void *Pool_malloc(Pool *self, size_t size)
#endif
{
BlockHeader *p;
BlockHeader *s;
#ifdef POOL_DEBUG
size_t i;
size_t alloc_block_size = ALIGN_UP(HEADER_SIZE + size + WALL_SIZE * 2);
#else
size_t alloc_block_size = ALIGN_UP(HEADER_SIZE + size);
#endif
if (self->init_flag != self) {
DEBUGLOG(("Pool_malloc: pool is not initialized: %s(%d)\n", file, line));
assert(0);
return 0;
}
#ifdef POOL_DEBUG
if (!size) {
DEBUGLOG(("Pool_malloc: alloc 0 bytes: %s(%d)\n", file, line));
}
if (self->fail_count >= 0) {
if (self->fail_count > 0) {
self->fail_count--;
} else {
DEBUGLOG(("Pool_malloc: return NULL by fail_count: %s(%d)\n", file, line));
return 0;
}
}
#endif
for (p = self->loop_p->next; p != self->loop_p; p = p->next) {
if (!p->occupied && p->size >= alloc_block_size) {
if (p->size >= alloc_block_size + HEADER_SIZE + SPLIT_MIN) {
s = (BlockHeader *) ((char *) p + alloc_block_size);
s->size = p->size - alloc_block_size;
s->occupied = 0;
s->magic = MAGIC_NO;
p->size = alloc_block_size;
s->next = p->next;
#ifndef POOL_SLIST
s->prev = p;
p->next->prev = s;
#endif
p->next = s;
}
p->occupied = 1;
#ifdef POOL_DEBUG
p->file = file;
p->line = line;
p->alloc_size = size;
for (i = 0; i < WALL_SIZE; i++) {
((char *) p)[HEADER_SIZE + i] = WALL_CHAR;
((char *) p)[HEADER_SIZE + WALL_SIZE + size + i] = WALL_CHAR;
}
#endif
self->loop_p = p;
#ifdef POOL_DEBUG
return (char *) p + HEADER_SIZE + WALL_SIZE;
#else
return (char *) p + HEADER_SIZE;
#endif
}
}
DEBUGLOG(("Pool_malloc: return NULL: %s(%d)\n", file, line));
return 0;
}
void PackArmPe::rebuildImports(upx_byte *& extrainfo)
{
if (ODADDR(PEDIR_IMPORT) == 0
|| ODSIZE(PEDIR_IMPORT) <= sizeof(import_desc))
return;
// const upx_byte * const idata = obuf + get_le32(extrainfo);
OPTR_C(const upx_byte, idata, obuf + get_le32(extrainfo));
const unsigned inamespos = get_le32(extrainfo + 4);
extrainfo += 8;
unsigned sdllnames = 0;
// const upx_byte *import = ibuf + IDADDR(PEDIR_IMPORT) - isection[2].vaddr;
// const upx_byte *p;
IPTR_I(const upx_byte, import, ibuf + IDADDR(PEDIR_IMPORT) - isection[2].vaddr);
OPTR(const upx_byte, p);
for (p = idata; get_le32(p) != 0; ++p)
{
const upx_byte *dname = get_le32(p) + import;
ICHECK(dname, 1);
const unsigned dlen = strlen(dname);
ICHECK(dname, dlen + 1);
sdllnames += dlen + 1;
for (p += 8; *p;)
if (*p == 1)
p += strlen(++p) + 1;
else if (*p == 0xff)
p += 3; // ordinal
else
p += 5;
}
sdllnames = ALIGN_UP(sdllnames,2u);
upx_byte * const Obuf = obuf - rvamin;
import_desc * const im0 = (import_desc*) (Obuf + ODADDR(PEDIR_IMPORT));
import_desc *im = im0;
upx_byte *dllnames = Obuf + inamespos;
upx_byte *importednames = dllnames + sdllnames;
upx_byte * const importednames_start = importednames;
for (p = idata; get_le32(p) != 0; ++p)
{
// restore the name of the dll
const upx_byte *dname = get_le32(p) + import;
ICHECK(dname, 1);
const unsigned dlen = strlen(dname);
ICHECK(dname, dlen + 1);
const unsigned iatoffs = get_le32(p + 4) + rvamin;
if (inamespos)
{
// now I rebuild the dll names
OCHECK(dllnames, dlen + 1);
strcpy(dllnames, dname);
im->dllname = ptr_diff(dllnames,Obuf);
//;;;printf("\ndll: %s:",dllnames);
dllnames += dlen + 1;
}
else
{
OCHECK(Obuf + im->dllname, dlen + 1);
strcpy(Obuf + im->dllname, dname);
}
im->oft = im->iat = iatoffs;
// LE32 *newiat = (LE32 *) (Obuf + iatoffs);
OPTR_I(LE32, newiat, (LE32 *) (Obuf + iatoffs));
// restore the imported names+ordinals
for (p += 8; *p; ++newiat)
if (*p == 1)
{
const unsigned ilen = strlen(++p) + 1;
if (inamespos)
{
if (ptr_diff(importednames, importednames_start) & 1)
importednames -= 1;
omemcpy(importednames + 2, p, ilen);
//;;;printf(" %s",importednames+2);
*newiat = ptr_diff(importednames, Obuf);
importednames += 2 + ilen;
}
else
{
OCHECK(Obuf + *newiat + 2, ilen + 1);
strcpy(Obuf + *newiat + 2, p);
}
p += ilen;
}
else if (*p == 0xff)
{
*newiat = get_le16(p + 1) + 0x80000000;
//;;;printf(" %x",(unsigned)*newiat);
p += 3;
}
else
{
*newiat = get_le32(get_le32(p + 1) + import);
assert(*newiat & 0x80000000);
p += 5;
}
*newiat = 0;
im++;
}
//memset(idata,0,p - idata);
}
void bootstrap(void)
{
version_print();
ofw_memmap(&bootinfo.memmap);
void *bootinfo_pa = ofw_translate(&bootinfo);
void *real_mode_pa = ofw_translate(&real_mode);
void *loader_address_pa = ofw_translate((void *) LOADER_ADDRESS);
printf("\nMemory statistics (total %llu MB)\n", bootinfo.memmap.total >> 20);
printf(" %p|%p: real mode trampoline\n", &real_mode, real_mode_pa);
printf(" %p|%p: boot info structure\n", &bootinfo, bootinfo_pa);
printf(" %p|%p: kernel entry point\n",
(void *) PA2KA(BOOT_OFFSET), (void *) BOOT_OFFSET);
printf(" %p|%p: loader entry point\n",
(void *) LOADER_ADDRESS, loader_address_pa);
size_t i;
for (i = 0; i < COMPONENTS; i++)
printf(" %p|%p: %s image (%zu/%zu bytes)\n", components[i].start,
ofw_translate(components[i].start), components[i].name,
components[i].inflated, components[i].size);
size_t dest[COMPONENTS];
size_t top = 0;
size_t cnt = 0;
bootinfo.taskmap.cnt = 0;
for (i = 0; i < min(COMPONENTS, TASKMAP_MAX_RECORDS); i++) {
top = ALIGN_UP(top, PAGE_SIZE);
if (i > 0) {
bootinfo.taskmap.tasks[bootinfo.taskmap.cnt].addr =
(void *) PA2KA(top);
bootinfo.taskmap.tasks[bootinfo.taskmap.cnt].size =
components[i].inflated;
str_cpy(bootinfo.taskmap.tasks[bootinfo.taskmap.cnt].name,
BOOTINFO_TASK_NAME_BUFLEN, components[i].name);
bootinfo.taskmap.cnt++;
}
dest[i] = top;
top += components[i].inflated;
cnt++;
}
void *balloc_base;
void *balloc_base_pa;
ofw_alloc("boot allocator area", &balloc_base, &balloc_base_pa,
BALLOC_MAX_SIZE, loader_address_pa);
printf(" %p|%p: boot allocator area\n", balloc_base, balloc_base_pa);
void *inflate_base;
void *inflate_base_pa;
ofw_alloc("inflate area", &inflate_base, &inflate_base_pa, top,
loader_address_pa);
printf(" %p|%p: inflate area\n", inflate_base, inflate_base_pa);
uintptr_t balloc_start = ALIGN_UP(top, PAGE_SIZE);
size_t pages = (balloc_start + ALIGN_UP(BALLOC_MAX_SIZE, PAGE_SIZE))
>> PAGE_WIDTH;
void *transtable;
void *transtable_pa;
ofw_alloc("translate table", &transtable, &transtable_pa,
pages * sizeof(void *), loader_address_pa);
printf(" %p|%p: translate table\n", transtable, transtable_pa);
check_overlap("boot allocator area", balloc_base_pa, pages);
check_overlap("inflate area", inflate_base_pa, pages);
check_overlap("translate table", transtable_pa, pages);
printf("\nInflating components ... ");
for (i = cnt; i > 0; i--) {
printf("%s ", components[i - 1].name);
int err = inflate(components[i - 1].start, components[i - 1].size,
inflate_base + dest[i - 1], components[i - 1].inflated);
if (err != EOK) {
printf("\n%s: Inflating error %d, halting.\n",
components[i - 1].name, err);
halt();
}
}
printf(".\n");
printf("Setting up boot allocator ...\n");
balloc_init(&bootinfo.ballocs, balloc_base, PA2KA(balloc_start),
BALLOC_MAX_SIZE);
printf("Setting up screens ...\n");
ofw_setup_screens();
printf("Canonizing OpenFirmware device tree ...\n");
bootinfo.ofw_root = ofw_tree_build();
printf("Setting up translate table ...\n");
for (i = 0; i < pages; i++) {
uintptr_t off = i << PAGE_WIDTH;
void *phys;
if (off < balloc_start)
phys = ofw_translate(inflate_base + off);
else
phys = ofw_translate(balloc_base + off - balloc_start);
((void **) transtable)[i] = phys;
}
printf("Booting the kernel...\n");
jump_to_kernel(bootinfo_pa, transtable_pa, pages, real_mode_pa);
}
static ssize_t relocate_fvh(uintptr_t new_addr, void *fsp, size_t fsp_size,
size_t fvh_offset, size_t *fih_offset)
{
EFI_FIRMWARE_VOLUME_HEADER *fvh;
EFI_FFS_FILE_HEADER *ffsfh;
EFI_COMMON_SECTION_HEADER *csh;
size_t offset;
size_t file_offset;
size_t size;
size_t fv_length;
offset = fvh_offset;
fvh = relative_offset(fsp, offset);
if (read_le32(&fvh->Signature) != EFI_FVH_SIGNATURE)
return -1;
fv_length = read_le64(&fvh->FvLength);
printk(FSP_DBG_LVL, "FVH length: %zx Offset: %zx Mapping length: %zx\n",
fv_length, offset, fsp_size);
if (fv_length + offset > fsp_size)
return -1;
/* Parse only this FV. However, the algorithm uses offsets into the
* entire FSP region so make size include the starting offset. */
size = fv_length + offset;
if (guid_compare(&fvh->FileSystemGuid, &ffs2_guid)) {
printk(BIOS_ERR, "FVH not an FFS2 type.\n");
return -1;
}
if (read_le16(&fvh->ExtHeaderOffset) != 0) {
EFI_FIRMWARE_VOLUME_EXT_HEADER *fveh;
offset += read_le16(&fvh->ExtHeaderOffset);
fveh = relative_offset(fsp, offset);
printk(FSP_DBG_LVL, "Extended Header Offset: %zx Size: %zx\n",
(size_t)read_le16(&fvh->ExtHeaderOffset),
(size_t)read_le32(&fveh->ExtHeaderSize));
offset += read_le32(&fveh->ExtHeaderSize);
/* FFS files are 8 byte aligned after extended header. */
offset = ALIGN_UP(offset, 8);
} else {
offset += read_le16(&fvh->HeaderLength);
}
file_offset = offset;
while (file_offset + sizeof(*ffsfh) < size) {
offset = file_offset;
printk(FSP_DBG_LVL, "file offset: %zx\n", file_offset);
/* First file and section should be FSP info header. */
if (fih_offset != NULL && *fih_offset == 0)
*fih_offset = file_offset;
ffsfh = relative_offset(fsp, file_offset);
printk(FSP_DBG_LVL, "file type = %x\n", read_le8(&ffsfh->Type));
printk(FSP_DBG_LVL, "file attribs = %x\n",
read_le8(&ffsfh->Attributes));
/* Exit FV relocation when empty space found */
if (read_le8(&ffsfh->Type) == EFI_FV_FILETYPE_FFS_MAX)
break;
/* Next file on 8 byte alignment. */
file_offset += ffs_file_size(ffsfh);
file_offset = ALIGN_UP(file_offset, 8);
/* Padding files have no section information. */
if (read_le8(&ffsfh->Type) == EFI_FV_FILETYPE_FFS_PAD)
continue;
offset += file_section_offset(ffsfh);
while (offset + sizeof(*csh) < file_offset) {
size_t data_size;
size_t data_offset;
csh = relative_offset(fsp, offset);
printk(FSP_DBG_LVL, "section offset: %zx\n", offset);
printk(FSP_DBG_LVL, "section type: %x\n",
read_le8(&csh->Type));
data_size = section_data_size(csh);
data_offset = section_data_offset(csh);
if (data_size + data_offset + offset > file_offset) {
printk(BIOS_ERR, "Section exceeds FV size.\n");
return -1;
}
/*
* The entire FSP 1.1 image can be thought of as one
* program with a single link address even though there
* are multiple TEs linked separately. The reason is
* that each TE is linked for XIP. So in order to
* relocate the TE properly we need to form the
* relocated address based on the TE offset within
* FSP proper.
