diff options
author | Thomas Hellstrom <unichrome@shipmail.org> | 2004-12-03 23:03:36 +0000 |
---|---|---|
committer | Thomas Hellstrom <unichrome@shipmail.org> | 2004-12-03 23:03:36 +0000 |
commit | 267e0645272720344eb7556a948e72112edbe2ec (patch) | |
tree | fa1ab572a213ca42e7938aa8af43791a0401af16 | |
parent | f197110e07b0be03d5d79329779b665ac46433d5 (diff) |
Added 3D functionality to the via command verifier. Modified the via
ring-buffer code somewhat to workaround hardware problems. Bumped via
minor version number.
-rw-r--r-- | linux-2.6/Makefile | 2 | ||||
-rw-r--r-- | linux-2.6/Makefile.kernel | 2 | ||||
-rw-r--r-- | linux-core/Makefile | 2 | ||||
-rw-r--r-- | linux-core/Makefile.kernel | 4 | ||||
-rw-r--r-- | shared-core/via_dma.c | 111 | ||||
-rw-r--r-- | shared-core/via_drv.c | 1 | ||||
-rw-r--r-- | shared-core/via_drv.h | 8 | ||||
-rw-r--r-- | shared-core/via_verifier.c | 671 | ||||
-rw-r--r-- | shared/via.h | 4 | ||||
-rw-r--r-- | shared/via_dma.c | 100 | ||||
-rw-r--r-- | shared/via_drv.h | 10 | ||||
-rw-r--r-- | shared/via_verifier.c | 672 |
12 files changed, 1436 insertions, 151 deletions
diff --git a/linux-2.6/Makefile b/linux-2.6/Makefile index d39bddda..7e86e84c 100644 --- a/linux-2.6/Makefile +++ b/linux-2.6/Makefile @@ -111,7 +111,7 @@ VIAHEADERS = via_drm.h via_drv.h via.h via_mm.h via_ds.h \ via_3d_reg.h $(DRMHEADERS) $(DRMTEMPLATES) VIASHARED = via_drm.h via_drv.h via.h via_mm.h via_ds.h \ via_3d_reg.h via_drv.c via_ds.c via_irq.c via_map.c \ - via_mm.c via_dma.c + via_mm.c via_dma.c via_verifier.c MACH64HEADERS = mach64.h mach64_drv.h mach64_drm.h $(DRMHEADERS) \ $(DRMTEMPLATES) MACH64SHARED = mach64.h mach64_drv.h mach64_drm.h mach64_dma.c \ diff --git a/linux-2.6/Makefile.kernel b/linux-2.6/Makefile.kernel index 5ff3d399..b75c81fd 100644 --- a/linux-2.6/Makefile.kernel +++ b/linux-2.6/Makefile.kernel @@ -17,7 +17,7 @@ radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o sis-objs := sis_drv.o sis_ds.o sis_mm.o ffb-objs := ffb_drv.o ffb_context.o savage-objs := savage_drv.o savage_dma.o -via-objs := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o via_dma.o +via-objs := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o via_dma.o via_verifier.o mach64-objs := mach64_drv.o mach64_dma.o mach64_irq.o mach64_state.o # Kernel version checks diff --git a/linux-core/Makefile b/linux-core/Makefile index 2d02baea..05854ea5 100644 --- a/linux-core/Makefile +++ b/linux-core/Makefile @@ -92,7 +92,7 @@ VIAHEADERS = via_drm.h via_drv.h via_mm.h via_ds.h \ via_3d_reg.h $(DRMHEADERS) VIASHARED = via_drm.h via_drv.h via_mm.h via_ds.h \ via_3d_reg.h via_drv.c via_ds.c via_irq.c via_map.c \ - via_mm.c via_dma.c + via_mm.c via_dma.c via_verifier.c MACH64HEADERS = mach64_drv.h mach64_drm.h $(DRMHEADERS) MACH64SHARED = mach64_drv.h mach64_drm.h mach64_dma.c \ mach64_irq.c mach64_state.c diff --git a/linux-core/Makefile.kernel b/linux-core/Makefile.kernel index e90c8fcf..34cbf53b 100644 --- a/linux-core/Makefile.kernel +++ b/linux-core/Makefile.kernel @@ -22,8 +22,8 @@ radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o sis-objs := sis_drv.o sis_ds.o sis_mm.o ffb-objs := ffb_drv.o ffb_context.o savage-objs := savage_drv.o savage_dma.o -via-objs := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o via_dma.o -mach64-objs := mach64_drv.o mach64_dma.o mach64_irq.o mach64_state.o +via-objs := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o via_dma.o via_verifier.o +mach64-objs := mach64_drv.o mach64_dma.o mach64_irq.o mach64_state.o obj-m += drm.o obj-$(CONFIG_DRM_TDFX) += tdfx.o diff --git a/shared-core/via_dma.c b/shared-core/via_dma.c index 179636ea..0c2ac470 100644 --- a/shared-core/via_dma.c +++ b/shared-core/via_dma.c @@ -1,13 +1,13 @@ /* via_dma.c -- DMA support for the VIA Unichrome/Pro */ /************************************************************************** - * + * * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. * All Rights Reserved. * * Copyright 2004 Digeo, Inc., Palo Alto, CA, U.S.A. * All Rights Reserved. - * + * * Copyright 2004 The Unichrome project. * All Rights Reserved. * @@ -36,57 +36,6 @@ static void via_cmdbuf_start(drm_via_private_t * dev_priv); static void via_cmdbuf_pause(drm_via_private_t * dev_priv); static void via_cmdbuf_reset(drm_via_private_t * dev_priv); static void via_cmdbuf_rewind(drm_via_private_t * dev_priv); -static int via_wait_idle(drm_via_private_t * dev_priv); - - -/* - * This function needs to be extended whenever a new command set - * is implemented. Currently it works only for the 2D engine - * command, which on the Unichrome allows writing to - * at least the 2D engine and the mpeg engine, but not the - * video engine. - * - * If you update this function with new commands, please also - * consider implementing these commands in - * via_parse_pci_cmdbuffer below. - * - * Carefully review this function for security holes - * after an update!!!!!!!!! - */ - -static int via_check_command_stream(const uint32_t * buf, unsigned int size) -{ - - uint32_t offset; - unsigned int i; - - if (size & 7) { - DRM_ERROR("Illegal command buffer size.\n"); - return DRM_ERR(EINVAL); - } - size >>= 3; - for (i = 0; i < size; ++i) { - offset = *buf; - buf += 2; - if ((offset > ((0x3FF >> 2) | HALCYON_HEADER1)) && - (offset < ((0xC00 >> 2) | HALCYON_HEADER1))) { - DRM_ERROR - ("Attempt to access Burst Command / 3D Area.\n"); - return DRM_ERR(EINVAL); - } else if (offset > ((0xDFF >> 2) | HALCYON_HEADER1)) { - DRM_ERROR("Attempt to access DMA or VGA registers.\n"); - return DRM_ERR(EINVAL); - } - - /* - * ... - * A volunteer should complete this to allow non-root - * usage of accelerated 3D OpenGL. - */ - - } - return 0; -} static inline int via_cmdbuf_wait(drm_via_private_t * dev_priv, unsigned int size) @@ -97,10 +46,10 @@ via_cmdbuf_wait(drm_via_private_t * dev_priv, unsigned int size) uint32_t count; hw_addr_ptr = dev_priv->hw_addr_ptr; cur_addr = dev_priv->dma_low; - next_addr = cur_addr + size; + next_addr = cur_addr + size + 512*1024; count = 1000000; /* How long is this? */ do { - hw_addr = *hw_addr_ptr - agp_base; + hw_addr = *hw_addr_ptr - agp_base; if (count-- == 0) { DRM_ERROR("via_cmdbuf_wait timed out hw %x cur_addr %x next_addr %x\n", hw_addr, cur_addr, next_addr); @@ -141,6 +90,7 @@ int via_dma_cleanup(drm_device_t * dev) drm_core_ioremapfree(&dev_priv->ring.map, dev); dev_priv->ring.virtual_start = NULL; } + } return 0; @@ -224,17 +174,32 @@ static int via_dispatch_cmdbuffer(drm_device_t * dev, drm_via_cmdbuffer_t * cmd) uint32_t *vb; int ret; + + if (cmd->size > pci_bufsiz && pci_bufsiz > 0) { + return DRM_ERR(ENOMEM); + } + vb = via_check_dma(dev_priv, cmd->size); if (vb == NULL) { return DRM_ERR(EAGAIN); } - if (DRM_COPY_FROM_USER(vb, cmd->buf, cmd->size)) { + + if (DRM_COPY_FROM_USER(pci_buf, cmd->buf, cmd->size)) return DRM_ERR(EFAULT); - } - if ((ret = via_check_command_stream(vb, cmd->size))) - return ret; + /* + * Running this function on AGP memory is dead slow. Therefore + * we run it on a temporary cacheable system memory buffer and + * copy it to AGP memory when ready. + */ + + if ((ret = via_verify_command_stream((uint32_t *)pci_buf, cmd->size, dev))) { + return ret; + } + + memcpy(vb, pci_buf, cmd->size); + dev_priv->dma_low += cmd->size; via_cmdbuf_pause(dev_priv); @@ -285,12 +250,12 @@ static int via_parse_pci_cmdbuffer(drm_device_t * dev, const char *buf, unsigned int size) { drm_via_private_t *dev_priv = dev->dev_private; - const uint32_t *regbuf = (uint32_t *) buf; + const uint32_t *regbuf = (const uint32_t *) buf; const uint32_t *regend = regbuf + (size >> 2); int ret; int check_2d_cmd = 1; - if ((ret = via_check_command_stream(regbuf, size))) + if ((ret = via_verify_command_stream(regbuf, size, dev))) return ret; while (regbuf != regend) { @@ -454,7 +419,7 @@ static int via_hook_segment(drm_via_private_t *dev_priv, -static int via_wait_idle(drm_via_private_t * dev_priv) +int via_wait_idle(drm_via_private_t * dev_priv) { int count = 10000000; while (count-- && (VIA_READ(VIA_REG_STATUS) & @@ -465,14 +430,16 @@ static int via_wait_idle(drm_via_private_t * dev_priv) static uint32_t *via_align_cmd(drm_via_private_t * dev_priv, uint32_t cmd_type, uint32_t addr, uint32_t *cmd_addr_hi, - uint32_t *cmd_addr_lo) + uint32_t *cmd_addr_lo, + int skip_wait) { uint32_t agp_base; uint32_t cmd_addr, addr_lo, addr_hi; uint32_t *vb; uint32_t qw_pad_count; - via_cmdbuf_wait(dev_priv, 2*CMDBUF_ALIGNMENT_SIZE); + if (!skip_wait) + via_cmdbuf_wait(dev_priv, 2*CMDBUF_ALIGNMENT_SIZE); vb = via_get_dma(dev_priv); VIA_OUT_RING_QW( HC_HEADER2 | ((VIA_REG_TRANSET >> 2) << 12) | @@ -519,7 +486,7 @@ static void via_cmdbuf_start(drm_via_private_t * dev_priv) dev_priv->last_pause_ptr = via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, - &pause_addr_hi, & pause_addr_lo) - 1; + &pause_addr_hi, & pause_addr_lo, 1) - 1; via_flush_write_combine(); while(! *dev_priv->last_pause_ptr); @@ -539,9 +506,10 @@ static inline void via_dummy_bitblt(drm_via_private_t * dev_priv) uint32_t *vb = via_get_dma(dev_priv); SetReg2DAGP(0x0C, (0 | (0 << 16))); SetReg2DAGP(0x10, 0 | (0 << 16)); - SetReg2DAGP(0x0, 0x1 | 0x2000 | 0xAA000000); + SetReg2DAGP(0x0, 0x1 | 0x2000 | 0xAA000000); } + static void via_cmdbuf_jump(drm_via_private_t * dev_priv) { uint32_t agp_base; @@ -551,7 +519,7 @@ static void via_cmdbuf_jump(drm_via_private_t * dev_priv) agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr; via_align_cmd(dev_priv, HC_HAGPBpID_JUMP, 0, &jump_addr_hi, - &jump_addr_lo); + &jump_addr_lo, 0); dev_priv->dma_low = 0; if (via_cmdbuf_wait(dev_priv, CMDBUF_ALIGNMENT_SIZE) != 0) { @@ -567,7 +535,7 @@ static void via_cmdbuf_jump(drm_via_private_t * dev_priv) via_dummy_bitblt(dev_priv); via_dummy_bitblt(dev_priv); last_pause_ptr = via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi, - &pause_addr_lo) -1; + &pause_addr_lo, 0) -1; /* * The regulator may still be suffering from the shock of the jump. @@ -576,13 +544,12 @@ static void via_cmdbuf_jump(drm_via_private_t * dev_priv) */ via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi, - &pause_addr_lo); + &pause_addr_lo, 0); *last_pause_ptr = pause_addr_lo; via_hook_segment( dev_priv, jump_addr_hi, jump_addr_lo, 0); } - static void via_cmdbuf_rewind(drm_via_private_t * dev_priv) { via_cmdbuf_jump(dev_priv); @@ -592,7 +559,7 @@ static void via_cmdbuf_flush(drm_via_private_t * dev_priv, uint32_t cmd_type) { uint32_t pause_addr_lo, pause_addr_hi; - via_align_cmd(dev_priv, cmd_type, 0, &pause_addr_hi, &pause_addr_lo); + via_align_cmd(dev_priv, cmd_type, 0, &pause_addr_hi, &pause_addr_lo, 0); via_hook_segment( dev_priv, pause_addr_hi, pause_addr_lo, 0); } @@ -607,3 +574,5 @@ static void via_cmdbuf_reset(drm_via_private_t * dev_priv) via_cmdbuf_flush(dev_priv, HC_HAGPBpID_STOP); via_wait_idle(dev_priv); } + +/************************************************************************/ diff --git a/shared-core/via_drv.c b/shared-core/via_drv.c index d22932ec..59690600 100644 --- a/shared-core/via_drv.c +++ b/shared-core/via_drv.c @@ -113,6 +113,7 @@ static int probe(struct pci_dev *pdev, const struct pci_device_id *ent) static int __init via_init(void) { + via_init_command_verifier(); return drm_init(&driver, pciidlist); } diff --git a/shared-core/via_drv.h b/shared-core/via_drv.h index 3bca72f3..13a93b78 100644 --- a/shared-core/via_drv.h +++ b/shared-core/via_drv.h @@ -28,10 +28,10 @@ #define DRIVER_NAME "via" #define DRIVER_DESC "VIA Unichrome" -#define DRIVER_DATE "20041127" +#define DRIVER_DATE "20041204" #define DRIVER_MAJOR 2 -#define DRIVER_MINOR 1 +#define DRIVER_MINOR 2 #define DRIVER_PATCHLEVEL 0 typedef struct drm_via_ring_buffer { @@ -81,7 +81,9 @@ extern void via_driver_irq_postinstall(drm_device_t * dev); extern void via_driver_irq_uninstall(drm_device_t * dev); extern int via_dma_cleanup(drm_device_t * dev); - +extern void via_init_command_verifier(void); +extern int via_verify_command_stream(const uint32_t * buf, unsigned int size, drm_device_t *dev); +extern int via_wait_idle(drm_via_private_t * dev_priv); #endif diff --git a/shared-core/via_verifier.c b/shared-core/via_verifier.c new file mode 100644 index 00000000..1ce28c88 --- /dev/null +++ b/shared-core/via_verifier.c @@ -0,0 +1,671 @@ +/* + * Copyright 2004 The Unichrome Project. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sub license, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE UNICHROME PROJECT, 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. + * + * Author: Thomas Hellström 2004. + * This code was written using docs obtained under NDA from VIA Inc. + * + * Don't run this code directly on an AGP buffer. Due to cache problems it will + * be very slow. + */ + + +#include "via_3d_reg.h" +#include "drmP.h" + +typedef enum{ + state_command, + state_header2, + state_header1, + state_error +} verifier_state_t; + +typedef enum{ + no_sequence = 0, + z_address, + dest_address, + tex_address +}sequence_t; + + +typedef enum{ + no_check = 0, + check_for_header2, + check_for_header1, + check_for_header2_err, + check_for_header1_err, + check_for_fire, + check_z_buffer_addr0, + check_z_buffer_addr1, + check_z_buffer_addr_mode, + check_destination_addr0, + check_destination_addr1, + check_destination_addr_mode, + check_for_dummy, + check_for_dd, + check_texture_addr0, + check_texture_addr1, + check_texture_addr2, + check_texture_addr3, + check_texture_addr4, + check_texture_addr5, + check_texture_addr6, + check_texture_addr7, + check_texture_addr8, + check_texture_addr_mode, + forbidden_command +}hazard_t; + +/* + * Associates each hazard above with a possible multi-command + * sequence. For example an address that is split over multiple + * commands and that needs to be checked at the first command + * that does not include any part of the address. + */ + +static sequence_t seqs[] = { + no_sequence, + no_sequence, + no_sequence, + no_sequence, + no_sequence, + no_sequence, + z_address, + z_address, + z_address, + dest_address, + dest_address, + dest_address, + no_sequence, + no_sequence, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + no_sequence +}; + +typedef struct{ + unsigned int code; + hazard_t hz; +} hz_init_t; + + + +static hz_init_t init_table1[] = { + {0xf2, check_for_header2_err}, + {0xf0, check_for_header1_err}, + {0xee, check_for_fire}, + {0xcc, check_for_dummy}, + {0xdd, check_for_dd}, + {0x00, no_check}, + {0x10, check_z_buffer_addr0}, + {0x11, check_z_buffer_addr1}, + {0x12, check_z_buffer_addr_mode}, + {0x13, no_check}, + {0x14, no_check}, + {0x15, no_check}, + {0x23, no_check}, + {0x24, no_check}, + {0x33, no_check}, + {0x34, no_check}, + {0x35, no_check}, + {0x36, no_check}, + {0x37, no_check}, + {0x38, no_check}, + {0x39, no_check}, + {0x3A, no_check}, + {0x3B, no_check}, + {0x3C, no_check}, + {0x3D, no_check}, + {0x3E, no_check}, + {0x40, check_destination_addr0}, + {0x41, check_destination_addr1}, + {0x42, check_destination_addr_mode}, + {0x43, no_check}, + {0x44, no_check}, + {0x50, no_check}, + {0x51, no_check}, + {0x52, no_check}, + {0x53, no_check}, + {0x54, no_check}, + {0x55, no_check}, + {0x56, no_check}, + {0x57, no_check}, + {0x58, no_check}, + {0x70, no_check}, + {0x71, no_check}, + {0x78, no_check}, + {0x79, no_check}, + {0x7A, no_check}, + {0x7B, no_check}, + {0x7C, no_check}, + {0x7D, no_check} +}; + + + +static hz_init_t init_table2[] = { + {0xf2, check_for_header2_err}, + {0xf0, check_for_header1_err}, + {0xee, check_for_fire}, + {0xcc, check_for_dummy}, + {0x00, check_texture_addr0}, + {0x01, check_texture_addr0}, + {0x02, check_texture_addr0}, + {0x03, check_texture_addr0}, + {0x04, check_texture_addr0}, + {0x05, check_texture_addr0}, + {0x06, check_texture_addr0}, + {0x07, check_texture_addr0}, + {0x08, check_texture_addr0}, + {0x09, check_texture_addr0}, + {0x20, check_texture_addr1}, + {0x21, check_texture_addr1}, + {0x22, check_texture_addr1}, + {0x23, check_texture_addr4}, + {0x2B, check_texture_addr3}, + {0x2C, check_texture_addr3}, + {0x2D, check_texture_addr3}, + {0x2E, check_texture_addr3}, + {0x2F, check_texture_addr3}, + {0x30, check_texture_addr3}, + {0x31, check_texture_addr3}, + {0x32, check_texture_addr3}, + {0x33, check_texture_addr3}, + {0x34, check_texture_addr3}, + {0x4B, check_texture_addr5}, + {0x4C, check_texture_addr6}, + {0x51, check_texture_addr7}, + {0x52, check_texture_addr8}, + {0x77, check_texture_addr2}, + {0x78, no_check}, + {0x79, no_check}, + {0x7A, no_check}, + {0x7B, check_texture_addr_mode}, + {0x7C, no_check}, + {0x7D, no_check}, + {0x7E, no_check}, + {0x7F, no_check}, + {0x80, no_check}, + {0x81, no_check}, + {0x82, no_check}, + {0x83, no_check}, + {0x85, no_check}, + {0x86, no_check}, + {0x87, no_check}, + {0x88, no_check}, + {0x89, no_check}, + {0x8A, no_check}, + {0x90, no_check}, + {0x91, no_check}, + {0x92, no_check}, + {0x93, no_check} +}; + +static hz_init_t init_table3[] = { + {0xf2, check_for_header2_err}, + {0xf0, check_for_header1_err}, + {0xcc, check_for_dummy}, + {0x00, no_check} +}; + + +static hazard_t table1[256]; +static hazard_t table2[256]; +static hazard_t table3[256]; + + + +typedef struct{ + unsigned texture; + uint32_t z_addr; + uint32_t d_addr; + uint32_t t_addr[2][10]; + uint32_t pitch[2][10]; + uint32_t height[2][10]; + uint32_t tex_level_lo[2]; + uint32_t tex_level_hi[2]; + sequence_t unfinished; + int agp_texture; + drm_device_t *dev; +} sequence_context_t; + +static sequence_context_t hc_sequence; + +/* + * stolen from drm_memory.h + */ + +static __inline__ drm_map_t * +via_drm_lookup_map (unsigned long offset, unsigned long size, drm_device_t *dev) +{ + struct list_head *list; + drm_map_list_t *r_list; + drm_map_t *map; + + list_for_each(list, &dev->maplist->head) { + r_list = (drm_map_list_t *) list; + map = r_list->map; + if (!map) + continue; + if (map->offset <= offset && (offset + size) <= (map->offset + map->size)) + return map; + } + return NULL; +} + + +/* + * Require that all AGP texture levels reside in the same map which should + * be mapped by the client. This is not a big restriction. + * FIXME: To actually enforce this security policy strictly, drm_unmap + * would have to wait for dma quiescent before unmapping an AGP page. + * The via_drm_lookup_map call in reality seems to take + * very little CPU time. + */ + + +static __inline__ int +finish_current_sequence(sequence_context_t *cur_seq) +{ + switch(cur_seq->unfinished) { + case z_address: + DRM_DEBUG("Z Buffer start address is 0x%x\n", cur_seq->z_addr); + break; + case dest_address: + DRM_DEBUG("Destination start address is 0x%x\n", cur_seq->d_addr); + break; + case tex_address: + if (cur_seq->agp_texture) { + unsigned start = cur_seq->tex_level_lo[cur_seq->texture]; + unsigned end = cur_seq->tex_level_hi[cur_seq->texture]; + unsigned long lo=~0, hi=0, tmp; + uint32_t *addr, *pitch, *height, tex; + unsigned i; + + if (end > 9) end = 9; + if (start > 9) start = 9; + + addr =&(cur_seq->t_addr[tex = cur_seq->texture][start]); + pitch = &(cur_seq->pitch[tex][start]); + height = &(cur_seq->height[tex][start]); + + for (i=start; i<= end; ++i) { + tmp = *addr++; + if (tmp < lo) lo = tmp; + tmp += (*height++ << *pitch++); + if (tmp > hi) hi = tmp; + } + + if (! via_drm_lookup_map (lo, hi - lo, cur_seq->dev)) { + DRM_ERROR("AGP texture is not in client map\n"); + return 2; + } + } + break; + default: + break; + } + cur_seq->unfinished = no_sequence; + return 0; +} + +static __inline__ int +investigate_hazard( uint32_t cmd, hazard_t hz, sequence_context_t *cur_seq) +{ + register uint32_t tmp, *tmp_addr; + + if (cur_seq->unfinished && (cur_seq->unfinished != seqs[hz])) { + int ret; + if ((ret = finish_current_sequence(cur_seq))) return ret; + } + + switch(hz) { + case