intel-gpu-tools/debugging: shader debugger

high level summary of the files:
  * debug_rdata - get current state from debug registers. Helpful
    when developing the debugger, and could serve some purpose in the
    future.
  * eudb - the debugger itself
  * eviction_macro - generate the proper macro to flush the EU render
    cache until I get control flow working
  * pre_cpp - an evaluating c preprocesser like thing, to be used before
    cpp
  * sr - the system routine, exception handler which runs on the EU
  * test - a very basic test system routine
  * debug.h
This commit is contained in:
Ben Widawsky 2011-06-24 11:42:03 -07:00
parent e39b13d7e4
commit 62873796c4
11 changed files with 1463 additions and 1 deletions

View File

@ -21,4 +21,8 @@
ACLOCAL_AMFLAGS = -I m4 ${ACLOCAL_FLAGS}
SUBDIRS = lib man tools scripts tests benchmarks
if HAVE_SHADER_DEBUGGER
SHADER_DEBUGGER_SUBDIR = debugger
endif
SUBDIRS = lib man tools scripts tests benchmarks $(SHADER_DEBUGGER_SUBDIR)

View File

@ -86,6 +86,16 @@ fi
AC_SUBST([WARN_CFLAGS])
AC_ARG_ENABLE(shader-debugger,
AS_HELP_STRING([--enable-shader-debugger],
[Enable shader debugging support [default=no]]),
[SHADER_DEBUGGER="$enableval"],
[SHADER_DEBUGGER=no])
AM_CONDITIONAL(HAVE_SHADER_DEBUGGER, [test "x$SHADER_DEBUGGER" = xyes])
if test "x$SHADER_DEBUGGER" = xyes; then
AC_DEFINE(HAVE_SHADER_DEBUGGER, 1, [Have shader debugging support])
fi
m4_ifndef([XORG_MACROS_VERSION], [AC_FATAL([must install xorg-macros 1.3 or later before running autoconf/autogen])])
XORG_MACROS_VERSION(1.3)
XORG_DEFAULT_OPTIONS
@ -98,5 +108,6 @@ AC_CONFIG_FILES([
scripts/Makefile
tests/Makefile
tools/Makefile
debugger/Makefile
])
AC_OUTPUT

12
debugger/Makefile.am Normal file
View File

@ -0,0 +1,12 @@
SUBDIRS=system_routine
bin_PROGRAMS = \
eudb \
debug_rdata \
$(NULL)
LDADD = ../lib/libintel_tools.la $(DRM_LIBS) $(PCIACCESS_LIBS)
AM_CFLAGS = $(DRM_CFLAGS) $(PCIACCESS_CFLAGS) $(WARN_CFLAGS) \
-I$(srcdir)/.. \
-I$(srcdir)/../lib

141
debugger/debug_rdata.c Normal file
View File

@ -0,0 +1,141 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
*
*/
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include "intel_gpu_tools.h"
struct eu_rdata {
union {
struct {
uint8_t sendc_dep : 1;
uint8_t swh_dep : 1;
uint8_t pwc_dep : 1;
uint8_t n2_dep : 1;
uint8_t n1_dep : 1;
uint8_t n0_dep : 1;
uint8_t flag1_dep : 1;
uint8_t flag0_dep : 1;
uint8_t indx_dep : 1;
uint8_t mrf_dep : 1;
uint8_t dst_dep : 1;
uint8_t src2_dep : 1;
uint8_t src1_dep : 1;
uint8_t src0_dep : 1;
uint8_t mp_dep_pin : 1;
uint8_t sp_dep_pin : 1;
uint8_t fftid : 8;
uint8_t ffid : 4;
uint8_t instruction_valid : 1;
uint8_t thread_status : 3;
};
uint32_t dword;
} ud0;
union {
struct {
uint8_t mrf_addr : 4;
uint8_t dst_addr : 7;
uint8_t src2_addr : 7;
uint8_t src1_addr : 7;
uint8_t src0_addr : 7;
};
uint32_t dword;
} ud1;
union {
struct {
uint16_t exip : 12;
uint8_t opcode : 7;
uint8_t pwc : 8;
uint8_t instruction_valid : 1;
uint8_t mbz : 4;
};
uint32_t dword;
} ud2;
};
const char *thread_status[] =
{"INVALID", "invalid/no thread", "standby (dependency)", "INVALID", "Executing",
"INVALID" , "INVALID" , "INVALID"};
static struct eu_rdata
collect_rdata(int eu, int tid) {
struct eu_rdata rdata;
intel_register_write(0x7800, eu << 16 | (3 * tid) << 8);
rdata.ud0.dword = intel_register_read(0x7840);
intel_register_write(0x7800, eu << 16 | (3 * tid + 1) << 8);
rdata.ud1.dword = intel_register_read(0x7840);
intel_register_write(0x7800, eu << 16 | (3 * tid + 2) << 8);
rdata.ud2.dword = intel_register_read(0x7840);
return rdata;
}
static void
print_rdata(struct eu_rdata rdata) {
printf("\t%s\n", thread_status[rdata.ud0.thread_status]);
printf("\tn1_dep: %d\n", rdata.ud0.n1_dep);
printf("\tpwc_dep: %d\n", rdata.ud0.pwc_dep);
printf("\tswh_dep: %d\n", rdata.ud0.swh_dep);
printf("\tsource 0 %x\n", rdata.ud1.src0_addr);
printf("\tsource 1 %x\n", rdata.ud1.src1_addr);
printf("\tsource 2 %x\n", rdata.ud1.src2_addr);
printf("\tdest %x\n", rdata.ud1.dst_addr);
printf("\tmrf %x\n", rdata.ud1.mrf_addr);
printf("\tIP: %x\n", rdata.ud2.exip);
printf("\topcode: %x\n", rdata.ud2.opcode);
}
static void
find_stuck_threads(void)
{
int i, j;
for (i = 0; i < 15; i++)
for (j = 0; j < 5; j++) {
struct eu_rdata rdata;
rdata = collect_rdata(i, j);
if (rdata.ud0.thread_status == 2 ||
rdata.ud0.thread_status == 4) {
printf("%d %d:\n", i, j);
print_rdata(rdata);
}
}
}
int main(int argc, char *argv[]) {
struct pci_device *pci_dev;
pci_dev = intel_get_pci_device();
intel_register_access_init(pci_dev, 1);
find_stuck_threads();
// collect_rdata(atoi(argv[1]), atoi(argv[2]));
return 0;
}