*/
if (read_le8(&csh->Type) == EFI_SECTION_TE) {
void *te;
size_t te_offset = offset + data_offset;
uintptr_t te_addr = new_addr + te_offset;
printk(FSP_DBG_LVL, "TE image at offset %zx\n",
te_offset);
te = relative_offset(fsp, te_offset);
te_relocate(te_addr, te);
}
offset += data_size + data_offset;
/* Sections are aligned to 4 bytes. */
offset = ALIGN_UP(offset, 4);
}
}
/* Return amount of buffer parsed: FV size. */
return fv_length;
}
static int __INIT__ flash_parse_part(struct flash_chip *host,
struct part_attr *part, const char *part_def)
{
int i, ret = -EINVAL, index = 0;
__u32 curr_base = 0;
char buff[128];
const char *p;
p = strchr(part_def, ':');
if (!p)
goto error;
p++;
while (*p && *p != ';') {
if (curr_base >= host->chip_size)
goto error;
while (' ' == *p) p++;
// part size
if (*p == '-') {
part->size = host->chip_size - curr_base;
p++;
} else {
for (i = 0; *p; i++, p++) {
if (*p == '@' || *p == '(' || *p == 'r' || *p == ',')
break;
buff[i] = *p;
}
buff[i] = '\0';
ret = hr_str_to_val(buff, (unsigned long *)&part->size);
if (ret < 0)
goto error;
ALIGN_UP(part->size, host->erase_size);
}
// part base
if (*p == '@') {
for (i = 0, p++; *p; i++, p++) {
if (*p == '(' || *p == 'r' || *p == ',')
break;
buff[i] = *p;
}
buff[i] = '\0';
ret = hr_str_to_val(buff, (unsigned long *)&part->base);
if (ret < 0)
goto error;
ALIGN_UP(part->base, host->erase_size);
curr_base = part->base;
} else {
part->base = curr_base;
}
// part label and image
i = 0;
if (*p == '(') {
for (p++; *p && *p != ')'; i++, p++)
part->label[i] = *p;
p++;
}
if (*p == 'r') {
p++;
if (*p != 'o') {
return -EINVAL;
}
part->flags |= BDF_RDONLY;
p++;
}
if (',' == *p)
p++;
part->label[i] = '\0';
curr_base += part->size;
index++;
part++;
}
return index;
error:
printf("%s(): invalid part definition \"%s\"\n", __func__, part_def);
return ret;
}
grub_err_t
SUFFIX (grub_freebsd_load_elfmodule_obj) (grub_file_t file, int argc,
char *argv[], grub_addr_t *kern_end)
{
Elf_Ehdr e;
Elf_Shdr *s;
char *shdr;
grub_addr_t curload, module;
grub_err_t err;
err = read_headers (file, &e, &shdr);
if (err)
return err;
curload = module = ALIGN_PAGE (*kern_end);
for (s = (Elf_Shdr *) shdr; s < (Elf_Shdr *) ((char *) shdr
+ e.e_shnum * e.e_shentsize);
s = (Elf_Shdr *) ((char *) s + e.e_shentsize))
{
if (s->sh_size == 0)
continue;
if (s->sh_addralign)
curload = ALIGN_UP (curload, s->sh_addralign);
s->sh_addr = curload;
grub_dprintf ("bsd", "loading section to %x, size %d, align %d\n",
(unsigned) curload, (int) s->sh_size,
(int) s->sh_addralign);
switch (s->sh_type)
{
default:
case SHT_PROGBITS:
err = load (file, UINT_TO_PTR (curload), s->sh_offset, s->sh_size);
if (err)
return err;
break;
case SHT_NOBITS:
if (curload + s->sh_size > grub_os_area_addr + grub_os_area_size)
return grub_error (GRUB_ERR_OUT_OF_RANGE,
"not enough memory for the module");
grub_memset (UINT_TO_PTR (curload), 0, s->sh_size);
break;
}
curload += s->sh_size;
}
*kern_end = ALIGN_PAGE (curload);
err = grub_freebsd_add_meta_module (argv[0], FREEBSD_MODTYPE_ELF_MODULE_OBJ,
argc - 1, argv + 1, module,
curload - module);
if (! err)
err = grub_freebsd_add_meta (FREEBSD_MODINFO_METADATA
| FREEBSD_MODINFOMD_ELFHDR,
&e, sizeof (e));
if (! err)
err = grub_freebsd_add_meta (FREEBSD_MODINFO_METADATA
| FREEBSD_MODINFOMD_SHDR,
shdr, e.e_shnum * e.e_shentsize);
return err;
}
static u32 alloc_mem(u32 bytes)
{
return alloc -= ALIGN_UP(bytes, 16);
}
static grub_err_t
grub_cbfs_find_file (struct grub_archelp_data *data, char **name,
grub_int32_t *mtime,
grub_uint32_t *mode)
{
grub_size_t offset;
for (;;
data->dofs = data->hofs + offset,
data->next_hofs = ALIGN_UP (data->dofs + data->size, data->cbfs_align))
{
struct cbfs_file hd;
grub_size_t namesize;
data->hofs = data->next_hofs;
if (data->hofs >= data->cbfs_end)
{
*mode = GRUB_ARCHELP_ATTR_END;
return GRUB_ERR_NONE;
}
if (grub_disk_read (data->disk, 0, data->hofs, sizeof (hd), &hd))
return grub_errno;
if (grub_memcmp (hd.magic, CBFS_FILE_MAGIC, sizeof (hd.magic)) != 0)
{
*mode = GRUB_ARCHELP_ATTR_END;
return GRUB_ERR_NONE;
}
data->size = grub_be_to_cpu32 (hd.len);
(void) mtime;
offset = grub_be_to_cpu32 (hd.offset);
*mode = GRUB_ARCHELP_ATTR_FILE | GRUB_ARCHELP_ATTR_NOTIME;
namesize = offset;
if (namesize >= sizeof (hd))
namesize -= sizeof (hd);
if (namesize == 0)
continue;
*name = grub_malloc (namesize + 1);
if (*name == NULL)
return grub_errno;
if (grub_disk_read (data->disk, 0, data->hofs + sizeof (hd),
namesize, *name))
{
grub_free (*name);
return grub_errno;
}
if ((*name)[0] == '\0')
{
grub_free (*name);
*name = NULL;
continue;
}
(*name)[namesize] = 0;
data->dofs = data->hofs + offset;
data->next_hofs = ALIGN_UP (data->dofs + data->size, data->cbfs_align);
return GRUB_ERR_NONE;
}
}
static int
load_of(uval64 rmo_start, uval64 rmo_size)
{
uval32 stub_addr = rmo_size - (1 << 20);
char mem_node[64];
uval64 prop[2];
int ret;
char *dum = NULL;
retry:
snprintf(mem_node, sizeof(mem_node), "/memory@0x%x", 0);
prop[0] = 0;
prop[1] = stub_addr;
of_set_prop(mem_node, "available", prop, sizeof (prop));
char oftree[512];
char ofstub[512];
char ofdimg[512];
snprintf(oftree, sizeof(oftree), HYPE_ROOT "/%s/of_tree", oh_pname);
snprintf(ofstub, sizeof(ofstub), HYPE_ROOT "/%s/of_image", oh_pname);
snprintf(ofdimg, sizeof(ofdimg), HYPE_ROOT "/%s/of_tree.ofd", oh_pname);
const char *const args[] = { "ofdfs", "pack", oftree, ofdimg, NULL };
run_command(args);
uval32 stub_size;
struct stat buf;
ret = stat(ofstub, &buf);
ASSERT(ret == 0, "Stat failure %s\n", ofstub);
stub_size = buf.st_size + 4 * PGSIZE;
uval32 data_size;
ret = stat(ofdimg, &buf);
ASSERT(ret == 0, "Stat failure %s\n", ofdimg);
data_size = buf.st_size;
uval32 bottom = ALIGN_DOWN(rmo_size - data_size - stub_size, 1 << 20);
if (stub_addr > bottom) {
stub_addr = bottom;
goto retry;
}
uval32 data_addr = ALIGN_UP(stub_addr + stub_size, PGSIZE);
int fd = open(ofstub, O_RDWR);
ASSERT(fd >= 0, "bad fd for %s\n", ofstub);
off_t offset = lseek(fd, 0x10, SEEK_SET);
ASSERT(offset != ((off_t) - 1), "seek failure on %s\n", ofstub);
write(fd, &data_addr, sizeof (data_addr));
close(fd);
msg("Loading OF stub: 0x%x[0x%x]\n", stub_addr, stub_size);
msg("Loading OF data: 0x%x[0x%x]\n", data_addr, data_size);
laddr_load(ofstub, rmo_start + stub_addr, &dum);
laddr_load(ofdimg, rmo_start + data_addr, &dum);
set_file_printf("r5", "0x%lx", stub_addr);
set_file_printf("r1", "0x%lx", stub_addr - PGSIZE);
return 0;
}
int
main(int argc, char **argv)
{
char scratch[128];
hcall_fd = hcall_init();
int ret = parse_args(argc, argv);
if (ret < 0)
return ret;
uval total = 0;
ASSERT(num_ranges > 0, "No memory ranges specified\n");
ret = get_file("state", scratch, sizeof(scratch));
if (ret <= 0 || strncmp(scratch, "READY", ret) != 0) {
bailout("Partition not ready\n");
}
uval rmo_start = lranges[0].lr_base;
uval rmo_size = lranges[0].lr_size;
uval laddr = 0;
uval img_base = 0;
uval img_offset = 0;
int count = 0;
while (count < 255) {
char image[256];
char image_laddr[256];
char data[64];
struct stat sbuf;
uval base;
snprintf(image, sizeof(image), HYPE_ROOT "/%s/image%02x",
oh_pname, count);
snprintf(image_laddr, sizeof(image_laddr), "image%02x_load", count);
++count;
