/*
* Copyright 2004 The Unichrome Project. All Rights Reserved.
* Copyright 2005 Thomas Hellstrom. 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 above copyright notice and this permission notice (including the
* next paragraph) shall be included in all copies or substantial portions
* of the Software.
*
* 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 AUTHOR(S), AND/OR THE COPYRIGHT HOLDER(S) 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.
*
* Author: Thomas Hellstrom 2004, 2005.
* This code was written using docs obtained under NDA from VIA Inc.
*
* Don't run this code directly on an AGP buffer. Due to cache problems it will
* be very slow.
*/
#include "via_3d_reg.h"
#include "drmP.h"
#include "drm.h"
#include "via_drm.h"
#include "via_verifier.h"
#include "via_drv.h"
typedef enum {
state_command,
state_header2,
state_header1,
state_vheader5,
state_vheader6,
state_error
} verifier_state_t;
typedef enum {
no_check = 0,
check_for_header2,
check_for_header1,
check_for_header2_err,
check_for_header1_err,
check_for_fire,
check_z_buffer_addr0,
check_z_buffer_addr1,
check_z_buffer_addr_mode,
check_destination_addr0,
check_destination_addr1,
check_destination_addr_mode,
check_for_dummy,
check_for_dd,
check_texture_addr0,
check_texture_addr1,
check_texture_addr2,
check_texture_addr3,
check_texture_addr4,
check_texture_addr5,
check_texture_addr6,
check_texture_addr7,
check_texture_addr8,
check_texture_addr_mode,
check_for_vertex_count,
check_number_texunits,
forbidden_command
} hazard_t;
/*
* Associates each hazard above with a possible multi-command
* sequence. For example an address that is split over multiple
* commands and that needs to be checked at the first command
* that does not include any part of the address.
*/
static drm_via_sequence_t seqs[] = {
no_sequence,
no_sequence,
no_sequence,
no_sequence,
no_sequence,
no_sequence,
z_address,
z_address,
z_address,
dest_address,
dest_address,
dest_address,
no_sequence,
no_sequence,
tex_address,
tex_address,
tex_address,
tex_address,
tex_address,
tex_address,
tex_address,
tex_address,
tex_address,
tex_address,
no_sequence
};
typedef struct {
unsigned int code;
hazard_t hz;
} hz_init_t;
static hz_init_t init_table1[] = {
{0xf2, check_for_header2_err},
{0xf0, check_for_header1_err},
{0xee, check_for_fire},
{0xcc, check_for_dummy},
{0xdd, check_for_dd},
{0x00, no_check},
{0x10, check_z_buffer_addr0},
{0x11, check_z_buffer_addr1},
{0x12, check_z_buffer_addr_mode},
{0x13, no_check},
{0x14, no_check},
{0x15, no_check},
{0x23, no_check},
{0x24, no_check},
{0x33, no_check},
{0x34, no_check},
{0x35, no_check},
{0x36, no_check},
{0x37, no_check},
{0x38, no_check},
{0x39, no_check},
{0x3A, no_check},
{0x3B, no_check},
{0x3C, no_check},
{0x3D, no_check},
{0x3E, no_check},
{0x40, check_destination_addr0},
{0x41, check_destination_addr1},
{0x42, check_destination_addr_mode},
{0x43, no_check},
{0x44, no_check},
{0x50, no_check},
{0x51, no_check},
{0x52, no_check},
{0x53, no_check},
{0x54, no_check},
{0x55, no_check},
{0x56, no_check},
{0x57, no_check},
{0x58, no_check},
{0x70, no_check},
{0x71, no_check},
{0x78, no_check},
{0x79, no_check},
{0x7A, no_check},
{0x7B, no_check},
{0x7C, no_check},
{0x7D, check_for_vertex_count}
};
static hz_init_t init_table2[] = {
{0xf2, check_for_header2_err},
{0xf0, check_for_header1_err},
{0xee, check_for_fire},
{0xcc, check_for_dummy},
{0x00, check_texture_addr0},
{0x01, check_texture_addr0},
{0x02, check_texture_addr0},
{0x03, check_texture_addr0},
{0x04, check_texture_addr0},
{0x05, check_texture_addr0},
{0x06, check_texture_addr0},
{0x07, check_texture_addr0},
{0x08, check_texture_addr0},
{0x09, check_texture_addr0},
{0x20, check_texture_addr1},
{0x21, check_texture_addr1},
{0x22, check_texture_addr1},
{0x23, check_texture_addr4},
{0x2B, check_texture_addr3},
{0x2C, check_texture_addr3},
{0x2D, check_texture_addr3},
{0x2E, check_texture_addr3},
{0x2F, check_texture_addr3},
{0x30, check_texture_addr3},
{0x31, check_texture_addr3},
{0x32, check_texture_addr3},
{0x33, check_texture_addr3},
{0x34, check_texture_addr3},
{0x4B, check_texture_addr5},
{0x4C, check_texture_addr6},
{0x51, check_texture_addr7},
{0x52, check_texture_addr8},
{0x77, check_texture_addr2},
{0x78, no_check},
{0x79, no_check},
{0x7A, no_check},
{0x7B, check_texture_addr_mode},
{0x7C, no_check},
{0x7D, no_check},
{0x7E, no_check},
{0x7F, no_check},
{0x80, no_check},
{0x81, no_check},
{0x82, no_check},
{0x83, no_check},
{0x85, no_check},
{0x86, no_check},
{0x87, no_check},
{0x88, no_check},
{0x89, no_check},
{0x8A, no_check},
{0x90, no_check},
{0x91, no_check},
{0x92, no_check},
{0x93, no_check}
};
static hz_init_t init_table3[] = {
{0xf2, check_for_header2_err},
{0xf0, check_for_header1_err},
{0xcc, check_for_dummy},
{0x00, check_number_texunits}
};
static hazard_t table1[256];
static hazard_t table2[256];
static hazard_t table3[256];
static __inline__ int
eat_words(const uint32_t ** buf, const uint32_t * buf_end, unsigned num_words)
{
if ((buf_end - *buf) >= num_words) {
*buf += num_words;
return 0;
}
DRM_ERROR("Illegal termination of DMA command buffer\n");
return 1;
}
/*
* Partially stolen from drm_memory.h
*/
static __inline__ drm_local_map_t *via_drm_lookup_agp_map(drm_via_state_t *seq,
unsigned long offset,
unsigned long size,
struct drm_device *dev)
{
#ifdef __linux__
struct drm_map_list *r_list;
#endif
drm_local_map_t *map = seq->map_cache;
if (map && map->offset <= offset
&& (offset + size) <= (map->offset + map->size)) {
return map;
}
#ifdef __linux__
list_for_each_entry(r_list, &dev->maplist, head) {
map = r_list->map;
if (!map)
continue;
#else
TAILQ_FOREACH(map, &dev->maplist, link) {
#endif
if (map->offset <= offset
&& (offset + size) <= (map->offset + map->size)
&& !(map->flags & _DRM_RESTRICTED)
&& (map->type == _DRM_AGP)) {
seq->map_cache = map;
return map;
}
}
return NULL;
}
/*
* Require that all AGP texture levels reside in the same AGP map which should
* be mappable by the client. This is not a big restriction.
* FIXME: To actually enforce this security policy strictly, drm_rmmap
* would have to wait for dma quiescent before removing an AGP map.
* The via_drm_lookup_agp_map call in reality seems to take
* very little CPU time.
