2000-02-22 08:43:59 -07:00
|
|
|
/* mga_dma.c -- DMA support for mga g200/g400 -*- linux-c -*-
|
|
|
|
* Created: Mon Dec 13 01:50:01 1999 by jhartmann@precisioninsight.com
|
2005-11-28 16:10:41 -07:00
|
|
|
*/
|
|
|
|
/* Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
|
2000-06-08 08:38:22 -06:00
|
|
|
* Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
|
2000-02-22 08:43:59 -07:00
|
|
|
* 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:
|
2000-09-06 14:56:34 -06:00
|
|
|
*
|
2000-02-22 08:43:59 -07:00
|
|
|
* The above copyright notice and this permission notice (including the next
|
|
|
|
* paragraph) shall be included in all copies or substantial portions of the
|
|
|
|
* Software.
|
2000-09-06 14:56:34 -06:00
|
|
|
*
|
2000-02-22 08:43:59 -07:00
|
|
|
* 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.
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* \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>
|
2000-02-22 08:43:59 -07:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include "drmP.h"
|
2002-04-09 15:54:56 -06:00
|
|
|
#include "drm.h"
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
#include "drm_sarea.h"
|
2002-04-09 15:54:56 -06:00
|
|
|
#include "mga_drm.h"
|
2000-02-22 08:43:59 -07:00
|
|
|
#include "mga_drv.h"
|
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
#define MGA_DEFAULT_USEC_TIMEOUT 10000
|
2001-03-20 20:29:23 -07:00
|
|
|
#define MGA_FREELIST_DEBUG 0
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2005-10-13 23:01:19 -06:00
|
|
|
#define MINIMAL_CLEANUP 0
|
|
|
|
#define FULL_CLEANUP 1
|
|
|
|
static int mga_do_cleanup_dma(drm_device_t * dev, int full_cleanup);
|
2005-02-01 04:08:31 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* ================================================================
|
|
|
|
* Engine control
|
|
|
|
*/
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_do_wait_for_idle(drm_mga_private_t * dev_priv)
|
2000-04-04 16:08:14 -06:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
u32 status = 0;
|
|
|
|
int i;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
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);
|
2001-02-20 13:16:56 -07:00
|
|
|
return 0;
|
|
|
|
}
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_UDELAY(1);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-03-21 06:10:27 -07:00
|
|
|
#if MGA_DMA_DEBUG
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_ERROR("failed!\n");
|
|
|
|
DRM_INFO(" status=0x%08x\n", status);
|
2001-03-21 06:10:27 -07:00
|
|
|
#endif
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EBUSY);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2005-02-01 04:08:31 -07:00
|
|
|
static int mga_do_dma_reset(drm_mga_private_t * dev_priv)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
|
|
|
|
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* The primary DMA stream should look like new right about now.
|
|
|
|
*/
|
|
|
|
primary->tail = 0;
|
|
|
|
primary->space = primary->size;
|
|
|
|
primary->last_flush = 0;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
sarea_priv->last_wrap = 0;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* FIXME: Reset counters, buffer ages etc...
|
|
|
|
*/
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* FIXME: What else do we need to reinitialize? WARP stuff?
|
|
|
|
*/
|
|
|
|
|
|
|
|
return 0;
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* ================================================================
|
|
|
|
* Primary DMA stream
|
|
|
|
*/
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
void mga_do_dma_flush(drm_mga_private_t * dev_priv)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
|
|
|
|
u32 head, tail;
|
2002-10-10 02:25:40 -06:00
|
|
|
u32 status = 0;
|
|
|
|
int i;
|
2004-09-30 15:12:10 -06:00
|
|
|
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);
|
2002-10-10 02:25:40 -06:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (primary->tail == primary->last_flush) {
|
|
|
|
DRM_DEBUG(" bailing out...\n");
|
2001-02-15 01:12:14 -07:00
|
|
|
return;
|
2000-05-25 15:06:02 -06:00
|
|
|
}
|
2000-09-27 15:32:19 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
tail = primary->tail + dev_priv->primary->offset;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* 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.
|
|
|
|
*/
|
2004-09-30 15:12:10 -06:00
|
|
|
BEGIN_DMA(1);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DMA_BLOCK(MGA_DMAPAD, 0x00000000,
|
|
|
|
MGA_DMAPAD, 0x00000000,
|
|
|
|
MGA_DMAPAD, 0x00000000, MGA_DMAPAD, 0x00000000);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
ADVANCE_DMA();
|
|
|
|
|
|
|
|
primary->last_flush = primary->tail;
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
head = MGA_READ(MGA_PRIMADDRESS);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (head <= tail) {
|
2001-02-15 01:12:14 -07:00
|
|
|
primary->space = primary->size - primary->tail;
|
2000-02-22 08:43:59 -07:00
|
|
|
} else {
|
2001-02-15 01:12:14 -07:00
|
|
|
primary->space = head - tail;
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
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);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
mga_flush_write_combine();
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
MGA_WRITE(MGA_PRIMEND, tail | dev_priv->dma_access);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("done.\n");
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
void mga_do_dma_wrap_start(drm_mga_private_t * dev_priv)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
|
|
|
|
u32 head, tail;
|
|
|
|
DMA_LOCALS;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
BEGIN_DMA_WRAP();
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DMA_BLOCK(MGA_DMAPAD, 0x00000000,
