Upto: Table of Contents of full book "Programming Wayland Clients"

This project hasn't been updated since mid-2017. Wayland has moved on since then, so the information here may be out of date, and there is no guarantee the programs still work.

Direct Rendering Manager

DRM (Direct Rendering Manager) allows frame buffers to be mapped into user space so that applications can write directly to the buffers. This is a critical piece of the software that allows Wayland to replace X.

The material in this chapter is not used elsewhere in this book, so may be skipped. I included it because I want to know what is happening.



To display graphics, and application needs to write to a framebuffer. This is a portion of video memory, and what is written there is displayed on a monitor. For many years, there was very little information about frame buffers or even how to access them. Early versions of X had complex configuration files to specify this information.

DRM (and an associated X driver) allow applications to directly access the framebuffers, and to find information about them. Over the years this has led to a progressive simplification of X configuration, and this may also be used by Wayland compositors, login screens, OpenGL, etc.

The Wayland route is: an application uses EGL to write content to a framebuffer, and the framebuffer is managed by DRM. Wayland has calls to talk to EGL, and EGL in turn uses the DRM API to manipulate the buffers.


The DRM HowTo contains superbly documented programs, and there is no point trying to rewrite their explanation. The simplest program modeset.c is simply included here

 * From https://github.com/dvdhrm/docs/blob/master/drm-howto/modeset.c
 * modeset - DRM Modesetting Example
 * Written 2012 by David Herrmann <dh.herrmann@googlemail.com>
 * Dedicated to the Public Domain.

 * DRM Modesetting Howto
 * This document describes the DRM modesetting API. Before we can use the DRM
 * API, we have to include xf86drm.h and xf86drmMode.h. Both are provided by
 * libdrm which every major distribution ships by default. It has no other
 * dependencies and is pretty small.
 * Please ignore all forward-declarations of functions which are used later. I
 * reordered the functions so you can read this document from top to bottom. If
 * you reimplement it, you would probably reorder the functions to avoid all the
 * nasty forward declarations.
 * For easier reading, we ignore all memory-allocation errors of malloc() and
 * friends here. However, we try to correctly handle all other kinds of errors
 * that may occur.
 * All functions and global variables are prefixed with "modeset_*" in this
 * file. So it should be clear whether a function is a local helper or if it is
 * provided by some external library.

#define _GNU_SOURCE
#include <errno.h>
#include <fcntl.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mman.h>
#include <time.h>
#include <unistd.h>
#include <xf86drm.h>
#include <xf86drmMode.h>

struct modeset_dev;
static int modeset_find_crtc(int fd, drmModeRes *res, drmModeConnector *conn,
			     struct modeset_dev *dev);
static int modeset_create_fb(int fd, struct modeset_dev *dev);
static int modeset_setup_dev(int fd, drmModeRes *res, drmModeConnector *conn,
			     struct modeset_dev *dev);
static int modeset_open(int *out, const char *node);
static int modeset_prepare(int fd);
static void modeset_draw(void);
static void modeset_cleanup(int fd);

 * When the linux kernel detects a graphics-card on your machine, it loads the
 * correct device driver (located in kernel-tree at ./drivers/gpu/drm/<xy>) and
 * provides two character-devices to control it. Udev (or whatever hotplugging
 * application you use) will create them as:
 * /dev/dri/card0
 * /dev/dri/controlID64
 * We only need the first one. You can hard-code this path into your application
 * like we do here, but it is recommended to use libudev with real hotplugging
 * and multi-seat support. However, this is beyond the scope of this document.
 * Also note that if you have multiple graphics-cards, there may also be
 * /dev/dri/card1, /dev/dri/card2, ...
 * We simply use /dev/dri/card0 here but the user can specify another path on
 * the command line.
 * modeset_open(out, node): This small helper function opens the DRM device
 * which is given as @node. The new fd is stored in @out on success. On failure,
 * a negative error code is returned.
 * After opening the file, we also check for the DRM_CAP_DUMB_BUFFER capability.
 * If the driver supports this capability, we can create simple memory-mapped
 * buffers without any driver-dependent code. As we want to avoid any radeon,
 * nvidia, intel, etc. specific code, we depend on DUMB_BUFFERs here.

