diff --git a/drivers/gpu/nvgpu/common/linux/channel.c b/drivers/gpu/nvgpu/common/linux/channel.c
index e2a106347..727b50674 100644
--- a/drivers/gpu/nvgpu/common/linux/channel.c
+++ b/drivers/gpu/nvgpu/common/linux/channel.c
@@ -692,10 +692,6 @@ static int gk20a_submit_append_gpfifo(struct channel_gk20a *c,
 		 * This path (from userspace to sysmem) is special in order to
 		 * avoid two copies unnecessarily (from user to pipe, then from
 		 * pipe to gpu sysmem buffer).
-		 *
-		 * As a special case, the pipe buffer exists if PRAMIN writes
-		 * are forced, although the buffers may not be in vidmem in
-		 * that case.
 		 */
 		if (end > gpfifo_size) {
 			/* wrap-around */
@@ -723,8 +719,7 @@ static int gk20a_submit_append_gpfifo(struct channel_gk20a *c,
 				0, num_entries);
 		goto out;
 	} else if (user_gpfifo) {
-		/* from userspace to vidmem or sysmem when pramin forced, use
-		 * the common copy path below */
+		/* from userspace to vidmem, use the common copy path below */
 		err = copy_from_user(c->gpfifo.pipe, user_gpfifo, len);
 		if (err)
 			return err;
diff --git a/drivers/gpu/nvgpu/common/linux/debug_mm.c b/drivers/gpu/nvgpu/common/linux/debug_mm.c
index f2c42e70d..5a3510bf5 100644
--- a/drivers/gpu/nvgpu/common/linux/debug_mm.c
+++ b/drivers/gpu/nvgpu/common/linux/debug_mm.c
@@ -19,8 +19,4 @@
 
 void gk20a_mm_debugfs_init(struct gk20a *g)
 {
-	struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g);
-
-	debugfs_create_bool("force_pramin", 0664, l->debugfs,
-			   &g->mm.force_pramin);
 }
diff --git a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
index 015295ba2..93925803a 100644
--- a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
+++ b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
@@ -54,7 +54,7 @@ int nvgpu_mem_begin(struct gk20a *g, struct nvgpu_mem *mem)
 		PAGE_KERNEL :
 		pgprot_writecombine(PAGE_KERNEL);
 
-	if (mem->aperture != APERTURE_SYSMEM || g->mm.force_pramin)
+	if (mem->aperture != APERTURE_SYSMEM)
 		return 0;
 
 	/*
@@ -91,7 +91,7 @@ int nvgpu_mem_begin(struct gk20a *g, struct nvgpu_mem *mem)
 
 void nvgpu_mem_end(struct gk20a *g, struct nvgpu_mem *mem)
 {
-	if (mem->aperture != APERTURE_SYSMEM || g->mm.force_pramin)
+	if (mem->aperture != APERTURE_SYSMEM)
 		return;
 
 	/*
@@ -134,7 +134,7 @@ u32 nvgpu_mem_rd32(struct gk20a *g, struct nvgpu_mem *mem, u32 w)
 {
 	u32 data = 0;
 
-	if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
+	if (mem->aperture == APERTURE_SYSMEM) {
 		u32 *ptr = mem->cpu_va;
 
 		WARN_ON(!ptr);
@@ -142,7 +142,7 @@ u32 nvgpu_mem_rd32(struct gk20a *g, struct nvgpu_mem *mem, u32 w)
 #ifdef CONFIG_TEGRA_SIMULATION_PLATFORM
 		nvgpu_log(g, gpu_dbg_mem, " %p = 0x%x", ptr + w, data);
 #endif
-	} else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
+	} else if (mem->aperture == APERTURE_VIDMEM) {
 		u32 value;
 		u32 *p = &value;
 
@@ -170,7 +170,7 @@ void nvgpu_mem_rd_n(struct gk20a *g, struct nvgpu_mem *mem,
 	WARN_ON(offset & 3);
 	WARN_ON(size & 3);
 
-	if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
+	if (mem->aperture == APERTURE_SYSMEM) {
 		u8 *src = (u8 *)mem->cpu_va + offset;
 
 		WARN_ON(!mem->cpu_va);
@@ -180,7 +180,7 @@ void nvgpu_mem_rd_n(struct gk20a *g, struct nvgpu_mem *mem,
 			nvgpu_log(g, gpu_dbg_mem, " %p = 0x%x ... [%d bytes]",
 					src, *dest, size);
 #endif
-	} else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
+	} else if (mem->aperture == APERTURE_VIDMEM) {
 		u32 *dest_u32 = dest;
 
 		nvgpu_pramin_access_batched(g, mem, offset, size,
@@ -210,7 +210,7 @@ static void pramin_access_batch_wr_n(struct gk20a *g, u32 start, u32 words, u32
 
 void nvgpu_mem_wr32(struct gk20a *g, struct nvgpu_mem *mem, u32 w, u32 data)
 {
-	if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
+	if (mem->aperture == APERTURE_SYSMEM) {
 		u32 *ptr = mem->cpu_va;
 