check_for_header2: + if (cmd == HALCYON_HEADER2) return 1; + return 0; + case check_for_header1: + if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1) return 1; + return 0; + case check_for_header2_err: + if (cmd == HALCYON_HEADER2) return 1; + DRM_ERROR("Illegal DMA HALCYON_HEADER2 command\n"); + break; + case check_for_header1_err: + if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1) return 1; + DRM_ERROR("Illegal DMA HALCYON_HEADER1 command\n"); + break; + case check_for_fire: + if ((cmd & HALCYON_FIREMASK) == HALCYON_FIRECMD) return 1; + DRM_ERROR("Illegal DMA HALCYON_FIRECMD command\n"); + break; + case check_for_dummy: + if (HC_DUMMY == cmd) return 0; + DRM_ERROR("Illegal DMA HC_DUMMY command\n"); + break; + case check_for_dd: + if (0xdddddddd == cmd) return 0; + DRM_ERROR("Illegal DMA 0xdddddddd command\n"); + break; + case check_z_buffer_addr0: + cur_seq->unfinished = z_address; + cur_seq->z_addr = (cur_seq->z_addr & 0xFF000000) | + (cmd & 0x00FFFFFF); + return 0; + case check_z_buffer_addr1: + cur_seq->unfinished = z_address; + cur_seq->z_addr = (cur_seq->z_addr & 0x00FFFFFF) | + ((cmd & 0xFF) << 24); + return 0; + case check_z_buffer_addr_mode: + cur_seq->unfinished = z_address; + if ((cmd & 0x0000C000) == 0) return 0; + DRM_ERROR("Attempt to place Z buffer in system memory\n"); + return 2; + case check_destination_addr0: + cur_seq->unfinished = dest_address; + cur_seq->d_addr = (cur_seq->d_addr & 0xFF000000) | + (cmd & 0x00FFFFFF); + return 0; + case check_destination_addr1: + cur_seq->unfinished = dest_address; + cur_seq->d_addr = (cur_seq->d_addr & 0x00FFFFFF) | + ((cmd & 0xFF) << 24); + return 0; + case check_destination_addr_mode: + cur_seq->unfinished = dest_address; + if ((cmd & 0x0000C000) == 0) return 0; + DRM_ERROR("Attempt to place 3D drawing buffer in system memory\n"); + return 2; + case check_texture_addr0: + cur_seq->unfinished = tex_address; + tmp = (cmd >> 24); + tmp_addr = &cur_seq->t_addr[cur_seq->texture][tmp]; + *tmp_addr = (*tmp_addr & 0xFF000000) | (cmd & 0x00FFFFFF); + return 0; + case check_texture_addr1: + cur_seq->unfinished = tex_address; + tmp = ((cmd >> 24) - 0x20); + tmp += tmp << 1; + tmp_addr = &cur_seq->t_addr[cur_seq->texture][tmp]; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF) << 24); + tmp_addr++; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF00) << 16); + tmp_addr++; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF0000) << 8); + return 0; + case check_texture_addr2: + cur_seq->unfinished = tex_address; + cur_seq->tex_level_lo[tmp = cur_seq->texture] = cmd & 0x3F; + cur_seq->tex_level_hi[tmp] = (cmd & 0xFC0) >> 6; + return 0; + case check_texture_addr3: + cur_seq->unfinished = tex_address; + tmp = ((cmd >> 24) - 0x2B); + cur_seq->pitch[cur_seq->texture][tmp] = (cmd & 0x00F00000) >> 20; + if (!tmp && (cmd & 0x000FFFFF)) { + DRM_ERROR("Unimplemented texture level 0 pitch mode.\n"); + return 2; + } + return 0; + case check_texture_addr4: + cur_seq->unfinished = tex_address; + tmp_addr = &cur_seq->t_addr[cur_seq->texture][9]; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF) << 24); + return 0; + case check_texture_addr5: + case check_texture_addr6: + cur_seq->unfinished = tex_address; + /* + * Texture width. We don't care since we have the pitch. + */ + return 0; + case check_texture_addr7: + cur_seq->unfinished = tex_address; + tmp_addr = &(cur_seq->height[cur_seq->texture][0]); + tmp_addr[5] = 1 << ((cmd & 0x00F00000) >> 20); + tmp_addr[4] = 1 << ((cmd & 0x000F0000) >> 16); + tmp_addr[3] = 1 << ((cmd & 0x0000F000) >> 12); + tmp_addr[2] = 1 << ((cmd & 0x00000F00) >> 8); + tmp_addr[1] = 1 << ((cmd & 0x000000F0) >> 4); + tmp_addr[0] = 1 << (cmd & 0x0000000F); + return 0; + case check_texture_addr8: + cur_seq->unfinished = tex_address; + tmp_addr = &(cur_seq->height[cur_seq->texture][0]); + tmp_addr[9] = 1 << ((cmd & 0x0000F000) >> 12); + tmp_addr[8] = 1 << ((cmd & 0x00000F00) >> 8); + tmp_addr[7] = 1 << ((cmd & 0x000000F0) >> 4); + tmp_addr[6] = 1 << (cmd & 0x0000000F); + return 0; + case check_texture_addr_mode: + cur_seq->unfinished = tex_address; + if ( 2 != (tmp = cmd & 0x00000003)) { + cur_seq->agp_texture = (tmp == 3); + return 0; + } + DRM_ERROR("Attempt to fetch texture from system memory.\n"); + return 2; + default: + DRM_ERROR("Illegal DMA data: 0x%x\n", cmd); + return 2; + } + return 2; +} + + + + +static __inline__ verifier_state_t +via_check_header2( uint32_t const **buffer, const uint32_t *buf_end ) +{ + uint32_t cmd; + int hz_mode; + hazard_t hz; + const uint32_t *buf = *buffer; + const hazard_t *hz_table; + + if ((buf_end - buf) < 2) { + DRM_ERROR("Illegal termination of DMA HALCYON_HEADER2 sequence.\n"); + return state_error; + } + buf++; + cmd = (*buf++ & 0xFFFF0000) >> 16; + + switch(cmd) { + case HC_ParaType_CmdVdata: + + /* + * Command vertex data. + * It is assumed that the command regulator remains in this state + * until it encounters a double fire command or a header2 data. + * CHECK: Could vertex data accidently be header2 or fire? + * CHECK: What does the regulator do if it encounters a header1 + * cmd? + */ + + while (buf < buf_end) { + if (*buf == HALCYON_HEADER2) break; + if ((*buf & HALCYON_FIREMASK) == HALCYON_FIRECMD) { + buf++; + if ((buf < buf_end) && + ((*buf & HALCYON_FIREMASK) == HALCYON_FIRECMD)) + buf++; + if ((buf < buf_end) && + ((*buf & HALCYON_CMDBMASK) != HC_ACMD_HCmdB)) + break; + } + buf++; + } + *buffer = buf; + return state_command; + + case HC_ParaType_NotTex: + hz_table = table1; + break; + case HC_ParaType_Tex: + hc_sequence.texture = 0; + hz_table = table2; + break; + case (HC_ParaType_Tex | (HC_SubType_Tex1 << 8)): + hc_sequence.texture = 1; + hz_table = table2; + break; + case (HC_ParaType_Tex | (HC_SubType_TexGeneral << 8)): + hz_table = table3; + break; + default: + + /* + * There are some unimplemented HC_ParaTypes here, that + * need to be implemented if the Mesa driver is extended. + */ + + DRM_ERROR("Invalid or unimplemented HALCYON_HEADER2 " + "DMA subcommand: 0x%x\n", cmd); + *buffer = buf; + return state_error; + } + + while(buf < buf_end) { + cmd = *buf++; + if ((hz = hz_table[cmd >> 24])) { + if ((hz_mode = investigate_hazard(cmd, hz, &hc_sequence))) { + if (hz_mode == 1) { + buf--; + break; + } + return state_error; + } + } else if (hc_sequence.unfinished && + finish_current_sequence(&hc_sequence)) { + return state_error; + } + } + if (hc_sequence.unfinished && finish_current_sequence(&hc_sequence)) { + return state_error; + } + *buffer = buf; + return state_command; +} + + +static __inline__ verifier_state_t +via_check_header1( uint32_t const **buffer, const uint32_t *buf_end ) +{ + uint32_t cmd; + const uint32_t *buf = *buffer; + verifier_state_t ret = state_command; + + while (buf < buf_end) { + cmd = *buf; + if ((cmd > ((0x3FF >> 2) | HALCYON_HEADER1)) && + (cmd < ((0xC00 >> 2) | HALCYON_HEADER1))) { + if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1) + break; + DRM_ERROR("Invalid HALCYON_HEADER1 command. " + "Attempt to access 3D- or command burst area.