path: root/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c

/*
 * Copyright © 2009 Corbin Simpson
 * Copyright © 2011 Marek Olšák <maraeo@gmail.com>
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining
 * a copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 */

#include "radeon_drm_bo.h"
#include "radeon_drm_cs.h"
#include "radeon_drm_public.h"

#include "util/u_cpu_detect.h"
#include "util/u_memory.h"
#include "util/u_hash_table.h"

#include <xf86drm.h>
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <fcntl.h>
#include <radeon_surface.h>

static struct util_hash_table *fd_tab = NULL;
static mtx_t fd_tab_mutex = _MTX_INITIALIZER_NP;

/* Enable/disable feature access for one command stream.
 * If enable == true, return true on success.
 * Otherwise, return false.
 *
 * We basically do the same thing kernel does, because we have to deal
 * with multiple contexts (here command streams) backed by one winsys. */
static bool radeon_set_fd_access(struct radeon_drm_cs *applier,
                                 struct radeon_drm_cs **owner,
                                 mtx_t *mutex,
                                 unsigned request, const char *request_name,
                                 bool enable)
{
    struct drm_radeon_info info;
    unsigned value = enable ? 1 : 0;

    memset(&info, 0, sizeof(info));

    mtx_lock(&*mutex);

    /* Early exit if we are sure the request will fail. */
    if (enable) {
        if (*owner) {
            mtx_unlock(&*mutex);
            return false;
        }
    } else {
        if (*owner != applier) {
            mtx_unlock(&*mutex);
            return false;
        }
    }

    /* Pass through the request to the kernel. */
    info.value = (unsigned long)&value;
    info.request = request;
    if (drmCommandWriteRead(applier->ws->fd, DRM_RADEON_INFO,
                            &info, sizeof(info)) != 0) {
        mtx_unlock(&*mutex);
        return false;
    }

    /* Update the rights in the winsys. */
    if (enable) {
        if (value) {
            *owner = applier;
            mtx_unlock(&*mutex);
            return true;
        }
    } else {
        *owner = NULL;
    }

    mtx_unlock(&*mutex);
    return false;
}

static bool radeon_get_drm_value(int fd, unsigned request,
                                 const char *errname, uint32_t *out)
{
    struct drm_radeon_info info;
    int retval;

    memset(&info, 0, sizeof(info));

    info.value = (unsigned long)out;
    info.request = request;

    retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));
    if (retval) {
        if (errname) {
            fprintf(stderr, "radeon: Failed to get %s, error number %d\n",
                    errname, retval);
        }
        return false;
    }
    return true;
}

/* Helper function to do the ioctls needed for setup and init. */
static bool do_winsys_init(struct radeon_drm_winsys *ws)
{
    struct drm_radeon_gem_info gem_info;
    int retval;
    drmVersionPtr version;

    memset(&gem_info, 0, sizeof(gem_info));

    /* We do things in a specific order here.
     *
     * DRM version first. We need to be sure we're running on a KMS chipset.
     * This is also for some features.
     *
     * Then, the PCI ID. This is essential and should return usable numbers
     * for all Radeons. If this fails, we probably got handed an FD for some
     * non-Radeon card.
     *
     * The GEM info is actually bogus on the kernel side, as well as our side
     * (see radeon_gem_info_ioctl in radeon_gem.c) but that's alright because
     * we don't actually use the info for anything yet.
     *
     * The GB and Z pipe requests should always succeed, but they might not
     * return sensical values for all chipsets, but that's alright because
     * the pipe drivers already know that.
     */

    /* Get DRM version. */
    version = drmGetVersion(ws->fd);
    if (version->version_major != 2 ||
        version->version_minor < 12) {
        fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is "
                "only compatible with 2.12.0 (kernel 3.2) or later.\n",
                __FUNCTION__,
                version->version_major,
                version->version_minor,
                version->version_patchlevel);
        drmFreeVersion(version);
        return false;
    }

    ws->info.drm_major = version->version_major;
    ws->info.drm_minor = version->version_minor;
    ws->info.drm_patchlevel = version->version_patchlevel;
    drmFreeVersion(version);

    /* Get PCI ID. */
    if (!radeon_get_drm_value(ws->fd, RADEON_INFO_DEVICE_ID, "PCI ID",
                              &ws->info.pci_id))
        return false;

    /* Check PCI ID. */
    switch (ws->info.pci_id) {
#define CHIPSET(pci_id, name, cfamily) case pci_id: ws->info.family = CHIP_##cfamily; ws->gen = DRV_R300; break;
#include "pci_ids/r300_pci_ids.h"
#undef CHIPSET