606
debugger/eudb.c Normal file
View File

@ -0,0 +1,606 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
*
* Notes:
*
*/
#include <signal.h>
#include <stdlib.h>
#include <fcntl.h>
#include <unistd.h>
#include <string.h>
#include <strings.h>
#include <assert.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <sys/un.h>
#include <sys/socket.h>
#include "drm.h"
#include "i915_drm.h"
#include "drmtest.h"
#include "intel_chipset.h"
#include "intel_bufmgr.h"
#include "intel_gpu_tools.h"
#include "intel_batchbuffer.h"
#include "intel_debug.h"
#include "debug.h"
#define EU_ATT 0x7810
#define EU_ATT_CLR 0x7830
#define RSVD_EU -1
#define RSVD_THREAD -1
#define RSVD_ID EUID(-1, -1, -1)
enum {
EBAD_SHMEM,
EBAD_PROTOCOL,
EBAD_MAGIC,
EBAD_WRITE
};
struct debuggee {
int euid;
int tid;
int fd;
int clr;
uint32_t reg;
};
struct debugger {
struct debuggee *debuggees;
int num_threads;
int real_num_threads;
int threads_per_eu;
} *eu_info;
drm_intel_bufmgr *bufmgr;
struct intel_batchbuffer *batch;
drm_intel_bo *scratch_bo;
int handle;
int drm_fd;
int debug_fd = 0;
const char *debug_file = "dump_debug.bin";
int debug;
int clear_waits;
int shutting_down = 0;
struct intel_debug_handshake dh;
int force_clear = 0;
uint32_t old_td_ctl;
/*
* The docs are wrong about the attention clear bits. The clear bits are
* provided as part of the structure in case they change in future generations.
*/
#define EUID(eu, td, clear) \
{ .euid = eu, .tid = td, .reg = EU_ATT, .fd = -1, .clr = clear }
#define EUID2(eu, td, clear) \
{ .euid = eu, .tid = td, .reg = EU_ATT + 4, .fd = -1, .clr = clear }
struct debuggee gt1_debug_ids[] = {
RSVD_ID, RSVD_ID,
RSVD_ID, EUID(6, 3, 28), EUID(6, 2, 27), EUID(6, 1, 26), EUID(6, 0, 25),
RSVD_ID, EUID(5, 3, 23), EUID(5, 2, 22), EUID(5, 1, 21), EUID(5, 0, 20),
RSVD_ID, EUID(4, 3, 18), EUID(4, 2, 17), EUID(4, 1, 16), EUID(4, 0, 15),
RSVD_ID, EUID(2, 3, 13), EUID(2, 2, 12), EUID(2, 1, 11), EUID(2, 0, 10),
RSVD_ID, EUID(1, 3, 8), EUID(1, 2, 7), EUID(1, 1, 6), EUID(1, 0, 5),
RSVD_ID, EUID(0, 3, 3), EUID(0, 2, 2), EUID(0, 1, 1), EUID(0, 0, 0)
};
struct debuggee gt2_debug_ids[] = {
EUID(8, 1, 31), EUID(8, 0, 30),
EUID(6, 4, 29), EUID(6, 3, 28), EUID(6, 2, 27), EUID(6, 1, 26), EUID(6, 0, 25),
EUID(5, 4, 24), EUID(5, 3, 23), EUID(5, 2, 22), EUID(5, 1, 21), EUID(5, 0, 20),
EUID(4, 4, 19), EUID(4, 3, 18), EUID(4, 2, 17), EUID(4, 1, 16), EUID(4, 0, 15),
EUID(2, 4, 14), EUID(2, 3, 13), EUID(2, 2, 12), EUID(2, 1, 11), EUID(2, 0, 10),
EUID(1, 4, 9), EUID(1, 3, 8), EUID(1, 2, 7), EUID(1, 1, 6), EUID(1, 0, 5),
EUID(0, 4, 4), EUID(0, 3, 3), EUID(0, 2, 2), EUID(0, 1, 1), EUID(0, 0, 0),
RSVD_ID, RSVD_ID, RSVD_ID, RSVD_ID,
EUID2(14, 4, 27), EUID2(14, 3, 26), EUID2(14, 2, 25), EUID2(14, 1, 24), EUID2(14, 0, 23),
EUID2(13, 4, 22), EUID2(13, 3, 21), EUID2(13, 2, 20), EUID2(13, 1, 19), EUID2(13, 0, 18),
EUID2(12, 4, 17), EUID2(12, 3, 16), EUID2(12, 2, 15), EUID2(12, 1, 14), EUID2(12, 0, 13),
EUID2(10, 4, 12), EUID2(10, 3, 11), EUID2(10, 2, 10), EUID2(10, 1, 9), EUID2(10, 0, 8),
EUID2(9, 4, 7), EUID2(9, 3, 6), EUID2(9, 2, 5), EUID2(9, 1, 4), EUID2(9, 0, 3),
EUID2(8, 4, 2), EUID2(8, 3, 1), EUID2(8, 2, 0)
};
struct debugger gt1 = {
.debuggees = gt1_debug_ids,
.num_threads = 32,
.real_num_threads = 24,
.threads_per_eu = 4
};
struct debugger gt2 = {
.debuggees = gt2_debug_ids,
.num_threads = 64,
.real_num_threads = 60,
.threads_per_eu = 5
};
static void
dump_debug(void *buf, size_t count) {
if (!debug_fd)
debug_fd = open(debug_file, O_CREAT | O_WRONLY | O_TRUNC, S_IRWXO);
write(debug_fd, buf, count);
}
static volatile void *
map_debug_buffer(void) {
int ret;
ret = drm_intel_bo_map(scratch_bo, 0);
assert(ret == 0);
return scratch_bo->virtual;
}
static void
unmap_debug_buffer(void) {
drm_intel_bo_unmap(scratch_bo);
}
static int
wait_for_attn(int timeout, int *out_bits) {
int step = 1;
int eus_waiting = 0;
int i,j;
if (timeout <= 0) {
timeout = 1;
step = 0;
}
for (i = 0; i < timeout; i += step) {
for (j = 0; j < 8; j += 4) {
uint32_t attn = intel_register_read(EU_ATT + j);
if (attn) {
int bit = 0;
while( (bit = ffs(attn)) != 0) {
