diff --git a/backend/drm/drm.c b/backend/drm/drm.c
index c6e29b86..2f64066e 100644
--- a/backend/drm/drm.c
+++ b/backend/drm/drm.c
@@ -92,13 +92,13 @@ static uint32_t get_fb_for_bo(int fd, struct gbm_bo *bo) {
 	return *id;
 }
 
-static void wlr_drm_output_begin(struct wlr_output_state *output) {
+static void wlr_drm_output_make_current(struct wlr_output_state *output) {
 	struct wlr_drm_renderer *renderer = output->renderer;
 	eglMakeCurrent(renderer->egl.display, output->egl,
 			output->egl, renderer->egl.context);
 }
 
-static void wlr_drm_output_end(struct wlr_output_state *output) {
+static void wlr_drm_output_swap_buffers(struct wlr_output_state *output) {
 	struct wlr_drm_renderer *renderer = output->renderer;
 
 	if (!eglSwapBuffers(renderer->egl.display, output->egl)) {
@@ -207,10 +207,10 @@ static void wlr_drm_output_enable(struct wlr_output_state *output, bool enable)
 			DRM_MODE_DPMS_ON);
 
 		// Start rendering loop again by drawing a black frame
-		wlr_drm_output_begin(output);
+		wlr_drm_output_make_current(output);
 		glClearColor(0.0, 0.0, 0.0, 1.0);
 		glClear(GL_COLOR_BUFFER_BIT);
-		wlr_drm_output_end(output);
+		wlr_drm_output_swap_buffers(output);
 	} else {
 		drmModeConnectorSetProperty(state->fd, output->connector, output->props.dpms,
 			DRM_MODE_DPMS_STANDBY);
@@ -342,8 +342,13 @@ static bool wlr_drm_output_set_cursor(struct wlr_output_state *output,
 		wlr_log(L_ERROR, "Failed to write cursor to bo");
 		return false;
 	}
-	return !drmModeSetCursor(state->fd, output->crtc,
-			gbm_bo_get_handle(bo).s32, width, height);
+
+	if (drmModeSetCursor(state->fd, output->crtc, gbm_bo_get_handle(bo).u32, width, height)) {
+		wlr_log_errno(L_INFO, "Failed to set hardware cursor");
+		return false;
+	}
+
+	return true;
 }
 
 static bool wlr_drm_output_move_cursor(struct wlr_output_state *output,
@@ -365,6 +370,8 @@ static struct wlr_output_impl output_impl = {
 	.set_cursor = wlr_drm_output_set_cursor,
 	.move_cursor = wlr_drm_output_move_cursor,
 	.destroy = wlr_drm_output_destroy,
+	.make_current = wlr_drm_output_make_current,
+	.swap_buffers = wlr_drm_output_swap_buffers,
 };
 
 static int32_t calculate_refresh_rate(drmModeModeInfo *mode) {
@@ -625,9 +632,7 @@ static void page_flip_handler(int fd, unsigned seq,
 
 	output->pageflip_pending = false;
 	if (output->state == DRM_OUTPUT_CONNECTED && state->session->active) {
-		wlr_drm_output_begin(output);
 		wl_signal_emit(&output->wlr_output->events.frame, output->wlr_output);
-		wlr_drm_output_end(output);
 	}
 }
 
diff --git a/example/pointer.c b/example/pointer.c
index 8dba5454..c7518841 100644
--- a/example/pointer.c
+++ b/example/pointer.c
@@ -21,9 +21,7 @@
 #include "cat.h"
 
 struct sample_state {
-	struct wlr_renderer *renderer;
-	struct wlr_surface *cat_texture;
-	int cur_x, cur_y;
+	double cur_x, cur_y;
 	float default_color[4];
 	float clear_color[4];
 };
@@ -33,18 +31,13 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 	struct sample_state *sample = state->data;
 	struct wlr_output *wlr_output = output->output;
 