|
|
|
|
MGA_DMAPAD, 0x00000000,
|
|
|
|
MGA_DMAPAD, 0x00000000, MGA_DMAPAD, 0x00000000);
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
ADVANCE_DMA();
|
|
|
|
|
|
|
|
tail = primary->tail + dev_priv->primary->offset;
|
|
|
|
|
|
|
|
primary->tail = 0;
|
|
|
|
primary->last_flush = 0;
|
|
|
|
primary->last_wrap++;
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
head = MGA_READ(MGA_PRIMADDRESS);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (head == dev_priv->primary->offset) {
|
2001-02-15 01:12:14 -07:00
|
|
|
primary->space = primary->size;
|
2000-02-22 08:43:59 -07:00
|
|
|
} else {
|
2001-02-15 01:12:14 -07:00
|
|
|
primary->space = head - dev_priv->primary->offset;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
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);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
mga_flush_write_combine();
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
MGA_WRITE(MGA_PRIMEND, tail | dev_priv->dma_access);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
set_bit(0, &primary->wrapped);
|
|
|
|
DRM_DEBUG("done.\n");
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
void mga_do_dma_wrap_end(drm_mga_private_t * dev_priv)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-03-21 06:10:27 -07:00
|
|
|
drm_mga_primary_buffer_t *primary = &dev_priv->prim;
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
|
|
|
|
u32 head = dev_priv->primary->offset;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
sarea_priv->last_wrap++;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG(" wrap = %d\n", sarea_priv->last_wrap);
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
mga_flush_write_combine();
|
2004-09-30 15:12:10 -06:00
|
|
|
MGA_WRITE(MGA_PRIMADDRESS, head | MGA_DMA_GENERAL);
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
clear_bit(0, &primary->wrapped);
|
|
|
|
DRM_DEBUG("done.\n");
|
2001-02-15 01:12:14 -07:00
|
|
|
}
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* ================================================================
|
|
|
|
* Freelist management
|
|
|
|
*/
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
#define MGA_BUFFER_USED ~0
|
|
|
|
#define MGA_BUFFER_FREE 0
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2001-03-20 20:29:23 -07:00
|
|
|
#if MGA_FREELIST_DEBUG
|
2004-09-30 15:12:10 -06:00
|
|
|
static void mga_freelist_print(drm_device_t * dev)
|
2001-02-15 01:12:14 -07:00
|
|
|
{
|
|
|
|
drm_mga_private_t *dev_priv = dev->dev_private;
|
|
|
|
drm_mga_freelist_t *entry;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
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);
|
2001-02-15 01:12:14 -07:00
|
|
|
}
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_INFO("\n");
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2001-03-20 20:29:23 -07:00
|
|
|
#endif
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
static int mga_freelist_init(drm_device_t * dev, drm_mga_private_t * dev_priv)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
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;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("count=%d\n", dma->buf_count);
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
dev_priv->head = drm_alloc(sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
|
|
|
|
if (dev_priv->head == NULL)
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(ENOMEM);
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
memset(dev_priv->head, 0, sizeof(drm_mga_freelist_t));
|
|
|
|
SET_AGE(&dev_priv->head->age, MGA_BUFFER_USED, 0);
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
for (i = 0; i < dma->buf_count; i++) {
|
2001-02-15 01:12:14 -07:00
|
|
|
buf = dma->buflist[i];
|
2004-09-30 15:12:10 -06:00
|
|
|
buf_priv = buf->dev_private;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
entry = drm_alloc(sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
|
|
|
|
if (entry == NULL)
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(ENOMEM);
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
memset(entry, 0, sizeof(drm_mga_freelist_t));
|
2000-09-27 15:32:19 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
entry->next = dev_priv->head->next;
|
|
|
|
entry->prev = dev_priv->head;
|
2004-09-30 15:12:10 -06:00
|
|
|
SET_AGE(&entry->age, MGA_BUFFER_FREE, 0);
|
2001-02-15 01:12:14 -07:00
|
|
|
entry->buf = buf;
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (dev_priv->head->next != NULL)
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->head->next->prev = entry;
|
2004-09-30 15:12:10 -06:00
|
|
|
if (entry->next == NULL)
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->tail = entry;
|
|
|
|
|
|
|
|
buf_priv->list_entry = entry;
|
|
|
|
buf_priv->discard = 0;
|
|
|
|
buf_priv->dispatched = 0;
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->head->next = entry;
|
|
|
|
}
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
return 0;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
static void mga_freelist_cleanup(drm_device_t * dev)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_mga_private_t *dev_priv = dev->dev_private;
|
|
|
|
drm_mga_freelist_t *entry;
|
|
|
|
drm_mga_freelist_t *next;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
entry = dev_priv->head;
|
2004-09-30 15:12:10 -06:00
|
|
|
while (entry) {
|
2001-02-15 01:12:14 -07:00
|
|
|
next = entry->next;
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_free(entry, sizeof(drm_mga_freelist_t), DRM_MEM_DRIVER);
|
2001-02-15 01:12:14 -07:00
|
|
|
entry = next;
|
|
|
|
}
|
|
|
|
|
|
|
|
dev_priv->head = dev_priv->tail = NULL;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
|
|
|
|
2001-03-20 20:29:23 -07:00
|
|
|
#if 0
|
|
|
|
/* FIXME: Still needed?