ret = stat(image, &sbuf);
if (ret < 0)
continue;
ret = get_file(image_laddr, data, sizeof(data));
if (ret >= 0) {
uval64 l = strtoull(data, NULL, 0);
ASSERT(errno != ERANGE, "Corrupted data %s\n",
image_laddr);
laddr = l;
}
int size = laddr_load(image, rmo_start + laddr, &base);
if (0 == img_base) {
img_base = base;
img_offset = laddr;
}
laddr = ALIGN_UP(laddr + size, PGSIZE);
}
char pinfo_buf[64];
uval64 pinfo;
ret = get_file("pinfo", pinfo_buf, sizeof(pinfo_buf));
if (ret <= 0 || (pinfo = strtoull(pinfo_buf, NULL, 0)) <= 0) {
pinfo = (uval64)-1;
}
uval lpid;
hargs.opcode = H_CREATE_PARTITION;
hargs.args[0] = rmo_start;
hargs.args[1] = rmo_size;
hargs.args[2] = pinfo;
ret = hcall(&hargs);
ASSERT(ret >= 0 && hargs.retval == 0,
"hcall failure: %d " UVAL_CHOOSE("0x%x", "0x%lx") "\n", ret,
hargs.retval);
lpid = hargs.args[0];
set_file_printf("of_tree/ibm,partition-no", "0x%llx", lpid);
set_file_printf("lpid", "0x%llx", lpid);
set_file("state", "CREATED", -1);
clean_vdevices();
of_add_memory(0, rmo_size);
total += rmo_size;
int i = 1;
while (i < num_ranges) {
uval64 base = lranges[i].lr_base;
uval64 size = lranges[i].lr_size;
add_memory(lpid, base, size);
total += size;
++i;
}
#ifdef __PPC__
add_htab(lpid, total);
#endif
if (get_file_numeric("res_console_srv", &console_ua) < 0) {
console_ua = 0;
} else {
vtys = 1;
}
uval64 vty0 = 0;
if (vtys == 0 && console_ua == 0) {
add_vty(0);
} else {
if (vtys > 0) {
/* add vty and force it to be vty@0 */
vty0 = add_vterm(0, lpid);
--vtys;
}
while (vtys > 0) {
/* add vty and let it be the proper liobn */
add_vterm(-1, lpid);
}
}
if (crq > 0) {
ret = add_vdev(lpid, crq);
if (ret == -1) {
fprintf(stderr, "vdev failed : 0x%lx\n", crq);
return 1;
}
}
if (console_ua) {
printf("Registering console vterm: "
UVAL_CHOOSE("0x%lx","0x%llx")" -> 0x%lx:0x%llx\n",
console_ua, lpid, vty0);
hargs.opcode = H_REGISTER_VTERM;
hargs.args[0] = console_ua;
hargs.args[1] = lpid;
hargs.args[2] = vty0;
ret = hcall(&hargs);
ASSERT(ret >= 0 && hargs.retval == 0,
"hcall failure: %d " UVAL_CHOOSE("0x%x", "0x%lx") "\n",
ret, hargs.retval);
}
add_llan(lpid);
#ifdef __PPC__
load_of(lranges[0].lr_base, lranges[0].lr_size);
#endif
#ifdef __i386__
if (0 != img_base) {
static struct partition_info part_info[2];
fill_pinfo(&part_info[1], lpid, rmo_size);
if (inject_pinfo(part_info, sizeof(part_info),
img_base, img_offset)) {
add_cmdlineargs(img_base,
0x1000, /* offset in image for data */
lpid);
}
}
#endif
hargs.opcode = H_SET_SCHED_PARAMS;
hargs.args[0] = lpid;
hargs.args[1] = 0;
hargs.args[2] = 1;
hargs.args[3] = 0;
ret = hcall(&hargs);
ASSERT(ret >= 0 && hargs.retval == 0,
"hcall failure: %d " UVAL_CHOOSE("0x%x", "0x%lx") "\n", ret,
hargs.retval);
hargs.opcode = H_START;
hargs.args[0] = lpid;
char buf[64];
ret = get_file("pc", buf, sizeof(buf));
if (ret >= 0) {
buf[ret] = 0;
hargs.args[1] = strtoull(buf, NULL, 0);
} else {
hargs.args[1] = 0;
}
int x = 2;
while (x < 8) {
int y = 0;
snprintf(buf, sizeof(buf), "r%d", x);
y = get_file(buf, buf, sizeof(buf));
if (y >= 0) {
buf[y] = 0;
hargs.args[x] = strtoull(buf, NULL, 0);
} else {
hargs.args[x] = 0;
}
++x;
}
if (wait_for_key) {
printf("waiting for keypress:");
(void)fgetc(stdin);
}
printf("Starting...\n");
hcall(&hargs);
ASSERT(ret >= 0 && hargs.retval == 0,
"hcall failure: %d " UVAL_CHOOSE("0x%x", "0x%lx") "\n", ret,
hargs.retval);
set_file("state", "RUNNING", -1);
return 0;
}
/* static */ SysStatusUval
SysVSharedMem::MakeRegion(sval32 key, uval size, uval flags)
{
SysStatus rc = 0;
uval addr;
ObjectHandle frOH;
RegionInfo *ri;
uval perm;
uval accMode;
if (flags & SHM_HUGETLB) {
rc = StubFRComputation::_CreateLargePage(frOH, LARGE_PAGES_SIZE);
size = ALIGN_UP(size, LARGE_PAGES_SIZE);
} else {
rc = StubFRComputation::_Create(frOH);
}
if (_FAILURE(rc)) return rc;
perm = (flags & (S_IRWXU|S_IRWXG|S_IRWXO));
if (!(perm & (S_IWUSR | S_IRUSR))) {
tassertWrn(0,"Region perms are neither read or write.\n");
perm |= S_IWUSR;
}
// FIXME: this is wrong
if (perm & S_IWUSR) {
accMode = AccessMode::writeUserWriteSup;
} else {
accMode = AccessMode::readUserReadSup;
}
// FIXME: The only reason to bind the region in the servers
// address space is to "reserve" at least one region somewhere
// since the shamt cannot fail due to absence of memory resources.
/*
* Note: We cannot assume that all processes have the reion mapped
* at the same address, so keeping track of this by our address
* value is stupid.
*/
rc = StubRegionDefault::_CreateFixedLenExt(
addr, size, 0, frOH, 0, accMode, 0, RegionType::K42Region);
if (_FAILURE(rc)) {
Obj::ReleaseAccess(frOH);
return rc;
}
ri = new RegionInfo;
if (ri == NULL) {
Obj::ReleaseAccess(frOH);
rc = DREFGOBJ(TheProcessRef)->regionDestroy(addr);
tassert(_SUCCESS(rc), err_printf("region destroy @ 0x%lx: failed.\n",
addr));
return _SERROR(2520, 0, ENOMEM);
}
ri->frOH = frOH;
ri->addr = addr;
// FIXME: need some sort of reuse policy!!
ri->shmID = FetchAndAddSynced(&(obj->shmID), 1);
tassert(ri->shmID > 0,
err_printf("Blew the max shmget(2) id values\n"));
ri->shmInfo.shm_perm.__key = key;
ri->shmInfo.shm_perm.uid = 0; // uid of owner
ri->shmInfo.shm_perm.gid = 0; // gid of owner
ri->shmInfo.shm_perm.cuid = 0; // uid of creator
ri->shmInfo.shm_perm.cgid = 0; // gid of creator
ri->shmInfo.shm_perm.mode = perm; // perms
ri->shmInfo.shm_perm.__seq = ri->shmID; // FIXME: I THINK
ri->shmInfo.shm_segsz = size; // size of segment (bytes)
ri->shmInfo.shm_atime = 0; // last attach time
ri->shmInfo.shm_dtime = 0; // last dettach time
// Should be time now
ri->shmInfo.shm_ctime = 2; // last change time by shmctl()
ri->shmInfo.shm_cpid = 1; // pid of creator
ri->shmInfo.shm_lpid = 1; // pid of last shm operation
ri->shmInfo.shm_nattch = 0; // number of current attaches
AutoLock<BLock> al(&obj->lock);
if(key != IPC_PRIVATE) {
obj->regionsByKey.add(key, ri);
}
obj->regionsByShmID.add(ri->shmID, ri);
obj->regionsByAddr.add(addr, ri);
return _SRETUVAL(ri->shmID);
}
mspace create_contiguous_mspace_with_base(size_t starting_capacity,
size_t max_capacity, int locked, void *base) {
struct mspace_contig_state *cs;
unsigned int pagesize;
mstate m;
init_mparams();
pagesize = PAGESIZE;
assert(starting_capacity <= max_capacity);
assert(((uintptr_t)base & (pagesize-1)) == 0);
assert(((uintptr_t)max_capacity & (pagesize-1)) == 0);
starting_capacity = (size_t)ALIGN_UP(starting_capacity, pagesize);
/* Make the first page read/write. dlmalloc needs to use that page.
*/
if (mprotect(base, starting_capacity, PROT_READ | PROT_WRITE) < 0) {
goto error;
}
/* Create the mspace, pointing to the memory given.
*/
m = create_mspace_with_base((char *)base + sizeof(*cs), starting_capacity,
locked);
if (m == (mspace)0) {
goto error;
}
/* Make sure that m is in the same page as base.
*/
assert(((uintptr_t)m & (uintptr_t)~(pagesize-1)) == (uintptr_t)base);
/* Use some space for the information that our MORECORE needs.
*/
cs = (struct mspace_contig_state *)base;
/* Find out exactly how much of the memory the mspace
* is using.
*/
cs->brk = m->seg.base + m->seg.size;
cs->top = (char *)base + max_capacity;
assert((char *)base <= cs->brk);
assert(cs->brk <= cs->top);
/* Prevent access to the memory we haven't handed out yet.