bit--; // ffs is 1 based
assert(bit >= 0);
out_bits[eus_waiting] = bit + (j * 8);
attn &= ~(1 << bit);
eus_waiting++;
}
}
}
if (intel_register_read(EU_ATT + 8) ||
intel_register_read(EU_ATT + 0xc)) {
fprintf(stderr, "Unknown attention bits\n");
}
if (eus_waiting || shutting_down)
break;
}
return eus_waiting;
}
#define eu_fd(bit) eu_info->debuggees[bit].fd
#define eu_id(bit) eu_info->debuggees[bit].euid
#define eu_tid(bit) eu_info->debuggees[bit].tid
static struct eu_state *
find_eu_shmem(int bit, volatile uint8_t *buf) {
struct per_thread_data {
uint8_t ____[dh.per_thread_scratch];
}__attribute__((packed)) *data;
struct eu_state *eu;
int mem_tid, mem_euid, i;
data = (struct per_thread_data *)buf;
for(i = 0; i < eu_info->num_threads; i++) {
eu = (struct eu_state *)&data[i];
mem_tid = eu->sr0 & 0x7;
mem_euid = (eu->sr0 >> 8) & 0xf;
if (mem_tid == eu_tid(bit) && mem_euid == eu_id(bit))
break;
eu = NULL;
}
return eu;
}
#define GRF_CMP(a, b) memcmp(a, b, sizeof(grf))
#define GRF_CPY(a, b) memcpy(a, b, sizeof(grf))
static int
verify(struct eu_state *eu) {
if (GRF_CMP(eu->version, protocol_version)) {
if (debug) {
printf("Bad EU protocol version %x %x\n",
((uint32_t *)&eu->version)[0],
DEBUG_PROTOCOL_VERSION);
dump_debug((void *)eu, sizeof(*eu));
}
return -EBAD_PROTOCOL;
}
if (GRF_CMP(eu->state_magic, eu_msg)) {
if (debug) {
printf("Bad EU state magic %x %x\n",
((uint32_t *)&eu->state_magic)[0],
((uint32_t *)&eu->state_magic)[1]);
dump_debug((void *)eu, sizeof(*eu));
}
return -EBAD_MAGIC;
} else {
GRF_CPY(eu->state_magic, cpu_ack);
}
eu->sr0 = RSVD_EU << 8 | RSVD_THREAD;
return 0;
}
static int
collect_data(int bit, volatile uint8_t *buf) {
struct eu_state *eu;
ssize_t num;
int ret;
assert(eu_id(bit) != RSVD_EU);
if (eu_fd(bit) == -1) {
char name[128];
sprintf(name, "dump_eu_%02d_%d.bin", eu_id(bit), eu_tid(bit));
eu_fd(bit) = open(name, O_CREAT | O_WRONLY | O_TRUNC, S_IRWXO);
if (eu_fd(bit) == -1)
return -1;
}
eu = find_eu_shmem(bit, buf);
if (eu == NULL) {
if (debug)
printf("Bad offset %d %d\n", eu_id(bit), eu_tid(bit));
return -EBAD_SHMEM;
}
ret = verify(eu);
if (ret)
return ret;
num = write(eu_fd(bit), (void *)eu, sizeof(*eu));
if (num != sizeof(*eu)) {
perror("unhandled write failure");
return EBAD_WRITE;
}
return 0;
}
static void
clear_attn(int bit) {
#if 0
/*
* This works but doesn't allow for easily changed clearing bits
*/
static void
clear_attn_old(int bit) {
int bit_to_clear = bit % 32;
bit_to_clear = 31 - bit_to_clear;
intel_register_write(0x7830 + (bit/32) * 4, 0);
intel_register_write(0x7830 + (bit/32) * 4, 1 << bit_to_clear);
}
#else
if (!force_clear) {
int bit_to_clear;
bit_to_clear = eu_info->debuggees[bit].clr;
intel_register_write(EU_ATT_CLR + (bit/32) * 4, 0);
intel_register_write(EU_ATT_CLR + (bit/32) * 4, 1 << bit_to_clear);
} else {
intel_register_write(EU_ATT_CLR + 0, 0);
intel_register_write(EU_ATT_CLR + 4, 0);
intel_register_write(EU_ATT_CLR + 0, 0xffffffff);
intel_register_write(EU_ATT_CLR + 4, 0xffffffff);
}
#endif
}
static void
db_shutdown(int sig) {
shutting_down = 1;
printf("Shutting down...\n");
}
static void
die(int reason) {
int i = 0;
intel_register_write(EU_ATT_CLR, 0);
intel_register_write(EU_ATT_CLR + 4, 0);
if (debug_fd)
close(debug_fd);
for (i = 0; i < eu_info->num_threads; i++) {
if (eu_info->debuggees[i].fd != -1)
close(eu_info->debuggees[i].fd);
}
unmap_debug_buffer();
if (old_td_ctl)
intel_register_write(TD_CTL, old_td_ctl);
intel_register_access_fini();
exit(reason);
}
static int
identify_device(int devid) {
switch(devid) {
case PCI_CHIP_SANDYBRIDGE_GT1:
case PCI_CHIP_SANDYBRIDGE_M_GT1:
case PCI_CHIP_SANDYBRIDGE_S:
eu_info = &gt1;
break;
case PCI_CHIP_SANDYBRIDGE_GT2:
case PCI_CHIP_SANDYBRIDGE_GT2_PLUS:
case PCI_CHIP_SANDYBRIDGE_M_GT2:
case PCI_CHIP_SANDYBRIDGE_M_GT2_PLUS:
eu_info = &gt2;
break;
default:
return 1;
}
return 0;
}
static void
parse_data(const char *file_name) {
struct eu_state *eu_state = NULL;
struct stat st;
int fd = -1;
int ret, i, elements;
fd = open(file_name, O_RDONLY);
if (fd == -1) {
perror("open");
goto out;
}