|
|
|
|
*/
|
2004-09-30 15:12:10 -06:00
|
|
|
static void mga_freelist_reset(drm_device_t * dev)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_device_dma_t *dma = dev->dma;
|
|
|
|
drm_buf_t *buf;
|
|
|
|
drm_mga_buf_priv_t *buf_priv;
|
|
|
|
int i;
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
for (i = 0; i < dma->buf_count; i++) {
|
2001-02-15 01:12:14 -07:00
|
|
|
buf = dma->buflist[i];
|
2004-09-30 15:12:10 -06:00
|
|
|
buf_priv = buf->dev_private;
|
|
|
|
SET_AGE(&buf_priv->list_entry->age, MGA_BUFFER_FREE, 0);
|
2001-02-15 01:12:14 -07:00
|
|
|
}
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2001-03-20 20:29:23 -07:00
|
|
|
#endif
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
static drm_buf_t *mga_freelist_get(drm_device_t * dev)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2001-02-15 01:12:14 -07:00
|
|
|
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;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
head = MGA_READ(MGA_PRIMADDRESS);
|
2001-02-15 01:12:14 -07:00
|
|
|
wrap = dev_priv->sarea_priv->last_wrap;
|
2000-09-27 15:32:19 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
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);
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (TEST_AGE(&tail->age, head, wrap)) {
|
2001-02-15 01:12:14 -07:00
|
|
|
prev = dev_priv->tail->prev;
|
|
|
|
next = dev_priv->tail;
|
|
|
|
prev->next = NULL;
|
|
|
|
next->prev = next->next = NULL;
|
|
|
|
dev_priv->tail = prev;
|
2004-09-30 15:12:10 -06:00
|
|
|
SET_AGE(&next->age, MGA_BUFFER_USED, 0);
|
2001-02-15 01:12:14 -07:00
|
|
|
return next->buf;
|
|
|
|
}
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("returning NULL!\n");
|
2001-02-15 01:12:14 -07:00
|
|
|
return NULL;
|
|
|
|
}
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_freelist_put(drm_device_t * dev, drm_buf_t * buf)
|
2001-02-15 01:12:14 -07:00
|
|
|
{
|
|
|
|
drm_mga_private_t *dev_priv = dev->dev_private;
|
|
|
|
drm_mga_buf_priv_t *buf_priv = buf->dev_private;
|
2001-03-12 11:18:06 -07:00
|
|
|
drm_mga_freelist_t *head, *entry, *prev;
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("age=0x%06lx wrap=%d\n",
|
|
|
|
buf_priv->list_entry->age.head -
|
|
|
|
dev_priv->primary->offset, buf_priv->list_entry->age.wrap);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2001-03-12 11:18:06 -07:00
|
|
|
entry = buf_priv->list_entry;
|
2001-02-15 01:12:14 -07:00
|
|
|
head = dev_priv->head;
|
2001-02-20 13:16:56 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (buf_priv->list_entry->age.head == MGA_BUFFER_USED) {
|
|
|
|
SET_AGE(&entry->age, MGA_BUFFER_FREE, 0);
|
2001-03-12 11:18:06 -07:00
|
|
|
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;
|
2001-02-20 13:16:56 -07:00
|
|
|
}
|
|
|
|
|
2000-04-04 16:08:14 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* ================================================================
|
|
|
|
* DMA initialization, cleanup
|
|
|
|
*/
|
|
|
|
|
2005-08-04 21:50:23 -06:00
|
|
|
int mga_driver_load(drm_device_t *dev, unsigned long flags)
|
2001-02-15 01:12:14 -07:00
|
|
|
{
|
2005-05-20 20:27:51 -06:00
|
|
|
drm_mga_private_t * dev_priv;
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
dev_priv = drm_alloc(sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
|
|
|
|
if (!dev_priv)
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(ENOMEM);
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2005-05-20 20:27:51 -06:00
|
|
|
dev->dev_private = (void *)dev_priv;
|
2004-09-30 15:12:10 -06:00
|
|
|
memset(dev_priv, 0, sizeof(drm_mga_private_t));
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->usec_timeout = MGA_DEFAULT_USEC_TIMEOUT;
|
2005-05-20 20:27:51 -06:00
|
|
|
dev_priv->chipset = flags;
|
|
|
|
|
2005-08-04 21:50:23 -06:00
|
|
|
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;
|
|
|
|
|
2005-05-20 20:27:51 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
/**
|
|
|
|
* 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;
|
2005-08-26 17:27:19 -06:00
|
|
|
unsigned int warp_size = mga_warp_microcode_size(dev_priv);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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;
|
2005-11-02 17:38:25 -07:00
|
|
|
drm_agp_buffer_t agp_req;
|
|
|
|
drm_agp_binding_t bind_req;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
|
|
|
|
/* Acquire AGP. */
|
|
|
|
err = drm_agp_acquire(dev);
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to acquire AGP: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = drm_agp_info(dev, &info);
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to get AGP info: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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);
|
2005-08-04 01:46:11 -06:00
|
|
|
} else {
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
MGA_WRITE(MGA_AGP_PLL, MGA_AGP2XPLL_DISABLE);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* Allocate and bind AGP memory. */
|
2005-11-02 17:38:25 -07:00
|
|
|
agp_req.size = agp_size;
|
|
|
|
agp_req.type = 0;
|
|
|
|
err = drm_agp_alloc( dev, & agp_req );
|
|
|
|
if (err) {
|
|
|
|
dev_priv->agp_size = 0;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
DRM_ERROR("Unable to allocate %uMB AGP memory\n",
|
|
|
|
dma_bs->agp_size);
|
2005-11-02 17:38:25 -07:00
|
|
|
return err;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
}
|
2005-11-02 17:38:25 -07:00
|
|
|
|
|
|
|
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 );
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to bind AGP memory: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2005-08-26 17:27:19 -06:00
|
|
|
/* 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;
|
|
|
|
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
offset = 0;
|
|
|
|
err = drm_addmap( dev, offset, warp_size,
|
|
|
|
_DRM_AGP, _DRM_READ_ONLY, & dev_priv->warp );
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to map WARP microcode: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
offset += warp_size;
|
|
|
|
err = drm_addmap( dev, offset, dma_bs->primary_size,
|
|
|
|
_DRM_AGP, _DRM_READ_ONLY, & dev_priv->primary );
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to map primary DMA region: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
offset += dma_bs->primary_size;
|
|
|
|
err = drm_addmap( dev, offset, secondary_size,
|
|
|
|
_DRM_AGP, 0, & dev->agp_buffer_map );
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to map secondary DMA region: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to add secondary DMA buffers: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2005-11-07 22:29:26 -07:00
|
|
|
#ifdef __linux__
|
2005-11-02 17:38:25 -07:00
|
|
|
{
|
|
|
|
drm_map_list_t *_entry;
|
|
|
|
unsigned long agp_token = 0;
|
|
|
|
|
|
|
|
list_for_each_entry(_entry, &dev->maplist->head, head) {
|
|
|
|
if (_entry->map == dev->agp_buffer_map)
|
|
|
|
agp_token = _entry->user_token;
|
|
|
|
}
|
|
|
|
if (!agp_token)
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
dev->agp_buffer_token = agp_token;
|
|
|
|
}
|
2005-11-07 22:29:26 -07:00
|
|
|
#endif
|
2005-11-02 17:38:25 -07:00
|
|
|
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
offset += secondary_size;
|
|
|
|
err = drm_addmap( dev, offset, agp_size - offset,
|
|
|
|
_DRM_AGP, 0, & dev_priv->agp_textures );
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to map AGP texture region: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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;
|
2005-08-26 17:27:19 -06:00
|
|
|
unsigned int warp_size = mga_warp_microcode_size(dev_priv);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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);
|
|
|
|
}
|
|
|
|
|
2005-08-26 17:27:19 -06:00
|
|
|
/* 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;
|
|
|
|
|
- Remove drm_initmap and replace its usage with drm_addmap. This reduces
code duplication, and it also hands you the map pointer so you don't
need to re-find it.