*/
static __inline__ int finish_current_sequence(drm_via_state_t * cur_seq)
{
switch (cur_seq->unfinished) {
case z_address:
DRM_DEBUG("Z Buffer start address is 0x%x\n", cur_seq->z_addr);
break;
case dest_address:
DRM_DEBUG("Destination start address is 0x%x\n",
cur_seq->d_addr);
break;
case tex_address:
if (cur_seq->agp_texture) {
unsigned start =
cur_seq->tex_level_lo[cur_seq->texture];
unsigned end = cur_seq->tex_level_hi[cur_seq->texture];
unsigned long lo = ~0, hi = 0, tmp;
uint32_t *addr, *pitch, *height, tex;
unsigned i;
int npot;
if (end > 9)
end = 9;
if (start > 9)
start = 9;
addr =
&(cur_seq->t_addr[tex = cur_seq->texture][start]);
pitch = &(cur_seq->pitch[tex][start]);
height = &(cur_seq->height[tex][start]);
npot = cur_seq->tex_npot[tex];
for (i = start; i <= end; ++i) {
tmp = *addr++;
if (tmp < lo)
lo = tmp;
if (i == 0 && npot)
tmp += (*height++ * *pitch++);
else
tmp += (*height++ << *pitch++);
if (tmp > hi)
hi = tmp;
}
if (!via_drm_lookup_agp_map
(cur_seq, lo, hi - lo, cur_seq->dev)) {
DRM_ERROR
("AGP texture is not in allowed map\n");
return 2;
}
}
break;
default:
break;
}
cur_seq->unfinished = no_sequence;
return 0;
}
static __inline__ int
investigate_hazard(uint32_t cmd, hazard_t hz, drm_via_state_t * cur_seq)
{
register uint32_t tmp, *tmp_addr;
if (cur_seq->unfinished && (cur_seq->unfinished != seqs[hz])) {
int ret;
if ((ret = finish_current_sequence(cur_seq)))
return ret;
}
switch (hz) {
case check_for_header2:
if (cmd == HALCYON_HEADER2)
return 1;
return 0;
case check_for_header1:
if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1)
return 1;
return 0;
case check_for_header2_err:
if (cmd == HALCYON_HEADER2)
return 1;
DRM_ERROR("Illegal DMA HALCYON_HEADER2 command\n");
break;
case check_for_header1_err:
if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1)
return 1;
DRM_ERROR("Illegal DMA HALCYON_HEADER1 command\n");
break;
case check_for_fire:
if ((cmd & HALCYON_FIREMASK) == HALCYON_FIRECMD)
return 1;
DRM_ERROR("Illegal DMA HALCYON_FIRECMD command\n");
break;
case check_for_dummy:
if (HC_DUMMY == cmd)
return 0;
DRM_ERROR("Illegal DMA HC_DUMMY command\n");
break;
case check_for_dd:
if (0xdddddddd == cmd)
return 0;
DRM_ERROR("Illegal DMA 0xdddddddd command\n");
break;
case check_z_buffer_addr0:
cur_seq->unfinished = z_address;
cur_seq->z_addr = (cur_seq->z_addr & 0xFF000000) |
(cmd & 0x00FFFFFF);
return 0;
case check_z_buffer_addr1:
cur_seq->unfinished = z_address;
cur_seq->z_addr = (cur_seq->z_addr & 0x00FFFFFF) |
((cmd & 0xFF) << 24);
return 0;
case check_z_buffer_addr_mode:
cur_seq->unfinished = z_address;
if ((cmd & 0x0000C000) == 0)
return 0;
DRM_ERROR("Attempt to place Z buffer in system memory\n");
return 2;
case check_destination_addr0:
cur_seq->unfinished = dest_address;
cur_seq->d_addr = (cur_seq->d_addr & 0xFF000000) |
(cmd & 0x00FFFFFF);
return 0;
case check_destination_addr1:
cur_seq->unfinished = dest_address;
cur_seq->d_addr = (cur_seq->d_addr & 0x00FFFFFF) |
((cmd & 0xFF) << 24);
return 0;
case check_destination_addr_mode:
cur_seq->unfinished = dest_address;
if ((cmd & 0x0000C000) == 0)
return 0;
DRM_ERROR
("Attempt to place 3D drawing buffer in system memory\n");
return 2;
case check_texture_addr0:
cur_seq->unfinished = tex_address;
tmp = (cmd >> 24);
tmp_addr = &cur_seq->t_addr[cur_seq->texture][tmp];
*tmp_addr = (*tmp_addr & 0xFF000000) | (cmd & 0x00FFFFFF);
return 0;
case check_texture_addr1:
cur_seq->unfinished = tex_address;
tmp = ((cmd >> 24) - 0x20);
tmp += tmp << 1;
tmp_addr = &cur_seq->t_addr[cur_seq->texture][tmp];
*tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF) << 24);
tmp_addr++;
*tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF00) << 16);
tmp_addr++;
*tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF0000) << 8);
return 0;
case check_texture_addr2:
cur_seq->unfinished = tex_address;
cur_seq->tex_level_lo[tmp = cur_seq->texture] = cmd & 0x3F;
cur_seq->tex_level_hi[tmp] = (cmd & 0xFC0) >> 6;
return 0;
case check_texture_addr3:
cur_seq->unfinished = tex_address;
tmp = ((cmd >> 24) - HC_SubA_HTXnL0Pit);
if (tmp == 0 &&
(cmd & HC_HTXnEnPit_MASK)) {
cur_seq->pitch[cur_seq->texture][tmp] =
(cmd & HC_HTXnLnPit_MASK);
cur_seq->tex_npot[cur_seq->texture] = 1;
} else {
cur_seq->pitch[cur_seq->texture][tmp] =
(cmd & HC_HTXnLnPitE_MASK) >> HC_HTXnLnPitE_SHIFT;
cur_seq->tex_npot[cur_seq->texture] = 0;
if (cmd & 0x000FFFFF) {
DRM_ERROR
("Unimplemented texture level 0 pitch mode.\n");
return 2;
}
}
return 0;
case check_texture_addr4:
cur_seq->unfinished = tex_address;
tmp_addr = &cur_seq->t_addr[cur_seq->texture][9];
*tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF) << 24);
return 0;
case check_texture_addr5:
case check_texture_addr6:
cur_seq->unfinished = tex_address;
/*
* Texture width. We don't care since we have the pitch.
*/
return 0;
case check_texture_addr7:
cur_seq->unfinished = tex_address;
tmp_addr = &(cur_seq->height[cur_seq->texture][0]);
tmp_addr[5] = 1 << ((cmd & 0x00F00000) >> 20);
tmp_addr[4] = 1 << ((cmd & 0x000F0000) >> 16);
tmp_addr[3] = 1 << ((cmd & 0x0000F000) >> 12);
tmp_addr[2] = 1 << ((cmd & 0x00000F00) >> 8);
tmp_addr[1] = 1 << ((cmd & 0x000000F0) >> 4);
tmp_addr[0] = 1 << (cmd & 0x0000000F);
return 0;
case check_texture_addr8:
cur_seq->unfinished = tex_address;
tmp_addr = &(cur_seq->height[cur_seq->texture][0]);
tmp_addr[9] = 1 << ((cmd & 0x0000F000) >> 12);
tmp_addr[8] = 1 << ((cmd & 0x00000F00) >> 8);
tmp_addr[7] = 1 << ((cmd & 0x000000F0) >> 4);
tmp_addr[6] = 1 << (cmd & 0x0000000F);
return 0;
case check_texture_addr_mode:
cur_seq->unfinished = tex_address;
if (2 == (tmp = cmd & 0x00000003)) {
DRM_ERROR
("Attempt to fetch texture from system memory.\n");
return 2;
}
cur_seq->agp_texture = (tmp == 3);
cur_seq->tex_palette_size[cur_seq->texture] =
(cmd >> 16) & 0x000000007;
return 0;
case check_for_vertex_count:
cur_seq->vertex_count = cmd & 0x0000FFFF;
return 0;
case check_number_texunits:
cur_seq->multitex = (cmd >> 3) & 1;
return 0;
default:
DRM_ERROR("Illegal DMA data: 0x%x\n", cmd);
return 2;
}
return 2;
}
static __inline__ int
via_check_prim_list(uint32_t const **buffer, const uint32_t * buf_end,
drm_via_state_t * cur_seq)
{
drm_via_private_t *dev_priv =
(drm_via_private_t *) cur_seq->dev->dev_private;
uint32_t a_fire, bcmd, dw_count;
int ret = 0;
int have_fire;
const uint32_t *buf = *buffer;
while (buf < buf_end) {
have_fire = 0;
if ((buf_end - buf) < 2) {
DRM_ERROR
("Unexpected termination of primitive list.\n");
ret = 1;
break;
}
if ((*buf & HC_ACMD_MASK) != HC_ACMD_HCmdB)
break;
bcmd = *buf++;
if ((*buf & HC_ACMD_MASK) != HC_ACMD_HCmdA) {
DRM_ERROR("Expected Vertex List A command, got 0x%x\n",
*buf);
ret = 1;
break;
}
a_fire =
*buf++ | HC_HPLEND_MASK | HC_HPMValidN_MASK |
HC_HE3Fire_MASK;
/*
* How many dwords per vertex ?