ret = fstat(fd, &st);
if (ret == -1) {
perror("fstat");
goto out;
}
elements = st.st_size / sizeof(struct eu_state);
if (elements == 0) {
fprintf(stderr, "File not big enough for 1 entry\n");
goto out;
}
eu_state = mmap(0, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
if (eu_state == MAP_FAILED) {
perror("mmap");
goto out;
}
for(i = 0; i < elements; i++) {
printf("AIP: ");
printf("%x\n", ((uint32_t *)eu_state[i].cr0)[2]);
}
out:
if (eu_state)
munmap(eu_state, st.st_size);
if (fd != -1)
close(fd);
}
static int
wait_for_scratch_bo(void) {
struct sockaddr_un addr;
uint32_t version;
int fd, ret, handle = -1;
assert(sizeof(version) == sizeof(dh.version));
fd = socket(AF_UNIX, SOCK_STREAM, 0);
if (fd == -1)
return -1;
/* Clean up previous runs */
remove(SHADER_DEBUG_SOCKET);
memset(&addr, 0, sizeof(addr));
addr.sun_family = AF_UNIX;
strncpy(addr.sun_path, SHADER_DEBUG_SOCKET, sizeof(addr.sun_path) - 1);
ret = bind(fd, (const struct sockaddr *)&addr, sizeof(addr));
if (ret == -1) {
perror("listen");
return -1;
}
ret = listen(fd, 1);
if (ret == -1) {
perror("listen");
goto done;
}
while(1) {
int client_fd;
size_t count;
char ack[] = DEBUG_HANDSHAKE_ACK;
client_fd = accept(fd, NULL, NULL);
if (client_fd == -1) {
perror("accept");
goto done;
}
count = read(client_fd, &version, sizeof(version));
if (count != sizeof(version)) {
perror("read version");
goto loop_out;
}
if (version != DEBUG_HANDSHAKE_VERSION) {
fprintf(stderr, "Bad debug handshake\n");
goto loop_out;
}
count = read(client_fd, ((char *)&dh) + 1, sizeof(dh) - 1);
if (count != sizeof(dh) - 1) {
perror("read handshake");
goto loop_out;
}
count = write(client_fd, ack, sizeof(ack));
if (count != sizeof(ack)) {
perror("write ack");
goto loop_out;
}
handle = dh.flink_handle;
if (debug > 0) {
printf("Handshake completed successfully\n"
"\tprotocol version = %d\n"
"\tflink handle = %d\n"
"\tper thread scratch = %x\n", version,
dh.flink_handle, dh.per_thread_scratch);
}
loop_out:
close(client_fd);
break;
}
done:
close(fd);
return handle;
}
static void
setup_hw_bits(void)
{
intel_register_write(INST_PM, GEN6_GLOBAL_DEBUG_ENABLE |
GEN6_GLOBAL_DEBUG_ENABLE << 16);
old_td_ctl = intel_register_read(GEN6_TD_CTL);
intel_register_write(GEN6_TD_CTL, GEN6_TD_CTL_FORCE_TD_BKPT);
}
int main(int argc, char* argv[]) {
struct pci_device *pci_dev;
volatile uint8_t *scratch = NULL;
int bits[64];
int devid = -1, opt;
while ((opt = getopt(argc, argv, "cdr:pf?h")) != -1) {
switch (opt) {
case 'c':
clear_waits = 1;
break;
case 'd':
debug = 1;
break;
case 'r':
parse_data(optarg);
exit(0);
break;
case 'p':
devid = atoi(optarg);
break;
case 'f':
force_clear = 1;
break;
case '?':
case 'h':
default:
exit(0);
}
}
pci_dev = intel_get_pci_device();
if (devid != -1);
devid = pci_dev->device_id;
if (identify_device(devid)) {
abort();
}
assert(intel_register_access_init(pci_dev, 1) == 0);
memset(bits, -1, sizeof(bits));
/*
* These events have to occur before the SR runs, or we need
* non-blocking versions of the functions.
*/
if (!clear_waits) {
int handle;
drm_fd = drm_open_any();
bufmgr = drm_intel_bufmgr_gem_init(drm_fd, 4096);
setup_hw_bits();
/* We are probably root, make files world friendly */
umask(0);
handle = wait_for_scratch_bo();
if (handle == -1) {
printf("No handle from mesa, please enter manually: ");
if (fscanf(stdin, "%1d", &handle) == 0)
exit(1);
}
scratch_bo = intel_bo_gem_create_from_name(bufmgr, "scratch", handle);
if (scratch_bo == NULL) {
fprintf(stderr, "Couldn't flink buffer\n");
abort();
}
signal(SIGINT, db_shutdown);
printf("Press Ctrl-C to stop\n");
} else {
int time = force_clear ? 0 : 20000;
while (wait_for_attn(time, bits)) {
clear_attn(bits[0]);
memset(bits, -1, sizeof(bits));
}
die(0);
}
scratch = map_debug_buffer();
while (shutting_down == 0) {
int num_events, i;
memset(bits, -1, sizeof(bits));
num_events = wait_for_attn(-1, bits);
if (num_events == 0)
break;
for (i = 0; i < num_events; i++) {
assert(bits[i] < 64 && bits[i] >= 0);
if (collect_data(bits[i], scratch)) {
bits[i] = -1;
continue;
}
clear_attn(bits[i]);
}
}
die(0);
return 0;
}