\n"); + ret = state_error; + break; + } else if (cmd > ((0xCFF >> 2) | HALCYON_HEADER1)) { + if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1) + break; + DRM_ERROR("Invalid HALCYON_HEADER1 command. " + "Attempt to access VGA registers.\n"); + ret = state_error; + break; + } else { + buf += 2; + } + } + *buffer = buf; + return ret; +} + + + +int +via_verify_command_stream(const uint32_t * buf, unsigned int size, drm_device_t *dev) +{ + + uint32_t cmd; + const uint32_t *buf_end = buf + ( size >> 2 ); + verifier_state_t state = state_command; + + hc_sequence.dev = dev; + hc_sequence.unfinished = no_sequence; + + while (buf < buf_end) { + switch (state) { + case state_header2: + state = via_check_header2( &buf, buf_end ); + break; + case state_header1: + state = via_check_header1( &buf, buf_end ); + break; + case state_command: + if (HALCYON_HEADER2 == (cmd = *buf)) + state = state_header2; + else if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1) + state = state_header1; + else { + DRM_ERROR("Invalid / Unimplemented DMA HEADER command. 0x%x\n", + cmd); + state = state_error; + } + break; + case state_error: + default: + return DRM_ERR(EINVAL); + } + } + return (state == state_error) ? DRM_ERR(EINVAL) : 0; +} + +static void +setup_hazard_table(hz_init_t init_table[], hazard_t table[], int size) +{ + int i; + + for(i=0; i<256; ++i) { + table[i] = forbidden_command; + } + + for(i=0; i<size; ++i) { + table[init_table[i].code] = init_table[i].hz; + } +} + +void +via_init_command_verifier( void ) +{ + hc_sequence.texture = 0; + setup_hazard_table(init_table1, table1, sizeof(init_table1) / sizeof(hz_init_t)); + setup_hazard_table(init_table2, table2, sizeof(init_table2) / sizeof(hz_init_t)); + setup_hazard_table(init_table3, table3, sizeof(init_table3) / sizeof(hz_init_t)); +} diff --git a/shared/via.h b/shared/via.h index 6b312210..423efc6b 100644 --- a/shared/via.h +++ b/shared/via.h @@ -30,10 +30,10 @@ #define DRIVER_NAME "via" #define DRIVER_DESC "VIA Unichrome" -#define DRIVER_DATE "20041127" +#define DRIVER_DATE "20041204" #define DRIVER_MAJOR 2 -#define DRIVER_MINOR 1 +#define DRIVER_MINOR 2 #define DRIVER_PATCHLEVEL 0 #define DRIVER_IOCTLS \ diff --git a/shared/via_dma.c b/shared/via_dma.c index 4b591fa4..fbd3f6cb 100644 --- a/shared/via_dma.c +++ b/shared/via_dma.c @@ -37,57 +37,6 @@ static void via_cmdbuf_start(drm_via_private_t * dev_priv); static void via_cmdbuf_pause(drm_via_private_t * dev_priv); static void via_cmdbuf_reset(drm_via_private_t * dev_priv); static void via_cmdbuf_rewind(drm_via_private_t * dev_priv); -static int via_wait_idle(drm_via_private_t * dev_priv); - - -/* - * This function needs to be extended whenever a new command set - * is implemented. Currently it works only for the 2D engine - * command, which on the Unichrome allows writing to - * at least the 2D engine and the mpeg engine, but not the - * video engine. - * - * If you update this function with new commands, please also - * consider implementing these commands in - * via_parse_pci_cmdbuffer below. - * - * Carefully review this function for security holes - * after an update!!!!!!!!! - */ - -static int via_check_command_stream(const uint32_t * buf, unsigned int size) -{ - - uint32_t offset; - unsigned int i; - - if (size & 7) { - DRM_ERROR("Illegal command buffer size.\n"); - return DRM_ERR(EINVAL); - } - size >>= 3; - for (i = 0; i < size; ++i) { - offset = *buf; - buf += 2; - if ((offset > ((0x3FF >> 2) | HALCYON_HEADER1)) && - (offset < ((0xC00 >> 2) | HALCYON_HEADER1))) { - DRM_ERROR - ("Attempt to access Burst Command / 3D Area.\n"); - return DRM_ERR(EINVAL); - } else if (offset > ((0xDFF >> 2) | HALCYON_HEADER1)) { - DRM_ERROR("Attempt to access DMA or VGA registers.\n"); - return DRM_ERR(EINVAL); - } - - /* - * ... - * A volunteer should complete this to allow non-root - * usage of accelerated 3D OpenGL. - */ - - } - return 0; -} static inline int via_cmdbuf_wait(drm_via_private_t * dev_priv, unsigned int size) @@ -98,7 +47,7 @@ via_cmdbuf_wait(drm_via_private_t * dev_priv, unsigned int size) uint32_t count; hw_addr_ptr = dev_priv->hw_addr_ptr; cur_addr = dev_priv->dma_low; - next_addr = cur_addr + size; + next_addr = cur_addr + size + 512*1024; count = 1000000; /* How long is this? */ do { hw_addr = *hw_addr_ptr - agp_base; @@ -142,6 +91,7 @@ int via_dma_cleanup(drm_device_t * dev) drm_core_ioremapfree(&dev_priv->ring.map, dev); dev_priv->ring.virtual_start = NULL; } + } return 0; @@ -225,17 +175,32 @@ static int via_dispatch_cmdbuffer(drm_device_t * dev, drm_via_cmdbuffer_t * cmd) uint32_t *vb; int ret; + + if (cmd->size > pci_bufsiz && pci_bufsiz > 0) { + return DRM_ERR(ENOMEM); + } + vb = via_check_dma(dev_priv, cmd->size); if (vb == NULL) { return DRM_ERR(EAGAIN); } - if (DRM_COPY_FROM_USER(vb, cmd->buf, cmd->size)) { + + if (DRM_COPY_FROM_USER(pci_buf, cmd->buf, cmd->size)) return DRM_ERR(EFAULT); - } - if ((ret = via_check_command_stream(vb, cmd->size))) - return ret; + /* + * Running this function on AGP memory is dead slow. Therefore + * we run it on a temporary cacheable system memory buffer and + * copy it to AGP memory when ready. + */ + + if ((ret = via_verify_command_stream((uint32_t *)pci_buf, cmd->size, dev))) { + return ret; + } + + memcpy(vb, pci_buf, cmd->size); + dev_priv->dma_low += cmd->size; via_cmdbuf_pause(dev_priv); @@ -286,12 +251,12 @@ static int via_parse_pci_cmdbuffer(drm_device_t * dev, const char *buf, unsigned int size) { drm_via_private_t *dev_priv = dev->dev_private; - const uint32_t *regbuf = (uint32_t *) buf; + const uint32_t *regbuf = (const uint32_t *) buf; const uint32_t *regend = regbuf + (size >> 2); int ret; int check_2d_cmd = 1; - if ((ret = via_check_command_stream(regbuf, size))) + if ((ret = via_verify_command_stream(regbuf, size, dev))) return ret; while (regbuf != regend) { @@ -455,7 +420,7 @@ static int via_hook_segment(drm_via_private_t *dev_priv, -static int via_wait_idle(drm_via_private_t * dev_priv) +int via_wait_idle(drm_via_private_t * dev_priv) { int count = 10000000; while (count-- && (VIA_READ(VIA_REG_STATUS) & @@ -466,14 +431,16 @@ static int via_wait_idle(drm_via_private_t * dev_priv) static uint32_t *via_align_cmd(drm_via_private_t * dev_priv, uint32_t cmd_type, uint32_t addr, uint32_t *cmd_addr_hi, - uint32_t *cmd_addr_lo) + uint32_t *cmd_addr_lo, + int skip_wait) { uint32_t agp_base; uint32_t cmd_addr, addr_lo, addr_hi; uint32_t *vb; uint32_t qw_pad_count; - via_cmdbuf_wait(dev_priv, 2*CMDBUF_ALIGNMENT_SIZE); + if (!skip_wait) + via_cmdbuf_wait(dev_priv, 2*CMDBUF_ALIGNMENT_SIZE); vb = via_get_dma(dev_priv); VIA_OUT_RING_QW( HC_HEADER2 | ((VIA_REG_TRANSET >> 2) << 12) | @@ -520,7 +487,7 @@ static void via_cmdbuf_start(drm_via_private_t * dev_priv) dev_priv->last_pause_ptr = via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, - &pause_addr_hi, & pause_addr_lo) - 1; + &pause_addr_hi, & pause_addr_lo, 1) - 1; via_flush_write_combine(); while(! *dev_priv->last_pause_ptr); @@ -553,7 +520,7 @@ static void via_cmdbuf_jump(drm_via_private_t * dev_priv) agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr; via_align_cmd(dev_priv, HC_HAGPBpID_JUMP, 0, &jump_addr_hi, - &jump_addr_lo); + &jump_addr_lo, 0); dev_priv->dma_low = 0; if (via_cmdbuf_wait(dev_priv, CMDBUF_ALIGNMENT_SIZE) != 0) { @@ -569,7 +536,7 @@ static void via_cmdbuf_jump(drm_via_private_t * dev_priv) via_dummy_bitblt(dev_priv); via_dummy_bitblt(dev_priv); last_pause_ptr = via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi, - &pause_addr_lo) -1; + &pause_addr_lo, 0) -1; /* * The regulator may still be suffering from the shock of the jump. @@ -578,13 +545,12 @@ static void via_cmdbuf_jump(drm_via_private_t * dev_priv) */ via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi, - &pause_addr_lo); + &pause_addr_lo, 0); *last_pause_ptr = pause_addr_lo; via_hook_segment( dev_priv, jump_addr_hi, jump_addr_lo, 0); } - static void via_cmdbuf_rewind(drm_via_private_t * dev_priv) { via_cmdbuf_jump(dev_priv); @@ -594,7 +560,7 @@ static void via_cmdbuf_flush(drm_via_private_t * dev_priv, uint32_t cmd_type) { uint32_t pause_addr_lo, pause_addr_hi; - via_align_cmd(dev_priv, cmd_type, 0, &pause_addr_hi, &pause_addr_lo); + via_align_cmd(dev_priv, cmd_type, 0, &pause_addr_hi, &pause_addr_lo, 0); via_hook_segment( dev_priv, pause_addr_hi, pause_addr_lo, 0); } diff --git a/shared/via_drv.h b/shared/via_drv.h index 9ed8cd3a..057956f0 100644 --- a/shared/via_drv.h +++ b/shared/via_drv.h @@ -44,7 +44,7 @@ typedef struct drm_via_private { unsigned int dma_offset; volatile uint32_t *last_pause_ptr; volatile uint32_t *hw_addr_ptr; - drm_via_ring_buffer_t ring; + drm_via_ring_buffer_t ring, fb_blit; struct timeval last_vblank; int last_vblank_valid; unsigned usec_per_vblank; @@ -73,7 +73,11 @@ extern void via_driver_irq_postinstall(drm_device_t * dev); extern void via_driver_irq_uninstall(drm_device_t * dev); extern int via_dma_cleanup(drm_device_t * dev); - - +extern int via_wait_idle(drm_via_private_t * dev_priv); +extern void via_init_command_verifier( void ); +extern int via_verify_command_stream(const uint32_t * buf, unsigned int size, + drm_device_t *dev); +extern int via_fb_free(drm_via_mem_t * mem); +extern int via_fb_alloc(drm_via_mem_t * mem); #endif diff --git a/shared/via_verifier.c b/shared/via_verifier.c new file mode 100644 index 00000000..e0f25546 --- /dev/null +++ b/shared/via_verifier.c @@ -0,0 +1,672 @@ +/* + * Copyright 2004 The Unichrome Project. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sub license, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE UNICHROME PROJECT, 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. + * + * Author: Thomas Hellström 2004. + * This code was written using docs obtained under NDA from VIA Inc. + * + * Don't run this code directly on an AGP buffer. Due to cache problems it will + * be very slow. + */ + + +#include "via_3d_reg.h" +#include "via.h" +#include "drmP.h" + +typedef enum{ + state_command, + state_header2, + state_header1, + state_error +} verifier_state_t; + +typedef enum{ + no_sequence = 0, + z_address, + dest_address, + tex_address +}sequence_t; + + +typedef enum{ + no_check = 0, + check_for_header2, + check_for_header1, + check_for_header2_err, + check_for_header1_err, + check_for_fire, + check_z_buffer_addr0, + check_z_buffer_addr1, + check_z_buffer_addr_mode, + check_destination_addr0, + check_destination_addr1, + check_destination_addr_mode, + check_for_dummy, + check_for_dd, + check_texture_addr0, + check_texture_addr1, + check_texture_addr2, + check_texture_addr3, + check_texture_addr4, + check_texture_addr5, + check_texture_addr6, + check_texture_addr7, + check_texture_addr8, + check_texture_addr_mode, + forbidden_command +}hazard_t; + +/* + * Associates each hazard above with a possible multi-command + * sequence. For example an address that is split over multiple + * commands and that needs to be checked at the first command + * that does not include any part of the address. + */ + +static sequence_t seqs[] = { + no_sequence, + no_sequence, + no_sequence, + no_sequence, + no_sequence, + no_sequence, + z_address, + z_address, + z_address, + dest_address, + dest_address, + dest_address, + no_sequence, + no_sequence, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + tex_address, + no_sequence +}; + +typedef struct{ + unsigned int code; + hazard_t hz; +} hz_init_t; + + + +static hz_init_t init_table1[] = { + {0xf2, check_for_header2_err}, + {0xf0, check_for_header1_err}, + {0xee, check_for_fire}, + {0xcc, check_for_dummy}, + {0xdd, check_for_dd}, + {0x00, no_check}, + {0x10, check_z_buffer_addr0}, + {0x11, check_z_buffer_addr1}, + {0x12, check_z_buffer_addr_mode}, + {0x13, no_check}, + {0x14, no_check}, + {0x15, no_check}, + {0x23, no_check}, + {0x24, no_check}, + {0x33, no_check}, + {0x34, no_check}, + {0x35, no_check}, + {0x36, no_check}, + {0x37, no_check}, + {0x38, no_check}, + {0x39, no_check}, + {0x3A, no_check}, + {0x3B, no_check}, + {0x3C, no_check}, + {0x3D, no_check}, + {0x3E, no_check}, + {0x40, check_destination_addr0}, + {0x41, check_destination_addr1}, + {0x42, check_destination_addr_mode}, + {0x43, no_check}, + {0x44, no_check}, + {0x50, no_check}, + {0x51, no_check}, + {0x52, no_check}, + {0x53, no_check}, + {0x54, no_check}, + {0x55, no_check}, + {0x56, no_check}, + {0x57, no_check}, + {0x58, no_check}, + {0x70, no_check}, + {0x71, no_check}, + {0x78, no_check}, + {0x79, no_check}, + {0x7A, no_check}, + {0x7B, no_check}, + {0x7C, no_check}, + {0x7D, no_check} +}; + + + +static hz_init_t