*/
if (cur_seq->agp && ((bcmd & (0xF << 11)) == 0)) {
DRM_ERROR("Illegal B command vertex data for AGP.\n");
ret = 1;
break;
}
dw_count = 0;
if (bcmd & (1 << 7))
dw_count += (cur_seq->multitex) ? 2 : 1;
if (bcmd & (1 << 8))
dw_count += (cur_seq->multitex) ? 2 : 1;
if (bcmd & (1 << 9))
dw_count++;
if (bcmd & (1 << 10))
dw_count++;
if (bcmd & (1 << 11))
dw_count++;
if (bcmd & (1 << 12))
dw_count++;
if (bcmd & (1 << 13))
dw_count++;
if (bcmd & (1 << 14))
dw_count++;
while (buf < buf_end) {
if (*buf == a_fire) {
if (dev_priv->num_fire_offsets >=
VIA_FIRE_BUF_SIZE) {
DRM_ERROR("Fire offset buffer full.\n");
ret = 1;
break;
}
dev_priv->fire_offsets[dev_priv->
num_fire_offsets++] =
buf;
have_fire = 1;
buf++;
if (buf < buf_end && *buf == a_fire)
buf++;
break;
}
if ((*buf == HALCYON_HEADER2) ||
((*buf & HALCYON_FIREMASK) == HALCYON_FIRECMD)) {
DRM_ERROR("Missing Vertex Fire command, "
"Stray Vertex Fire command or verifier "
"lost sync.\n");
ret = 1;
break;
}
if ((ret = eat_words(&buf, buf_end, dw_count)))
break;
}
if (buf >= buf_end && !have_fire) {
DRM_ERROR("Missing Vertex Fire command or verifier "
"lost sync.\n");
ret = 1;
break;
}
if (cur_seq->agp && ((buf - cur_seq->buf_start) & 0x01)) {
DRM_ERROR("AGP Primitive list end misaligned.\n");
ret = 1;
break;
}
}
*buffer = buf;
return ret;
}
static __inline__ verifier_state_t
via_check_header2(uint32_t const **buffer, const uint32_t * buf_end,
drm_via_state_t * hc_state)
{
uint32_t cmd;
int hz_mode;
hazard_t hz;
const uint32_t *buf = *buffer;
const hazard_t *hz_table;
if ((buf_end - buf) < 2) {
DRM_ERROR
("Illegal termination of DMA HALCYON_HEADER2 sequence.\n");
return state_error;
}
buf++;
cmd = (*buf++ & 0xFFFF0000) >> 16;
switch (cmd) {
case HC_ParaType_CmdVdata:
if (via_check_prim_list(&buf, buf_end, hc_state))
return state_error;
*buffer = buf;
return state_command;
case HC_ParaType_NotTex:
hz_table = table1;
break;
case HC_ParaType_Tex:
hc_state->texture = 0;
hz_table = table2;
break;
case (HC_ParaType_Tex | (HC_SubType_Tex1 << 8)):
hc_state->texture = 1;
hz_table = table2;
break;
case (HC_ParaType_Tex | (HC_SubType_TexGeneral << 8)):
hz_table = table3;
break;
case HC_ParaType_Auto:
if (eat_words(&buf, buf_end, 2))
return state_error;
*buffer = buf;
return state_command;
case (HC_ParaType_Palette | (HC_SubType_Stipple << 8)):
if (eat_words(&buf, buf_end, 32))
return state_error;
*buffer = buf;
return state_command;
case (HC_ParaType_Palette | (HC_SubType_TexPalette0 << 8)):
case (HC_ParaType_Palette | (HC_SubType_TexPalette1 << 8)):
DRM_ERROR("Texture palettes are rejected because of "
"lack of info how to determine their size.\n");
return state_error;
case (HC_ParaType_Palette | (HC_SubType_FogTable << 8)):
DRM_ERROR("Fog factor palettes are rejected because of "
"lack of info how to determine their size.\n");
return state_error;
default:
/*
* There are some unimplemented HC_ParaTypes here, that
* need to be implemented if the Mesa driver is extended.
*/
DRM_ERROR("Invalid or unimplemented HALCYON_HEADER2 "
"DMA subcommand: 0x%x. Previous dword: 0x%x\n",
cmd, *(buf - 2));
*buffer = buf;
return state_error;
}
while (buf < buf_end) {
cmd = *buf++;
if ((hz = hz_table[cmd >> 24])) {
if ((hz_mode = investigate_hazard(cmd, hz, hc_state))) {
if (hz_mode == 1) {
buf--;
break;
}
return state_error;
}
} else if (hc_state->unfinished &&
finish_current_sequence(hc_state)) {
return state_error;
}
}
if (hc_state->unfinished && finish_current_sequence(hc_state)) {
return state_error;
}
*buffer = buf;
return state_command;
}
static __inline__ verifier_state_t
via_parse_header2(drm_via_private_t * dev_priv, uint32_t const **buffer,
const uint32_t * buf_end, int *fire_count)
{
uint32_t cmd;
const uint32_t *buf = *buffer;
const uint32_t *next_fire;
int burst = 0;
next_fire = dev_priv->fire_offsets[*fire_count];
buf++;
cmd = (*buf & 0xFFFF0000) >> 16;
VIA_WRITE(HC_REG_TRANS_SET + HC_REG_BASE, *buf++);
switch (cmd) {
case HC_ParaType_CmdVdata:
while ((buf < buf_end) &&
(*fire_count < dev_priv->num_fire_offsets) &&
(*buf & HC_ACMD_MASK) == HC_ACMD_HCmdB) {
while (buf <= next_fire) {
VIA_WRITE(HC_REG_TRANS_SPACE + HC_REG_BASE +
(burst & 63), *buf++);
burst += 4;
}
if ((buf < buf_end)
&& ((*buf & HALCYON_FIREMASK) == HALCYON_FIRECMD))
buf++;
if (++(*fire_count) < dev_priv->num_fire_offsets)
next_fire = dev_priv->fire_offsets[*fire_count];
}
break;
default:
while (buf < buf_end) {
if (*buf == HC_HEADER2 ||
(*buf & HALCYON_HEADER1MASK) == HALCYON_HEADER1 ||
(*buf & VIA_VIDEOMASK) == VIA_VIDEO_HEADER5 ||
(*buf & VIA_VIDEOMASK) == VIA_VIDEO_HEADER6)
break;
VIA_WRITE(HC_REG_TRANS_SPACE + HC_REG_BASE +
(burst & 63), *buf++);
burst += 4;
}
}
*buffer = buf;
return state_command;
}
static __inline__ int verify_mmio_address(uint32_t address)
{
if ((address > 0x3FF) && (address < 0xC00)) {
DRM_ERROR("Invalid VIDEO DMA command. "
"Attempt to access 3D- or command burst area.\n");
return 1;
} else if ((address > 0xCFF) && (address < 0x1300)) {
DRM_ERROR("Invalid VIDEO DMA command. "
"Attempt to access PCI DMA area.\n");
return 1;
} else if (address > 0x13FF) {
DRM_ERROR("Invalid VIDEO DMA command. "
"Attempt to access VGA registers.\n");
return 1;
}
return 0;
}
static __inline__ int
verify_video_tail(uint32_t const **buffer, const uint32_t * buf_end,
uint32_t dwords)
{
const uint32_t *buf = *buffer;
if (buf_end - buf < dwords) {
DRM_ERROR("Illegal termination of video command.\n");
return 1;
}
while (dwords--) {
if (*buf++) {
DRM_ERROR("Illegal video command tail.\n");
return 1;
}
}
*buffer = buf;
return 0;
}
static __inline__ verifier_state_t
via_check_header1(uint32_t const **buffer, const uint32_t * buf_end)
{
uint32_t cmd;
const uint32_t *buf = *buffer;
verifier_state_t ret = state_command;
while (buf < buf_end) {
cmd = *buf;
if ((cmd > ((0x3FF >> 2) | HALCYON_HEADER1)) &&
(cmd < ((0xC00 >> 2) | HALCYON_HEADER1))) {
if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1)
break;
DRM_ERROR("Invalid HALCYON_HEADER1 command. "
"Attempt to access 3D- or command burst area.\n");
ret = state_error;
break;
} else if (cmd > ((0xCFF >> 2) | HALCYON_HEADER1)) {
if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1)
break;
DRM_ERROR("Invalid HALCYON_HEADER1 command. "
"Attempt to access VGA registers.\n");
ret = state_error;
break;
} else {
buf += 2;
}
}
*buffer = buf;
return ret;
}
static __inline__ verifier_state_t
via_parse_header1(drm_via_private_t * dev_priv, uint32_t const **buffer,
const uint32_t * buf_end)
{
register uint32_t cmd;
const uint32_t *buf = *buffer;
while (buf < buf_end) {
cmd = *buf;
if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1)