*/
if (cs->brk != cs->top) {
/* mprotect() requires page-aligned arguments, but it's possible
* for cs->brk not to be page-aligned at this point.
*/
char *prot_brk = (char *)ALIGN_UP(cs->brk, pagesize);
if ((mprotect(base, prot_brk - (char *)base, PROT_READ | PROT_WRITE) < 0) ||
(mprotect(prot_brk, cs->top - prot_brk, PROT_NONE) < 0)) {
goto error;
}
}
cs->m = m;
cs->magic = CONTIG_STATE_MAGIC;
return (mspace)m;
error:
return (mspace)0;
}
void *Pool_realloc(Pool *self, void *ptr, size_t newsize)
#endif
{
size_t i;
BlockHeader *p;
BlockHeader *s;
size_t new_alloc_block_size;
if (self->init_flag != self) {
DEBUGLOG(("Pool_realloc: pool is not initialized: %s(%d)\n", file, line));
assert(0);
return 0;
}
if (!ptr) {
#ifdef POOL_DEBUG
return Pool_malloc_debug(self, newsize, file, line);
#else
return Pool_malloc(self, newsize);
#endif
}
if (!newsize) {
#ifdef POOL_DEBUG
Pool_free_debug(self, ptr, file, line);
#else
Pool_free(self, ptr);
#endif
return 0;
}
#ifdef POOL_DEBUG
p = (BlockHeader *) ((char *) ptr - (HEADER_SIZE + WALL_SIZE));
#else
p = (BlockHeader *) ((char *) ptr - HEADER_SIZE);
#endif
if (p->magic != MAGIC_NO || !p->occupied) {
DEBUGLOG(("Pool_realloc: NG pointer: %s(%d)\n", file, line));
assert(0);
return 0;
}
#ifdef POOL_DEBUG
new_alloc_block_size = ALIGN_UP(HEADER_SIZE + newsize + WALL_SIZE * 2);
#else
new_alloc_block_size = ALIGN_UP(HEADER_SIZE + newsize);
#endif
if (p->size >= new_alloc_block_size) {
if (p->size >= 2 * new_alloc_block_size && new_alloc_block_size >= HEADER_SIZE + SPLIT_MIN) {
/* 縮んだ分を空き領域として回収 */
s = (BlockHeader *) ((char *) p + new_alloc_block_size);
s->occupied = 0;
s->magic = MAGIC_NO;
#ifndef POOL_SLIST
s->prev = p;
#endif
if (p->next->occupied) {
s->size = p->size - new_alloc_block_size;
s->next = p->next;
#ifndef POOL_SLIST
p->next->prev = s;
#endif
} else {
if (self->loop_p == p->next) {
self->loop_p = p->next->next;
}
s->size = p->size - new_alloc_block_size + p->next->size;
s->next = p->next->next;
#ifndef POOL_SLIST
p->next->next->prev = s;
#endif
}
p->size = new_alloc_block_size;
p->next = s;
}
#ifdef POOL_DEBUG
p->alloc_size = newsize;
for (i = 0; i < WALL_SIZE; i++) {
((char *) ptr)[newsize + i] = WALL_CHAR;
}
#endif
return ptr;
}
/* 以下、拡張 */
if (p->next->occupied || p->size + p->next->size < new_alloc_block_size) {
#ifdef POOL_DEBUG
void *newptr = Pool_malloc_debug(self, newsize, file, line);
#else
void *newptr = Pool_malloc(self, newsize);
#endif
if (!newptr) {
DEBUGLOG(("Pool_realloc: return NULL: %s(%d)\n", file, line));
return 0;
}
for (i = 0;
#ifdef POOL_DEBUG
i < p->alloc_size;
#else
i < p->size - HEADER_SIZE;
#endif
i++) {
((char *) newptr)[i] = ((char *) ptr)[i];
}
#ifdef POOL_DEBUG
Pool_free_debug(self, ptr, file, line);
#else
Pool_free(self, ptr);
#endif
return newptr;
}
unsigned PackArmPe::processImports() // pass 1
{
static const unsigned char kernel32dll[] = "COREDLL.dll";
static const char llgpa[] = "\x0\x0""LoadLibraryW\x0\x0"
"GetProcAddressA\x0\x0\x0"
"CacheSync";
//static const char exitp[] = "ExitProcess\x0\x0\x0";
unsigned dllnum = 0;
import_desc *im = (import_desc*) (ibuf + IDADDR(PEDIR_IMPORT));
import_desc * const im_save = im;
if (IDADDR(PEDIR_IMPORT))
{
while (im->dllname)
dllnum++, im++;
im = im_save;
}
struct udll
{
const upx_byte *name;
const upx_byte *shname;
unsigned ordinal;
unsigned iat;
LE32 *lookupt;
unsigned npos;
unsigned original_position;
bool isk32;
static int __acc_cdecl_qsort compare(const void *p1, const void *p2)
{
const udll *u1 = * (const udll * const *) p1;
const udll *u2 = * (const udll * const *) p2;
if (u1->isk32) return -1;
if (u2->isk32) return 1;
if (!*u1->lookupt) return 1;
if (!*u2->lookupt) return -1;
int rc = strcasecmp(u1->name,u2->name);
if (rc) return rc;
if (u1->ordinal) return -1;
if (u2->ordinal) return 1;
if (!u1->shname) return 1;
if (!u2->shname) return -1;
return strlen(u1->shname) - strlen(u2->shname);
}
};
// +1 for dllnum=0
Array(struct udll, dlls, dllnum+1);
Array(struct udll *, idlls, dllnum+1);
soimport = 1024; // safety
unsigned ic,k32o;
for (ic = k32o = 0; dllnum && im->dllname; ic++, im++)
{
idlls[ic] = dlls + ic;
dlls[ic].name = ibuf + im->dllname;
dlls[ic].shname = NULL;
dlls[ic].ordinal = 0;
dlls[ic].iat = im->iat;
dlls[ic].lookupt = (LE32*) (ibuf + (im->oft ? im->oft : im->iat));
dlls[ic].npos = 0;
dlls[ic].original_position = ic;
dlls[ic].isk32 = strcasecmp(kernel32dll,dlls[ic].name) == 0;
soimport += strlen(dlls[ic].name) + 1 + 4;
for (IPTR_I(LE32, tarr, dlls[ic].lookupt); *tarr; tarr += 1)
{
if (*tarr & 0x80000000)
{
importbyordinal = true;
soimport += 2; // ordinal num: 2 bytes
dlls[ic].ordinal = *tarr & 0xffff;
//if (dlls[ic].isk32)
// kernel32ordinal = true,k32o++;
}
else
{
IPTR_I(const upx_byte, n, ibuf + *tarr + 2);
unsigned len = strlen(n);
soimport += len + 1;
if (dlls[ic].shname == NULL || len < strlen (dlls[ic].shname))
dlls[ic].shname = n;
}
soimport++; // separator
}
}
oimport = new upx_byte[soimport];
memset(oimport,0,soimport);
oimpdlls = new upx_byte[soimport];
memset(oimpdlls,0,soimport);
qsort(idlls,dllnum,sizeof (udll*),udll::compare);
unsigned dllnamelen = sizeof (kernel32dll);
unsigned dllnum2 = 1;
for (ic = 0; ic < dllnum; ic++)
if (!idlls[ic]->isk32 && (ic == 0 || strcasecmp(idlls[ic - 1]->name,idlls[ic]->name)))
{
dllnum2++;
dllnamelen += strlen(idlls[ic]->name) + 1;
}
//fprintf(stderr,"dllnum=%d dllnum2=%d soimport=%d\n",dllnum,dllnum2,soimport); //
info("Processing imports: %d DLLs", dllnum);
// create the new import table
im = (import_desc*) oimpdlls;
LE32 *ordinals = (LE32*) (oimpdlls + (dllnum2 + 1) * sizeof(import_desc));
LE32 *lookuptable = ordinals + 4;// + k32o + (isdll ? 0 : 1);
upx_byte *dllnames = ((upx_byte*) lookuptable) + (dllnum2 - 1) * 8;
upx_byte *importednames = dllnames + (dllnamelen &~ 1);
unsigned k32namepos = ptr_diff(dllnames,oimpdlls);
memcpy(importednames, llgpa, ALIGN_UP((unsigned) sizeof(llgpa), 2u));
strcpy(dllnames,kernel32dll);
im->dllname = k32namepos;
im->iat = ptr_diff(ordinals,oimpdlls);
*ordinals++ = ptr_diff(importednames,oimpdlls); // LoadLibraryW
*ordinals++ = ptr_diff(importednames,oimpdlls) + 14; // GetProcAddressA
*ordinals++ = ptr_diff(importednames,oimpdlls) + 14 + 18; // CacheSync
dllnames += sizeof(kernel32dll);
importednames += sizeof(llgpa);
im++;
for (ic = 0; ic < dllnum; ic++)
if (idlls[ic]->isk32)
{
idlls[ic]->npos = k32namepos;
/*
if (idlls[ic]->ordinal)
for (LE32 *tarr = idlls[ic]->lookupt; *tarr; tarr++)
if (*tarr & 0x80000000)
*ordinals++ = *tarr;
*/
}
else if (ic && strcasecmp(idlls[ic-1]->name,idlls[ic]->name) == 0)
idlls[ic]->npos = idlls[ic-1]->npos;
else
{
im->dllname = idlls[ic]->npos = ptr_diff(dllnames,oimpdlls);
im->iat = ptr_diff(lookuptable,oimpdlls);
strcpy(dllnames,idlls[ic]->name);
dllnames += strlen(idlls[ic]->name)+1;
if (idlls[ic]->ordinal)
*lookuptable = idlls[ic]->ordinal + 0x80000000;
else if (idlls[ic]->shname)
{
if (ptr_diff(importednames,oimpdlls) & 1)
importednames--;
*lookuptable = ptr_diff(importednames,oimpdlls);
importednames += 2;
strcpy(importednames,idlls[ic]->shname);
importednames += strlen(idlls[ic]->shname) + 1;
}
lookuptable += 2;
im++;
}
soimpdlls = ALIGN_UP(ptr_diff(importednames,oimpdlls),4);
Interval names(ibuf),iats(ibuf),lookups(ibuf);
// create the preprocessed data
//ordinals -= k32o;
upx_byte *ppi = oimport; // preprocessed imports
for (ic = 0; ic < dllnum; ic++)
{
LE32 *tarr = idlls[ic]->lookupt;
#if 0 && ENABLE_THIS_AND_UNCOMPRESSION_WILL_BREAK
if (!*tarr) // no imports from this dll
continue;
#endif
set_le32(ppi,idlls[ic]->npos);
set_le32(ppi+4,idlls[ic]->iat - rvamin);
ppi += 8;
for (; *tarr; tarr++)
if (*tarr & 0x80000000)
{
/*if (idlls[ic]->isk32)
{
*ppi++ = 0xfe; // signed + odd parity
set_le32(ppi,ptr_diff(ordinals,oimpdlls));
ordinals++;
ppi += 4;
}
else*/
{
*ppi++ = 0xff;
set_le16(ppi,*tarr & 0xffff);
ppi += 2;
}
}
else
{
*ppi++ = 1;
unsigned len = strlen(ibuf + *tarr + 2) + 1;
memcpy(ppi,ibuf + *tarr + 2,len);
ppi += len;
names.add(*tarr,len + 2 + 1);
}
ppi++;
unsigned esize = ptr_diff((char *)tarr, (char *)idlls[ic]->lookupt);
lookups.add(idlls[ic]->lookupt,esize);
if (ptr_diff(ibuf + idlls[ic]->iat, (char *)idlls[ic]->lookupt))
{
memcpy(ibuf + idlls[ic]->iat, idlls[ic]->lookupt, esize);
iats.add(idlls[ic]->iat,esize);
}
names.add(idlls[ic]->name,strlen(idlls[ic]->name) + 1 + 1);
}
ppi += 4;
assert(ppi < oimport+soimport);
soimport = ptr_diff(ppi,oimport);
if (soimport == 4)
soimport = 0;
//OutputFile::dump("x0.imp", oimport, soimport);
//OutputFile::dump("x1.imp", oimpdlls, soimpdlls);
unsigned ilen = 0;
names.flatten();
if (names.ivnum > 1)
{
// The area occupied by the dll and imported names is not continuous
// so to still support uncompression, I can't zero the iat area.