 		WARN_ON(!ptr);
@@ -218,7 +218,7 @@ void nvgpu_mem_wr32(struct gk20a *g, struct nvgpu_mem *mem, u32 w, u32 data)
 		nvgpu_log(g, gpu_dbg_mem, " %p = 0x%x", ptr + w, data);
 #endif
 		ptr[w] = data;
-	} else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
+	} else if (mem->aperture == APERTURE_VIDMEM) {
 		u32 value = data;
 		u32 *p = &value;
 
@@ -243,7 +243,7 @@ void nvgpu_mem_wr_n(struct gk20a *g, struct nvgpu_mem *mem, u32 offset,
 	WARN_ON(offset & 3);
 	WARN_ON(size & 3);
 
-	if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
+	if (mem->aperture == APERTURE_SYSMEM) {
 		u8 *dest = (u8 *)mem->cpu_va + offset;
 
 		WARN_ON(!mem->cpu_va);
@@ -253,7 +253,7 @@ void nvgpu_mem_wr_n(struct gk20a *g, struct nvgpu_mem *mem, u32 offset,
 					dest, *src, size);
 #endif
 		memcpy(dest, src, size);
-	} else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
+	} else if (mem->aperture == APERTURE_VIDMEM) {
 		u32 *src_u32 = src;
 
 		nvgpu_pramin_access_batched(g, mem, offset, size,
@@ -290,7 +290,7 @@ void nvgpu_memset(struct gk20a *g, struct nvgpu_mem *mem, u32 offset,
 
 	c &= 0xff;
 
-	if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
+	if (mem->aperture == APERTURE_SYSMEM) {
 		u8 *dest = (u8 *)mem->cpu_va + offset;
 
 		WARN_ON(!mem->cpu_va);
@@ -300,7 +300,7 @@ void nvgpu_memset(struct gk20a *g, struct nvgpu_mem *mem, u32 offset,
 				dest, c, size);
 #endif
 		memset(dest, c, size);
-	} else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
+	} else if (mem->aperture == APERTURE_VIDMEM) {
 		u32 repeat_value = c | (c << 8) | (c << 16) | (c << 24);
 		u32 *p = &repeat_value;
 
diff --git a/drivers/gpu/nvgpu/common/pramin.c b/drivers/gpu/nvgpu/common/pramin.c
index 98565140b..ed961965b 100644
--- a/drivers/gpu/nvgpu/common/pramin.c
+++ b/drivers/gpu/nvgpu/common/pramin.c
@@ -27,13 +27,6 @@
 
 #include "gk20a/gk20a.h"
 
-/*
- * Flip this to force all gk20a_mem* accesses via PRAMIN from the start of the
- * boot, even for buffers that would work via cpu_va. In runtime, the flag is
- * in debugfs, called "force_pramin".
- */
-#define GK20A_FORCE_PRAMIN_DEFAULT false
-
 /*
  * The PRAMIN range is 1 MB, must change base addr if a buffer crosses that.
  * This same loop is used for read/write/memset. Offset and size in bytes.
@@ -96,5 +89,4 @@ void nvgpu_init_pramin(struct mm_gk20a *mm)
 {
 	mm->pramin_window = 0;
 	nvgpu_spinlock_init(&mm->pramin_window_lock);
-	mm->force_pramin = GK20A_FORCE_PRAMIN_DEFAULT;
 }
diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 45ce0e77a..551e8b041 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -1124,7 +1124,7 @@ int gk20a_channel_alloc_gpfifo(struct channel_gk20a *c,
 		goto clean_up_usermode;
 	}
 
-	if (c->gpfifo.mem.aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
+	if (c->gpfifo.mem.aperture == APERTURE_VIDMEM) {
 		c->gpfifo.pipe = nvgpu_big_malloc(g,
 				gpfifo_size * gpfifo_entry_size);
 		if (!c->gpfifo.pipe) {
diff --git a/drivers/gpu/nvgpu/include/nvgpu/mm.h b/drivers/gpu/nvgpu/include/nvgpu/mm.h
index 4999df2e2..3c9da601f 100644
--- a/drivers/gpu/nvgpu/include/nvgpu/mm.h
+++ b/drivers/gpu/nvgpu/include/nvgpu/mm.h
@@ -151,7 +151,6 @@ struct mm_gk20a {
 
 	u32 pramin_window;
 	struct nvgpu_spinlock pramin_window_lock;
-	bool force_pramin; /* via debugfs */
 
 	struct {
 		size_t size;