break;
VIA_WRITE((cmd & ~HALCYON_HEADER1MASK) << 2, *++buf);
buf++;
}
*buffer = buf;
return state_command;
}
static __inline__ verifier_state_t
via_check_vheader5(uint32_t const **buffer, const uint32_t * buf_end)
{
uint32_t data;
const uint32_t *buf = *buffer;
if (buf_end - buf < 4) {
DRM_ERROR("Illegal termination of video header5 command\n");
return state_error;
}
data = *buf++ & ~VIA_VIDEOMASK;
if (verify_mmio_address(data))
return state_error;
data = *buf++;
if (*buf++ != 0x00F50000) {
DRM_ERROR("Illegal header5 header data\n");
return state_error;
}
if (*buf++ != 0x00000000) {
DRM_ERROR("Illegal header5 header data\n");
return state_error;
}
if (eat_words(&buf, buf_end, data))
return state_error;
if ((data & 3) && verify_video_tail(&buf, buf_end, 4 - (data & 3)))
return state_error;
*buffer = buf;
return state_command;
}
static __inline__ verifier_state_t
via_parse_vheader5(drm_via_private_t * dev_priv, uint32_t const **buffer,
const uint32_t * buf_end)
{
uint32_t addr, count, i;
const uint32_t *buf = *buffer;
addr = *buf++ & ~VIA_VIDEOMASK;
i = count = *buf;
buf += 3;
while (i--) {
VIA_WRITE(addr, *buf++);
}
if (count & 3)
buf += 4 - (count & 3);
*buffer = buf;
return state_command;
}
static __inline__ verifier_state_t
via_check_vheader6(uint32_t const **buffer, const uint32_t * buf_end)
{
uint32_t data;
const uint32_t *buf = *buffer;
uint32_t i;
if (buf_end - buf < 4) {
DRM_ERROR("Illegal termination of video header6 command\n");
return state_error;
}
buf++;
data = *buf++;
if (*buf++ != 0x00F60000) {
DRM_ERROR("Illegal header6 header data\n");
return state_error;
}
if (*buf++ != 0x00000000) {
DRM_ERROR("Illegal header6 header data\n");
return state_error;
}
if ((buf_end - buf) < (data << 1)) {
DRM_ERROR("Illegal termination of video header6 command\n");
return state_error;
}
for (i = 0; i < data; ++i) {
if (verify_mmio_address(*buf++))
return state_error;
buf++;
}
data <<= 1;
if ((data & 3) && verify_video_tail(&buf, buf_end, 4 - (data & 3)))
return state_error;
*buffer = buf;
return state_command;
}
static __inline__ verifier_state_t
via_parse_vheader6(drm_via_private_t * dev_priv, uint32_t const **buffer,
const uint32_t * buf_end)
{
uint32_t addr, count, i;
const uint32_t *buf = *buffer;
i = count = *++buf;
buf += 3;
while (i--) {
addr = *buf++;
VIA_WRITE(addr, *buf++);
}
count <<= 1;
if (count & 3)
buf += 4 - (count & 3);
*buffer = buf;
return state_command;
}
int
via_verify_command_stream(const uint32_t * buf, unsigned int size,
struct drm_device * dev, int agp)
{
drm_via_private_t *dev_priv = (drm_via_private_t *) dev->dev_private;
drm_via_state_t *hc_state = &dev_priv->hc_state;
drm_via_state_t saved_state = *hc_state;
uint32_t cmd;
const uint32_t *buf_end = buf + (size >> 2);
verifier_state_t state = state_command;
int cme_video;
int supported_3d;
cme_video = (dev_priv->chipset == VIA_PRO_GROUP_A ||
dev_priv->chipset == VIA_DX9_0);
supported_3d = dev_priv->chipset != VIA_DX9_0;
hc_state->dev = dev;
hc_state->unfinished = no_sequence;
hc_state->map_cache = NULL;
hc_state->agp = agp;
hc_state->buf_start = buf;
dev_priv->num_fire_offsets = 0;
while (buf < buf_end) {
switch (state) {
case state_header2:
state = via_check_header2(&buf, buf_end, hc_state);
break;
case state_header1:
state = via_check_header1(&buf, buf_end);
break;
case state_vheader5:
state = via_check_vheader5(&buf, buf_end);
break;
case state_vheader6:
state = via_check_vheader6(&buf, buf_end);
break;
case state_command:
if ((HALCYON_HEADER2 == (cmd = *buf)) &&
supported_3d)
state = state_header2;
else if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1)
state = state_header1;
else if (cme_video
&& (cmd & VIA_VIDEOMASK) == VIA_VIDEO_HEADER5)
state = state_vheader5;
else if (cme_video
&& (cmd & VIA_VIDEOMASK) == VIA_VIDEO_HEADER6)
state = state_vheader6;
else if ((cmd == HALCYON_HEADER2) && !supported_3d) {
DRM_ERROR("Accelerated 3D is not supported on this chipset yet.\n");
state = state_error;
} else {
DRM_ERROR
("Invalid / Unimplemented DMA HEADER command. 0x%x\n",
cmd);
state = state_error;
}
break;
case state_error:
default:
*hc_state = saved_state;
return -EINVAL;
}
}
if (state == state_error) {
*hc_state = saved_state;
return -EINVAL;
}
return 0;
}
int
via_parse_command_stream(struct drm_device * dev, const uint32_t * buf,
unsigned int size)
{
drm_via_private_t *dev_priv = (drm_via_private_t *) dev->dev_private;
uint32_t cmd;
const uint32_t *buf_end = buf + (size >> 2);
verifier_state_t state = state_command;
int fire_count = 0;
while (buf < buf_end) {
switch (state) {
case state_header2:
state =
via_parse_header2(dev_priv, &buf, buf_end,
&fire_count);
break;
case state_header1:
state = via_parse_header1(dev_priv, &buf, buf_end);
break;
case state_vheader5:
state = via_parse_vheader5(dev_priv, &buf, buf_end);
break;
case state_vheader6:
state = via_parse_vheader6(dev_priv, &buf, buf_end);
break;
case state_command:
if (HALCYON_HEADER2 == (cmd = *buf))
state = state_header2;
else if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1)
state = state_header1;
else if ((cmd & VIA_VIDEOMASK) == VIA_VIDEO_HEADER5)
state = state_vheader5;
else if ((cmd & VIA_VIDEOMASK) == VIA_VIDEO_HEADER6)
state = state_vheader6;
else {
DRM_ERROR
("Invalid / Unimplemented DMA HEADER command. 0x%x\n",
cmd);
state = state_error;
}
break;
case state_error:
default:
return -EINVAL;
}
}
if (state == state_error) {
return -EINVAL;
}
return 0;
}
static void
setup_hazard_table(hz_init_t init_table[], hazard_t table[], int size)
{
int i;
for (i = 0; i < 256; ++i) {
table[i] = forbidden_command;
}
for (i = 0; i < size; ++i) {
table[init_table[i].code] = init_table[i].hz;
}
}
void via_init_command_verifier(void)
{
setup_hazard_table(init_table1, table1,
sizeof(init_table1) / sizeof(hz_init_t));
setup_hazard_table(init_table2, table2,
sizeof(init_table2) / sizeof(hz_init_t));
setup_hazard_table(init_table3, table3,
sizeof(init_table3) / sizeof(hz_init_t));
}
d='n754' href='#n754'>754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
/* mga_dma.c -- DMA support for mga g200/g400 -*- linux-c -*-
* Created: Mon Dec 13 01:50:01 1999 by jhartmann@precisioninsight.com
*/
/* Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
* Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
* 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, sublicense,
* 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 above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* 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 NONINFRINGEMENT. IN NO EVENT SHALL
* PRECISION INSIGHT 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.