// This decreases compression ratio, so FIXME somehow.
infoWarning("can't remove unneeded imports");
ilen += sizeof(import_desc) * dllnum;
#if defined(DEBUG)
if (opt->verbose > 3)
names.dump();
#endif
// do some work for the unpacker
im = im_save;
for (ic = 0; ic < dllnum; ic++, im++)
{
memset(im,FILLVAL,sizeof(*im));
im->dllname = ptr_diff(dlls[idlls[ic]->original_position].name,ibuf);
}
}
else
{
iats.add(im_save,sizeof(import_desc) * dllnum);
// zero unneeded data
iats.clear();
lookups.clear();
}
names.clear();
iats.add(&names);
iats.add(&lookups);
iats.flatten();
for (ic = 0; ic < iats.ivnum; ic++)
ilen += iats.ivarr[ic].len;
info("Imports: original size: %u bytes, preprocessed size: %u bytes",ilen,soimport);
return names.ivnum == 1 ? names.ivarr[0].start : 0;
}
static void* reserve_code(struct jit* jit, lua_State* L, size_t sz)
{
struct page* page;
size_t off = (jit->pagenum > 0) ? jit->pages[jit->pagenum-1]->off : 0;
size_t size = (jit->pagenum > 0) ? jit->pages[jit->pagenum-1]->size : 0;
if (off + sz >= size) {
int i;
uint8_t* pdata;
cfunction func;
/* need to create a new page */
jit->pages = (struct page**) realloc(jit->pages, (++jit->pagenum) * sizeof(jit->pages[0]));
size = ALIGN_UP(sz + LINKTABLE_MAX_SIZE + sizeof(struct page), jit->align_page_size);
page = (struct page*) AllocPage(size);
jit->pages[jit->pagenum-1] = page;
pdata = (uint8_t*) page;
page->size = size;
page->off = sizeof(struct page);
lua_newtable(L);
#define ADDFUNC(DLL, NAME) \
lua_pushliteral(L, #NAME); \
func = DLL ? (cfunction) GetProcAddressA(DLL, #NAME) : NULL; \
func = func ? func : (cfunction) &NAME; \
lua_pushcfunction(L, (lua_CFunction) func); \
lua_rawset(L, -3)
ADDFUNC(NULL, check_double);
ADDFUNC(NULL, check_float);
ADDFUNC(NULL, check_uint64);
ADDFUNC(NULL, check_int64);
ADDFUNC(NULL, check_int32);
ADDFUNC(NULL, check_uint32);
ADDFUNC(NULL, check_uintptr);
ADDFUNC(NULL, check_enum);
ADDFUNC(NULL, check_typed_pointer);
ADDFUNC(NULL, check_typed_cfunction);
ADDFUNC(NULL, check_complex_double);
ADDFUNC(NULL, check_complex_float);
ADDFUNC(NULL, unpack_varargs_stack);
ADDFUNC(NULL, unpack_varargs_stack_skip);
ADDFUNC(NULL, unpack_varargs_reg);
ADDFUNC(NULL, unpack_varargs_float);
ADDFUNC(NULL, unpack_varargs_int);
ADDFUNC(NULL, push_cdata);
ADDFUNC(NULL, push_int);
ADDFUNC(NULL, push_uint);
ADDFUNC(NULL, push_float);
ADDFUNC(jit->kernel32_dll, SetLastError);
ADDFUNC(jit->kernel32_dll, GetLastError);
ADDFUNC(jit->lua_dll, luaL_error);
ADDFUNC(jit->lua_dll, lua_pushnumber);
ADDFUNC(jit->lua_dll, lua_pushboolean);
ADDFUNC(jit->lua_dll, lua_gettop);
ADDFUNC(jit->lua_dll, lua_rawgeti);
ADDFUNC(jit->lua_dll, lua_pushnil);
ADDFUNC(jit->lua_dll, lua_callk);
ADDFUNC(jit->lua_dll, lua_settop);
lua_pushliteral(L, "lua_remove");
lua_pushcfunction(L, (lua_CFunction) lua_removef);
lua_rawset(L, -3);
#undef ADDFUNC
for (i = 0; extnames[i] != NULL; i++) {
if (strcmp(extnames[i], "FUNCTION") == 0) {
shred(pdata + page->off, 0, JUMP_SIZE);
jit->function_extern = i;
} else {
lua_getfield(L, -1, extnames[i]);
func = (cfunction) lua_tocfunction(L, -1);
if (func == NULL) {
luaL_error(L, "internal error: missing link for %s", extnames[i]);
}
compile_extern_jump(jit, L, func, pdata + page->off);
lua_pop(L, 1);
}
page->off += JUMP_SIZE;
}
page->freed = page->off;
lua_pop(L, 1);
} else {
page = jit->pages[jit->pagenum-1];
EnableWrite(page, page->size);
}
return (uint8_t*) page + page->off;
}
void PackArmPe::pack(OutputFile *fo)
{
// FIXME: we need to think about better support for --exact
if (opt->exact)
throwCantPackExact();
const unsigned objs = ih.objects;
isection = new pe_section_t[objs];
fi->seek(pe_offset+sizeof(ih),SEEK_SET);
fi->readx(isection,sizeof(pe_section_t)*objs);
rvamin = isection[0].vaddr;
infoHeader("[Processing %s, format %s, %d sections]", fn_basename(fi->getName()), getName(), objs);
// check the PE header
// FIXME: add more checks
if (!opt->force && (
(ih.cpu != 0x1c0 && ih.cpu != 0x1c2)
|| (ih.opthdrsize != 0xe0)
|| ((ih.flags & EXECUTABLE) == 0)
|| (ih.subsystem != 9)
|| (ih.entry == 0 /*&& !isdll*/)
|| (ih.ddirsentries != 16)
// || IDSIZE(PEDIR_EXCEPTION) // is this used on arm?
// || IDSIZE(PEDIR_COPYRIGHT)
))
throwCantPack("unexpected value in PE header (try --force)");
if (IDSIZE(PEDIR_SEC))
IDSIZE(PEDIR_SEC) = IDADDR(PEDIR_SEC) = 0;
// throwCantPack("compressing certificate info is not supported");
if (IDSIZE(PEDIR_COMRT))
throwCantPack(".NET files (win32/net) are not yet supported");
if (isdll)
opt->win32_pe.strip_relocs = false;
else if (opt->win32_pe.strip_relocs < 0)
opt->win32_pe.strip_relocs = (ih.imagebase >= 0x10000);
if (opt->win32_pe.strip_relocs)
{
if (ih.imagebase < 0x10000)
throwCantPack("--strip-relocs is not allowed when imagebase < 0x10000");
else
ih.flags |= RELOCS_STRIPPED;
}
if (memcmp(isection[0].name,"UPX",3) == 0)
throwAlreadyPackedByUPX();
if (!opt->force && IDSIZE(15))
throwCantPack("file is possibly packed/protected (try --force)");
if (ih.entry && ih.entry < rvamin)
throwCantPack("run a virus scanner on this file!");
if (!opt->force && ih.subsystem == 1)
throwCantPack("subsystem 'native' is not supported (try --force)");
if (ih.filealign < 0x200)
throwCantPack("filealign < 0x200 is not yet supported");
handleStub(fi,fo,pe_offset);
const unsigned usize = ih.imagesize;
const unsigned xtrasize = UPX_MAX(ih.datasize, 65536u) + IDSIZE(PEDIR_IMPORT) + IDSIZE(PEDIR_BOUNDIM) + IDSIZE(PEDIR_IAT) + IDSIZE(PEDIR_DELAYIMP) + IDSIZE(PEDIR_RELOC);
ibuf.alloc(usize + xtrasize);
// BOUND IMPORT support. FIXME: is this ok?
fi->seek(0,SEEK_SET);
fi->readx(ibuf,isection[0].rawdataptr);
Interval holes(ibuf);
unsigned ic,jc,overlaystart = 0;
ibuf.clear(0, usize);
for (ic = jc = 0; ic < objs; ic++)
{
if (isection[ic].rawdataptr && overlaystart < isection[ic].rawdataptr + isection[ic].size)
overlaystart = ALIGN_UP(isection[ic].rawdataptr + isection[ic].size,ih.filealign);
if (isection[ic].vsize == 0)
isection[ic].vsize = isection[ic].size;
if ((isection[ic].flags & PEFL_BSS) || isection[ic].rawdataptr == 0
|| (isection[ic].flags & PEFL_INFO))
{
holes.add(isection[ic].vaddr,isection[ic].vsize);
continue;
}
if (isection[ic].vaddr + isection[ic].size > usize)
throwCantPack("section size problem");
if (((isection[ic].flags & (PEFL_WRITE|PEFL_SHARED))
== (PEFL_WRITE|PEFL_SHARED)))
if (!opt->force)
throwCantPack("writable shared sections not supported (try --force)");
if (jc && isection[ic].rawdataptr - jc > ih.filealign)
throwCantPack("superfluous data between sections");
fi->seek(isection[ic].rawdataptr,SEEK_SET);
jc = isection[ic].size;
if (jc > isection[ic].vsize)
jc = isection[ic].vsize;
if (isection[ic].vsize == 0) // hack for some tricky programs - may this break other progs?
jc = isection[ic].vsize = isection[ic].size;
if (isection[ic].vaddr + jc > ibuf.getSize())
throwInternalError("buffer too small 1");
fi->readx(ibuf + isection[ic].vaddr,jc);
jc += isection[ic].rawdataptr;
}
// check for NeoLite
if (find(ibuf + ih.entry, 64+7, "NeoLite", 7) >= 0)
throwCantPack("file is already compressed with another packer");
unsigned overlay = file_size - stripDebug(overlaystart);
if (overlay >= (unsigned) file_size)
{
#if 0
if (overlay < file_size + ih.filealign)
overlay = 0;
else if (!opt->force)
throwNotCompressible("overlay problem (try --force)");
#endif
overlay = 0;
}
checkOverlay(overlay);
Resource res;
Interval tlsiv(ibuf);
Export xport((char*)(unsigned char*)ibuf);
const unsigned dllstrings = processImports();
processTls(&tlsiv); // call before processRelocs!!