#define CHIPSET(pci_id, name, cfamily) case pci_id: ws->info.family = CHIP_##cfamily; ws->gen = DRV_R600; break;
#include "pci_ids/r600_pci_ids.h"
#undef CHIPSET

#define CHIPSET(pci_id, cfamily) \
    case pci_id: \
        ws->info.family = CHIP_##cfamily; \
        ws->info.name = #cfamily; \
        ws->gen = DRV_SI; \
        break;
#include "pci_ids/radeonsi_pci_ids.h"
#undef CHIPSET

    default:
        fprintf(stderr, "radeon: Invalid PCI ID.\n");
        return false;
    }

    switch (ws->info.family) {
    default:
    case CHIP_UNKNOWN:
        fprintf(stderr, "radeon: Unknown family.\n");
        return false;
    case CHIP_R300:
    case CHIP_R350:
    case CHIP_RV350:
    case CHIP_RV370:
    case CHIP_RV380:
    case CHIP_RS400:
    case CHIP_RC410:
    case CHIP_RS480:
        ws->info.chip_class = R300;
        break;
    case CHIP_R420:     /* R4xx-based cores. */
    case CHIP_R423:
    case CHIP_R430:
    case CHIP_R480:
    case CHIP_R481:
    case CHIP_RV410:
    case CHIP_RS600:
    case CHIP_RS690:
    case CHIP_RS740:
        ws->info.chip_class = R400;
        break;
    case CHIP_RV515:    /* R5xx-based cores. */
    case CHIP_R520:
    case CHIP_RV530:
    case CHIP_R580:
    case CHIP_RV560:
    case CHIP_RV570:
        ws->info.chip_class = R500;
        break;
    case CHIP_R600:
    case CHIP_RV610:
    case CHIP_RV630:
    case CHIP_RV670:
    case CHIP_RV620:
    case CHIP_RV635:
    case CHIP_RS780:
    case CHIP_RS880:
        ws->info.chip_class = R600;
        break;
    case CHIP_RV770:
    case CHIP_RV730:
    case CHIP_RV710:
    case CHIP_RV740:
        ws->info.chip_class = R700;
        break;
    case CHIP_CEDAR:
    case CHIP_REDWOOD:
    case CHIP_JUNIPER:
    case CHIP_CYPRESS:
    case CHIP_HEMLOCK:
    case CHIP_PALM:
    case CHIP_SUMO:
    case CHIP_SUMO2:
    case CHIP_BARTS:
    case CHIP_TURKS:
    case CHIP_CAICOS:
        ws->info.chip_class = EVERGREEN;
        break;
    case CHIP_CAYMAN:
    case CHIP_ARUBA:
        ws->info.chip_class = CAYMAN;
        break;
    case CHIP_TAHITI:
    case CHIP_PITCAIRN:
    case CHIP_VERDE:
    case CHIP_OLAND:
    case CHIP_HAINAN:
        ws->info.chip_class = SI;
        break;
    case CHIP_BONAIRE:
    case CHIP_KAVERI:
    case CHIP_KABINI:
    case CHIP_HAWAII:
    case CHIP_MULLINS:
        ws->info.chip_class = CIK;
        break;
    }

    /* Set which chips don't have dedicated VRAM. */
    switch (ws->info.family) {
    case CHIP_RS400:
    case CHIP_RC410:
    case CHIP_RS480:
    case CHIP_RS600:
    case CHIP_RS690:
    case CHIP_RS740:
    case CHIP_RS780:
    case CHIP_RS880:
    case CHIP_PALM:
    case CHIP_SUMO:
    case CHIP_SUMO2:
    case CHIP_ARUBA:
    case CHIP_KAVERI:
    case CHIP_KABINI:
    case CHIP_MULLINS:
       ws->info.has_dedicated_vram = false;
       break;

    default:
       ws->info.has_dedicated_vram = true;
    }

    /* Check for dma */
    ws->info.num_sdma_rings = 0;
    /* DMA is disabled on R700. There is IB corruption and hangs. */
    if (ws->info.chip_class >= EVERGREEN && ws->info.drm_minor >= 27) {
        ws->info.num_sdma_rings = 1;
    }