- Remove the permanent maps flag. Instead, for register and framebuffer
maps, we always check whether there's already a map of that type and
offset around. Move the Radeon map initialization into presetup (first
open) so it happens again after every takedown.
- Remove the split cleanup of maps between driver takedown (last close) and
cleanup (module unload). Instead, always tear down maps on takedown,
and drivers can recreate them on first open.
- Make MGA always use addmap, instead of allocating consistent memory in
the PCI case and then faking up a map for it, which accomplished nearly
the same thing, in a different order. Note that the maps are exposed to
the user again: we may want to expose a flag to avoid this, but it's
not a security concern, and saves us a lot of code.
- Remove rmmaps in the MGA driver. Since the function is only called during
takedown anyway, we can let them die a natural death.
- Make removal of maps happen in one function, which is called by both
drm_takedown and drm_rmmap_ioctl.
Reviewed by: idr (previous revision) Tested on: mga (old/new/pci dma),
radeon, savage
2005-06-28 14:58:34 -06:00
|
|
|
/* 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) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to create mapping for WARP microcode: %d\n",
|
|
|
|
err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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 ) {
|
- Remove drm_initmap and replace its usage with drm_addmap. This reduces
code duplication, and it also hands you the map pointer so you don't
need to re-find it.
- Remove the permanent maps flag. Instead, for register and framebuffer
maps, we always check whether there's already a map of that type and
offset around. Move the Radeon map initialization into presetup (first
open) so it happens again after every takedown.
- Remove the split cleanup of maps between driver takedown (last close) and
cleanup (module unload). Instead, always tear down maps on takedown,
and drivers can recreate them on first open.
- Make MGA always use addmap, instead of allocating consistent memory in
the PCI case and then faking up a map for it, which accomplished nearly
the same thing, in a different order. Note that the maps are exposed to
the user again: we may want to expose a flag to avoid this, but it's
not a security concern, and saves us a lot of code.
- Remove rmmaps in the MGA driver. Since the function is only called during
takedown anyway, we can let them die a natural death.
- Make removal of maps happen in one function, which is called by both
drm_takedown and drm_rmmap_ioctl.
Reviewed by: idr (previous revision) Tested on: mga (old/new/pci dma),
radeon, savage
2005-06-28 14:58:34 -06:00
|
|
|
/* 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)
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
- Remove drm_initmap and replace its usage with drm_addmap. This reduces
code duplication, and it also hands you the map pointer so you don't
need to re-find it.
- Remove the permanent maps flag. Instead, for register and framebuffer
maps, we always check whether there's already a map of that type and
offset around. Move the Radeon map initialization into presetup (first
open) so it happens again after every takedown.
- Remove the split cleanup of maps between driver takedown (last close) and
cleanup (module unload). Instead, always tear down maps on takedown,
and drivers can recreate them on first open.
- Make MGA always use addmap, instead of allocating consistent memory in
the PCI case and then faking up a map for it, which accomplished nearly
the same thing, in a different order. Note that the maps are exposed to
the user again: we may want to expose a flag to avoid this, but it's
not a security concern, and saves us a lot of code.
- Remove rmmaps in the MGA driver. Since the function is only called during
takedown anyway, we can let them die a natural death.
- Make removal of maps happen in one function, which is called by both
drm_takedown and drm_rmmap_ioctl.
Reviewed by: idr (previous revision) Tested on: mga (old/new/pci dma),
radeon, savage
2005-06-28 14:58:34 -06:00
|
|
|
if (err != 0) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to allocate primary DMA region: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return DRM_ERR(ENOMEM);
|
|
|
|
}
|
|
|
|
|
- Remove drm_initmap and replace its usage with drm_addmap. This reduces
code duplication, and it also hands you the map pointer so you don't
need to re-find it.
- Remove the permanent maps flag. Instead, for register and framebuffer
maps, we always check whether there's already a map of that type and
offset around. Move the Radeon map initialization into presetup (first
open) so it happens again after every takedown.
- Remove the split cleanup of maps between driver takedown (last close) and
cleanup (module unload). Instead, always tear down maps on takedown,
and drivers can recreate them on first open.
- Make MGA always use addmap, instead of allocating consistent memory in
the PCI case and then faking up a map for it, which accomplished nearly
the same thing, in a different order. Note that the maps are exposed to
the user again: we may want to expose a flag to avoid this, but it's
not a security concern, and saves us a lot of code.
- Remove rmmaps in the MGA driver. Since the function is only called during
takedown anyway, we can let them die a natural death.
- Make removal of maps happen in one function, which is called by both
drm_takedown and drm_rmmap_ioctl.