-	wlr_renderer_begin(sample->renderer, wlr_output);
+	wlr_output_make_current(wlr_output);
+
 	glClearColor(sample->clear_color[0], sample->clear_color[1],
 			sample->clear_color[2], sample->clear_color[3]);
 	glClear(GL_COLOR_BUFFER_BIT);
 
-	float matrix[16];
-	wlr_surface_get_matrix(sample->cat_texture, &matrix,
-		&wlr_output->transform_matrix, sample->cur_x, sample->cur_y);
-	wlr_render_with_matrix(sample->renderer,
-			sample->cat_texture, &matrix);
-
-	wlr_renderer_end(sample->renderer);
+	wlr_output_swap_buffers(wlr_output);
 }
 
 static void handle_pointer_motion(struct pointer_state *pstate,
@@ -52,6 +45,11 @@ static void handle_pointer_motion(struct pointer_state *pstate,
 	struct sample_state *state = pstate->compositor->data;
 	state->cur_x += d_x;
 	state->cur_y += d_y;
+
+	struct output_state *output;
+	wl_list_for_each(output, &pstate->compositor->outputs, link) {
+		wlr_output_move_cursor(output->output, state->cur_x, state->cur_y);
+	}
 }
 
 static void handle_pointer_motion_absolute(struct pointer_state *pstate,
@@ -59,6 +57,11 @@ static void handle_pointer_motion_absolute(struct pointer_state *pstate,
 	struct sample_state *state = pstate->compositor->data;
 	state->cur_x = x;
 	state->cur_y = y;
+
+	struct output_state *output;
+	wl_list_for_each(output, &pstate->compositor->outputs, link) {
+		wlr_output_move_cursor(output->output, state->cur_x, state->cur_y);
+	}
 }
 
 static void handle_pointer_button(struct pointer_state *pstate,
@@ -95,9 +98,8 @@ static void handle_pointer_axis(struct pointer_state *pstate,
 
 static void handle_output_add(struct output_state *ostate) {
 	struct wlr_output *wlr_output = ostate->output;
-	int width = 16, height = 16;
 	if (!wlr_output_set_cursor(wlr_output, cat_tex.pixel_data,
-			width * 4, width, height)) {
+			cat_tex.width, cat_tex.width, cat_tex.height)) {
 		wlr_log(L_DEBUG, "Failed to set hardware cursor");
 		return;
 	}
@@ -121,13 +123,5 @@ int main(int argc, char *argv[]) {
 	compositor.pointer_axis_cb = handle_pointer_axis;
 	compositor_init(&compositor);
 
-	state.renderer = wlr_gles2_renderer_init();
-	state.cat_texture = wlr_render_surface_init(state.renderer);
-	wlr_surface_attach_pixels(state.cat_texture, GL_RGBA,
-		cat_tex.width, cat_tex.height, cat_tex.pixel_data);
-
 	compositor_run(&compositor);
-
-	wlr_surface_destroy(state.cat_texture);
-	wlr_renderer_destroy(state.renderer);
 }
diff --git a/example/rotation.c b/example/rotation.c
index b8c43d4c..2dd148d0 100644
--- a/example/rotation.c
+++ b/example/rotation.c
@@ -46,6 +46,7 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 	int32_t width, height;
 	wlr_output_effective_resolution(wlr_output, &width, &height);
 
+	wlr_output_make_current(wlr_output);
 	wlr_renderer_begin(sample->renderer, wlr_output);
 
 	float matrix[16];
@@ -59,6 +60,7 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 	}
 
 	wlr_renderer_end(sample->renderer);
+	wlr_output_swap_buffers(wlr_output);
 
 	long ms = (ts->tv_sec - output->last_frame.tv_sec) * 1000 +
 		(ts->tv_nsec - output->last_frame.tv_nsec) / 1000000;
@@ -205,7 +207,7 @@ int main(int argc, char *argv[]) {
 	state.renderer = wlr_gles2_renderer_init();
 	state.cat_texture = wlr_render_surface_init(state.renderer);
 	wlr_surface_attach_pixels(state.cat_texture, GL_RGBA,
-		cat_tex.width, cat_tex.height, cat_tex.pixel_data);
+		cat_tex.width, cat_tex.width, cat_tex.height, cat_tex.pixel_data);
 
 	compositor_run(&compositor);
 
diff --git a/example/simple.c b/example/simple.c
index 4bbdf399..45cb2b93 100644
--- a/example/simple.c
+++ b/example/simple.c
@@ -34,8 +34,12 @@ void handle_output_frame(struct output_state *output, struct timespec *ts) {
 		sample->dec = inc;
 	}
 
