn(g,
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
index 5d90cbf6..94342818 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
@@ -256,8 +256,6 @@ struct mm_gk20a {
         bool disable_bigpage;
 #endif
         bool has_physical_mode;
-        /* false if vidmem aperture actually points to sysmem */
-        bool vidmem_is_vidmem;
 
         struct nvgpu_mem sysmem_flush;
 
diff --git a/drivers/gpu/nvgpu/gk20a/platform_gk20a.h b/drivers/gpu/nvgpu/gk20a/platform_gk20a.h
index 452b868d..9499f91a 100644
--- a/drivers/gpu/nvgpu/gk20a/platform_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/platform_gk20a.h
@@ -215,8 +215,10 @@ struct gk20a_platform {
         /* soc name for finding firmware files */
         const char *soc_name;
 
-        /* if vidmem aperture actually points to vidmem*/
-        bool vidmem_is_vidmem;
+        /* false if vidmem aperture actually points to sysmem */
+        bool honors_aperture;
+        /* unified or split memory with separate vidmem? */
+        bool unified_memory;
 
         /* minimum supported VBIOS version */
         u32 vbios_min_version;
diff --git a/drivers/gpu/nvgpu/gp10b/fifo_gp10b.c b/drivers/gpu/nvgpu/gp10b/fifo_gp10b.c
index 386318e7..cfc2eb8d 100644
--- a/drivers/gpu/nvgpu/gp10b/fifo_gp10b.c
+++ b/drivers/gpu/nvgpu/gp10b/fifo_gp10b.c
@@ -68,8 +68,8 @@ int channel_gp10b_commit_userd(struct channel_gk20a *c)
 
         nvgpu_mem_wr32(g, &c->inst_block,
                        ram_in_ramfc_w() + ram_fc_userd_w(),
-                       (g->mm.vidmem_is_vidmem ?
-                        pbdma_userd_target_sys_mem_ncoh_f() :
+                       nvgpu_aperture_mask(g, &g->fifo.userd,
+                        pbdma_userd_target_sys_mem_ncoh_f(),
                         pbdma_userd_target_vid_mem_f()) |
                        pbdma_userd_addr_f(addr_lo));
 
diff --git a/drivers/gpu/nvgpu/include/nvgpu/enabled.h b/drivers/gpu/nvgpu/include/nvgpu/enabled.h
index a0b809ca..78dea193 100644
--- a/drivers/gpu/nvgpu/include/nvgpu/enabled.h
+++ b/drivers/gpu/nvgpu/include/nvgpu/enabled.h
@@ -32,6 +32,10 @@ struct gk20a;
  * MM flags.
  */
 #define NVGPU_MM_UNIFY_ADDRESS_SPACES           16
+/* false if vidmem aperture actually points to sysmem */
+#define NVGPU_MM_HONORS_APERTURE                17
+/* unified or split memory with separate vidmem? */
+#define NVGPU_MM_UNIFIED_MEMORY                 18
 
 /*
  * Must be greater than the largest bit offset in the above list.
diff --git a/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c b/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c
index 96312a00..8a83f7d0 100644
--- a/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c
+++ b/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c
@@ -934,8 +934,6 @@ static int gk20a_tegra_probe(struct device *dev)
         if (tegra_get_chip_id() == TEGRA132)
                 platform->soc_name = "tegra13x";
 
-        platform->g->mm.vidmem_is_vidmem = platform->vidmem_is_vidmem;
-
         gk20a_tegra_get_clocks(dev);
         nvgpu_linux_init_clk_support(platform->g);
         gk20a_tegra_init_secure_alloc(platform->g);
@@ -1051,7 +1049,7 @@ struct gk20a_platform gk20a_tegra_platform = {
 
         .soc_name = "tegra12x",
 
-        .vidmem_is_vidmem = false,
+        .unified_memory = true,
 };
 
 struct gk20a_platform gm20b_tegra_platform = {
@@ -1123,5 +1121,5 @@ struct gk20a_platform gm20b_tegra_platform = {
 
         .soc_name = "tegra21x",
 
-        .vidmem_is_vidmem = false,
+        .unified_memory = true,
 };
diff --git a/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c b/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c
index b7cbf56e..8b530fac 100644
--- a/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c
+++ b/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c
@@ -161,8 +161,6 @@ static int gp10b_tegra_probe(struct device *dev)
         platform->g->gr.t18x.ctx_vars.force_preemption_gfxp = false;
         platform->g->gr.t18x.ctx_vars.force_preemption_cilp = false;
 
-        platform->g->mm.vidmem_is_vidmem = platform->vidmem_is_vidmem;
-
         gp10b_tegra_get_clocks(dev);
         nvgpu_linux_init_clk_support(platform->g);
         gk20a_tegra_init_secure_alloc(platform->g);
@@ -436,7 +434,7 @@ struct gk20a_platform gp10b_tegra_platform = {
 
         .soc_name = "tegra18x",
 
-        .vidmem_is_vidmem = false,
+        .unified_memory = true,
 };