*/
/**
* \file mga_dma.c
* DMA support for MGA G200 / G400.
*
* \author Rickard E. (Rik) Faith <faith@valinux.com>
* \author Jeff Hartmann <jhartmann@valinux.com>
* \author Keith Whitwell <keith@tungstengraphics.com>
* \author Gareth Hughes <gareth@valinux.com>
*/
#include "drmP.h"
#include "drm.h"
#include "drm_sarea.h"
#include "mga_drm.h"
#include "mga_drv.h"
#define MGA_DEFAULT_USEC_TIMEOUT 10000
#define MGA_FREELIST_DEBUG 0
#define MINIMAL_CLEANUP 0
#define FULL_CLEANUP 1
static int mga_do_cleanup_dma(drm_device_t * dev, int full_cleanup);
/* ================================================================
* Engine control
*/
int mga_do_wait_for_idle(drm_mga_private_t * dev_priv)
{
u32 status = 0;
int i;
DRM_DEBUG("\n");
for (i = 0; i < dev_priv->usec_timeout; i++) {
status = MGA_READ(MGA_STATUS) & MGA_ENGINE_IDLE_MASK;
if (status == MGA_ENDPRDMASTS) {
MGA_WRITE8(MGA_CRTC_INDEX, 0);
return 0;
}
DRM_UDELAY(1);
}
#if MGA_DMA_DEBUG
DRM_ERROR("failed!\n");
DRM_INFO(" status=0x%08x\n", status);
#endif
return DRM_ERR(EBUSY);
}
static int mga_do_dma_reset(drm_mga_private_t * dev_priv)
{
drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
DRM_DEBUG("\n");
/* The primary DMA stream should look like new right about now.
*/
primary->tail = 0;
primary->space = primary->size;
primary->last_flush = 0;
sarea_priv->last_wrap = 0;
/* FIXME: Reset counters, buffer ages etc...
*/
/* FIXME: What else do we need to reinitialize? WARP stuff?
*/
return 0;
}
/* ================================================================
* Primary DMA stream
*/
void mga_do_dma_flush(drm_mga_private_t * dev_priv)
{
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
u32 head, tail;
u32 status = 0;
int i;
DMA_LOCALS;
DRM_DEBUG("\n");
/* We need to wait so that we can do an safe flush */
for (i = 0; i < dev_priv->usec_timeout; i++) {
status = MGA_READ(MGA_STATUS) & MGA_ENGINE_IDLE_MASK;
if (status == MGA_ENDPRDMASTS)
break;
DRM_UDELAY(1);
}
if (primary->tail == primary->last_flush) {
DRM_DEBUG(" bailing out...\n");
return;
}
tail = primary->tail + dev_priv->primary->offset;
/* We need to pad the stream between flushes, as the card
* actually (partially?) reads the first of these commands.
* See page 4-16 in the G400 manual, middle of the page or so.
*/
BEGIN_DMA(1);
DMA_BLOCK(MGA_DMAPAD, 0x00000000,
MGA_DMAPAD, 0x00000000,
MGA_DMAPAD, 0x00000000, MGA_DMAPAD, 0x00000000);
ADVANCE_DMA();
primary->last_flush = primary->tail;
head = MGA_READ(MGA_PRIMADDRESS);
if (head <= tail) {
primary->space = primary->size - primary->tail;
} else {
primary->space = head - tail;
}
DRM_DEBUG(" head = 0x%06lx\n", head - dev_priv->primary->offset);
DRM_DEBUG(" tail = 0x%06lx\n", tail - dev_priv->primary->offset);
DRM_DEBUG(" space = 0x%06x\n", primary->space);
mga_flush_write_combine();
MGA_WRITE(MGA_PRIMEND, tail | dev_priv->dma_access);
DRM_DEBUG("done.\n");
}
void mga_do_dma_wrap_start(drm_mga_private_t * dev_priv)
{
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
u32 head, tail;
DMA_LOCALS;
DRM_DEBUG("\n");
BEGIN_DMA_WRAP();
DMA_BLOCK(MGA_DMAPAD, 0x00000000,
MGA_DMAPAD, 0x00000000,
MGA_DMAPAD, 0x00000000, MGA_DMAPAD, 0x00000000);
ADVANCE_DMA();
tail = primary->tail + dev_priv->primary->offset;
primary->tail = 0;
primary->last_flush = 0;
primary->last_wrap++;
head = MGA_READ(MGA_PRIMADDRESS);
if (head == dev_priv->primary->offset) {
primary->space = primary->size;
} else {
primary->space = head - dev_priv->primary->offset;
}
DRM_DEBUG(" head = 0x%06lx\n", head - dev_priv->primary->offset);
DRM_DEBUG(" tail = 0x%06x\n", primary->tail);
DRM_DEBUG(" wrap = %d\n", primary->last_wrap);
DRM_DEBUG(" space = 0x%06x\n", primary->space);
mga_flush_write_combine();
MGA_WRITE(MGA_PRIMEND, tail | dev_priv->dma_access);
set_bit(0, &primary->wrapped);
DRM_DEBUG("done.\n");
}
void mga_do_dma_wrap_end(drm_mga_private_t * dev_priv)
{
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
u32 head = dev_priv->primary->offset;
DRM_DEBUG("\n");
sarea_priv->last_wrap++;
DRM_DEBUG(" wrap = %d\n", sarea_priv->last_wrap);
mga_flush_write_combine();
MGA_WRITE(MGA_PRIMADDRESS, head | MGA_DMA_GENERAL);
clear_bit(0, &primary->wrapped);
DRM_DEBUG("done.\n");
}
/* ================================================================
* Freelist management
*/
#define MGA_BUFFER_USED ~0
#define MGA_BUFFER_FREE 0
#if MGA_FREELIST_DEBUG
static void mga_freelist_print(drm_device_t * dev)
{
drm_mga_private_t *dev_priv = dev->dev_private;
drm_mga_freelist_t *entry;
DRM_INFO("\n");
DRM_INFO("current dispatch: last=0x%x done=0x%x\n",
dev_priv->sarea_priv->last_dispatch,
(unsigned int)(MGA_READ(MGA_PRIMADDRESS) -
dev_priv->primary->offset));
DRM_INFO("current freelist:\n");
for (entry = dev_priv->head->next; entry; entry = entry->next) {
DRM_INFO(" %p idx=%2d age=0x%x 0x%06lx\n",
entry, entry->buf->idx, entry->age.head,
entry->age.head - dev_priv->primary->offset);
}
DRM_INFO("\n");
}
#endif
static int mga_freelist_init(drm_device_t * dev, drm_mga_private_t * dev_priv)
{
drm_device_dma_t *dma = dev->dma;
drm_buf_t *buf;
drm_mga_buf_priv_t *buf_priv;
drm_mga_freelist_t *entry;
int i;
DRM_DEBUG("count=%d\n", dma->buf_count);
dev_priv->head = drm_alloc(sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
if (dev_priv->head == NULL)
return DRM_ERR(ENOMEM);
memset(dev_priv->head, 0, sizeof(drm_mga_freelist_t));
SET_AGE(&dev_priv->head->age, MGA_BUFFER_USED, 0);
for (i = 0; i < dma->buf_count; i++) {
buf = dma->buflist[i];
buf_priv = buf->dev_private;
entry = drm_alloc(sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
if (entry == NULL)
return DRM_ERR(ENOMEM);
memset(entry, 0, sizeof(drm_mga_freelist_t));
entry->next = dev_priv->head->next;
entry->prev = dev_priv->head;
SET_AGE(&entry->age, MGA_BUFFER_FREE, 0);
entry->buf = buf;
if (dev_priv->head->next != NULL)
dev_priv->head->next->prev = entry;
if (entry->next == NULL)
dev_priv->tail = entry;
buf_priv->list_entry = entry;
buf_priv->discard = 0;
buf_priv->dispatched = 0;
dev_priv->head->next = entry;
}
return 0;
}
static void mga_freelist_cleanup(drm_device_t * dev)
{
drm_mga_private_t *dev_priv = dev->dev_private;
drm_mga_freelist_t *entry;
drm_mga_freelist_t *next;
DRM_DEBUG("\n");
entry = dev_priv->head;
while (entry) {
next = entry->next;
drm_free(entry, sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
entry = next;
}
dev_priv->head = dev_priv->tail = NULL;
}
#if 0
/* FIXME: Still needed?