    /* Check for UVD and VCE */
    ws->info.has_hw_decode = false;
    ws->info.vce_fw_version = 0x00000000;
    if (ws->info.drm_minor >= 32) {
	uint32_t value = RADEON_CS_RING_UVD;
        if (radeon_get_drm_value(ws->fd, RADEON_INFO_RING_WORKING,
                                 "UVD Ring working", &value))
            ws->info.has_hw_decode = value;

        value = RADEON_CS_RING_VCE;
        if (radeon_get_drm_value(ws->fd, RADEON_INFO_RING_WORKING,
                                 NULL, &value) && value) {

            if (radeon_get_drm_value(ws->fd, RADEON_INFO_VCE_FW_VERSION,
                                     "VCE FW version", &value))
                ws->info.vce_fw_version = value;
	}
    }

    /* Check for userptr support. */
    {
        struct drm_radeon_gem_userptr args = {0};

        /* If the ioctl doesn't exist, -EINVAL is returned.
         *
         * If the ioctl exists, it should return -EACCES
         * if RADEON_GEM_USERPTR_READONLY or RADEON_GEM_USERPTR_REGISTER
         * aren't set.
         */
        ws->info.has_userptr =
            drmCommandWriteRead(ws->fd, DRM_RADEON_GEM_USERPTR,
                                &args, sizeof(args)) == -EACCES;
    }

    /* Get GEM info. */
    retval = drmCommandWriteRead(ws->fd, DRM_RADEON_GEM_INFO,
            &gem_info, sizeof(gem_info));
    if (retval) {
        fprintf(stderr, "radeon: Failed to get MM info, error number %d\n",
                retval);
        return false;
    }
    ws->info.gart_size = gem_info.gart_size;
    ws->info.vram_size = gem_info.vram_size;
    ws->info.vram_vis_size = gem_info.vram_visible;
    /* Older versions of the kernel driver reported incorrect values, and
     * didn't support more than 256MB of visible VRAM anyway
     */
    if (ws->info.drm_minor < 49)
        ws->info.vram_vis_size = MIN2(ws->info.vram_vis_size, 256*1024*1024);

    /* Radeon allocates all buffers contiguously, which makes large allocations
     * unlikely to succeed. */
    if (ws->info.has_dedicated_vram)
	    ws->info.max_alloc_size = ws->info.vram_size * 0.7;
    else
	    ws->info.max_alloc_size = ws->info.gart_size * 0.7;

    if (ws->info.drm_minor < 40)
        ws->info.max_alloc_size = MIN2(ws->info.max_alloc_size, 256*1024*1024);
    /* Both 32-bit and 64-bit address spaces only have 4GB. */
    ws->info.max_alloc_size = MIN2(ws->info.max_alloc_size, 3ull*1024*1024*1024);

    /* Get max clock frequency info and convert it to MHz */
    radeon_get_drm_value(ws->fd, RADEON_INFO_MAX_SCLK, NULL,
                         &ws->info.max_shader_clock);
    ws->info.max_shader_clock /= 1000;

    ws->num_cpus = sysconf(_SC_NPROCESSORS_ONLN);

    /* Generation-specific queries. */
    if (ws->gen == DRV_R300) {
        if (!radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_GB_PIPES,
                                  "GB pipe count",
                                  &ws->info.r300_num_gb_pipes))
            return false;

        if (!radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_Z_PIPES,
                                  "Z pipe count",
                                  &ws->info.r300_num_z_pipes))
            return false;
    }
    else if (ws->gen >= DRV_R600) {
        uint32_t tiling_config = 0;

        if (!radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_BACKENDS,
                                  "num backends",
                                  &ws->info.num_render_backends))
            return false;

        /* get the GPU counter frequency, failure is not fatal */
        radeon_get_drm_value(ws->fd, RADEON_INFO_CLOCK_CRYSTAL_FREQ, NULL,
                             &ws->info.clock_crystal_freq);

        radeon_get_drm_value(ws->fd, RADEON_INFO_TILING_CONFIG, NULL,
                             &tiling_config);

        ws->info.r600_num_banks =
            ws->info.chip_class >= EVERGREEN ?
                4 << ((tiling_config & 0xf0) >> 4) :
                4 << ((tiling_config & 0x30) >> 4);

        ws->info.pipe_interleave_bytes =
            ws->info.chip_class >= EVERGREEN ?
                256 << ((tiling_config & 0xf00) >> 8) :
                256 << ((tiling_config & 0xc0) >> 6);

        if (!ws->info.pipe_interleave_bytes)
            ws->info.pipe_interleave_bytes =
                ws->info.chip_class >= EVERGREEN ? 512 : 256;

        radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_TILE_PIPES, NULL,
                             &ws->info.num_tile_pipes);