init_table2[] = { + {0xf2, check_for_header2_err}, + {0xf0, check_for_header1_err}, + {0xee, check_for_fire}, + {0xcc, check_for_dummy}, + {0x00, check_texture_addr0}, + {0x01, check_texture_addr0}, + {0x02, check_texture_addr0}, + {0x03, check_texture_addr0}, + {0x04, check_texture_addr0}, + {0x05, check_texture_addr0}, + {0x06, check_texture_addr0}, + {0x07, check_texture_addr0}, + {0x08, check_texture_addr0}, + {0x09, check_texture_addr0}, + {0x20, check_texture_addr1}, + {0x21, check_texture_addr1}, + {0x22, check_texture_addr1}, + {0x23, check_texture_addr4}, + {0x2B, check_texture_addr3}, + {0x2C, check_texture_addr3}, + {0x2D, check_texture_addr3}, + {0x2E, check_texture_addr3}, + {0x2F, check_texture_addr3}, + {0x30, check_texture_addr3}, + {0x31, check_texture_addr3}, + {0x32, check_texture_addr3}, + {0x33, check_texture_addr3}, + {0x34, check_texture_addr3}, + {0x4B, check_texture_addr5}, + {0x4C, check_texture_addr6}, + {0x51, check_texture_addr7}, + {0x52, check_texture_addr8}, + {0x77, check_texture_addr2}, + {0x78, no_check}, + {0x79, no_check}, + {0x7A, no_check}, + {0x7B, check_texture_addr_mode}, + {0x7C, no_check}, + {0x7D, no_check}, + {0x7E, no_check}, + {0x7F, no_check}, + {0x80, no_check}, + {0x81, no_check}, + {0x82, no_check}, + {0x83, no_check}, + {0x85, no_check}, + {0x86, no_check}, + {0x87, no_check}, + {0x88, no_check}, + {0x89, no_check}, + {0x8A, no_check}, + {0x90, no_check}, + {0x91, no_check}, + {0x92, no_check}, + {0x93, no_check} +}; + +static hz_init_t init_table3[] = { + {0xf2, check_for_header2_err}, + {0xf0, check_for_header1_err}, + {0xcc, check_for_dummy}, + {0x00, no_check} +}; + + +static hazard_t table1[256]; +static hazard_t table2[256]; +static hazard_t table3[256]; + + + +typedef struct{ + unsigned texture; + uint32_t z_addr; + uint32_t d_addr; + uint32_t t_addr[2][10]; + uint32_t pitch[2][10]; + uint32_t height[2][10]; + uint32_t tex_level_lo[2]; + uint32_t tex_level_hi[2]; + sequence_t unfinished; + int agp_texture; + drm_device_t *dev; +} sequence_context_t; + +static sequence_context_t hc_sequence; + +/* + * stolen from drm_memory.h + */ + +static __inline__ drm_map_t * +via_drm_lookup_map (unsigned long offset, unsigned long size, drm_device_t *dev) +{ + struct list_head *list; + drm_map_list_t *r_list; + drm_map_t *map; + + list_for_each(list, &dev->maplist->head) { + r_list = (drm_map_list_t *) list; + map = r_list->map; + if (!map) + continue; + if (map->offset <= offset && (offset + size) <= (map->offset + map->size)) + return map; + } + return NULL; +} + + +/* + * Require that all AGP texture levels reside in the same map which should + * be mapped by the client. This is not a big restriction. + * FIXME: To actually enforce this security policy strictly, drm_unmap + * would have to wait for dma quiescent before unmapping an AGP page. + * The via_drm_lookup_map call in reality seems to take + * very little CPU time. + */ + + +static __inline__ int +finish_current_sequence(sequence_context_t *cur_seq) +{ + switch(cur_seq->unfinished) { + case z_address: + DRM_DEBUG("Z Buffer start address is 0x%x\n", cur_seq->z_addr); + break; + case dest_address: + DRM_DEBUG("Destination start address is 0x%x\n", cur_seq->d_addr); + break; + case tex_address: + if (cur_seq->agp_texture) { + unsigned start = cur_seq->tex_level_lo[cur_seq->texture]; + unsigned end = cur_seq->tex_level_hi[cur_seq->texture]; + unsigned long lo=~0, hi=0, tmp; + uint32_t *addr, *pitch, *height, tex; + unsigned i; + + if (end > 9) end = 9; + if (start > 9) start = 9; + + addr =&(cur_seq->t_addr[tex = cur_seq->texture][start]); + pitch = &(cur_seq->pitch[tex][start]); + height = &(cur_seq->height[tex][start]); + + for (i=start; i<= end; ++i) { + tmp = *addr++; + if (tmp < lo) lo = tmp; + tmp += (*height++ << *pitch++); + if (tmp > hi) hi = tmp; + } + + if (! via_drm_lookup_map (lo, hi - lo, cur_seq->dev)) { + DRM_ERROR("AGP texture is not in client map\n"); + return 2; + } + } + break; + default: + break; + } + cur_seq->unfinished = no_sequence; + return 0; +} + +static __inline__ int +investigate_hazard( uint32_t cmd, hazard_t hz, sequence_context_t *cur_seq) +{ + register uint32_t tmp, *tmp_addr; + + if (cur_seq->unfinished && (cur_seq->unfinished != seqs[hz])) { + int ret; + if ((ret = finish_current_sequence(cur_seq))) return ret; + } + + switch(hz) { + case check_for_header2: + if (cmd == HALCYON_HEADER2) return 1; + return 0; + case check_for_header1: + if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1) return 1; + return 0; + case check_for_header2_err: + if (cmd == HALCYON_HEADER2) return 1; + DRM_ERROR("Illegal DMA HALCYON_HEADER2 command\n"); + break; + case check_for_header1_err: + if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1) return 1; + DRM_ERROR("Illegal DMA HALCYON_HEADER1 command\n"); + break; + case check_for_fire: + if ((cmd & HALCYON_FIREMASK) == HALCYON_FIRECMD) return 1; + DRM_ERROR("Illegal DMA HALCYON_FIRECMD command\n"); + break; + case check_for_dummy: + if (HC_DUMMY == cmd) return 0; + DRM_ERROR("Illegal DMA HC_DUMMY command\n"); + break; + case check_for_dd: + if (0xdddddddd == cmd) return 0; + DRM_ERROR("Illegal DMA 0xdddddddd command\n"); + break; + case check_z_buffer_addr0: + cur_seq->unfinished = z_address; + cur_seq->z_addr = (cur_seq->z_addr & 0xFF000000) | + (cmd & 0x00FFFFFF); + return 0; + case check_z_buffer_addr1: + cur_seq->unfinished = z_address; + cur_seq->z_addr = (cur_seq->z_addr & 0x00FFFFFF) | + ((cmd & 0xFF) << 24); + return 0; + case check_z_buffer_addr_mode: + cur_seq->unfinished = z_address; + if ((cmd & 0x0000C000) == 0) return 0; + DRM_ERROR("Attempt to place Z buffer in system memory\n"); + return 2; + case check_destination_addr0: + cur_seq->unfinished = dest_address; + cur_seq->d_addr = (cur_seq->d_addr & 0xFF000000) | + (cmd & 0x00FFFFFF); + return 0; + case check_destination_addr1: + cur_seq->unfinished = dest_address; + cur_seq->d_addr = (cur_seq->d_addr & 0x00FFFFFF) | + ((cmd & 0xFF) << 24); + return 0; + case check_destination_addr_mode: + cur_seq->unfinished = dest_address; + if ((cmd & 0x0000C000) == 0) return 0; + DRM_ERROR("Attempt to place 3D drawing buffer in system memory\n"); + return 2; + case check_texture_addr0: + cur_seq->unfinished = tex_address; + tmp = (cmd >> 24); + tmp_addr = &cur_seq->t_addr[cur_seq->texture][tmp]; + *tmp_addr = (*tmp_addr & 0xFF000000) | (cmd & 0x00FFFFFF); + return 0; + case check_texture_addr1: + cur_seq->unfinished = tex_address; + tmp = ((cmd >> 24) - 0x20); + tmp += tmp << 1; + tmp_addr = &cur_seq->t_addr[cur_seq->texture][tmp]; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF) << 24); + tmp_addr++; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF00) << 16); + tmp_addr++; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF0000) << 8); + return 0; + case check_texture_addr2: + cur_seq->unfinished = tex_address; + cur_seq->tex_level_lo[tmp = cur_seq->texture] = cmd & 0x3F; + cur_seq->tex_level_hi[tmp] = (cmd & 0xFC0) >> 6; + return 0; + case check_texture_addr3: + cur_seq->unfinished = tex_address; + tmp = ((cmd >> 24) - 0x2B); + cur_seq->pitch[cur_seq->texture][tmp] = (cmd & 0x00F00000) >> 20; + if (!tmp && (cmd & 0x000FFFFF)) { + DRM_ERROR("Unimplemented texture level 0 pitch mode.