*/
static void mga_freelist_reset(drm_device_t * dev)
{
drm_device_dma_t *dma = dev->dma;
drm_buf_t *buf;
drm_mga_buf_priv_t *buf_priv;
int i;
for (i = 0; i < dma->buf_count; i++) {
buf = dma->buflist[i];
buf_priv = buf->dev_private;
SET_AGE(&buf_priv->list_entry->age, MGA_BUFFER_FREE, 0);
}
}
#endif
static drm_buf_t *mga_freelist_get(drm_device_t * dev)
{
drm_mga_private_t *dev_priv = dev->dev_private;
drm_mga_freelist_t *next;
drm_mga_freelist_t *prev;
drm_mga_freelist_t *tail = dev_priv->tail;
u32 head, wrap;
DRM_DEBUG("\n");
head = MGA_READ(MGA_PRIMADDRESS);
wrap = dev_priv->sarea_priv->last_wrap;
DRM_DEBUG(" tail=0x%06lx %d\n",
tail->age.head ?
tail->age.head - dev_priv->primary->offset : 0,
tail->age.wrap);
DRM_DEBUG(" head=0x%06lx %d\n",
head - dev_priv->primary->offset, wrap);
if (TEST_AGE(&tail->age, head, wrap)) {
prev = dev_priv->tail->prev;
next = dev_priv->tail;
prev->next = NULL;
next->prev = next->next = NULL;
dev_priv->tail = prev;
SET_AGE(&next->age, MGA_BUFFER_USED, 0);
return next->buf;
}
DRM_DEBUG("returning NULL!\n");
return NULL;
}
int mga_freelist_put(drm_device_t * dev, drm_buf_t * buf)
{
drm_mga_private_t *dev_priv = dev->dev_private;
drm_mga_buf_priv_t *buf_priv = buf->dev_private;
drm_mga_freelist_t *head, *entry, *prev;
DRM_DEBUG("age=0x%06lx wrap=%d\n",
buf_priv->list_entry->age.head -
dev_priv->primary->offset, buf_priv->list_entry->age.wrap);
entry = buf_priv->list_entry;
head = dev_priv->head;
if (buf_priv->list_entry->age.head == MGA_BUFFER_USED) {
SET_AGE(&entry->age, MGA_BUFFER_FREE, 0);
prev = dev_priv->tail;
prev->next = entry;
entry->prev = prev;
entry->next = NULL;
} else {
prev = head->next;
head->next = entry;
prev->prev = entry;
entry->prev = head;
entry->next = prev;
}
return 0;
}
/* ================================================================
* DMA initialization, cleanup
*/
int mga_driver_load(drm_device_t *dev, unsigned long flags)
{
drm_mga_private_t * dev_priv;
dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
if (!dev_priv)
return DRM_ERR(ENOMEM);
dev->dev_private = (void *)dev_priv;
memset(dev_priv, 0, sizeof(drm_mga_private_t));
dev_priv->usec_timeout = MGA_DEFAULT_USEC_TIMEOUT;
dev_priv->chipset = flags;
dev_priv->mmio_base = drm_get_resource_start(dev, 1);
dev_priv->mmio_size = drm_get_resource_len(dev, 1);
dev->counters += 3;
dev->types[6] = _DRM_STAT_IRQ;
dev->types[7] = _DRM_STAT_PRIMARY;
dev->types[8] = _DRM_STAT_SECONDARY;
return 0;
}
/**
* Bootstrap the driver for AGP DMA.
*
* \todo
* Investigate whether there is any benifit to storing the WARP microcode in
* AGP memory. If not, the microcode may as well always be put in PCI
* memory.
*
* \todo
* This routine needs to set dma_bs->agp_mode to the mode actually configured
* in the hardware. Looking just at the Linux AGP driver code, I don't see
* an easy way to determine this.
*
* \sa mga_do_dma_bootstrap, mga_do_pci_dma_bootstrap
*/
static int mga_do_agp_dma_bootstrap(drm_device_t * dev,
drm_mga_dma_bootstrap_t * dma_bs)
{
drm_mga_private_t * const dev_priv = (drm_mga_private_t *) dev->dev_private;
unsigned int warp_size = mga_warp_microcode_size(dev_priv);
int err;
unsigned offset;
const unsigned secondary_size = dma_bs->secondary_bin_count
* dma_bs->secondary_bin_size;
const unsigned agp_size = (dma_bs->agp_size << 20);
drm_buf_desc_t req;
drm_agp_mode_t mode;
drm_agp_info_t info;
drm_agp_buffer_t agp_req;
drm_agp_binding_t bind_req;
/* Acquire AGP. */
err = drm_agp_acquire(dev);
if (err) {
DRM_ERROR("Unable to acquire AGP: %d\n", err);
return err;
}
err = drm_agp_info(dev, &info);
if (err) {
DRM_ERROR("Unable to get AGP info: %d\n", err);
return err;
}
mode.mode = (info.mode & ~0x07) | dma_bs->agp_mode;
err = drm_agp_enable(dev, mode);
if (err) {
DRM_ERROR("Unable to enable AGP (mode = 0x%lx)\n", mode.mode);
return err;
}
/* In addition to the usual AGP mode configuration, the G200 AGP cards
* need to have the AGP mode "manually" set.
*/
if (dev_priv->chipset == MGA_CARD_TYPE_G200) {
if (mode.mode & 0x02) {
MGA_WRITE(MGA_AGP_PLL, MGA_AGP2XPLL_ENABLE);
} else {
MGA_WRITE(MGA_AGP_PLL, MGA_AGP2XPLL_DISABLE);
}
}
/* Allocate and bind AGP memory. */
agp_req.size = agp_size;
agp_req.type = 0;
err = drm_agp_alloc( dev, & agp_req );
if (err) {
dev_priv->agp_size = 0;
DRM_ERROR("Unable to allocate %uMB AGP memory\n",
dma_bs->agp_size);
return err;
}
dev_priv->agp_size = agp_size;
dev_priv->agp_handle = agp_req.handle;
bind_req.handle = agp_req.handle;
bind_req.offset = 0;
err = drm_agp_bind( dev, &bind_req );
if (err) {
DRM_ERROR("Unable to bind AGP memory: %d\n", err);
return err;
}
/* Make drm_addbufs happy by not trying to create a mapping for less
* than a page.