        /* "num_tiles_pipes" must be equal to the number of pipes (Px) in the
         * pipe config field of the GB_TILE_MODE array. Only one card (Tahiti)
         * reports a different value (12). Fix it by setting what's in the
         * GB_TILE_MODE array (8).
         */
        if (ws->gen == DRV_SI && ws->info.num_tile_pipes == 12)
            ws->info.num_tile_pipes = 8;

        if (radeon_get_drm_value(ws->fd, RADEON_INFO_BACKEND_MAP, NULL,
                                  &ws->info.r600_gb_backend_map))
            ws->info.r600_gb_backend_map_valid = true;

        /* Default value. */
        ws->info.enabled_rb_mask = u_bit_consecutive(0, ws->info.num_render_backends);
        /*
         * This fails (silently) on non-GCN or older kernels, overwriting the
         * default enabled_rb_mask with the result of the last query.
        */
        if (ws->gen >= DRV_SI)
            radeon_get_drm_value(ws->fd, RADEON_INFO_SI_BACKEND_ENABLED_MASK, NULL,
                                 &ws->info.enabled_rb_mask);

        ws->info.r600_has_virtual_memory = false;
        if (ws->info.drm_minor >= 13) {
            uint32_t ib_vm_max_size;

            ws->info.r600_has_virtual_memory = true;
            if (!radeon_get_drm_value(ws->fd, RADEON_INFO_VA_START, NULL,
                                      &ws->va_start))
                ws->info.r600_has_virtual_memory = false;
            if (!radeon_get_drm_value(ws->fd, RADEON_INFO_IB_VM_MAX_SIZE, NULL,
                                      &ib_vm_max_size))
                ws->info.r600_has_virtual_memory = false;
            radeon_get_drm_value(ws->fd, RADEON_INFO_VA_UNMAP_WORKING, NULL,
                                 &ws->va_unmap_working);
        }
	if (ws->gen == DRV_R600 && !debug_get_bool_option("RADEON_VA", false))
		ws->info.r600_has_virtual_memory = false;
    }

    /* Get max pipes, this is only needed for compute shaders.  All evergreen+
     * chips have at least 2 pipes, so we use 2 as a default. */
    ws->info.r600_max_quad_pipes = 2;
    radeon_get_drm_value(ws->fd, RADEON_INFO_MAX_PIPES, NULL,
                         &ws->info.r600_max_quad_pipes);

    /* All GPUs have at least one compute unit */
    ws->info.num_good_compute_units = 1;
    radeon_get_drm_value(ws->fd, RADEON_INFO_ACTIVE_CU_COUNT, NULL,
                         &ws->info.num_good_compute_units);

    radeon_get_drm_value(ws->fd, RADEON_INFO_MAX_SE, NULL,
                         &ws->info.max_se);

    switch (ws->info.family) {
    case CHIP_HAINAN:
    case CHIP_KABINI:
    case CHIP_MULLINS:
        ws->info.num_tcc_blocks = 2;
        break;
    case CHIP_VERDE:
    case CHIP_OLAND:
    case CHIP_BONAIRE:
    case CHIP_KAVERI:
        ws->info.num_tcc_blocks = 4;
        break;
    case CHIP_PITCAIRN:
        ws->info.num_tcc_blocks = 8;
        break;
    case CHIP_TAHITI:
        ws->info.num_tcc_blocks = 12;
        break;
    case CHIP_HAWAII:
        ws->info.num_tcc_blocks = 16;
        break;
    default:
        ws->info.num_tcc_blocks = 0;
        break;
    }

    if (!ws->info.max_se) {
        switch (ws->info.family) {
        default:
            ws->info.max_se = 1;
            break;
        case CHIP_CYPRESS:
        case CHIP_HEMLOCK:
        case CHIP_BARTS:
        case CHIP_CAYMAN:
        case CHIP_TAHITI:
        case CHIP_PITCAIRN:
        case CHIP_BONAIRE:
            ws->info.max_se = 2;
            break;
        case CHIP_HAWAII:
            ws->info.max_se = 4;
            break;
        }
    }

    radeon_get_drm_value(ws->fd, RADEON_INFO_MAX_SH_PER_SE, NULL,
                         &ws->info.max_sh_per_se);
    if (ws->gen == DRV_SI) {
        ws->info.num_good_cu_per_sh = ws->info.num_good_compute_units /
                                      (ws->info.max_se * ws->info.max_sh_per_se);
    }