processResources(&res);
processExports(&xport);
processRelocs();
//OutputFile::dump("x1", ibuf, usize);
// some checks for broken linkers - disable filter if necessary
bool allow_filter = true;
if (ih.codebase == ih.database
|| ih.codebase + ih.codesize > ih.imagesize
|| (isection[virta2objnum(ih.codebase,isection,objs)].flags & PEFL_CODE) == 0)
allow_filter = false;
const unsigned oam1 = ih.objectalign - 1;
// FIXME: disabled: the uncompressor would not allocate enough memory
//objs = tryremove(IDADDR(PEDIR_RELOC),objs);
// FIXME: if the last object has a bss then this won't work
// newvsize = (isection[objs-1].vaddr + isection[objs-1].size + oam1) &~ oam1;
// temporary solution:
unsigned newvsize = (isection[objs-1].vaddr + isection[objs-1].vsize + oam1) &~ oam1;
//fprintf(stderr,"newvsize=%x objs=%d\n",newvsize,objs);
if (newvsize + soimport + sorelocs > ibuf.getSize())
throwInternalError("buffer too small 2");
memcpy(ibuf+newvsize,oimport,soimport);
memcpy(ibuf+newvsize+soimport,orelocs,sorelocs);
cimports = newvsize - rvamin; // rva of preprocessed imports
crelocs = cimports + soimport; // rva of preprocessed fixups
ph.u_len = newvsize + soimport + sorelocs;
// some extra data for uncompression support
unsigned s = 0;
upx_byte * const p1 = ibuf + ph.u_len;
memcpy(p1 + s,&ih,sizeof (ih));
s += sizeof (ih);
memcpy(p1 + s,isection,ih.objects * sizeof(*isection));
s += ih.objects * sizeof(*isection);
if (soimport)
{
set_le32(p1 + s,cimports);
set_le32(p1 + s + 4,dllstrings);
s += 8;
}
if (sorelocs)
{
set_le32(p1 + s,crelocs);
p1[s + 4] = (unsigned char) (big_relocs & 6);
s += 5;
}
if (soresources)
{
set_le16(p1 + s,icondir_count);
s += 2;
}
// end of extra data
set_le32(p1 + s,ptr_diff(p1,ibuf) - rvamin);
s += 4;
ph.u_len += s;
obuf.allocForCompression(ph.u_len);
// prepare packheader
ph.u_len -= rvamin;
// prepare filter
Filter ft(ph.level);
ft.buf_len = ih.codesize;
ft.addvalue = ih.codebase - rvamin;
// compress
int filter_strategy = allow_filter ? 0 : -3;
// disable filters for files with broken headers
if (ih.codebase + ih.codesize > ph.u_len)
{
ft.buf_len = 1;
filter_strategy = -3;
}
// limit stack size needed for runtime decompression
upx_compress_config_t cconf; cconf.reset();
cconf.conf_lzma.max_num_probs = 1846 + (768 << 4); // ushort: ~28 KiB stack
compressWithFilters(&ft, 2048, &cconf, filter_strategy,
ih.codebase, rvamin, 0, NULL, 0);
// info: see buildLoader()
newvsize = (ph.u_len + rvamin + ph.overlap_overhead + oam1) &~ oam1;
/*
if (tlsindex && ((newvsize - ph.c_len - 1024 + oam1) &~ oam1) > tlsindex + 4)
tlsindex = 0;
*/
const unsigned lsize = getLoaderSize();
int identsize = 0;
const unsigned codesize = getLoaderSection("IDENTSTR",&identsize);
assert(identsize > 0);
getLoaderSection("UPX1HEAD",(int*)&ic);
identsize += ic;
pe_section_t osection[4];
// section 0 : bss
// 1 : [ident + header] + packed_data + unpacker + tls
// 2 : not compressed data
// 3 : resource data -- wince 5 needs a new section for this
// identsplit - number of ident + (upx header) bytes to put into the PE header
int identsplit = pe_offset + sizeof(osection) + sizeof(oh);
if ((identsplit & 0x1ff) == 0)
identsplit = 0;
else if (((identsplit + identsize) ^ identsplit) < 0x200)
identsplit = identsize;
else
identsplit = ALIGN_GAP(identsplit, 0x200);
ic = identsize - identsplit;
const unsigned c_len = ((ph.c_len + ic) & 15) == 0 ? ph.c_len : ph.c_len + 16 - ((ph.c_len + ic) & 15);
obuf.clear(ph.c_len, c_len - ph.c_len);
const unsigned s1size = ALIGN_UP(ic + c_len + codesize,4u) + sotls;
const unsigned s1addr = (newvsize - (ic + c_len) + oam1) &~ oam1;
const unsigned ncsection = (s1addr + s1size + oam1) &~ oam1;
const unsigned upxsection = s1addr + ic + c_len;
Reloc rel(1024); // new relocations are put here
static const char* symbols_to_relocate[] = {
"ONAM", "BIMP", "BREL", "FIBE", "FIBS", "ENTR", "DST0", "SRC0"
};
for (unsigned s2r = 0; s2r < TABLESIZE(symbols_to_relocate); s2r++)
{
unsigned off = linker->getSymbolOffset(symbols_to_relocate[s2r]);
if (off != 0xdeaddead)
rel.add(off + upxsection, 3);
}
// new PE header
memcpy(&oh,&ih,sizeof(oh));
oh.filealign = 0x200; // identsplit depends on this
memset(osection,0,sizeof(osection));
oh.entry = upxsection;
oh.objects = 4;
oh.chksum = 0;
// fill the data directory
ODADDR(PEDIR_DEBUG) = 0;
ODSIZE(PEDIR_DEBUG) = 0;
ODADDR(PEDIR_IAT) = 0;
ODSIZE(PEDIR_IAT) = 0;
ODADDR(PEDIR_BOUNDIM) = 0;
ODSIZE(PEDIR_BOUNDIM) = 0;
// tls is put into section 1
ic = s1addr + s1size - sotls;
super::processTls(&rel,&tlsiv,ic);
ODADDR(PEDIR_TLS) = sotls ? ic : 0;
ODSIZE(PEDIR_TLS) = sotls ? 0x18 : 0;
ic += sotls;
// these are put into section 2
ic = ncsection;
// wince wants relocation data at the beginning of a section
processRelocs(&rel);
ODADDR(PEDIR_RELOC) = soxrelocs ? ic : 0;
ODSIZE(PEDIR_RELOC) = soxrelocs;
ic += soxrelocs;
processImports(ic, linker->getSymbolOffset("IATT") + upxsection);
ODADDR(PEDIR_IMPORT) = ic;
ODSIZE(PEDIR_IMPORT) = soimpdlls;
ic += soimpdlls;
processExports(&xport,ic);
ODADDR(PEDIR_EXPORT) = soexport ? ic : 0;
ODSIZE(PEDIR_EXPORT) = soexport;
if (!isdll && opt->win32_pe.compress_exports)
{
ODADDR(PEDIR_EXPORT) = IDADDR(PEDIR_EXPORT);
ODSIZE(PEDIR_EXPORT) = IDSIZE(PEDIR_EXPORT);
}
ic += soexport;
ic = (ic + oam1) &~ oam1;
const unsigned res_start = ic;
if (soresources)
processResources(&res,ic);
ODADDR(PEDIR_RESOURCE) = soresources ? ic : 0;
ODSIZE(PEDIR_RESOURCE) = soresources;
ic += soresources;
const unsigned onam = ncsection + soxrelocs + ih.imagebase;
linker->defineSymbol("start_of_dll_names", onam);
linker->defineSymbol("start_of_imports", ih.imagebase + rvamin + cimports);
linker->defineSymbol("start_of_relocs", crelocs + rvamin + ih.imagebase);
linker->defineSymbol("filter_buffer_end", ih.imagebase + ih.codebase + ih.codesize);
linker->defineSymbol("filter_buffer_start", ih.imagebase + ih.codebase);
linker->defineSymbol("original_entry", ih.entry + ih.imagebase);
linker->defineSymbol("uncompressed_length", ph.u_len);
linker->defineSymbol("start_of_uncompressed", ih.imagebase + rvamin);
linker->defineSymbol("compressed_length", ph.c_len);
linker->defineSymbol("start_of_compressed", ih.imagebase + s1addr + identsize - identsplit);
defineDecompressorSymbols();
relocateLoader();
MemBuffer loader(lsize);
memcpy(loader, getLoader(), lsize);
patchPackHeader(loader, lsize);
// this is computed here, because soxrelocs changes some lines above
const unsigned ncsize = soxrelocs + soimpdlls + soexport;
const unsigned fam1 = oh.filealign - 1;
// fill the sections
strcpy(osection[0].name,"UPX0");
strcpy(osection[1].name,"UPX1");
strcpy(osection[2].name, "UPX2");
strcpy(osection[3].name, ".rsrc");
osection[0].vaddr = rvamin;
osection[1].vaddr = s1addr;
osection[2].vaddr = ncsection;
osection[3].vaddr = res_start;
osection[0].size = 0;
osection[1].size = (s1size + fam1) &~ fam1;
osection[2].size = (ncsize + fam1) &~ fam1;
osection[3].size = (soresources + fam1) &~ fam1;
osection[0].vsize = osection[1].vaddr - osection[0].vaddr;
//osection[1].vsize = (osection[1].size + oam1) &~ oam1;
//osection[2].vsize = (osection[2].size + oam1) &~ oam1;
osection[1].vsize = osection[1].size;
osection[2].vsize = osection[2].size;
osection[3].vsize = osection[3].size;
osection[0].rawdataptr = 0;
osection[1].rawdataptr = (pe_offset + sizeof(oh) + sizeof(osection) + fam1) &~ fam1;
osection[2].rawdataptr = osection[1].rawdataptr + osection[1].size;
osection[3].rawdataptr = osection[2].rawdataptr + osection[2].size;
osection[0].flags = (unsigned) (PEFL_BSS|PEFL_EXEC|PEFL_WRITE|PEFL_READ);
osection[1].flags = (unsigned) (PEFL_DATA|PEFL_EXEC|PEFL_WRITE|PEFL_READ);
osection[2].flags = (unsigned) (PEFL_DATA|PEFL_READ);
osection[3].flags = (unsigned) (PEFL_DATA|PEFL_READ);
oh.imagesize = (osection[3].vaddr + osection[3].vsize + oam1) &~ oam1;
oh.bsssize = osection[0].vsize;
oh.datasize = osection[2].vsize + osection[3].vsize;
oh.database = osection[2].vaddr;
oh.codesize = osection[1].vsize;
oh.codebase = osection[1].vaddr;
oh.headersize = osection[1].rawdataptr;
if (rvamin < osection[0].rawdataptr)
throwCantPack("object alignment too small");
if (opt->win32_pe.strip_relocs && !isdll)
oh.flags |= RELOCS_STRIPPED;
//for (ic = 0; ic < oh.filealign; ic += 4)
// set_le32(ibuf + ic,get_le32("UPX "));
ibuf.clear(0, oh.filealign);
info("Image size change: %u -> %u KiB",
ih.imagesize / 1024, oh.imagesize / 1024);
infoHeader("[Writing compressed file]");
if (soresources == 0)
{
oh.objects = 3;
memset(&osection[3], 0, sizeof(osection[3]));
}
// write loader + compressed file
fo->write(&oh,sizeof(oh));
fo->write(osection,sizeof(osection));
// some alignment
if (identsplit == identsize)
{
unsigned n = osection[1].rawdataptr - fo->getBytesWritten() - identsize;
assert(n <= oh.filealign);
fo->write(ibuf, n);
}
fo->write(loader + codesize,identsize);
infoWriting("loader", fo->getBytesWritten());
fo->write(obuf,c_len);
infoWriting("compressed data", c_len);
fo->write(loader,codesize);
if (opt->debug.dump_stub_loader)
OutputFile::dump(opt->debug.dump_stub_loader, loader, codesize);
if ((ic = fo->getBytesWritten() & 3) != 0)
fo->write(ibuf,4 - ic);
fo->write(otls,sotls);
if ((ic = fo->getBytesWritten() & fam1) != 0)
fo->write(ibuf,oh.filealign - ic);
fo->write(oxrelocs,soxrelocs);
fo->write(oimpdlls,soimpdlls);
fo->write(oexport,soexport);
if ((ic = fo->getBytesWritten() & fam1) != 0)
fo->write(ibuf,oh.filealign - ic);
fo->write(oresources,soresources);
if ((ic = fo->getBytesWritten() & fam1) != 0)
fo->write(ibuf,oh.filealign - ic);
#if 0
printf("%-13s: program hdr : %8ld bytes\n", getName(), (long) sizeof(oh));
printf("%-13s: sections : %8ld bytes\n", getName(), (long) sizeof(osection));
printf("%-13s: ident : %8ld bytes\n", getName(), (long) identsize);
printf("%-13s: compressed : %8ld bytes\n", getName(), (long) c_len);
printf("%-13s: decompressor : %8ld bytes\n", getName(), (long) codesize);
printf("%-13s: tls : %8ld bytes\n", getName(), (long) sotls);
printf("%-13s: resources : %8ld bytes\n", getName(), (long) soresources);
printf("%-13s: imports : %8ld bytes\n", getName(), (long) soimpdlls);
printf("%-13s: exports : %8ld bytes\n", getName(), (long) soexport);
printf("%-13s: relocs : %8ld bytes\n", getName(), (long) soxrelocs);
#endif
// verify
verifyOverlappingDecompression();
// copy the overlay
copyOverlay(fo, overlay, &obuf);
// finally check the compression ratio
if (!checkFinalCompressionRatio(fo))
throwNotCompressible();
}
/*
* Provide a variant that takes just the vtable for small fixed-size objects.