*/
if (warp_size < PAGE_SIZE)
warp_size = PAGE_SIZE;
offset = 0;
err = drm_addmap( dev, offset, warp_size,
_DRM_AGP, _DRM_READ_ONLY, & dev_priv->warp );
if (err) {
DRM_ERROR("Unable to map WARP microcode: %d\n", err);
return err;
}
offset += warp_size;
err = drm_addmap( dev, offset, dma_bs->primary_size,
_DRM_AGP, _DRM_READ_ONLY, & dev_priv->primary );
if (err) {
DRM_ERROR("Unable to map primary DMA region: %d\n", err);
return err;
}
offset += dma_bs->primary_size;
err = drm_addmap( dev, offset, secondary_size,
_DRM_AGP, 0, & dev->agp_buffer_map );
if (err) {
DRM_ERROR("Unable to map secondary DMA region: %d\n", err);
return err;
}
(void) memset( &req, 0, sizeof(req) );
req.count = dma_bs->secondary_bin_count;
req.size = dma_bs->secondary_bin_size;
req.flags = _DRM_AGP_BUFFER;
req.agp_start = offset;
err = drm_addbufs_agp( dev, & req );
if (err) {
DRM_ERROR("Unable to add secondary DMA buffers: %d\n", err);
return err;
}
#ifdef __linux__
{
drm_map_list_t *_entry;
unsigned long agp_token = 0;
list_for_each_entry(_entry, &dev->maplist, head) {
if (_entry->map == dev->agp_buffer_map)
agp_token = _entry->user_token;
}
if (!agp_token)
return -EFAULT;
dev->agp_buffer_token = agp_token;
}
#endif
offset += secondary_size;
err = drm_addmap( dev, offset, agp_size - offset,
_DRM_AGP, 0, & dev_priv->agp_textures );
if (err) {
DRM_ERROR("Unable to map AGP texture region: %d\n", err);
return err;
}
drm_core_ioremap(dev_priv->warp, dev);
drm_core_ioremap(dev_priv->primary, dev);
drm_core_ioremap(dev->agp_buffer_map, dev);
if (!dev_priv->warp->handle ||
!dev_priv->primary->handle || !dev->agp_buffer_map->handle) {
DRM_ERROR("failed to ioremap agp regions! (%p, %p, %p)\n",
dev_priv->warp->handle, dev_priv->primary->handle,
dev->agp_buffer_map->handle);
return DRM_ERR(ENOMEM);
}
dev_priv->dma_access = MGA_PAGPXFER;
dev_priv->wagp_enable = MGA_WAGP_ENABLE;
DRM_INFO("Initialized card for AGP DMA.\n");
return 0;
}
/**
* Bootstrap the driver for PCI DMA.
*
* \todo
* The algorithm for decreasing the size of the primary DMA buffer could be
* better. The size should be rounded up to the nearest page size, then
* decrease the request size by a single page each pass through the loop.
*
* \todo
* Determine whether the maximum address passed to drm_pci_alloc is correct.
* The same goes for drm_addbufs_pci.
*
* \sa mga_do_dma_bootstrap, mga_do_agp_dma_bootstrap
*/
static int mga_do_pci_dma_bootstrap(drm_device_t * dev,
drm_mga_dma_bootstrap_t * dma_bs)
{
drm_mga_private_t * const dev_priv = (drm_mga_private_t *) dev->dev_private;
unsigned int warp_size = mga_warp_microcode_size(dev_priv);
unsigned int primary_size;
unsigned int bin_count;
int err;
drm_buf_desc_t req;
if (dev->dma == NULL) {
DRM_ERROR("dev->dma is NULL\n");
return DRM_ERR(EFAULT);
}
/* Make drm_addbufs happy by not trying to create a mapping for less
* than a page.
*/
if (warp_size < PAGE_SIZE)
warp_size = PAGE_SIZE;
/* The proper alignment is 0x100 for this mapping */
err = drm_addmap(dev, 0, warp_size, _DRM_CONSISTENT,
_DRM_READ_ONLY, &dev_priv->warp);
if (err != 0) {
DRM_ERROR("Unable to create mapping for WARP microcode: %d\n",
err);
return err;
}
/* Other than the bottom two bits being used to encode other
* information, there don't appear to be any restrictions on the
* alignment of the primary or secondary DMA buffers.
*/
for ( primary_size = dma_bs->primary_size
; primary_size != 0
; primary_size >>= 1 ) {
/* The proper alignment for this mapping is 0x04 */
err = drm_addmap(dev, 0, primary_size, _DRM_CONSISTENT,
_DRM_READ_ONLY, &dev_priv->primary);
if (!err)
break;
}
if (err != 0) {
DRM_ERROR("Unable to allocate primary DMA region: %d\n", err);
return DRM_ERR(ENOMEM);
}
if (dev_priv->primary->size != dma_bs->primary_size) {
DRM_INFO("Primary DMA buffer size reduced from %u to %u.\n",
dma_bs->primary_size,
(unsigned) dev_priv->primary->size);
dma_bs->primary_size = dev_priv->primary->size;
}
for ( bin_count = dma_bs->secondary_bin_count
; bin_count > 0
; bin_count-- ) {
(void) memset( &req, 0, sizeof(req) );
req.count = bin_count;
req.size = dma_bs->secondary_bin_size;
err = drm_addbufs_pci( dev, & req );
if (!err) {
break;
}
}
if (bin_count == 0) {
DRM_ERROR("Unable to add secondary DMA buffers: %d\n", err);
return err;
}
if (bin_count != dma_bs->secondary_bin_count) {
DRM_INFO("Secondary PCI DMA buffer bin count reduced from %u "
"to %u.\n", dma_bs->secondary_bin_count, bin_count);
dma_bs->secondary_bin_count = bin_count;
}
dev_priv->dma_access = 0;
dev_priv->wagp_enable = 0;
dma_bs->agp_mode = 0;
DRM_INFO("Initialized card for PCI DMA.\n");
return 0;
}
static int mga_do_dma_bootstrap(drm_device_t * dev,
drm_mga_dma_bootstrap_t * dma_bs)
{
const int is_agp = (dma_bs->agp_mode != 0) && drm_device_is_agp(dev);
int err;
drm_mga_private_t * const dev_priv =
(drm_mga_private_t *) dev->dev_private;
dev_priv->used_new_dma_init = 1;
/* The first steps are the same for both PCI and AGP based DMA. Map
* the cards MMIO registers and map a status page.
*/
err = drm_addmap( dev, dev_priv->mmio_base, dev_priv->mmio_size,
_DRM_REGISTERS, _DRM_READ_ONLY, & dev_priv->mmio );
if (err) {
DRM_ERROR("Unable to map MMIO region: %d\n", err);
return err;
}
err = drm_addmap( dev, 0, SAREA_MAX, _DRM_SHM,
_DRM_READ_ONLY | _DRM_LOCKED | _DRM_KERNEL,
& dev_priv->status );
if (err) {
DRM_ERROR("Unable to map status region: %d\n", err);
return err;
}
/* The DMA initialization procedure is slightly different for PCI and
* AGP cards. AGP cards just allocate a large block of AGP memory and
* carve off portions of it for internal uses. The remaining memory
* is returned to user-mode to be used for AGP textures.
*/
if (is_agp) {
err = mga_do_agp_dma_bootstrap(dev, dma_bs);
}
/* If we attempted to initialize the card for AGP DMA but failed,
* clean-up any mess that may have been created.
*/
if (err) {
mga_do_cleanup_dma(dev, MINIMAL_CLEANUP);
}
/* Not only do we want to try and initialized PCI cards for PCI DMA,
* but we also try to initialized AGP cards that could not be
* initialized for AGP DMA. This covers the case where we have an AGP
* card in a system with an unsupported AGP chipset. In that case the
* card will be detected as AGP, but we won't be able to allocate any
* AGP memory, etc.
*/
if (!is_agp || err) {
err = mga_do_pci_dma_bootstrap(dev, dma_bs);
}
return err;
}
int mga_dma_bootstrap(DRM_IOCTL_ARGS)
{
DRM_DEVICE;
drm_mga_dma_bootstrap_t bootstrap;
int err;
static const int modes[] = { 0, 1, 2, 2, 4, 4, 4, 4 };
const drm_mga_private_t * const dev_priv =
(drm_mga_private_t *) dev->dev_private;
DRM_COPY_FROM_USER_IOCTL(bootstrap,
(drm_mga_dma_bootstrap_t __user *) data,
sizeof(bootstrap));
err = mga_do_dma_bootstrap(dev, & bootstrap);
if (err) {
mga_do_cleanup_dma(dev, FULL_CLEANUP);
return err;
}
if (dev_priv->agp_textures != NULL) {
bootstrap.texture_handle = dev_priv->agp_textures->offset;
bootstrap.texture_size = dev_priv->agp_textures->size;
} else {
bootstrap.texture_handle = 0;
bootstrap.texture_size = 0;
}
bootstrap.agp_mode = modes[bootstrap.agp_mode & 0x07];
DRM_COPY_TO_USER_IOCTL((drm_mga_dma_bootstrap_t __user *)data,
bootstrap, sizeof(bootstrap));
return 0;
}
static int mga_do_init_dma(drm_device_t * dev, drm_mga_init_t * init)
{
drm_mga_private_t *dev_priv;
int ret;
DRM_DEBUG("\n");
dev_priv = dev->dev_private;
if (init->sgram) {
dev_priv->clear_cmd = MGA_DWGCTL_CLEAR | MGA_ATYPE_BLK;
} else {
dev_priv->clear_cmd = MGA_DWGCTL_CLEAR | MGA_ATYPE_RSTR;
}
dev_priv->maccess = init->maccess;
dev_priv->fb_cpp = init->fb_cpp;
dev_priv->front_offset = init->front_offset;
dev_priv->front_pitch = init->front_pitch;
dev_priv->back_offset = init->back_offset;
dev_priv->back_pitch = init->back_pitch;
dev_priv->depth_cpp = init->depth_cpp;
dev_priv->depth_offset = init->depth_offset;
dev_priv->depth_pitch = init->depth_pitch;
/* FIXME: Need to support AGP textures...