    radeon_get_drm_value(ws->fd, RADEON_INFO_ACCEL_WORKING2, NULL,
                         &ws->accel_working2);
    if (ws->info.family == CHIP_HAWAII && ws->accel_working2 < 2) {
        fprintf(stderr, "radeon: GPU acceleration for Hawaii disabled, "
                "returned accel_working2 value %u is smaller than 2. "
                "Please install a newer kernel.\n",
                ws->accel_working2);
        return false;
    }

    if (ws->info.chip_class == CIK) {
        if (!radeon_get_drm_value(ws->fd, RADEON_INFO_CIK_MACROTILE_MODE_ARRAY, NULL,
                                  ws->info.cik_macrotile_mode_array)) {
            fprintf(stderr, "radeon: Kernel 3.13 is required for CIK support.\n");
            return false;
        }
    }

    if (ws->info.chip_class >= SI) {
        if (!radeon_get_drm_value(ws->fd, RADEON_INFO_SI_TILE_MODE_ARRAY, NULL,
                                  ws->info.si_tile_mode_array)) {
            fprintf(stderr, "radeon: Kernel 3.10 is required for SI support.\n");
            return false;
        }
    }

    /* Hawaii with old firmware needs type2 nop packet.
     * accel_working2 with value 3 indicates the new firmware.
     */
    ws->info.gfx_ib_pad_with_type2 = ws->info.chip_class <= SI ||
				     (ws->info.family == CHIP_HAWAII &&
				      ws->accel_working2 < 3);
    ws->info.tcc_cache_line_size = 64; /* TC L2 line size on GCN */
    ws->info.ib_start_alignment = 4096;
    ws->info.kernel_flushes_hdp_before_ib = ws->info.drm_minor >= 40;
    /* HTILE is broken with 1D tiling on old kernels and CIK. */
    ws->info.htile_cmask_support_1d_tiling = ws->info.chip_class != CIK ||
                                             ws->info.drm_minor >= 38;
    ws->info.si_TA_CS_BC_BASE_ADDR_allowed = ws->info.drm_minor >= 48;
    ws->info.has_bo_metadata = false;
    ws->info.has_gpu_reset_status_query = false;
    ws->info.has_gpu_reset_counter_query = ws->info.drm_minor >= 43;
    ws->info.has_eqaa_surface_allocator = false;
    ws->info.has_format_bc1_through_bc7 = ws->info.drm_minor >= 31;
    ws->info.kernel_flushes_tc_l2_after_ib = true;
    /* Old kernels disallowed register writes via COPY_DATA
     * that are used for indirect compute dispatches. */
    ws->info.has_indirect_compute_dispatch = ws->info.chip_class == CIK ||
                                             (ws->info.chip_class == SI &&
                                              ws->info.drm_minor >= 45);
    /* SI doesn't support unaligned loads. */
    ws->info.has_unaligned_shader_loads = ws->info.chip_class == CIK &&
                                          ws->info.drm_minor >= 50;
    ws->info.has_sparse_vm_mappings = false;
    /* 2D tiling on CIK is supported since DRM 2.35.0 */
    ws->info.has_2d_tiling = ws->info.chip_class <= SI || ws->info.drm_minor >= 35;
    ws->info.has_read_registers_query = ws->info.drm_minor >= 42;
    ws->info.max_alignment = 1024*1024;

    ws->check_vm = strstr(debug_get_option("R600_DEBUG", ""), "check_vm") != NULL;

    return true;
}

static void radeon_winsys_destroy(struct radeon_winsys *rws)
{
    struct radeon_drm_winsys *ws = (struct radeon_drm_winsys*)rws;

    if (util_queue_is_initialized(&ws->cs_queue))
        util_queue_destroy(&ws->cs_queue);

    mtx_destroy(&ws->hyperz_owner_mutex);
    mtx_destroy(&ws->cmask_owner_mutex);

    if (ws->info.r600_has_virtual_memory)
        pb_slabs_deinit(&ws->bo_slabs);
    pb_cache_deinit(&ws->bo_cache);

    if (ws->gen >= DRV_R600) {
        radeon_surface_manager_free(ws->surf_man);
    }

    util_hash_table_destroy(ws->bo_names);
    util_hash_table_destroy(ws->bo_handles);
    util_hash_table_destroy(ws->bo_vas);
    mtx_destroy(&ws->bo_handles_mutex);
    mtx_destroy(&ws->vm32.mutex);
    mtx_destroy(&ws->vm64.mutex);
    mtx_destroy(&ws->bo_fence_lock);

    if (ws->fd >= 0)
        close(ws->fd);