static int modeset_open(int *out, const char *node)
    int fd, ret;
    uint64_t has_dumb;

    fd = open(node, O_RDWR | O_CLOEXEC);
    if (fd < 0) {
	ret = -errno;
	fprintf(stderr, "cannot open '%s': %m\n", node);
	return ret;

    if (drmGetCap(fd, DRM_CAP_DUMB_BUFFER, &has_dumb) < 0 ||
	!has_dumb) {
	fprintf(stderr, "drm device '%s' does not support dumb buffers\n",
	return -EOPNOTSUPP;

    *out = fd;
    return 0;

 * As a next step we need to find our available display devices. libdrm provides
 * a drmModeRes structure that contains all the needed information. We can
 * retrieve it via drmModeGetResources(fd) and free it via
 * drmModeFreeResources(res) again.
 * A physical connector on your graphics card is called a "connector". You can
 * plug a monitor into it and control what is displayed. We are definitely
 * interested in what connectors are currently used, so we simply iterate
 * through the list of connectors and try to display a test-picture on each
 * available monitor.
 * However, this isn't as easy as it sounds. First, we need to check whether the
 * connector is actually used (a monitor is plugged in and turned on). Then we
 * need to find a CRTC that can control this connector. CRTCs are described
 * later on. After that we create a framebuffer object. If we have all this, we
 * can mmap() the framebuffer and draw a test-picture into it. Then we can tell
 * the DRM device to show the framebuffer on the given CRTC with the selected
 * connector.
 * As we want to draw moving pictures on the framebuffer, we actually have to
 * remember all these settings. Therefore, we create one "struct modeset_dev"
 * object for each connector+crtc+framebuffer pair that we successfully
 * initialized and push it into the global device-list.
 * Each field of this structure is described when it is first used. But as a
 * summary:
 * "struct modeset_dev" contains: {
 * - @next: points to the next device in the single-linked list
 * - @width: width of our buffer object
 * - @height: height of our buffer object
 * - @stride: stride value of our buffer object
 * - @size: size of the memory mapped buffer
 * - @handle: a DRM handle to the buffer object that we can draw into
 * - @map: pointer to the memory mapped buffer
 * - @mode: the display mode that we want to use
 * - @fb: a framebuffer handle with our buffer object as scanout buffer
 * - @conn: the connector ID that we want to use with this buffer
 * - @crtc: the crtc ID that we want to use with this connector
 * - @saved_crtc: the configuration of the crtc before we changed it. We use it
 * so we can restore the same mode when we exit.
 * }

struct modeset_dev {
    struct modeset_dev *next;

    uint32_t width;
    uint32_t height;
    uint32_t stride;
    uint32_t size;
    uint32_t handle;
    uint8_t *map;

    drmModeModeInfo mode;
    uint32_t fb;
    uint32_t conn;
    uint32_t crtc;
    drmModeCrtc *saved_crtc;

static struct modeset_dev *modeset_list = NULL;

 * So as next step we need to actually prepare all connectors that we find. We
 * do this in this little helper function:
 * modeset_prepare(fd): This helper function takes the DRM fd as argument and
 * then simply retrieves the resource-info from the device. It then iterates
 * through all connectors and calls other helper functions to initialize this
 * connector (described later on).
 * If the initialization was successful, we simply add this object as new device
 * into the global modeset device list.
 * The resource-structure contains a list of all connector-IDs. We use the
 * helper function drmModeGetConnector() to retrieve more information on each
 * connector. After we are done with it, we free it again with
 * drmModeFreeConnector().
 * Our helper modeset_setup_dev() returns -ENOENT if the connector is currently
 * unused and no monitor is plugged in. So we can ignore this connector.

static int modeset_prepare(int fd)
    drmModeRes *res;
    drmModeConnector *conn;
    unsigned int i;
    struct modeset_dev *dev;
    int ret;

    /* retrieve resources */
    res = drmModeGetResources(fd);
    if (!res) {
	fprintf(stderr, "cannot retrieve DRM resources (%d): %m\n",
	return -errno;

    /* iterate all connectors */
    for (i = 0; i < res->count_connectors; ++i) {
	/* get information for each connector */
	conn = drmModeGetConnector(fd, res->connectors[i]);
	if (!conn) {
	    fprintf(stderr, "cannot retrieve DRM connector %u:%u (%d): %m\n",
		    i, res->connectors[i], errno);