*/
dev_priv->texture_offset = init->texture_offset[0];
dev_priv->texture_size = init->texture_size[0];
dev_priv->sarea = drm_getsarea(dev);
if (!dev_priv->sarea) {
DRM_ERROR("failed to find sarea!\n");
return DRM_ERR(EINVAL);
}
if (! dev_priv->used_new_dma_init) {
dev_priv->dma_access = MGA_PAGPXFER;
dev_priv->wagp_enable = MGA_WAGP_ENABLE;
dev_priv->status = drm_core_findmap(dev, init->status_offset);
if (!dev_priv->status) {
DRM_ERROR("failed to find status page!\n");
return DRM_ERR(EINVAL);
}
dev_priv->mmio = drm_core_findmap(dev, init->mmio_offset);
if (!dev_priv->mmio) {
DRM_ERROR("failed to find mmio region!\n");
return DRM_ERR(EINVAL);
}
dev_priv->warp = drm_core_findmap(dev, init->warp_offset);
if (!dev_priv->warp) {
DRM_ERROR("failed to find warp microcode region!\n");
return DRM_ERR(EINVAL);
}
dev_priv->primary = drm_core_findmap(dev, init->primary_offset);
if (!dev_priv->primary) {
DRM_ERROR("failed to find primary dma region!\n");
return DRM_ERR(EINVAL);
}
dev->agp_buffer_token = init->buffers_offset;
dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
if (!dev->agp_buffer_map) {
DRM_ERROR("failed to find dma buffer region!\n");
return DRM_ERR(EINVAL);
}
drm_core_ioremap(dev_priv->warp, dev);
drm_core_ioremap(dev_priv->primary, dev);
drm_core_ioremap(dev->agp_buffer_map, dev);
}
dev_priv->sarea_priv =
(drm_mga_sarea_t *) ((u8 *) dev_priv->sarea->handle +
init->sarea_priv_offset);
if (!dev_priv->warp->handle ||
!dev_priv->primary->handle ||
((dev_priv->dma_access != 0) &&
((dev->agp_buffer_map == NULL) ||
(dev->agp_buffer_map->handle == NULL)))) {
DRM_ERROR("failed to ioremap agp regions!\n");
return DRM_ERR(ENOMEM);
}
ret = mga_warp_install_microcode(dev_priv);
if (ret != 0) {
DRM_ERROR("failed to install WARP ucode: %d!\n", ret);
return ret;
}
ret = mga_warp_init(dev_priv);
if (ret != 0) {
DRM_ERROR("failed to init WARP engine: %d!\n", ret);
return ret;
}
dev_priv->prim.status = (u32 *) dev_priv->status->handle;
mga_do_wait_for_idle(dev_priv);
/* Init the primary DMA registers.
*/
MGA_WRITE(MGA_PRIMADDRESS, dev_priv->primary->offset | MGA_DMA_GENERAL);
#if 0
MGA_WRITE(MGA_PRIMPTR, virt_to_bus((void *)dev_priv->prim.status) | MGA_PRIMPTREN0 | /* Soft trap, SECEND, SETUPEND */
MGA_PRIMPTREN1); /* DWGSYNC */
#endif
dev_priv->prim.start = (u8 *) dev_priv->primary->handle;
dev_priv->prim.end = ((u8 *) dev_priv->primary->handle
+ dev_priv->primary->size);
dev_priv->prim.size = dev_priv->primary->size;
dev_priv->prim.tail = 0;
dev_priv->prim.space = dev_priv->prim.size;
dev_priv->prim.wrapped = 0;
dev_priv->prim.last_flush = 0;
dev_priv->prim.last_wrap = 0;
dev_priv->prim.high_mark = 256 * DMA_BLOCK_SIZE;
dev_priv->prim.status[0] = dev_priv->primary->offset;
dev_priv->prim.status[1] = 0;
dev_priv->sarea_priv->last_wrap = 0;
dev_priv->sarea_priv->last_frame.head = 0;
dev_priv->sarea_priv->last_frame.wrap = 0;
if (mga_freelist_init(dev, dev_priv) < 0) {
DRM_ERROR("could not initialize freelist\n");
return DRM_ERR(ENOMEM);
}
return 0;
}
static int mga_do_cleanup_dma(drm_device_t * dev, int full_cleanup)
{
int err = 0;
DRM_DEBUG("\n");
/* Make sure interrupts are disabled here because the uninstall ioctl
* may not have been called from userspace and after dev_private
* is freed, it's too late.
*/
if (dev->irq_enabled)
drm_irq_uninstall(dev);
if (dev->dev_private) {
drm_mga_private_t *dev_priv = dev->dev_private;
if ((dev_priv->warp != NULL)
&& (dev_priv->warp->type != _DRM_CONSISTENT))
drm_core_ioremapfree(dev_priv->warp, dev);
if ((dev_priv->primary != NULL)
&& (dev_priv->primary->type != _DRM_CONSISTENT))
drm_core_ioremapfree(dev_priv->primary, dev);
if (dev->agp_buffer_map != NULL)
drm_core_ioremapfree(dev->agp_buffer_map, dev);
if (dev_priv->used_new_dma_init) {
if (dev_priv->agp_handle != 0) {
drm_agp_binding_t unbind_req;
drm_agp_buffer_t free_req;
unbind_req.handle = dev_priv->agp_handle;
drm_agp_unbind(dev, &unbind_req);
free_req.handle = dev_priv->agp_handle;
drm_agp_free(dev, &free_req);
dev_priv->agp_textures = NULL;
dev_priv->agp_size = 0;
dev_priv->agp_handle = 0;
}
if ((dev->agp != NULL) && dev->agp->acquired) {
err = drm_agp_release(dev);
}
}
dev_priv->warp = NULL;
dev_priv->primary = NULL;
dev_priv->sarea = NULL;
dev_priv->sarea_priv = NULL;
dev->agp_buffer_map = NULL;
if (full_cleanup) {
dev_priv->mmio = NULL;
dev_priv->status = NULL;
dev_priv->used_new_dma_init = 0;
}
memset(&dev_priv->prim, 0, sizeof(dev_priv->prim));
dev_priv->warp_pipe = 0;
memset(dev_priv->warp_pipe_phys, 0, sizeof(dev_priv->warp_pipe_phys));
if (dev_priv->head != NULL) {
mga_freelist_cleanup(dev);
}
}
return 0;
}
int mga_dma_init(DRM_IOCTL_ARGS)
{
DRM_DEVICE;
drm_mga_init_t init;
int err;
LOCK_TEST_WITH_RETURN(dev, filp);
DRM_COPY_FROM_USER_IOCTL(init, (drm_mga_init_t __user *) data,
sizeof(init));
switch (init.func) {
case MGA_INIT_DMA:
err = mga_do_init_dma(dev, &init);
if (err) {
(void) mga_do_cleanup_dma(dev, FULL_CLEANUP);
}
return err;
case MGA_CLEANUP_DMA:
return mga_do_cleanup_dma(dev, FULL_CLEANUP);
}
return DRM_ERR(EINVAL);
}
/* ================================================================
* Primary DMA stream management
*/
int mga_dma_flush(DRM_IOCTL_ARGS)
{
DRM_DEVICE;
drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
drm_lock_t lock;
|