* The aligned size is already computed and stored in vt->gc_descr.
* Note: every SGEN_SCAN_START_SIZE or so we are given the chance to do some special
* processing. We can keep track of where objects start, for example,
* so when we scan the thread stacks for pinned objects, we can start
* a search for the pinned object in SGEN_SCAN_START_SIZE chunks.
*/
GCObject*
sgen_alloc_obj_nolock (GCVTable vtable, size_t size)
{
/* FIXME: handle OOM */
void **p;
char *new_next;
size_t real_size = size;
TLAB_ACCESS_INIT;
CANARIFY_SIZE(size);
HEAVY_STAT (++stat_objects_alloced);
if (real_size <= SGEN_MAX_SMALL_OBJ_SIZE)
HEAVY_STAT (stat_bytes_alloced += size);
else
HEAVY_STAT (stat_bytes_alloced_los += size);
size = ALIGN_UP (size);
SGEN_ASSERT (6, sgen_vtable_get_descriptor (vtable), "VTable without descriptor");
if (G_UNLIKELY (has_per_allocation_action)) {
static int alloc_count;
int current_alloc = InterlockedIncrement (&alloc_count);
if (collect_before_allocs) {
if (((current_alloc % collect_before_allocs) == 0) && nursery_section) {
sgen_perform_collection (0, GENERATION_NURSERY, "collect-before-alloc-triggered", TRUE);
if (!degraded_mode && sgen_can_alloc_size (size) && real_size <= SGEN_MAX_SMALL_OBJ_SIZE) {
// FIXME:
g_assert_not_reached ();
}
}
} else if (verify_before_allocs) {
if ((current_alloc % verify_before_allocs) == 0)
sgen_check_whole_heap_stw ();
}
}
/*
* We must already have the lock here instead of after the
* fast path because we might be interrupted in the fast path
* (after confirming that new_next < TLAB_TEMP_END) by the GC,
* and we'll end up allocating an object in a fragment which
* no longer belongs to us.
*
* The managed allocator does not do this, but it's treated
* specially by the world-stopping code.
*/
if (real_size > SGEN_MAX_SMALL_OBJ_SIZE) {
p = (void **)sgen_los_alloc_large_inner (vtable, ALIGN_UP (real_size));
} else {
/* tlab_next and tlab_temp_end are TLS vars so accessing them might be expensive */
p = (void**)TLAB_NEXT;
/* FIXME: handle overflow */
new_next = (char*)p + size;
TLAB_NEXT = new_next;
if (G_LIKELY (new_next < TLAB_TEMP_END)) {
/* Fast path */
/*
* FIXME: We might need a memory barrier here so the change to tlab_next is
* visible before the vtable store.
*/
CANARIFY_ALLOC(p,real_size);
SGEN_LOG (6, "Allocated object %p, vtable: %p (%s), size: %zd", p, vtable, sgen_client_vtable_get_name (vtable), size);
binary_protocol_alloc (p , vtable, size, sgen_client_get_provenance ());
g_assert (*p == NULL);
mono_atomic_store_seq (p, vtable);
return (GCObject*)p;
}
/* Slow path */
/* there are two cases: the object is too big or we run out of space in the TLAB */
/* we also reach here when the thread does its first allocation after a minor
* collection, since the tlab_ variables are initialized to NULL.
* there can be another case (from ORP), if we cooperate with the runtime a bit:
* objects that need finalizers can have the high bit set in their size
* so the above check fails and we can readily add the object to the queue.
* This avoids taking again the GC lock when registering, but this is moot when
* doing thread-local allocation, so it may not be a good idea.
*/
if (TLAB_NEXT >= TLAB_REAL_END) {
int available_in_tlab;
/*
* Run out of space in the TLAB. When this happens, some amount of space
* remains in the TLAB, but not enough to satisfy the current allocation
* request. Currently, we retire the TLAB in all cases, later we could
* keep it if the remaining space is above a treshold, and satisfy the
* allocation directly from the nursery.
*/
TLAB_NEXT -= size;
/* when running in degraded mode, we continue allocing that way
* for a while, to decrease the number of useless nursery collections.
*/
if (degraded_mode && degraded_mode < DEFAULT_NURSERY_SIZE)
return alloc_degraded (vtable, size, FALSE);
available_in_tlab = (int)(TLAB_REAL_END - TLAB_NEXT);//We'll never have tlabs > 2Gb
if (size > tlab_size || available_in_tlab > SGEN_MAX_NURSERY_WASTE) {
/* Allocate directly from the nursery */
p = (void **)sgen_nursery_alloc (size);
if (!p) {
/*
* We couldn't allocate from the nursery, so we try
* collecting. Even after the collection, we might
* still not have enough memory to allocate the
* object. The reason will most likely be that we've
* run out of memory, but there is the theoretical
* possibility that other threads might have consumed
* the freed up memory ahead of us.
*
* What we do in this case is allocate degraded, i.e.,
* from the major heap.
*
* Ideally we'd like to detect the case of other
* threads allocating ahead of us and loop (if we
* always loop we will loop endlessly in the case of
* OOM).