+	wlr_output_make_current(output->output);
+
 	glClearColor(sample->color[0], sample->color[1], sample->color[2], 1.0);
 	glClear(GL_COLOR_BUFFER_BIT);
+
+	wlr_output_swap_buffers(output->output);
 }
 
 int main() {
diff --git a/example/tablet.c b/example/tablet.c
index 0343bd0a..d7f7d042 100644
--- a/example/tablet.c
+++ b/example/tablet.c
@@ -41,6 +41,7 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 	int32_t width, height;
 	wlr_output_effective_resolution(wlr_output, &width, &height);
 
+	wlr_output_make_current(wlr_output);
 	wlr_renderer_begin(sample->renderer, wlr_output);
 
 	float matrix[16], view[16];
@@ -74,6 +75,7 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 	}
 
 	wlr_renderer_end(sample->renderer);
+	wlr_output_swap_buffers(wlr_output);
 }
 
 static void handle_tool_axis(struct tablet_tool_state *tstate,
diff --git a/example/touch.c b/example/touch.c
index 29380ed4..aca86739 100644
--- a/example/touch.c
+++ b/example/touch.c
@@ -38,6 +38,8 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 
 	int32_t width, height;
 	wlr_output_effective_resolution(wlr_output, &width, &height);
+
+	wlr_output_make_current(wlr_output);
 	wlr_renderer_begin(sample->renderer, wlr_output);
 
 	float matrix[16];
@@ -52,6 +54,7 @@ static void handle_output_frame(struct output_state *output, struct timespec *ts
 	}
 
 	wlr_renderer_end(sample->renderer);
+	wlr_output_swap_buffers(wlr_output);
 }
 
 static void handle_touch_down(struct touch_state *tstate, int32_t slot,
@@ -104,7 +107,7 @@ int main(int argc, char *argv[]) {
 	state.renderer = wlr_gles2_renderer_init();
 	state.cat_texture = wlr_render_surface_init(state.renderer);
 	wlr_surface_attach_pixels(state.cat_texture, GL_RGBA,
-		cat_tex.width, cat_tex.height, cat_tex.pixel_data);
+		cat_tex.width, cat_tex.width, cat_tex.height, cat_tex.pixel_data);
 
 	compositor_run(&compositor);
 
diff --git a/include/wlr/interfaces/wlr_output.h b/include/wlr/interfaces/wlr_output.h
index 0949bca3..5f9bfdd2 100644
--- a/include/wlr/interfaces/wlr_output.h
+++ b/include/wlr/interfaces/wlr_output.h
@@ -13,6 +13,8 @@ struct wlr_output_impl {
 		const uint8_t *buf, int32_t stride, uint32_t width, uint32_t height);
 	bool (*move_cursor)(struct wlr_output_state *state, int x, int y);
 	void (*destroy)(struct wlr_output_state *state);
+	void (*make_current)(struct wlr_output_state *state);
+	void (*swap_buffers)(struct wlr_output_state *state);
 };
 
 struct wlr_output *wlr_output_create(struct wlr_output_impl *impl,
diff --git a/include/wlr/render.h b/include/wlr/render.h
index 90967dd9..53df01d2 100644
--- a/include/wlr/render.h
+++ b/include/wlr/render.h
@@ -59,7 +59,7 @@ struct wlr_surface {
  * calling this function.
  */
 bool wlr_surface_attach_pixels(struct wlr_surface *surf, uint32_t format,
-		int width, int height, const unsigned char *pixels);
+		int stride, int width, int height, const unsigned char *pixels);
 /**
  * Attaches pixels from a wl_shm_buffer to this surface. The shm buffer may be
  * invalidated after calling this function.
diff --git a/include/wlr/render/interface.h b/include/wlr/render/interface.h
index 222b0c4d..a9c86f2c 100644
--- a/include/wlr/render/interface.h
+++ b/include/wlr/render/interface.h
@@ -31,7 +31,7 @@ struct wlr_renderer *wlr_renderer_init(struct wlr_renderer_state *state,
 