	/* create a device structure */
	dev = malloc(sizeof(*dev));
	memset(dev, 0, sizeof(*dev));
	dev->conn = conn->connector_id;

	/* call helper function to prepare this connector */
	ret = modeset_setup_dev(fd, res, conn, dev);
	if (ret) {
	    if (ret != -ENOENT) {
		errno = -ret;
		fprintf(stderr, "cannot setup device for connector %u:%u (%d): %m\n",
			i, res->connectors[i], errno);

	/* free connector data and link device into global list */
	dev->next = modeset_list;
	modeset_list = dev;

    /* free resources again */
    return 0;

 * Now we dig deeper into setting up a single connector. As described earlier,
 * we need to check several things first:
 * * If the connector is currently unused, that is, no monitor is plugged in,
 * then we can ignore it.
 * * We have to find a suitable resolution and refresh-rate. All this is
 * available in drmModeModeInfo structures saved for each crtc. We simply
 * use the first mode that is available. This is always the mode with the
 * highest resolution.
 * A more sophisticated mode-selection should be done in real applications,
 * though.
 * * Then we need to find an CRTC that can drive this connector. A CRTC is an
 * internal resource of each graphics-card. The number of CRTCs controls how
 * many connectors can be controlled indepedently. That is, a graphics-cards
 * may have more connectors than CRTCs, which means, not all monitors can be
 * controlled independently.
 * There is actually the possibility to control multiple connectors via a
 * single CRTC if the monitors should display the same content. However, we
 * do not make use of this here.
 * So think of connectors as pipelines to the connected monitors and the
 * CRTCs are the controllers that manage which data goes to which pipeline.
 * If there are more pipelines than CRTCs, then we cannot control all of
 * them at the same time.
 * * We need to create a framebuffer for this connector. A framebuffer is a
 * memory buffer that we can write XRGB32 data into. So we use this to
 * render our graphics and then the CRTC can scan-out this data from the
 * framebuffer onto the monitor.

static int modeset_setup_dev(int fd, drmModeRes *res, drmModeConnector *conn,
			     struct modeset_dev *dev)
    int ret;

    /* check if a monitor is connected */
    if (conn->connection != DRM_MODE_CONNECTED) {
	fprintf(stderr, "ignoring unused connector %u\n",
	return -ENOENT;

    /* check if there is at least one valid mode */
    if (conn->count_modes == 0) {
	fprintf(stderr, "no valid mode for connector %u\n",
	return -EFAULT;

    /* copy the mode information into our device structure */
    memcpy(&dev->mode, &conn->modes[0], sizeof(dev->mode));
    dev->width = conn->modes[0].hdisplay;
    dev->height = conn->modes[0].vdisplay;
    fprintf(stderr, "mode for connector %u is %ux%u\n",
	    conn->connector_id, dev->width, dev->height);

    /* find a crtc for this connector */
    ret = modeset_find_crtc(fd, res, conn, dev);
    if (ret) {
	fprintf(stderr, "no valid crtc for connector %u\n",
	return ret;

    /* create a framebuffer for this CRTC */
    ret = modeset_create_fb(fd, dev);
    if (ret) {
	fprintf(stderr, "cannot create framebuffer for connector %u\n",
	return ret;

    return 0;

 * modeset_find_crtc(fd, res, conn, dev): This small helper tries to find a
 * suitable CRTC for the given connector. We have actually have to introduce one
 * more DRM object to make this more clear: Encoders.
 * Encoders help the CRTC to convert data from a framebuffer into the right
 * format that can be used for the chosen connector. We do not have to
 * understand any more of these conversions to make use of it. However, you must
 * know that each connector has a limited list of encoders that it can use. And
 * each encoder can only work with a limited list of CRTCs. So what we do is
 * trying each encoder that is available and looking for a CRTC that this
 * encoder can work with. If we find the first working combination, we are happy
 * and write it into the @dev structure.
 * But before iterating all available encoders, we first try the currently
 * active encoder+crtc on a connector to avoid a full modeset.
 * However, before we can use a CRTC we must make sure that no other device,
 * that we setup previously, is already using this CRTC. Remember, we can only
 * drive one connector per CRTC! So we simply iterate through the "modeset_list"
 * of previously setup devices and check that this CRTC wasn't used before.
 * Otherwise, we continue with the next CRTC/Encoder combination.