Reviewed by: idr (previous revision) Tested on: mga (old/new/pci dma),
radeon, savage
2005-06-28 14:58:34 -06:00
|
|
|
if (dev_priv->primary->size != dma_bs->primary_size) {
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
DRM_INFO("Primary DMA buffer size reduced from %u to %u.\n",
|
|
|
|
dma_bs->primary_size,
|
- Remove drm_initmap and replace its usage with drm_addmap. This reduces
code duplication, and it also hands you the map pointer so you don't
need to re-find it.
- Remove the permanent maps flag. Instead, for register and framebuffer
maps, we always check whether there's already a map of that type and
offset around. Move the Radeon map initialization into presetup (first
open) so it happens again after every takedown.
- Remove the split cleanup of maps between driver takedown (last close) and
cleanup (module unload). Instead, always tear down maps on takedown,
and drivers can recreate them on first open.
- Make MGA always use addmap, instead of allocating consistent memory in
the PCI case and then faking up a map for it, which accomplished nearly
the same thing, in a different order. Note that the maps are exposed to
the user again: we may want to expose a flag to avoid this, but it's
not a security concern, and saves us a lot of code.
- Remove rmmaps in the MGA driver. Since the function is only called during
takedown anyway, we can let them die a natural death.
- Make removal of maps happen in one function, which is called by both
drm_takedown and drm_rmmap_ioctl.
Reviewed by: idr (previous revision) Tested on: mga (old/new/pci dma),
radeon, savage
2005-06-28 14:58:34 -06:00
|
|
|
(unsigned) dev_priv->primary->size);
|
|
|
|
dma_bs->primary_size = dev_priv->primary->size;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
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) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to add secondary DMA buffers: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to map MMIO region: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
err = drm_addmap( dev, 0, SAREA_MAX, _DRM_SHM,
|
|
|
|
_DRM_READ_ONLY | _DRM_LOCKED | _DRM_KERNEL,
|
|
|
|
& dev_priv->status );
|
|
|
|
if (err) {
|
2005-08-26 17:27:19 -06:00
|
|
|
DRM_ERROR("Unable to map status region: %d\n", err);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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) {
|
2005-10-13 23:01:19 -06:00
|
|
|
mga_do_cleanup_dma(dev, MINIMAL_CLEANUP);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* 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;
|
2005-08-04 01:42:01 -06:00
|
|
|
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;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
|
|
|
|
|
|
|
|
DRM_COPY_FROM_USER_IOCTL(bootstrap,
|
|
|
|
(drm_mga_dma_bootstrap_t __user *) data,
|
|
|
|
sizeof(bootstrap));
|
|
|
|
|
|
|
|
err = mga_do_dma_bootstrap(dev, & bootstrap);
|
2005-08-04 01:42:01 -06:00
|
|
|
if (err) {
|
2005-10-13 23:01:19 -06:00
|
|
|
mga_do_cleanup_dma(dev, FULL_CLEANUP);
|
2005-08-04 01:42:01 -06:00
|
|
|
return err;
|
|
|
|
}
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
|
2005-08-04 01:42:01 -06:00
|
|
|
if (dev_priv->agp_textures != NULL) {
|
|
|
|
bootstrap.texture_handle = dev_priv->agp_textures->offset;
|
|
|
|
bootstrap.texture_size = dev_priv->agp_textures->size;
|
2005-08-04 01:46:11 -06:00
|
|
|
} else {
|
2005-08-04 01:42:01 -06:00
|
|
|
bootstrap.texture_handle = 0;
|
|
|
|
bootstrap.texture_size = 0;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
}
|
|
|
|
|
2005-08-04 01:42:01 -06:00
|
|
|
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;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-05-20 20:27:51 -06:00
|
|
|
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;
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (init->sgram) {
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->clear_cmd = MGA_DWGCTL_CLEAR | MGA_ATYPE_BLK;
|
|
|
|
} else {
|
|
|
|
dev_priv->clear_cmd = MGA_DWGCTL_CLEAR | MGA_ATYPE_RSTR;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2004-09-30 15:12:10 -06:00
|
|
|
dev_priv->maccess = init->maccess;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
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;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
dev_priv->depth_cpp = init->depth_cpp;
|
|
|
|
dev_priv->depth_offset = init->depth_offset;
|
|
|
|
dev_priv->depth_pitch = init->depth_pitch;
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2001-03-20 20:29:23 -07:00
|
|
|
/* FIXME: Need to support AGP textures...
|
|
|
|
*/
|
|
|
|
dev_priv->texture_offset = init->texture_offset[0];
|
|
|
|
dev_priv->texture_size = init->texture_size[0];
|
|
|
|
|
2002-07-05 02:31:11 -06:00
|
|
|
DRM_GETSAREA();
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (!dev_priv->sarea) {
|
|
|
|
DRM_ERROR("failed to find sarea!\n");
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EINVAL);
|
2001-08-07 12:15:10 -06:00
|
|
|
}
|
2001-02-15 01:12:14 -07:00
|
|
|
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
if (! dev_priv->used_new_dma_init) {
|
2005-10-19 19:54:38 -06:00
|
|
|
|
2005-10-19 20:35:08 -06:00
|
|
|
dev_priv->dma_access = MGA_PAGPXFER;
|
2005-10-19 19:54:38 -06:00
|
|
|
dev_priv->wagp_enable = MGA_WAGP_ENABLE;
|
|
|
|
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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);
|
|
|
|
}
|
2005-08-16 06:51:57 -06:00
|
|
|
dev->agp_buffer_token = init->buffers_offset;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
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);
|
2001-08-07 12:15:10 -06:00
|
|
|
}
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
dev_priv->sarea_priv =
|
2004-09-30 15:12:10 -06:00
|
|
|
(drm_mga_sarea_t *) ((u8 *) dev_priv->sarea->handle +
|
|
|
|
init->sarea_priv_offset);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (!dev_priv->warp->handle ||
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
!dev_priv->primary->handle ||
|
|
|
|
((dev_priv->dma_access != 0) &&
|
|
|
|
((dev->agp_buffer_map == NULL) ||
|
|
|
|
(dev->agp_buffer_map->handle == NULL)))) {
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_ERROR("failed to ioremap agp regions!\n");
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(ENOMEM);
|
2001-08-07 12:15:10 -06:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
ret = mga_warp_install_microcode(dev_priv);
|
2005-08-26 17:27:19 -06:00
|
|
|
if (ret != 0) {
|
|
|
|
DRM_ERROR("failed to install WARP ucode: %d!\n", ret);
|
2001-02-15 01:12:14 -07:00
|
|
|
return ret;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
ret = mga_warp_init(dev_priv);
|
2005-08-26 17:27:19 -06:00
|
|
|
if (ret != 0) {
|
|
|
|
DRM_ERROR("failed to init WARP engine: %d!\n", ret);
|
2001-02-15 01:12:14 -07:00
|
|
|
return ret;
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
dev_priv->prim.status = (u32 *) dev_priv->status->handle;
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
mga_do_wait_for_idle(dev_priv);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
/* Init the primary DMA registers.