 struct wlr_surface_impl {
 	bool (*attach_pixels)(struct wlr_surface_state *state, uint32_t format,
-		int width, int height, const unsigned char *pixels);
+		int stride, int width, int height, const unsigned char *pixels);
 	bool (*attach_shm)(struct wlr_surface_state *state, uint32_t format,
 		struct wl_shm_buffer *shm);
 	// TODO: egl
diff --git a/include/wlr/types/wlr_output.h b/include/wlr/types/wlr_output.h
index 71e1d0fe..da5a7b07 100644
--- a/include/wlr/types/wlr_output.h
+++ b/include/wlr/types/wlr_output.h
@@ -42,6 +42,14 @@ struct wlr_output {
 		struct wl_signal frame;
 		struct wl_signal resolution;
 	} events;
+
+	struct {
+		bool is_sw;
+		int32_t x, y;
+		uint32_t width, height;
+		struct wlr_renderer *renderer;
+		struct wlr_surface *texture;
+	} cursor;
 };
 
 void wlr_output_enable(struct wlr_output *output, bool enable);
@@ -55,5 +63,7 @@ bool wlr_output_move_cursor(struct wlr_output *output, int x, int y);
 void wlr_output_destroy(struct wlr_output *output);
 void wlr_output_effective_resolution(struct wlr_output *output,
 		int *width, int *height);
+void wlr_output_make_current(struct wlr_output *output);
+void wlr_output_swap_buffers(struct wlr_output *output);
 
 #endif
diff --git a/render/CMakeLists.txt b/render/CMakeLists.txt
index 70ebced2..4058b50e 100644
--- a/render/CMakeLists.txt
+++ b/render/CMakeLists.txt
@@ -1,5 +1,5 @@
-add_library(wlr-render STATIC
-	matrix.c
+add_library(wlr-render
+    matrix.c
     wlr_renderer.c
     wlr_surface.c
     gles2/shaders.c
diff --git a/render/gles2/surface.c b/render/gles2/surface.c
index 278b3dc3..c9be6ac2 100644
--- a/render/gles2/surface.c
+++ b/render/gles2/surface.c
@@ -2,6 +2,7 @@
 #include <stdlib.h>
 #include <assert.h>
 #include <GLES2/gl2.h>
+#include <GLES2/gl2ext.h>
 #include <wayland-util.h>
 #include <wayland-server-protocol.h>
 #include <wlr/render.h>
@@ -10,15 +11,17 @@
 #include "render/gles2.h"
 
 static bool gles2_surface_attach_pixels(struct wlr_surface_state *surface,
-		uint32_t format, int width, int height, const unsigned char *pixels) {
+		uint32_t format, int stride, int width, int height, const unsigned char *pixels) {
 	assert(surface);
 	surface->wlr_surface->width = width;
 	surface->wlr_surface->height = height;
 	surface->wlr_surface->format = format;
 	GL_CALL(glGenTextures(1, &surface->tex_id));
 	GL_CALL(glBindTexture(GL_TEXTURE_2D, surface->tex_id));
+	GL_CALL(glPixelStorei(GL_UNPACK_ROW_LENGTH_EXT, stride));
 	GL_CALL(glTexImage2D(GL_TEXTURE_2D, 0, format, width, height, 0,
 			format, GL_UNSIGNED_BYTE, pixels));
+	GL_CALL(glPixelStorei(GL_UNPACK_ROW_LENGTH_EXT, 0));
 	surface->wlr_surface->valid = true;
 	return true;
 }
diff --git a/render/wlr_surface.c b/render/wlr_surface.c
index 46a12205..4e4f4101 100644
--- a/render/wlr_surface.c
+++ b/render/wlr_surface.c
@@ -20,9 +20,9 @@ void wlr_surface_bind(struct wlr_surface *surface) {
 }
 