\n"); + return 2; + } + return 0; + case check_texture_addr4: + cur_seq->unfinished = tex_address; + tmp_addr = &cur_seq->t_addr[cur_seq->texture][9]; + *tmp_addr = (*tmp_addr & 0x00FFFFFF) | ((cmd & 0xFF) << 24); + return 0; + case check_texture_addr5: + case check_texture_addr6: + cur_seq->unfinished = tex_address; + /* + * Texture width. We don't care since we have the pitch. + */ + return 0; + case check_texture_addr7: + cur_seq->unfinished = tex_address; + tmp_addr = &(cur_seq->height[cur_seq->texture][0]); + tmp_addr[5] = 1 << ((cmd & 0x00F00000) >> 20); + tmp_addr[4] = 1 << ((cmd & 0x000F0000) >> 16); + tmp_addr[3] = 1 << ((cmd & 0x0000F000) >> 12); + tmp_addr[2] = 1 << ((cmd & 0x00000F00) >> 8); + tmp_addr[1] = 1 << ((cmd & 0x000000F0) >> 4); + tmp_addr[0] = 1 << (cmd & 0x0000000F); + return 0; + case check_texture_addr8: + cur_seq->unfinished = tex_address; + tmp_addr = &(cur_seq->height[cur_seq->texture][0]); + tmp_addr[9] = 1 << ((cmd & 0x0000F000) >> 12); + tmp_addr[8] = 1 << ((cmd & 0x00000F00) >> 8); + tmp_addr[7] = 1 << ((cmd & 0x000000F0) >> 4); + tmp_addr[6] = 1 << (cmd & 0x0000000F); + return 0; + case check_texture_addr_mode: + cur_seq->unfinished = tex_address; + if ( 2 != (tmp = cmd & 0x00000003)) { + cur_seq->agp_texture = (tmp == 3); + return 0; + } + DRM_ERROR("Attempt to fetch texture from system memory.\n"); + return 2; + default: + DRM_ERROR("Illegal DMA data: 0x%x\n", cmd); + return 2; + } + return 2; +} + + + + +static __inline__ verifier_state_t +via_check_header2( uint32_t const **buffer, const uint32_t *buf_end ) +{ + uint32_t cmd; + int hz_mode; + hazard_t hz; + const uint32_t *buf = *buffer; + const hazard_t *hz_table; + + if ((buf_end - buf) < 2) { + DRM_ERROR("Illegal termination of DMA HALCYON_HEADER2 sequence.\n"); + return state_error; + } + buf++; + cmd = (*buf++ & 0xFFFF0000) >> 16; + + switch(cmd) { + case HC_ParaType_CmdVdata: + + /* + * Command vertex data. + * It is assumed that the command regulator remains in this state + * until it encounters a double fire command or a header2 data. + * CHECK: Could vertex data accidently be header2 or fire? + * CHECK: What does the regulator do if it encounters a header1 + * cmd? + */ + + while (buf < buf_end) { + if (*buf == HALCYON_HEADER2) break; + if ((*buf & HALCYON_FIREMASK) == HALCYON_FIRECMD) { + buf++; + if ((buf < buf_end) && + ((*buf & HALCYON_FIREMASK) == HALCYON_FIRECMD)) + buf++; + if ((buf < buf_end) && + ((*buf & HALCYON_CMDBMASK) != HC_ACMD_HCmdB)) + break; + } + buf++; + } + *buffer = buf; + return state_command; + + case HC_ParaType_NotTex: + hz_table = table1; + break; + case HC_ParaType_Tex: + hc_sequence.texture = 0; + hz_table = table2; + break; + case (HC_ParaType_Tex | (HC_SubType_Tex1 << 8)): + hc_sequence.texture = 1; + hz_table = table2; + break; + case (HC_ParaType_Tex | (HC_SubType_TexGeneral << 8)): + hz_table = table3; + break; + default: + + /* + * There are some unimplemented HC_ParaTypes here, that + * need to be implemented if the Mesa driver is extended. + */ + + DRM_ERROR("Invalid or unimplemented HALCYON_HEADER2 " + "DMA subcommand: 0x%x\n", cmd); + *buffer = buf; + return state_error; + } + + while(buf < buf_end) { + cmd = *buf++; + if ((hz = hz_table[cmd >> 24])) { + if ((hz_mode = investigate_hazard(cmd, hz, &hc_sequence))) { + if (hz_mode == 1) { + buf--; + break; + } + return state_error; + } + } else if (hc_sequence.unfinished && + finish_current_sequence(&hc_sequence)) { + return state_error; + } + } + if (hc_sequence.unfinished && finish_current_sequence(&hc_sequence)) { + return state_error; + } + *buffer = buf; + return state_command; +} + + +static __inline__ verifier_state_t +via_check_header1( uint32_t const **buffer, const uint32_t *buf_end ) +{ + uint32_t cmd; + const uint32_t *buf = *buffer; + verifier_state_t ret = state_command; + + while (buf < buf_end) { + cmd = *buf; + if ((cmd > ((0x3FF >> 2) | HALCYON_HEADER1)) && + (cmd < ((0xC00 >> 2) | HALCYON_HEADER1))) { + if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1) + break; + DRM_ERROR("Invalid HALCYON_HEADER1 command. " + "Attempt to access 3D- or command burst area.\n"); + ret = state_error; + break; + } else if (cmd > ((0xCFF >> 2) | HALCYON_HEADER1)) { + if ((cmd & HALCYON_HEADER1MASK) != HALCYON_HEADER1) + break; + DRM_ERROR("Invalid HALCYON_HEADER1 command. " + "Attempt to access VGA registers.\n"); + ret = state_error; + break; + } else { + buf += 2; + } + } + *buffer = buf; + return ret; +} + + + +int +via_verify_command_stream(const uint32_t * buf, unsigned int size, drm_device_t *dev) +{ + + uint32_t cmd; + const uint32_t *buf_end = buf + ( size >> 2 ); + verifier_state_t state = state_command; + + hc_sequence.dev = dev; + hc_sequence.unfinished = no_sequence; + + while (buf < buf_end) { + switch (state) { + case state_header2: + state = via_check_header2( &buf, buf_end ); + break; + case state_header1: + state = via_check_header1( &buf, buf_end ); + break; + case state_command: + if (HALCYON_HEADER2 == (cmd = *buf)) + state = state_header2; + else if ((cmd & HALCYON_HEADER1MASK) == HALCYON_HEADER1) + state = state_header1; + else { + DRM_ERROR("Invalid / Unimplemented DMA HEADER command. 0x%x\n", + cmd); + state = state_error; + } + break; + case state_error: + default: + return DRM_ERR(EINVAL); + } + } + return (state == state_error) ? DRM_ERR(EINVAL) : 0; +} + +static void +setup_hazard_table(hz_init_t init_table[], hazard_t table[], int size) +{ + int i; + + for(i=0; i<256; ++i) { + table[i] = forbidden_command; + } + + for(i=0; i<size; ++i) { + table[init_table[i].code] = init_table[i].hz; + } +} + +void +via_init_command_verifier( void ) +{ + hc_sequence.texture = 0; + setup_hazard_table(init_table1, table1, sizeof(init_table1) / sizeof(hz_init_t)); + setup_hazard_table(init_table2, table2, sizeof(init_table2) / sizeof(hz_init_t)); + setup_hazard_table(init_table3, table3, sizeof(init_table3) / sizeof(hz_init_t)); +} |