    FREE(rws);
}

static void radeon_query_info(struct radeon_winsys *rws,
                              struct radeon_info *info)
{
    *info = ((struct radeon_drm_winsys *)rws)->info;
}

static bool radeon_cs_request_feature(struct radeon_cmdbuf *rcs,
                                      enum radeon_feature_id fid,
                                      bool enable)
{
    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);

    switch (fid) {
    case RADEON_FID_R300_HYPERZ_ACCESS:
        return radeon_set_fd_access(cs, &cs->ws->hyperz_owner,
                                    &cs->ws->hyperz_owner_mutex,
                                    RADEON_INFO_WANT_HYPERZ, "Hyper-Z",
                                    enable);

    case RADEON_FID_R300_CMASK_ACCESS:
        return radeon_set_fd_access(cs, &cs->ws->cmask_owner,
                                    &cs->ws->cmask_owner_mutex,
                                    RADEON_INFO_WANT_CMASK, "AA optimizations",
                                    enable);
    }
    return false;
}

static uint64_t radeon_query_value(struct radeon_winsys *rws,
                                   enum radeon_value_id value)
{
    struct radeon_drm_winsys *ws = (struct radeon_drm_winsys*)rws;
    uint64_t retval = 0;

    switch (value) {
    case RADEON_REQUESTED_VRAM_MEMORY:
        return ws->allocated_vram;
    case RADEON_REQUESTED_GTT_MEMORY:
        return ws->allocated_gtt;
    case RADEON_MAPPED_VRAM:
       return ws->mapped_vram;
    case RADEON_MAPPED_GTT:
       return ws->mapped_gtt;
    case RADEON_BUFFER_WAIT_TIME_NS:
        return ws->buffer_wait_time;
    case RADEON_NUM_MAPPED_BUFFERS:
        return ws->num_mapped_buffers;
    case RADEON_TIMESTAMP:
        if (ws->info.drm_minor < 20 || ws->gen < DRV_R600) {
            assert(0);
            return 0;
        }

        radeon_get_drm_value(ws->fd, RADEON_INFO_TIMESTAMP, "timestamp",
                             (uint32_t*)&retval);
        return retval;
    case RADEON_NUM_GFX_IBS:
        return ws->num_gfx_IBs;
    case RADEON_NUM_SDMA_IBS:
        return ws->num_sdma_IBs;
    case RADEON_NUM_BYTES_MOVED:
        radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_BYTES_MOVED,
                             "num-bytes-moved", (uint32_t*)&retval);
        return retval;
    case RADEON_NUM_EVICTIONS:
    case RADEON_NUM_VRAM_CPU_PAGE_FAULTS:
    case RADEON_VRAM_VIS_USAGE:
    case RADEON_GFX_BO_LIST_COUNTER:
    case RADEON_GFX_IB_SIZE_COUNTER:
        return 0; /* unimplemented */
    case RADEON_VRAM_USAGE:
        radeon_get_drm_value(ws->fd, RADEON_INFO_VRAM_USAGE,
                             "vram-usage", (uint32_t*)&retval);
        return retval;
    case RADEON_GTT_USAGE:
        radeon_get_drm_value(ws->fd, RADEON_INFO_GTT_USAGE,
                             "gtt-usage", (uint32_t*)&retval);
        return retval;
    case RADEON_GPU_TEMPERATURE:
        radeon_get_drm_value(ws->fd, RADEON_INFO_CURRENT_GPU_TEMP,
                             "gpu-temp", (uint32_t*)&retval);
        return retval;
    case RADEON_CURRENT_SCLK:
        radeon_get_drm_value(ws->fd, RADEON_INFO_CURRENT_GPU_SCLK,
                             "current-gpu-sclk", (uint32_t*)&retval);
        return retval;
    case RADEON_CURRENT_MCLK:
        radeon_get_drm_value(ws->fd, RADEON_INFO_CURRENT_GPU_MCLK,
                             "current-gpu-mclk", (uint32_t*)&retval);
        return retval;
    case RADEON_GPU_RESET_COUNTER:
        radeon_get_drm_value(ws->fd, RADEON_INFO_GPU_RESET_COUNTER,
                             "gpu-reset-counter", (uint32_t*)&retval);
        return retval;
    case RADEON_CS_THREAD_TIME:
        return util_queue_get_thread_time_nano(&ws->cs_queue, 0);
    }
    return 0;
}