 bool wlr_surface_attach_pixels(struct wlr_surface *surface, uint32_t format,
-		int width, int height, const unsigned char *pixels) {
+		int stride, int width, int height, const unsigned char *pixels) {
 	return surface->impl->attach_pixels(surface->state,
-			format, width, height, pixels);
+			format, stride, width, height, pixels);
 }
 
 bool wlr_surface_attach_shm(struct wlr_surface *surface, uint32_t format,
diff --git a/types/CMakeLists.txt b/types/CMakeLists.txt
index 72d885c9..648db820 100644
--- a/types/CMakeLists.txt
+++ b/types/CMakeLists.txt
@@ -15,5 +15,6 @@ add_library(wlr-types
 
 target_link_libraries(wlr-types
     wlr-util
+    wlr-render
     ${WAYLAND_LIBRARIES}
 )
diff --git a/types/wlr_output.c b/types/wlr_output.c
index 4d2e7778..b64fd827 100644
--- a/types/wlr_output.c
+++ b/types/wlr_output.c
@@ -7,6 +7,10 @@
 #include <wlr/interfaces/wlr_output.h>
 #include <wlr/util/list.h>
 #include <wlr/util/log.h>
+#include <GLES2/gl2.h>
+#include <wlr/render/matrix.h>
+#include <wlr/render/gles2.h>
+#include <wlr/render.h>
 
 static void wl_output_send_to_resource(struct wl_resource *resource) {
 	assert(resource);
@@ -182,16 +186,44 @@ void wlr_output_transform(struct wlr_output *output,
 
 bool wlr_output_set_cursor(struct wlr_output *output,
 		const uint8_t *buf, int32_t stride, uint32_t width, uint32_t height) {
-	if (!output->impl || !output->impl->set_cursor) {
-		return false;
+	if (output->impl->set_cursor && output->impl->set_cursor(output->state, buf,
+			stride, width, height)) {
+		output->cursor.is_sw = false;
+		return true;
+	}
+
+	wlr_log(L_INFO, "Falling back to software cursor");
+
+	output->cursor.is_sw = true;
+	output->cursor.width = width;
+	output->cursor.height = height;
+
+	if (!output->cursor.renderer) {
+		output->cursor.renderer = wlr_gles2_renderer_init();
 	}
-	return output->impl->set_cursor(output->state, buf, stride, width, height);
+
+	if (!output->cursor.texture) {
+		output->cursor.texture = wlr_render_surface_init(output->cursor.renderer);
+	}
+
+	wlr_surface_attach_pixels(output->cursor.texture, GL_RGBA,
+		stride, width, height, buf);
+
+	return true;
 }
 
 bool wlr_output_move_cursor(struct wlr_output *output, int x, int y) {
-	if (!output->impl || !output->impl->move_cursor) {
+	output->cursor.x = x;
+	output->cursor.y = y;
+
+	if (output->cursor.is_sw) {
+		return true;
+	}
+
+	if (!output->impl->move_cursor) {
 		return false;
 	}
+
 	return output->impl->move_cursor(output->state, x, y);
 }
 
@@ -216,3 +248,18 @@ void wlr_output_effective_resolution(struct wlr_output *output,
 		*height = output->height;
 	}
 }
+
+void wlr_output_make_current(struct wlr_output *output) {
+	output->impl->make_current(output->state);
+}
+
+void wlr_output_swap_buffers(struct wlr_output *output) {
+	if (output->cursor.is_sw) {
+		float matrix[16];
+		wlr_surface_get_matrix(output->cursor.texture, &matrix, &output->transform_matrix,
+			output->cursor.x, output->cursor.y);
+		wlr_render_with_matrix(output->cursor.renderer, output->cursor.texture, &matrix);
+	}
+
+	output->impl->swap_buffers(output->state);
+}