|
|
|
|
*/
|
2004-09-30 15:12:10 -06:00
|
|
|
MGA_WRITE(MGA_PRIMADDRESS, dev_priv->primary->offset | MGA_DMA_GENERAL);
|
2001-03-19 05:04:12 -07:00
|
|
|
#if 0
|
2004-09-30 15:12:10 -06:00
|
|
|
MGA_WRITE(MGA_PRIMPTR, virt_to_bus((void *)dev_priv->prim.status) | MGA_PRIMPTREN0 | /* Soft trap, SECEND, SETUPEND */
|
|
|
|
MGA_PRIMPTREN1); /* DWGSYNC */
|
2001-03-19 05:04:12 -07:00
|
|
|
#endif
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
dev_priv->prim.start = (u8 *) dev_priv->primary->handle;
|
|
|
|
dev_priv->prim.end = ((u8 *) dev_priv->primary->handle
|
2001-02-15 01:12:14 -07:00
|
|
|
+ dev_priv->primary->size);
|
|
|
|
dev_priv->prim.size = dev_priv->primary->size;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->prim.tail = 0;
|
|
|
|
dev_priv->prim.space = dev_priv->prim.size;
|
2001-03-21 06:10:27 -07:00
|
|
|
dev_priv->prim.wrapped = 0;
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->prim.last_flush = 0;
|
|
|
|
dev_priv->prim.last_wrap = 0;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->prim.high_mark = 256 * DMA_BLOCK_SIZE;
|
2000-09-07 14:42:32 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->prim.status[0] = dev_priv->primary->offset;
|
|
|
|
dev_priv->prim.status[1] = 0;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
dev_priv->sarea_priv->last_wrap = 0;
|
|
|
|
dev_priv->sarea_priv->last_frame.head = 0;
|
|
|
|
dev_priv->sarea_priv->last_frame.wrap = 0;
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (mga_freelist_init(dev, dev_priv) < 0) {
|
|
|
|
DRM_ERROR("could not initialize freelist\n");
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(ENOMEM);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-10-13 23:01:19 -06:00
|
|
|
static int mga_do_cleanup_dma(drm_device_t * dev, int full_cleanup)
|
2001-02-15 01:12:14 -07:00
|
|
|
{
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
int err = 0;
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("\n");
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2003-04-26 16:28:56 -06:00
|
|
|
/* 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.
|
|
|
|
*/
|
2004-09-30 15:12:10 -06:00
|
|
|
if (dev->irq_enabled)
|
|
|
|
drm_irq_uninstall(dev);
|
2003-04-26 16:28:56 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (dev->dev_private) {
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_mga_private_t *dev_priv = dev->dev_private;
|
|
|
|
|
2005-10-13 19:25:38 -06:00
|
|
|
if ((dev_priv->warp != NULL)
|
|
|
|
&& (dev_priv->warp->type != _DRM_CONSISTENT))
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_core_ioremapfree(dev_priv->warp, dev);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
|
|
|
|
if ((dev_priv->primary != NULL)
|
|
|
|
&& (dev_priv->primary->type != _DRM_CONSISTENT))
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_core_ioremapfree(dev_priv->primary, dev);
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
if (dev->agp_buffer_map != NULL)
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_core_ioremapfree(dev->agp_buffer_map, dev);
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
if (dev_priv->used_new_dma_init) {
|
2005-11-02 17:38:25 -07:00
|
|
|
if (dev_priv->agp_handle != 0) {
|
|
|
|
drm_agp_binding_t unbind_req;
|
|
|
|
drm_agp_buffer_t free_req;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
|
2005-11-02 17:38:25 -07:00
|
|
|
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;
|
Adds support for PCI cards to MGA DRM
This patch adds serveral new ioctls and a new query to get_param query to
support PCI MGA cards.
Two ioctls were added to implement interrupt based waiting. With this
change, the client-side driver no longer needs to map the primary DMA
region or the MMIO region. Previously, end-of-frame waiting was done by
busy waiting in the client-side driver until one of the MMIO registers
(the current DMA pointer) matched a pointer to the end of primary DMA
space. By using interrupts, the busy waiting and the extra mappings are
removed.
A third ioctl was added to bootstrap DMA. This ioctl, which is used by the
X-server, moves a *LOT* of code from the X-server into the kernel. This
allows the kernel to do whatever needs to be done to setup DMA buffers.
The entire process and the locations of the buffers are hidden from
user-mode.