static bool radeon_read_registers(struct radeon_winsys *rws,
                                  unsigned reg_offset,
                                  unsigned num_registers, uint32_t *out)
{
    struct radeon_drm_winsys *ws = (struct radeon_drm_winsys*)rws;
    unsigned i;

    for (i = 0; i < num_registers; i++) {
        uint32_t reg = reg_offset + i*4;

        if (!radeon_get_drm_value(ws->fd, RADEON_INFO_READ_REG, NULL, &reg))
            return false;
        out[i] = reg;
    }
    return true;
}

static unsigned hash_fd(void *key)
{
    int fd = pointer_to_intptr(key);
    struct stat stat;
    fstat(fd, &stat);

    return stat.st_dev ^ stat.st_ino ^ stat.st_rdev;
}

static int compare_fd(void *key1, void *key2)
{
    int fd1 = pointer_to_intptr(key1);
    int fd2 = pointer_to_intptr(key2);
    struct stat stat1, stat2;
    fstat(fd1, &stat1);
    fstat(fd2, &stat2);

    return stat1.st_dev != stat2.st_dev ||
           stat1.st_ino != stat2.st_ino ||
           stat1.st_rdev != stat2.st_rdev;
}

DEBUG_GET_ONCE_BOOL_OPTION(thread, "RADEON_THREAD", true)

static bool radeon_winsys_unref(struct radeon_winsys *ws)
{
    struct radeon_drm_winsys *rws = (struct radeon_drm_winsys*)ws;
    bool destroy;

    /* When the reference counter drops to zero, remove the fd from the table.
     * This must happen while the mutex is locked, so that
     * radeon_drm_winsys_create in another thread doesn't get the winsys
     * from the table when the counter drops to 0. */
    mtx_lock(&fd_tab_mutex);

    destroy = pipe_reference(&rws->reference, NULL);
    if (destroy && fd_tab) {
        util_hash_table_remove(fd_tab, intptr_to_pointer(rws->fd));
        if (util_hash_table_count(fd_tab) == 0) {
           util_hash_table_destroy(fd_tab);
           fd_tab = NULL;
        }
    }

    mtx_unlock(&fd_tab_mutex);
    return destroy;
}

#define PTR_TO_UINT(x) ((unsigned)((intptr_t)(x)))

static unsigned handle_hash(void *key)
{
    return PTR_TO_UINT(key);
}

static int handle_compare(void *key1, void *key2)
{
    return PTR_TO_UINT(key1) != PTR_TO_UINT(key2);
}

static void radeon_pin_threads_to_L3_cache(struct radeon_winsys *ws,
                                           unsigned cache)
{
    struct radeon_drm_winsys *rws = (struct radeon_drm_winsys*)ws;

    if (util_queue_is_initialized(&rws->cs_queue)) {
        util_pin_thread_to_L3(rws->cs_queue.threads[0], cache,
                              util_cpu_caps.cores_per_L3);
    }
}

PUBLIC struct radeon_winsys *
radeon_drm_winsys_create(int fd, const struct pipe_screen_config *config,
			 radeon_screen_create_t screen_create)
{
    struct radeon_drm_winsys *ws;

    mtx_lock(&fd_tab_mutex);
    if (!fd_tab) {
        fd_tab = util_hash_table_create(hash_fd, compare_fd);
    }

    ws = util_hash_table_get(fd_tab, intptr_to_pointer(fd));
    if (ws) {
        pipe_reference(NULL, &ws->reference);
        mtx_unlock(&fd_tab_mutex);
        return &ws->base;
    }

    ws = CALLOC_STRUCT(radeon_drm_winsys);
    if (!ws) {
        mtx_unlock(&fd_tab_mutex);
        return NULL;
    }

    ws->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3);

    if (!do_winsys_init(ws))
        goto fail1;

    pb_cache_init(&ws->bo_cache, RADEON_MAX_CACHED_HEAPS,
                  500000, ws->check_vm ? 1.0f : 2.0f, 0,
                  MIN2(ws->info.vram_size, ws->info.gart_size),
                  radeon_bo_destroy,
                  radeon_bo_can_reclaim);

    if (ws->info.r600_has_virtual_memory) {
        /* There is no fundamental obstacle to using slab buffer allocation
         * without GPUVM, but enabling it requires making sure that the drivers
         * honor the address offset.
         */
        if (!pb_slabs_init(&ws->bo_slabs,
                           RADEON_SLAB_MIN_SIZE_LOG2, RADEON_SLAB_MAX_SIZE_LOG2,
                           RADEON_MAX_SLAB_HEAPS,
                           ws,
                           radeon_bo_can_reclaim_slab,
                           radeon_bo_slab_alloc,
                           radeon_bo_slab_free))
            goto fail_cache;

        ws->info.min_alloc_size = 1 << RADEON_SLAB_MIN_SIZE_LOG2;
    } else {
        ws->info.min_alloc_size = ws->info.gart_page_size;
    }

    if (ws->gen >= DRV_R600) {
        ws->surf_man = radeon_surface_manager_new(ws->fd);
        if (!ws->surf_man)
            goto fail_slab;
    }