View File

@ -0,0 +1,84 @@
# Copyright © 2011 Intel Corporation
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice (including the next
# paragraph) shall be included in all copies or substantial portions of the
# Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS 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.
#
# Authors:
# Ben Widawsky <ben@bwidawsk.net>
OBJCOPY?=objcopy
PRECPP=./pre_cpp.py
CPP_FLAGS=-x assembler-with-cpp -P
GEN_AS?=~/intel-gfx/intel-gen4asm/src/intel-gen4asm
GEN_AS_FLAGS?=-g6 -a -b
TEMP:=$(shell mktemp)
TEMP2:=$(shell mktemp)
GPU?=SANDYBRIDGE
DEFINES+=-DGEN_ASM -D$(GPU) -I. -I../..//lib
all: sr
sr.c: sr.g4a eviction_macro evict.h
$(PRECPP) $^ > $(TEMP)
$(CPP) $(CPP_FLAGS) $(DEFINES) -o $(TEMP2) $(TEMP)
$(GEN_AS) $(GEN_AS_FLAGS) $(TEMP2) -o $@
sr : sr.o
$(OBJCOPY) -O binary -K gen_eu_bytes $^ $@
# Test files are useful for development. The system routine kernel is very
# simple and doesn't depend on any external communication to run.
test.c: test.g4a
$(PRECPP) $^ > $(TEMP)
$(CPP) $(CPP_FLAGS) $(DEFINES) -o $(TEMP2) $(TEMP)
$(GEN_AS) $(GEN_AS_FLAGS) $(TEMP2) -o $@
test : test.o
$(OBJCOPY) -O binary -K gen_eu_bytes $^ $@
helper: sr.g4a eviction_macro
$(PRECPP) $^ > help
$(CPP) $(CPP_FLAGS) $(DEFINES) -o help2 help
$(GEN_AS) $(GEN_AS_FLAGS) help2 -o $@
eviction_macro :
evict.h : eviction_macro
$(shell ./eviction_macro >| evict.h)
.PHONY : clean
clean :
$(RM) *.o sr.c sr test.c test evict.h eviction_macro
.PHONY : clean
distclean: clean
$(RM) help*
maintainer-clean: clean
EMPTY_AUTOMAKE_TARGETS = install install-data install-exec uninstall \
install-dvi install-html install-info install-ps \
install-pdf installdirs check installcheck \
mostlyclean dvi pdf ps info html tags ctags
.PHONY: $(EMPTY_AUTOMAKE_TARGETS)
$(EMPTY_AUTOMAKE_TARGETS):

View File

@ -0,0 +1,48 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
*
*/
#include <stdio.h>
#define START 0x100
#define END ((128 << 10) / 4)
int main(int argc, char *argv[]) {
int i;
printf("#ifdef SANDYBRIDGE\n");
printf("#define EVICT_CACHE \\\n");
printf("\tmov (1) m0.5:ud g0.5:ud FLAGS; \\\n");
for (i = START; i < END - 8; i+=0x8) {
printf("\tmov (1) m0.2:ud 0x%04x:ud FLAGS; \\\n", i);
printf("\tWRITE_SCRATCH4(m0); \\\n");
}
printf("\tmov (1) m0.2:ud 0x%04x:ud FLAGS; \\\n", i);
printf("\tWRITE_SCRATCH4(m0)\n");
printf("#else\n");
printf("#define EVICT_CACHE\n");
printf("#endif\n");
}