Additionally, a get_param query was added to differentiate between G4x0
cards and G550 cards. A gap was left in the numbering sequence so that,
if needed, G450 cards could be distinguished from G400 cards. According
to Ville Syrjälä, the G4x0 cards and the G550 cards handle
anisotropic filtering differently. This seems the most compatible way
to let the client-side driver know which card it's own. Doing this very
small change now eliminates the need to bump the DRM minor version
twice.
http://marc.theaimsgroup.com/?l=dri-devel&m=106625815319773&w=2
A number of ioctl handlers in linux-core were also modified so that they
could be called in-kernel. In these cases, the in-kernel callable
version kept the existing name (e.g., drm_agp_acquire) and the ioctl
handler added _ioctl to the name (e.g., drm_agp_acquire_ioctl).
This patch also replaces the drm_agp_do_release function with
drm_agp_release. drm_agp_release (drm_core_agp_release in the previous
patch) is very similar to drm_agp_do_release, and I saw no reason to
have both.
This commit *breaks the build* on BSD. Eric said that he would make the
required updates to the BSD side soon.
Xorg bug: 3259 Reviewed by: Eric Anholt
2005-06-14 16:34:11 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
if ((dev->agp != NULL) && dev->agp->acquired) {
|
|
|
|
err = drm_agp_release(dev);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
dev_priv->warp = NULL;
|
|
|
|
dev_priv->primary = NULL;
|
|
|
|
dev_priv->sarea = NULL;
|
|
|
|
dev_priv->sarea_priv = NULL;
|
|
|
|
dev->agp_buffer_map = NULL;
|
|
|
|
|
2005-10-13 23:01:19 -06:00
|
|
|
if (full_cleanup) {
|
|
|
|
dev_priv->mmio = NULL;
|
|
|
|
dev_priv->status = NULL;
|
|
|
|
dev_priv->used_new_dma_init = 0;
|
|
|
|
}
|
|
|
|
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
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));
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (dev_priv->head != NULL) {
|
|
|
|
mga_freelist_cleanup(dev);
|
2001-02-15 01:12:14 -07:00
|
|
|
}
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2000-04-04 16:08:14 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_dma_init(DRM_IOCTL_ARGS)
|
2000-04-04 16:08:14 -06:00
|
|
|
{
|
2002-07-05 02:31:11 -06:00
|
|
|
DRM_DEVICE;
|
2000-04-04 16:08:14 -06:00
|
|
|
drm_mga_init_t init;
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
int err;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
LOCK_TEST_WITH_RETURN(dev, filp);
|
2003-04-26 16:28:56 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_COPY_FROM_USER_IOCTL(init, (drm_mga_init_t __user *) data,
|
|
|
|
sizeof(init));
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
switch (init.func) {
|
2000-04-04 16:08:14 -06:00
|
|
|
case MGA_INIT_DMA:
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
err = mga_do_init_dma(dev, &init);
|
|
|
|
if (err) {
|
2005-10-13 23:01:19 -06:00
|
|
|
(void) mga_do_cleanup_dma(dev, FULL_CLEANUP);
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
}
|
|
|
|
return err;
|
2000-04-04 16:08:14 -06:00
|
|
|
case MGA_CLEANUP_DMA:
|
2005-10-13 23:01:19 -06:00
|
|
|
return mga_do_cleanup_dma(dev, FULL_CLEANUP);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EINVAL);
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* ================================================================
|
|
|
|
* Primary DMA stream management
|
|
|
|
*/
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_dma_flush(DRM_IOCTL_ARGS)
|
2000-02-22 08:43:59 -07:00
|
|
|
{
|
2002-07-05 02:31:11 -06:00
|
|
|
DRM_DEVICE;
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_lock_t lock;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
LOCK_TEST_WITH_RETURN(dev, filp);
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_COPY_FROM_USER_IOCTL(lock, (drm_lock_t __user *) data,
|
|
|
|
sizeof(lock));
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_DEBUG("%s%s%s\n",
|
|
|
|
(lock.flags & _DRM_LOCK_FLUSH) ? "flush, " : "",
|
|
|
|
(lock.flags & _DRM_LOCK_FLUSH_ALL) ? "flush all, " : "",
|
|
|
|
(lock.flags & _DRM_LOCK_QUIESCENT) ? "idle, " : "");
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
WRAP_WAIT_WITH_RETURN(dev_priv);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (lock.flags & (_DRM_LOCK_FLUSH | _DRM_LOCK_FLUSH_ALL)) {
|
|
|
|
mga_do_dma_flush(dev_priv);
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2001-03-21 06:10:27 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (lock.flags & _DRM_LOCK_QUIESCENT) {
|
2001-03-21 06:10:27 -07:00
|
|
|
#if MGA_DMA_DEBUG
|
2004-09-30 15:12:10 -06:00
|
|
|
int ret = mga_do_wait_for_idle(dev_priv);
|
|
|
|
if (ret < 0)
|
|
|
|
DRM_INFO("%s: -EBUSY\n", __FUNCTION__);
|
2001-03-21 06:10:27 -07:00
|
|
|
return ret;
|
|
|
|
#else
|
2004-09-30 15:12:10 -06:00
|
|
|
return mga_do_wait_for_idle(dev_priv);
|
2001-03-21 06:10:27 -07:00
|
|
|
#endif
|
2001-02-15 01:12:14 -07:00
|
|
|
} else {
|
|
|
|
return 0;
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_dma_reset(DRM_IOCTL_ARGS)
|
2000-04-04 16:08:14 -06:00
|
|
|
{
|
2002-07-05 02:31:11 -06:00
|
|
|
DRM_DEVICE;
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
LOCK_TEST_WITH_RETURN(dev, filp);
|
2001-02-15 01:12:14 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
return mga_do_dma_reset(dev_priv);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* ================================================================