    /* init reference */
    pipe_reference_init(&ws->reference, 1);

    /* Set functions. */
    ws->base.unref = radeon_winsys_unref;
    ws->base.destroy = radeon_winsys_destroy;
    ws->base.query_info = radeon_query_info;
    ws->base.pin_threads_to_L3_cache = radeon_pin_threads_to_L3_cache;
    ws->base.cs_request_feature = radeon_cs_request_feature;
    ws->base.query_value = radeon_query_value;
    ws->base.read_registers = radeon_read_registers;

    radeon_drm_bo_init_functions(ws);
    radeon_drm_cs_init_functions(ws);
    radeon_surface_init_functions(ws);

    (void) mtx_init(&ws->hyperz_owner_mutex, mtx_plain);
    (void) mtx_init(&ws->cmask_owner_mutex, mtx_plain);

    ws->bo_names = util_hash_table_create(handle_hash, handle_compare);
    ws->bo_handles = util_hash_table_create(handle_hash, handle_compare);
    ws->bo_vas = util_hash_table_create(handle_hash, handle_compare);
    (void) mtx_init(&ws->bo_handles_mutex, mtx_plain);
    (void) mtx_init(&ws->vm32.mutex, mtx_plain);
    (void) mtx_init(&ws->vm64.mutex, mtx_plain);
    (void) mtx_init(&ws->bo_fence_lock, mtx_plain);
    list_inithead(&ws->vm32.holes);
    list_inithead(&ws->vm64.holes);

    /* The kernel currently returns 8MB. Make sure this doesn't change. */
    if (ws->va_start > 8 * 1024 * 1024) {
        /* Not enough 32-bit address space. */
        radeon_winsys_destroy(&ws->base);
        mtx_unlock(&fd_tab_mutex);
        return NULL;
    }

    ws->vm32.start = ws->va_start;
    ws->vm32.end = 1ull << 32;

    /* The maximum is 8GB of virtual address space limited by the kernel.
     * It's obviously not enough for bigger cards, like Hawaiis with 4GB
     * and 8GB of physical memory and 4GB of GART.
     *
     * Older kernels set the limit to 4GB, which is even worse, so they only
     * have 32-bit address space.
     */
    if (ws->info.drm_minor >= 41) {
        ws->vm64.start = 1ull << 32;
        ws->vm64.end = 1ull << 33;
    }

    /* TTM aligns the BO size to the CPU page size */
    ws->info.gart_page_size = sysconf(_SC_PAGESIZE);

    if (ws->num_cpus > 1 && debug_get_option_thread())
        util_queue_init(&ws->cs_queue, "rcs", 8, 1, 0);

    /* Create the screen at the end. The winsys must be initialized
     * completely.
     *
     * Alternatively, we could create the screen based on "ws->gen"
     * and link all drivers into one binary blob. */
    ws->base.screen = screen_create(&ws->base, config);
    if (!ws->base.screen) {
        radeon_winsys_destroy(&ws->base);
        mtx_unlock(&fd_tab_mutex);
        return NULL;
    }

    util_hash_table_set(fd_tab, intptr_to_pointer(ws->fd), ws);

    /* We must unlock the mutex once the winsys is fully initialized, so that
     * other threads attempting to create the winsys from the same fd will
     * get a fully initialized winsys and not just half-way initialized. */
    mtx_unlock(&fd_tab_mutex);

    return &ws->base;

fail_slab:
    if (ws->info.r600_has_virtual_memory)
        pb_slabs_deinit(&ws->bo_slabs);
fail_cache:
    pb_cache_deinit(&ws->bo_cache);
fail1:
    mtx_unlock(&fd_tab_mutex);
    if (ws->surf_man)
        radeon_surface_manager_free(ws->surf_man);
    if (ws->fd >= 0)
        close(ws->fd);

    FREE(ws);
    return NULL;
}