View File

@ -0,0 +1,123 @@
#!/usr/bin/env python3
# Copyright © 2011 Intel Corporation
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice (including the next
# paragraph) shall be included in all copies or substantial portions of the
# Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS 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.
#
# Authors:
# Ben Widawsky <ben@bwidawsk.net>
#very limited C-like preprocessor
#limitations:
# no macro substitutions
# no multiline definitions
# divide operator is //
import sys,re
file = open(sys.argv[1], "r")
lines = file.readlines()
len(lines)
out = dict()
defines = dict()
count = 0
#create a dict for our output
for line in lines:
out[count] = line
count = count + 1
#done is considered #define <name> <number>
def is_done(string):
m = re.match("#define\s+(\w+?)\s+([a-fA-F0-9\-]+?)\s*$", string)
return m
#skip macros, the real cpp will handle it
def skip(string):
#macro
m = re.match("#define\s+\w+\(.+", string)
return m != None
#put contants which are done being evaluated into the dictionary
def easy_constants():
ret = 0
for lineno, string in out.items():
if skip(string):
continue
m = is_done(string)
if m != None:
key = m.group(1)
value = m.group(2)
if not key in defines:
defines[key] = int(eval(value))
ret = 1
return ret
#replace names with dictionary values
def simple_replace():
ret = 0
for lineno, string in out.items():
if skip(string):
continue
for key, value in defines.items():
if is_done(string):
continue
s = re.subn(key, repr(value), string)
if s[1] > 0:
out[lineno] = s[0]
ret = s[1]
return ret
#evaluate expressions to try to simplify them
def collapse_constants():
ret = 0
for lineno, string in out.items():
if skip(string):
continue
if is_done(string):
continue
m = re.match("#define\s+(.+?)\s+(.+)$", string)
if m != None:
try:
out[lineno] = "#define " + m.group(1) + " " + repr(eval(m.group(2)))
ret = 1
except NameError as ne:
#this happens before a variable is resolved in simple_replace
continue
except SyntaxError:
#this happens with something like #define foo bar, which the
#regular cpp can handle
continue
except:
raise KeyboardInterrupt
return ret;
while True:
ret = 0
ret += easy_constants()
ret += simple_replace()
ret += collapse_constants()
if ret == 0:
break;
for lineno, string in out.items():
print(string.rstrip())