static int modeset_find_crtc(int fd, drmModeRes *res, drmModeConnector *conn,
			     struct modeset_dev *dev)
    drmModeEncoder *enc;
    unsigned int i, j;
    int32_t crtc;
    struct modeset_dev *iter;

    /* first try the currently conected encoder+crtc */
    if (conn->encoder_id)
	enc = drmModeGetEncoder(fd, conn->encoder_id);
	enc = NULL;

    if (enc) {
	if (enc->crtc_id) {
	    crtc = enc->crtc_id;
	    for (iter = modeset_list; iter; iter = iter->next) {
		if (iter->crtc == crtc) {
		    crtc = -1;

	    if (crtc >= 0) {
		dev->crtc = crtc;
		return 0;


    /* If the connector is not currently bound to an encoder or if the
     * encoder+crtc is already used by another connector (actually unlikely
     * but lets be safe), iterate all other available encoders to find a
     * matching CRTC. */
    for (i = 0; i < conn->count_encoders; ++i) {
	enc = drmModeGetEncoder(fd, conn->encoders[i]);
	if (!enc) {
	    fprintf(stderr, "cannot retrieve encoder %u:%u (%d): %m\n",
		    i, conn->encoders[i], errno);

	/* iterate all global CRTCs */
	for (j = 0; j < res->count_crtcs; ++j) {
	    /* check whether this CRTC works with the encoder */
	    if (!(enc->possible_crtcs & (1 << j)))

	    /* check that no other device already uses this CRTC */
	    crtc = res->crtcs[j];
	    for (iter = modeset_list; iter; iter = iter->next) {
		if (iter->crtc == crtc) {
		    crtc = -1;

	    /* we have found a CRTC, so save it and return */
	    if (crtc >= 0) {
		dev->crtc = crtc;
		return 0;


    fprintf(stderr, "cannot find suitable CRTC for connector %u\n",
    return -ENOENT;

 * modeset_create_fb(fd, dev): After we have found a crtc+connector+mode
 * combination, we need to actually create a suitable framebuffer that we can
 * use with it. There are actually two ways to do that:
 * * We can create a so called "dumb buffer". This is a buffer that we can
 * memory-map via mmap() and every driver supports this. We can use it for
 * unaccelerated software rendering on the CPU.
 * * We can use libgbm to create buffers available for hardware-acceleration.
 * libgbm is an abstraction layer that creates these buffers for each
 * available DRM driver. As there is no generic API for this, each driver
 * provides its own way to create these buffers.
 * We can then use such buffers to create OpenGL contexts with the mesa3D
 * library.
 * We use the first solution here as it is much simpler and doesn't require any
 * external libraries. However, if you want to use hardware-acceleration via
 * OpenGL, it is actually pretty easy to create such buffers with libgbm and
 * libEGL. But this is beyond the scope of this document.
 * So what we do is requesting a new dumb-buffer from the driver. We specify the
 * same size as the current mode that we selected for the connector.
 * Then we request the driver to prepare this buffer for memory mapping. After
 * that we perform the actual mmap() call. So we can now access the framebuffer
 * memory directly via the dev->map memory map.

static int modeset_create_fb(int fd, struct modeset_dev *dev)
    struct drm_mode_create_dumb creq;
    struct drm_mode_destroy_dumb dreq;
    struct drm_mode_map_dumb mreq;
    int ret;

    /* create dumb buffer */
    memset(&creq, 0, sizeof(creq));
    creq.width = dev->width;
    creq.height = dev->height;
    creq.bpp = 32;
    ret = drmIoctl(fd, DRM_IOCTL_MODE_CREATE_DUMB, &creq);
    if (ret < 0) {
	fprintf(stderr, "cannot create dumb buffer (%d): %m\n",
	return -errno;
    dev->stride = creq.pitch;
    dev->size = creq.size;
    dev->handle = creq.handle;