|
|
|
|
* DMA buffer management
|
|
|
|
*/
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
static int mga_dma_get_buffers(DRMFILE filp, drm_device_t * dev, drm_dma_t * d)
|
2001-02-15 01:12:14 -07:00
|
|
|
{
|
|
|
|
drm_buf_t *buf;
|
|
|
|
int i;
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
for (i = d->granted_count; i < d->request_count; i++) {
|
|
|
|
buf = mga_freelist_get(dev);
|
|
|
|
if (!buf)
|
|
|
|
return DRM_ERR(EAGAIN);
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2003-03-28 07:27:37 -07:00
|
|
|
buf->filp = filp;
|
2000-02-22 08:43:59 -07:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (DRM_COPY_TO_USER(&d->request_indices[i],
|
|
|
|
&buf->idx, sizeof(buf->idx)))
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EFAULT);
|
2004-09-30 15:12:10 -06:00
|
|
|
if (DRM_COPY_TO_USER(&d->request_sizes[i],
|
|
|
|
&buf->total, sizeof(buf->total)))
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EFAULT);
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
d->granted_count++;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2001-02-15 01:12:14 -07:00
|
|
|
return 0;
|
2000-02-22 08:43:59 -07:00
|
|
|
}
|
2000-09-06 14:56:34 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_dma_buffers(DRM_IOCTL_ARGS)
|
2000-04-04 16:08:14 -06:00
|
|
|
{
|
2002-07-05 02:31:11 -06:00
|
|
|
DRM_DEVICE;
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_device_dma_t *dma = dev->dma;
|
2004-09-30 15:12:10 -06:00
|
|
|
drm_mga_private_t *dev_priv = (drm_mga_private_t *) dev->dev_private;
|
2004-07-25 02:47:38 -06:00
|
|
|
drm_dma_t __user *argp = (void __user *)data;
|
2001-02-15 01:12:14 -07:00
|
|
|
drm_dma_t d;
|
|
|
|
int ret = 0;
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
LOCK_TEST_WITH_RETURN(dev, filp);
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_COPY_FROM_USER_IOCTL(d, argp, sizeof(d));
|
2000-04-04 16:08:14 -06:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* Please don't send us buffers.
|
|
|
|
*/
|
2004-09-30 15:12:10 -06:00
|
|
|
if (d.send_count != 0) {
|
|
|
|
DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
|
|
|
|
DRM_CURRENTPID, d.send_count);
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EINVAL);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
/* We'll send you buffers.
|
|
|
|
*/
|
2004-09-30 15:12:10 -06:00
|
|
|
if (d.request_count < 0 || d.request_count > dma->buf_count) {
|
|
|
|
DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
|
|
|
|
DRM_CURRENTPID, d.request_count, dma->buf_count);
|
2002-07-05 02:31:11 -06:00
|
|
|
return DRM_ERR(EINVAL);
|
2001-02-15 01:12:14 -07:00
|
|
|
}
|
2000-07-11 05:41:07 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
WRAP_TEST_WITH_RETURN(dev_priv);
|
2001-03-21 06:10:27 -07:00
|
|
|
|
2001-02-15 01:12:14 -07:00
|
|
|
d.granted_count = 0;
|
2000-05-25 15:06:02 -06:00
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
if (d.request_count) {
|
|
|
|
ret = mga_dma_get_buffers(filp, dev, &d);
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
DRM_COPY_TO_USER_IOCTL(argp, d, sizeof(d));
|
2001-02-15 01:12:14 -07:00
|
|
|
|
|
|
|
return ret;
|
2000-04-04 16:08:14 -06:00
|
|
|
}
|
2004-08-17 07:10:05 -06:00
|
|
|
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
/**
|
|
|
|
* Called just before the module is unloaded.
|
|
|
|
*/
|
2005-08-04 21:50:23 -06:00
|
|
|
int mga_driver_unload(drm_device_t * dev)
|
2004-08-17 07:10:05 -06:00
|
|
|
{
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
drm_free(dev->dev_private, sizeof(drm_mga_private_t), DRM_MEM_DRIVER);
|
|
|
|
dev->dev_private = NULL;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2005-06-03 16:45:21 -06:00
|
|
|
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
/**
|
|
|
|
* Called when the last opener of the device is closed.
|
|
|
|
*/
|
2005-08-04 21:50:23 -06:00
|
|
|
void mga_driver_lastclose(drm_device_t * dev)
|
Completely re-initialize DMA settings
There were two problems. First, the 'warp' and 'primary' pointers weren't
cleared, so mga_do_cleanup_dma, which gets called multiple times, would
try to ioremapfree them multiple times. This resulted in the new error
messages to syslog. The second problem was the, since the dev_private
structure isn't reallocated and cleaned out in mga_do_init_dma, when
the server is reloaded idle-waits would wait for impossible values.
I have given this patch some more riggorous testing. This includes:
- Load module, start server, run GL app, stop server, unload module.
- Load module, start server, run GL app, stop server, unload module, reload
module, restart server, run GL app.
- Load module, start server, run GL app, stop server, restart server, run
GL app, stop server, unload module.
In all three cases, everything worked as expected. Please let me know if
there are any further regressions with this patch.
Xorg bug: 3408 Reported by: Chris Rankin
2005-06-09 15:18:56 -06:00
|
|
|
{
|
2005-10-13 23:01:19 -06:00
|
|
|
mga_do_cleanup_dma(dev, FULL_CLEANUP);
|
2004-08-17 07:10:05 -06:00
|
|
|
}
|
|
|
|
|
2004-09-30 15:12:10 -06:00
|
|
|
int mga_driver_dma_quiescent(drm_device_t * dev)
|
2004-08-17 07:10:05 -06:00
|
|
|
{
|
|
|
|
drm_mga_private_t *dev_priv = dev->dev_private;
|
2004-09-30 15:12:10 -06:00
|
|
|
return mga_do_wait_for_idle(dev_priv);
|
2004-08-17 07:10:05 -06:00
|
|
|
}
|