View File

@ -0,0 +1,277 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
*
*/
#include "debug.h"
#include "evict.h"
#define CR0_0_ME_STATE_CTRL (1 << 31)
#define CR0_0_BP_SUPPRESS (1 << 15)
#define CR0_0_SPF_EN (1 << 2)
#define CR0_0_ACC_DIS (1 << 1)
#define CR0_1_BES_CTRL (1 << 31)
#define CR0_1_HALT_CTRL (1 << 30)
#define CR0_1_SOFT_EXCEPTION_CTRL (1 << 29)
#define CR0_1_ILLGL_OP_STS (1 << 28)
#define CR0_1_STACK_OVRFLW_STS (1 << 27)
#define CR0_0_ENTRY_UNMASK (CR0_0_SPF_EN | CR0_0_ACC_DIS)
// TODO: Need to fix this for non breakpoint case
#define CR0_1_ENTRY_UNMASK ~(CR0_1_BES_CTRL)
#define CR0_0_RETURN_MASK ~(CR0_0_ME_STATE_CTRL | CR0_0_SPF_EN | CR0_0_ACC_DIS)
// TODO: not sure how to make this not hardcoded
#define PER_THREAD_SCRATCH_SIZE (1 << 20)
#define PER_THREAD_QWORDS (PER_THREAD_SCRATCH_SIZE >> 4)
/* Should get this from brw_defines.h */
#define BRW_DATAPORT_OWORD_BLOCK_2_OWORDS 2
#define BRW_DATAPORT_OWORD_BLOCK_4_OWORDS 3
#define BRW_DATAPORT_OWORD_BLOCK_8_OWORDS 4
#define GEN6_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE 8
#define BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ 0
/* desc field, ie. dword3 6.3.66.2 and 2.11.2.1.4 */
#define SEND_MLEN_5 (5<<25)
#define SEND_MLEN_3 (3<<25)
#define SEND_MLEN_2 (2<<25)
#define SEND_MLEN_1 (1<<25)
#define SEND_RLEN_1 (1<<20)
#define SEND_RLEN_0 (0<<20)
#define SEND_HEADER_PRESENT (1<<19)
#define SEND_WRITE_COMMIT (1<<17)
#define SEND_TYPE_WRITE (GEN6_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE<<13)
#define SEND_TYPE_READ (BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ<<13)
#define SEND_BLOCK_SIZE1 (BRW_DATAPORT_OWORD_BLOCK_2_OWORDS<<8)
#define SEND_BLOCK_SIZE2 (BRW_DATAPORT_OWORD_BLOCK_4_OWORDS<<8)
#define SEND_BLOCK_SIZE4 (BRW_DATAPORT_OWORD_BLOCK_8_OWORDS<<8)
#define SEND_BINDING_TABLE (255<<0)
// No write commit
#define WRITE_DESC1_XXX SEND_BINDING_TABLE | SEND_BLOCK_SIZE1 | SEND_TYPE_WRITE | SEND_HEADER_PRESENT | SEND_MLEN_2
#define WRITE_DESC1_WC SEND_BINDING_TABLE | SEND_BLOCK_SIZE1 | SEND_TYPE_WRITE | SEND_HEADER_PRESENT | SEND_MLEN_2 | SEND_WRITE_COMMIT
#define WRITE_DESC2 SEND_BINDING_TABLE | SEND_BLOCK_SIZE2 | SEND_TYPE_WRITE | SEND_HEADER_PRESENT | SEND_MLEN_3
#define WRITE_DESC4 SEND_BINDING_TABLE | SEND_BLOCK_SIZE4 | SEND_TYPE_WRITE | SEND_HEADER_PRESENT | SEND_MLEN_5
#define RECV_DESC1 SEND_BINDING_TABLE | SEND_BLOCK_SIZE1 | SEND_TYPE_READ | SEND_HEADER_PRESENT | SEND_MLEN_1 | SEND_RLEN_1
//#define SEND_DESC1 0x40902FF
#define SEND_DESC1_WC 0x40b02FF
/* ex_desc field 6.3.66.2 */
#define SEND_DP_RENDER_CACHE (5<<0)
#define SEND_EOT (1<<5)
#define SEND_EX_DESC SEND_DP_RENDER_CACHE
/**
* WRITE_SCRATCH1 - Write 2 owords.
* cdst.2 - offset
* cdst.5 - per thread scratch base, relative to gsba??
* cdst+1 - data to be written.
*/
#define WRITE_SCRATCH1(cdst) \
send (16) null cdst SEND_EX_DESC WRITE_DESC1_XXX FLAGS
#define WRITE_SCRATCH1_WC(cdst) \
send (16) g1 cdst SEND_EX_DESC WRITE_DESC1_WC FLAGS
#define WRITE_SCRATCH2(cdst) \
send (16) null cdst SEND_EX_DESC WRITE_DESC2 FLAGS
#define WRITE_SCRATCH4(cdst) \
send (16) null cdst SEND_EX_DESC WRITE_DESC4 FLAGS
/**
* READ_SCRATCH1 - Read 2 owords.
* cdst.2 - offset
* cdst.5 - per thread scratch base, relative to gsba??
* grf - register where read data is populated.
*/
#define READ_SCRATCH1(grf, cdst) \
send (16) grf:ud cdst SEND_EX_DESC RECV_DESC1 FLAGS
/**
* SET_OFFSET - setup mrf for the given offset prior to a send instruction.
* mrf - message register to be used as the header.
* offset - offset.
*
* If a WRITE_SCRATCH follows, mrf+1 -> mrf+1+n should contain the data to be
* written.
*/
#define SET_OFFSET(mrf, offset) \
mov (1) mrf.5:ud g0.5:ud FLAGS; \
mov (1) mrf.2:ud offset:ud FLAGS
/**
* SAVE_CRF - save the control register
* clobbers: m0.2, m0.5
*/
#define CR_OFFSET 0x40
#define SAVE_CRF \
SET_OFFSET(m0, CR_OFFSET); \
mov (8) m1:ud 0xdeadbeef:ud FLAGS; \
mov (1) m1.0:ud cr0.0 FLAGS; \
mov (1) m1.1:ud cr0.1 FLAGS; \
mov (1) m1.2:ud cr0.2 FLAGS; \
mov (1) m1.3:ud sr0:ud FLAGS; \
WRITE_SCRATCH1(m0)
/*
* clobbers: m0.2, m0.5
*/
#define STORE_GRF(grf, offset) \
SET_OFFSET(m0, offset); \
mov (8) m1:ud grf:ud FLAGS; \
WRITE_SCRATCH1(m0)
/*
* clobbers: m0.2, m0.5
*/
#define LOAD_GRF(grf, offset) \
SET_OFFSET(m0, offset); \
READ_SCRATCH1(grf, m0)
/*
* clobbers: mrf.2 mrf.5
*/
#define STORE_MRF(mrf, offset) \
SET_OFFSET(mrf, offset); \
WRITE_SCRATCH1(mrf)
/*
* non-quirky semantics, unlike STORE_MRF
* clobbers: g1
*/
#define LOAD_MRF(mrf, offset) \
LOAD_GRF(g1, offset); \
mov (8) mrf:ud g1:ud FLAGS
#define SAVE_ALL_MRF \
/* m1 is saved already */ \
STORE_MRF(m1, 0x2); \
STORE_MRF(m2, 0x4); \
STORE_MRF(m3, 0x6); \
STORE_MRF(m4, 0x8); \
STORE_MRF(m5, 0xa); \
STORE_MRF(m6, 0xc); \
STORE_MRF(m7, 0xe); \
STORE_MRF(m8, 0x10); \
STORE_MRF(m9, 0x12); \
STORE_MRF(m10, 0x14); \
STORE_MRF(m11, 0x16); \
STORE_MRF(m12, 0x18); \
STORE_MRF(m13, 0x1a); \
STORE_MRF(m14, 0x1c)
#define RESTORE_ALL_MRF \
LOAD_MRF(m15, 0x1c); \
LOAD_MRF(m14, 0x1a); \
LOAD_MRF(m13, 0x18); \
LOAD_MRF(m12, 0x16); \
LOAD_MRF(m11, 0x14); \
LOAD_MRF(m10, 0x12); \
LOAD_MRF(m9, 0x10); \
LOAD_MRF(m8, 0xe); \
LOAD_MRF(m7, 0xc); \
LOAD_MRF(m6, 0xa); \
LOAD_MRF(m5, 0x8); \
LOAD_MRF(m4, 0x6); \
LOAD_MRF(m3, 0x4); \
LOAD_MRF(m2, 0x2); \
LOAD_MRF(m1, 0x0)
#ifndef SANDYBRIDGE
#error Only SandyBridge is supported
#endif
/* Default flags for an instruction */
#define FLAGS { ALIGN1, SWITCH, MASK_DISABLE, ACCWRCTRL}
/*
* We can clobber m0, and g0.4, everything else must be saved.
*/
Enter:
nop;
or (1) cr0.0 cr0.0 CR0_0_ENTRY_UNMASK:ud FLAGS;
/*
* g0.5 has the per thread scratch space when running in FS or VS.
* If we don't have a valid g0.5, we can calculate a per thread scratch offset
* using the system registers. The problem is we do not have a good way to know
* the offset from GSBA. The system routine will have to be hardcoded or
* dynamically patched with the correct offset.
* TID is in sr0.0[2:0]
* EUID is in sr0.0[11:8]
*/
#ifdef GPGPU
mov (1) g0.4:ud 0:ud FLAGS;
#if 0
/* This should work according to the docs, the add blows up */
shr (1) g0.8:uw sr0.0:uw 5 FLAGS;
add (1) g0.16:ub gr0.16:ub sr0.0:ub FLAGS;
#else
shr (1) g0.8:uw sr0.0:uw 5 FLAGS;
mov (1) g0.9:uw sr0.0:uw FLAGS;
and (1) g0.9:uw g0.9:uw 0x7:uw FLAGS;
add (1) g0.8:uw g0.8:uw g0.9:uw FLAGS;
mov (1) g0.9:uw 0:uw FLAGS;
mul (1) g0.4:ud g0.4:ud PER_THREAD_QWORDS FLAGS;
#endif
#endif
mov (8) m0:ud 0:ud FLAGS;
/* Saves must occur in order so as not to clobber the next register */
STORE_MRF(m0, 0);
STORE_GRF(g0, 0x20);
STORE_GRF(g1, 0x22);
SAVE_ALL_MRF;
mov (8) g1:ud STATE_EU_MSG:ud FLAGS;
STORE_GRF(g1, STATE_QWORD);
mov (8) g1:ud DEBUG_PROTOCOL_VERSION:ud FLAGS;
STORE_GRF(g1, COMMUNICATION_QWORD);
SAVE_CRF;
EVICT_CACHE;
wait n1:ud;
EVICT_CACHE;
/* Using this to try to keep coherency */
LOAD_GRF(g1, CR_OFFSET);
LOAD_GRF(g1, COMMUNICATION_QWORD);
LOAD_GRF(g1, STATE_QWORD);
RESTORE_ALL_MRF;
LOAD_GRF(g1, 0x22);
LOAD_GRF(g0, 0x20);
/* Clear breakpoint status */
and (1) cr0.1 cr0.1 CR0_1_ENTRY_UNMASK:ud FLAGS;
/* set breakpoint suppress this should be conditional on bes */
or (1) cr0.0 cr0.0 CR0_0_BP_SUPPRESS:ud FLAGS;
and (1) cr0.0 cr0.0 CR0_0_RETURN_MASK:ud FLAGS;
nop;