    /* create framebuffer object for the dumb-buffer */
    ret = drmModeAddFB(fd, dev->width, dev->height, 24, 32, dev->stride,
		       dev->handle, &dev->fb);
    if (ret) {
	fprintf(stderr, "cannot create framebuffer (%d): %m\n",
	ret = -errno;
	goto err_destroy;

    /* prepare buffer for memory mapping */
    memset(&mreq, 0, sizeof(mreq));
    mreq.handle = dev->handle;
    ret = drmIoctl(fd, DRM_IOCTL_MODE_MAP_DUMB, &mreq);
    if (ret) {
	fprintf(stderr, "cannot map dumb buffer (%d): %m\n",
	ret = -errno;
	goto err_fb;

    /* perform actual memory mapping */
    dev->map = mmap(0, dev->size, PROT_READ | PROT_WRITE, MAP_SHARED,
		    fd, mreq.offset);
    if (dev->map == MAP_FAILED) {
	fprintf(stderr, "cannot mmap dumb buffer (%d): %m\n",
	ret = -errno;
	goto err_fb;

    /* clear the framebuffer to 0 */
    memset(dev->map, 0, dev->size);

    return 0;

    drmModeRmFB(fd, dev->fb);
    memset(&dreq, 0, sizeof(dreq));
    dreq.handle = dev->handle;
    drmIoctl(fd, DRM_IOCTL_MODE_DESTROY_DUMB, &dreq);
    return ret;

 * Finally! We have a connector with a suitable CRTC. We know which mode we want
 * to use and we have a framebuffer of the correct size that we can write to.
 * There is nothing special left to do. We only have to program the CRTC to
 * connect each new framebuffer to each selected connector for each combination
 * that we saved in the global modeset_list.
 * This is done with a call to drmModeSetCrtc().
 * So we are ready for our main() function. First we check whether the user
 * specified a DRM device on the command line, otherwise we use the default
 * /dev/dri/card0. Then we open the device via modeset_open(). modeset_prepare()
 * prepares all connectors and we can loop over "modeset_list" and call
 * drmModeSetCrtc() on every CRTC/connector combination.
 * But printing empty black pages is boring so we have another helper function
 * modeset_draw() that draws some colors into the framebuffer for 5 seconds and
 * then returns. And then we have all the cleanup functions which correctly free
 * all devices again after we used them. All these functions are described below
 * the main() function.
 * As a side note: drmModeSetCrtc() actually takes a list of connectors that we
 * want to control with this CRTC. We pass only one connector, though. As
 * explained earlier, if we used multiple connectors, then all connectors would
 * have the same controlling framebuffer so the output would be cloned. This is
 * most often not what you want so we avoid explaining this feature here.
 * Furthermore, all connectors will have to run with the same mode, which is
 * also often not guaranteed. So instead, we only use one connector per CRTC.
 * Before calling drmModeSetCrtc() we also save the current CRTC configuration.
 * This is used in modeset_cleanup() to restore the CRTC to the same mode as was
 * before we changed it.
 * If we don't do this, the screen will stay blank after we exit until another
 * application performs modesetting itself.

int main(int argc, char **argv)
    int ret, fd;
    const char *card;
    struct modeset_dev *iter;

    /* check which DRM device to open */
    if (argc > 1)
	card = argv[1];
	card = "/dev/dri/card0";

    fprintf(stderr, "using card '%s'\n", card);

    /* open the DRM device */
    ret = modeset_open(&fd, card);
    if (ret)
	goto out_return;

    /* prepare all connectors and CRTCs */
    ret = modeset_prepare(fd);
    if (ret)
	goto out_close;

    /* perform actual modesetting on each found connector+CRTC */
    for (iter = modeset_list; iter; iter = iter->next) {
	iter->saved_crtc = drmModeGetCrtc(fd, iter->crtc);
	ret = drmModeSetCrtc(fd, iter->crtc, iter->fb, 0, 0,
			     &iter->conn, 1, &iter->mode);
	if (ret)
	    fprintf(stderr, "cannot set CRTC for connector %u (%d): %m\n",
		    iter->conn, errno);

    /* draw some colors for 5seconds */

    /* cleanup everything */

    ret = 0;

    if (ret) {
	errno = -ret;
	fprintf(stderr, "modeset failed with error %d: %m\n", errno);
    } else {
	fprintf(stderr, "exiting\n");
    return ret;

 * A short helper function to compute a changing color value. No need to
 * understand it.

static uint8_t next_color(bool *up, uint8_t cur, unsigned int mod)
    uint8_t next;

    next = cur + (*up ? 1 : -1) * (rand() % mod);
    if ((*up && next < cur) || (!*up && next > cur)) {
	*up = !*up;
	next = cur;

    return next;

 * modeset_draw(): This draws a solid color into all configured framebuffers.
 * Every 100ms the color changes to a slightly different color so we get some
 * kind of smoothly changing color-gradient.
 * The color calculation can be ignored as it is pretty boring. So the
 * interesting stuff is iterating over "modeset_list" and then through all lines
 * and width. We then set each pixel individually to the current color.
 * We do this 50 times as we sleep 100ms after each redraw round. This makes
 * 50*100ms = 5000ms = 5s so it takes about 5seconds to finish this loop.
 * Please note that we draw directly into the framebuffer. This means that you
 * will see flickering as the monitor might refresh while we redraw the screen.
 * To avoid this you would need to use two framebuffers and a call to
 * drmModeSetCrtc() to switch between both buffers.
 * You can also use drmModePageFlip() to do a vsync'ed pageflip. But this is
 * beyond the scope of this document.

static void modeset_draw(void)
    uint8_t r, g, b;
    bool r_up, g_up, b_up;
    unsigned int i, j, k, off;
    struct modeset_dev *iter;

    r = rand() % 0xff;
    g = rand() % 0xff;
    b = rand() % 0xff;
    r_up = g_up = b_up = true;

    for (i = 0; i < 50; ++i) {
	r = next_color(&r_up, r, 20);
	g = next_color(&g_up, g, 10);
	b = next_color(&b_up, b, 5);

	for (iter = modeset_list; iter; iter = iter->next) {
	    for (j = 0; j < iter->height; ++j) {
		for (k = 0; k < iter->width; ++k) {
		    off = iter->stride * j + k * 4;
		    *(uint32_t*)&iter->map[off] =
			(r << 16) | (g << 8) | b;


 * modeset_cleanup(fd): This cleans up all the devices we created during
 * modeset_prepare(). It resets the CRTCs to their saved states and deallocates
 * all memory.
 * It should be pretty obvious how all of this works.

static void modeset_cleanup(int fd)
    struct modeset_dev *iter;
    struct drm_mode_destroy_dumb dreq;

    while (modeset_list) {
	/* remove from global list */
	iter = modeset_list;
	modeset_list = iter->next;

	/* restore saved CRTC configuration */

	/* unmap buffer */
	munmap(iter->map, iter->size);

	/* delete framebuffer */
	drmModeRmFB(fd, iter->fb);

	/* delete dumb buffer */
	memset(&dreq, 0, sizeof(dreq));
	dreq.handle = iter->handle;
	drmIoctl(fd, DRM_IOCTL_MODE_DESTROY_DUMB, &dreq);

	/* free allocated memory */

 * I hope this was a short but easy overview of the DRM modesetting API. The DRM
 * API offers much more capabilities including:
 * - double-buffering or tripple-buffering (or whatever you want)
 * - vsync'ed page-flips
 * - hardware-accelerated rendering (for example via OpenGL)
 * - output cloning
 * - graphics-clients plus authentication
 * - DRM planes/overlays/sprites
 * - ...
 * If you are interested in these topics, I can currently only redirect you to
 * existing implementations, including:
 * - plymouth (which uses dumb-buffers like this example; very easy to understand)
 * - kmscon (which uses libuterm to do this)
 * - wayland (very sophisticated DRM renderer; hard to understand fully as it
 * uses more complicated techniques like DRM planes)
 * - xserver (very hard to understand as it is split across many files/projects)
 * But understanding how modesetting (as described in this document) works, is
 * essential to understand all further DRM topics.
 * Any feedback is welcome. Feel free to use this code freely for your own
 * documentation or projects.
 * - Hosted on http://github.com/dvdhrm/docs
 * - Written by David Herrmann <dh.herrmann@googlemail.com>



Copyright © Jan Newmarch, jan@newmarch.name

If you like this book, please contribute using Flattr
or donate using PayPal