View File

@ -0,0 +1,64 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
*
*/
#include "debug.h"
#define CR0_0_ME_STATE_CTRL (1 << 31)
#define CR0_0_BP_SUPPRESS (1 << 15)
#define CR0_0_SPF_EN (1 << 2)
#define CR0_0_ACC_DIS (1 << 1)
#define CR0_1_BES_CTRL (1 << 31)
#define CR0_1_HALT_CTRL (1 << 30)
#define CR0_1_SOFT_EXCEPTION_CTRL (1 << 29)
#define CR0_1_ILLGL_OP_STS (1 << 28)
#define CR0_1_STACK_OVRFLW_STS (1 << 27)
#define CR0_0_ENTRY_UNMASK (CR0_0_SPF_EN | CR0_0_ACC_DIS)
// TODO: Need to fix this for non breakpoint case
#define CR0_1_ENTRY_UNMASK ~(CR0_1_BES_CTRL)
#define CR0_0_RETURN_MASK ~(CR0_0_ME_STATE_CTRL | CR0_0_SPF_EN | CR0_0_ACC_DIS)
#ifndef SANDYBRIDGE
#error Only SandyBridge is supported
#endif
/* Default flags for an instruction */
#define FLAGS { ALIGN1, SWITCH, MASK_DISABLE, ACCWRCTRL}
Enter:
nop;
or (1) cr0.0 cr0.0 CR0_0_ENTRY_UNMASK:ud FLAGS;
/* Clear breakpoint status */
and (1) cr0.1 cr0.1 CR0_1_ENTRY_UNMASK:ud FLAGS;
/* set breakpoint suppress this should be conditional on bes */
or (1) cr0.0 cr0.0 CR0_0_BP_SUPPRESS:ud FLAGS;
and (1) cr0.0 cr0.0 CR0_0_RETURN_MASK:ud FLAGS;
nop;

92
lib/debug.h Normal file
View File

@ -0,0 +1,92 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
*
*/
#ifndef _DEBUG_H_
#define _DEBUG_H_
#define DEBUG_PROTOCOL_VERSION 1
#define COMMUNICATION_OFFSET 0xc00
#define COMMUNICATION_QWORD 0xc0
#define STATE_EU_MSG 0x47534d65 /* eMSG */
#define STATE_CPU_ACK 0x4b434163 /* cACK */
#define STATE_OFFSET 0xc20
#define STATE_QWORD 0xc2
#define TX_OFFSET 0xc40
#define TX_QWORD 0xc4
#define RX_OFFSET 0xc60
#define RX_QWORD 0xc6
#ifndef GEN_ASM
typedef uint32_t grf[8];
typedef uint32_t mrf[8];
typedef uint8_t cr[12];
typedef uint32_t sr;
#define DWORD8(x) {x, x, x, x, x, x, x, x}
const static grf protocol_version = DWORD8(DEBUG_PROTOCOL_VERSION);
const static grf eu_msg = DWORD8(STATE_EU_MSG);
const static grf cpu_ack = DWORD8(STATE_CPU_ACK);
struct eu_state {
mrf m_regs[15];
grf g_regs[16];
grf pad;
/* 0x400 */
cr cr0;
sr sr0;
uint32_t beef_pad[4];
uint8_t pad2[992 + 1024];
/* 0xc00 COMMUNICATION_OFFSET */
grf version;
grf state_magic;
grf eu_tx;
grf eu_rx;
uint8_t pad3[896];
} __attribute__((packed));
static inline void
print_reg(uint8_t reg[32]) {
uint32_t *dwords = (uint32_t *)reg;
printf("%08x %08x %08x %08x %08x %08x %08x %08x",
dwords[7], dwords[6], dwords[5], dwords[4],
dwords[3], dwords[2], dwords[1], dwords[0]);
}
static inline void
print_creg(uint8_t reg[12]) {
uint32_t *dwords = (uint32_t *)reg;
printf("%08x %08x %08x", dwords[2], dwords[1], dwords[0]);
}
#endif
#endif