*/
sgen_ensure_free_space (real_size, GENERATION_NURSERY);
if (!degraded_mode)
p = (void **)sgen_nursery_alloc (size);
}
if (!p)
return alloc_degraded (vtable, size, FALSE);
zero_tlab_if_necessary (p, size);
} else {
size_t alloc_size = 0;
if (TLAB_START)
SGEN_LOG (3, "Retire TLAB: %p-%p [%ld]", TLAB_START, TLAB_REAL_END, (long)(TLAB_REAL_END - TLAB_NEXT - size));
sgen_nursery_retire_region (p, available_in_tlab);
p = (void **)sgen_nursery_alloc_range (tlab_size, size, &alloc_size);
if (!p) {
/* See comment above in similar case. */
sgen_ensure_free_space (tlab_size, GENERATION_NURSERY);
if (!degraded_mode)
p = (void **)sgen_nursery_alloc_range (tlab_size, size, &alloc_size);
}
if (!p)
return alloc_degraded (vtable, size, FALSE);
/* Allocate a new TLAB from the current nursery fragment */
TLAB_START = (char*)p;
TLAB_NEXT = TLAB_START;
TLAB_REAL_END = TLAB_START + alloc_size;
TLAB_TEMP_END = TLAB_START + MIN (SGEN_SCAN_START_SIZE, alloc_size);
zero_tlab_if_necessary (TLAB_START, alloc_size);
/* Allocate from the TLAB */
p = (void **)TLAB_NEXT;
TLAB_NEXT += size;
sgen_set_nursery_scan_start ((char*)p);
}
} else {
/* Reached tlab_temp_end */
/* record the scan start so we can find pinned objects more easily */
sgen_set_nursery_scan_start ((char*)p);
/* we just bump tlab_temp_end as well */
TLAB_TEMP_END = MIN (TLAB_REAL_END, TLAB_NEXT + SGEN_SCAN_START_SIZE);
SGEN_LOG (5, "Expanding local alloc: %p-%p", TLAB_NEXT, TLAB_TEMP_END);
}
CANARIFY_ALLOC(p,real_size);
}
if (G_LIKELY (p)) {
SGEN_LOG (6, "Allocated object %p, vtable: %p (%s), size: %zd", p, vtable, sgen_client_vtable_get_name (vtable), size);
binary_protocol_alloc (p, vtable, size, sgen_client_get_provenance ());
mono_atomic_store_seq (p, vtable);
}
return (GCObject*)p;
}
bool CMMALRenderer::Configure(unsigned int width, unsigned int height, unsigned int d_width, unsigned int d_height, float fps, unsigned flags, ERenderFormat format, unsigned extended_format, unsigned int orientation)
{
ReleaseBuffers();
m_sourceWidth = width;
m_sourceHeight = height;
m_renderOrientation = orientation;
m_fps = fps;
m_iFlags = flags;
m_format = format;
CLog::Log(LOGDEBUG, "%s::%s - %dx%d->%dx%d@%.2f flags:%x format:%d ext:%x orient:%d", CLASSNAME, __func__, width, height, d_width, d_height, fps, flags, format, extended_format, orientation);
m_RenderUpdateCallBackFn = NULL;
m_RenderUpdateCallBackCtx = NULL;
if ((m_format == RENDER_FMT_BYPASS) && g_application.GetCurrentPlayer())
{
m_renderFeatures.clear();
m_scalingMethods.clear();
m_deinterlaceModes.clear();
m_deinterlaceMethods.clear();
if (m_RenderFeaturesCallBackFn)
{
(*m_RenderFeaturesCallBackFn)(m_RenderFeaturesCallBackCtx, m_renderFeatures);
// after setting up m_renderFeatures, we are done with the callback
m_RenderFeaturesCallBackFn = NULL;
m_RenderFeaturesCallBackCtx = NULL;
}
g_application.m_pPlayer->GetRenderFeatures(m_renderFeatures);
g_application.m_pPlayer->GetDeinterlaceMethods(m_deinterlaceMethods);
g_application.m_pPlayer->GetDeinterlaceModes(m_deinterlaceModes);
g_application.m_pPlayer->GetScalingMethods(m_scalingMethods);
}
// calculate the input frame aspect ratio
CalculateFrameAspectRatio(d_width, d_height);
ChooseBestResolution(fps);
m_destWidth = g_graphicsContext.GetResInfo(m_resolution).iWidth;
m_destHeight = g_graphicsContext.GetResInfo(m_resolution).iHeight;
SetViewMode(CMediaSettings::Get().GetCurrentVideoSettings().m_ViewMode);
ManageDisplay();
if (m_format == RENDER_FMT_MMAL|| m_format == RENDER_FMT_YUV420P)
{
MMAL_ES_FORMAT_T *es_format = mmal_format_alloc();
es_format->type = MMAL_ES_TYPE_VIDEO;
es_format->es->video.crop.width = m_sourceWidth;
es_format->es->video.crop.height = m_sourceHeight;
if (m_format == RENDER_FMT_MMAL)
{
es_format->encoding = MMAL_ENCODING_OPAQUE;
es_format->es->video.width = m_sourceWidth;
es_format->es->video.height = m_sourceHeight;
}
else if (m_format == RENDER_FMT_YUV420P)
{
const int pitch = ALIGN_UP(m_sourceWidth, 32);
const int aligned_height = ALIGN_UP(m_sourceHeight, 16);
es_format->encoding = MMAL_ENCODING_I420;
es_format->es->video.width = pitch;
es_format->es->video.height = aligned_height;
if (CONF_FLAGS_YUVCOEF_MASK(m_iFlags) == CONF_FLAGS_YUVCOEF_BT709)
es_format->es->video.color_space = MMAL_COLOR_SPACE_ITUR_BT709;
else if (CONF_FLAGS_YUVCOEF_MASK(m_iFlags) == CONF_FLAGS_YUVCOEF_BT601)
es_format->es->video.color_space = MMAL_COLOR_SPACE_ITUR_BT601;
else if (CONF_FLAGS_YUVCOEF_MASK(m_iFlags) == CONF_FLAGS_YUVCOEF_240M)
es_format->es->video.color_space = MMAL_COLOR_SPACE_SMPTE240M;
}
if (m_bConfigured)
UnInit();
m_bConfigured = init_vout(es_format);
mmal_format_free(es_format);
}
else
m_bConfigured = true;
return m_bConfigured;
}
GCObject*
sgen_try_alloc_obj_nolock (GCVTable vtable, size_t size)
{
void **p;
char *new_next;
size_t real_size = size;
TLAB_ACCESS_INIT;
CANARIFY_SIZE(size);
size = ALIGN_UP (size);
SGEN_ASSERT (9, real_size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE, "Object too small");
SGEN_ASSERT (6, sgen_vtable_get_descriptor (vtable), "VTable without descriptor");
if (real_size > SGEN_MAX_SMALL_OBJ_SIZE)
return NULL;
if (G_UNLIKELY (size > tlab_size)) {
/* Allocate directly from the nursery */
p = (void **)sgen_nursery_alloc (size);
if (!p)
return NULL;
sgen_set_nursery_scan_start ((char*)p);
/*FIXME we should use weak memory ops here. Should help specially on x86. */
zero_tlab_if_necessary (p, size);
} else {
int available_in_tlab;
char *real_end;
/* tlab_next and tlab_temp_end are TLS vars so accessing them might be expensive */
p = (void**)TLAB_NEXT;
/* FIXME: handle overflow */
new_next = (char*)p + size;
real_end = TLAB_REAL_END;
available_in_tlab = (int)(real_end - (char*)p);//We'll never have tlabs > 2Gb
if (G_LIKELY (new_next < real_end)) {
TLAB_NEXT = new_next;
/* Second case, we overflowed temp end */
if (G_UNLIKELY (new_next >= TLAB_TEMP_END)) {
sgen_set_nursery_scan_start (new_next);
/* we just bump tlab_temp_end as well */
TLAB_TEMP_END = MIN (TLAB_REAL_END, TLAB_NEXT + SGEN_SCAN_START_SIZE);
SGEN_LOG (5, "Expanding local alloc: %p-%p", TLAB_NEXT, TLAB_TEMP_END);
}
} else if (available_in_tlab > SGEN_MAX_NURSERY_WASTE) {
/* Allocate directly from the nursery */
p = (void **)sgen_nursery_alloc (size);
if (!p)
return NULL;
zero_tlab_if_necessary (p, size);
} else {
size_t alloc_size = 0;
sgen_nursery_retire_region (p, available_in_tlab);
new_next = (char *)sgen_nursery_alloc_range (tlab_size, size, &alloc_size);
p = (void**)new_next;
if (!p)
return NULL;
TLAB_START = (char*)new_next;
TLAB_NEXT = new_next + size;
TLAB_REAL_END = new_next + alloc_size;
TLAB_TEMP_END = new_next + MIN (SGEN_SCAN_START_SIZE, alloc_size);
sgen_set_nursery_scan_start ((char*)p);
zero_tlab_if_necessary (new_next, alloc_size);
}
}
HEAVY_STAT (++stat_objects_alloced);
HEAVY_STAT (stat_bytes_alloced += size);
CANARIFY_ALLOC(p,real_size);
SGEN_LOG (6, "Allocated object %p, vtable: %p (%s), size: %zd", p, vtable, sgen_client_vtable_get_name (vtable), size);
binary_protocol_alloc (p, vtable, size, sgen_client_get_provenance ());
g_assert (*p == NULL); /* FIXME disable this in non debug builds */
mono_atomic_store_seq (p, vtable);
return (GCObject*)p;
}
grub_err_t
SUFFIX (grub_freebsd_load_elf_meta) (grub_file_t file, grub_addr_t *kern_end)
{
grub_err_t err;
Elf_Ehdr e;
Elf_Shdr *s;
char *shdr;
unsigned symoff, stroff, symsize, strsize;
grub_addr_t curload;
grub_freebsd_addr_t symstart, symend, symentsize, dynamic;
Elf_Sym *sym;
const char *str;
unsigned i;
err = read_headers (file, &e, &shdr);
if (err)
return err;
err = grub_freebsd_add_meta (FREEBSD_MODINFO_METADATA |
FREEBSD_MODINFOMD_ELFHDR, &e,
sizeof (e));
if (err)
return err;
for (s = (Elf_Shdr *) shdr; s < (Elf_Shdr *) (shdr
+ e.e_shnum * e.e_shentsize);
s = (Elf_Shdr *) ((char *) s + e.e_shentsize))
if (s->sh_type == SHT_SYMTAB)
break;
if (s >= (Elf_Shdr *) ((char *) shdr
+ e.e_shnum * e.e_shentsize))
return grub_error (GRUB_ERR_BAD_OS, "no symbol table");
symoff = s->sh_offset;
symsize = s->sh_size;
symentsize = s->sh_entsize;
s = (Elf_Shdr *) (shdr + e.e_shentsize * s->sh_link);
stroff = s->sh_offset;
strsize = s->sh_size;
if (*kern_end + 4 * sizeof (grub_freebsd_addr_t) + symsize + strsize
> grub_os_area_addr + grub_os_area_size)
return grub_error (GRUB_ERR_OUT_OF_RANGE,
"not enough memory for kernel symbols");
symstart = curload = ALIGN_UP (*kern_end, sizeof (grub_freebsd_addr_t));
*((grub_freebsd_addr_t *) UINT_TO_PTR (curload)) = symsize;
curload += sizeof (grub_freebsd_addr_t);
if (grub_file_seek (file, symoff) == (grub_off_t) -1)
return grub_errno;
sym = (Elf_Sym *) UINT_TO_PTR (curload);
if (grub_file_read (file, UINT_TO_PTR (curload), symsize) !=
(grub_ssize_t) symsize)
{
if (! grub_errno)
return grub_error (GRUB_ERR_BAD_OS, "invalid ELF");
return grub_errno;
}
curload += symsize;
*((grub_freebsd_addr_t *) UINT_TO_PTR (curload)) = strsize;
curload += sizeof (grub_freebsd_addr_t);
if (grub_file_seek (file, stroff) == (grub_off_t) -1)
return grub_errno;
str = (char *) UINT_TO_PTR (curload);
if (grub_file_read (file, UINT_TO_PTR (curload), strsize)
!= (grub_ssize_t) strsize)
{
if (! grub_errno)
return grub_error (GRUB_ERR_BAD_OS, "invalid ELF");
return grub_errno;
}
curload += strsize;
curload = ALIGN_UP (curload, sizeof (grub_freebsd_addr_t));
symend = curload;
for (i = 0;
i * symentsize < symsize;
i++, sym = (Elf_Sym *) ((char *) sym + symentsize))
{
const char *name = str + sym->st_name;
if (grub_strcmp (name, "_DYNAMIC") == 0)
break;
}
if (i * symentsize < symsize)
{
dynamic = sym->st_value;
grub_dprintf ("bsd", "dynamic = %llx\n", (unsigned long long) dynamic);
err = grub_freebsd_add_meta (FREEBSD_MODINFO_METADATA |
FREEBSD_MODINFOMD_DYNAMIC, &dynamic,
sizeof (dynamic));
if (err)
return err;
}
err = grub_freebsd_add_meta (FREEBSD_MODINFO_METADATA |
FREEBSD_MODINFOMD_SSYM, &symstart,
sizeof (symstart));
if (err)
return err;
err = grub_freebsd_add_meta (FREEBSD_MODINFO_METADATA |
FREEBSD_MODINFOMD_ESYM, &symend,
sizeof (symend));
if (err)
return err;
*kern_end = ALIGN_PAGE (curload);
return GRUB_ERR_NONE;
}
void rk_display_init(device_t dev, u32 lcdbase,
unsigned long fb_size)
{
struct edid edid;
uint32_t val;
struct soc_rockchip_rk3288_config *conf = dev->chip_info;
uint32_t lower = ALIGN_DOWN(lcdbase, MiB);
uint32_t upper = ALIGN_UP(lcdbase + fb_size, MiB);
enum vop_modes detected_mode = VOP_MODE_UNKNOWN;
printk(BIOS_SPEW, "LCD framebuffer @%p\n", (void *)(lcdbase));
memset((void *)lcdbase, 0, fb_size); /* clear the framebuffer */
dcache_clean_invalidate_by_mva((void *)lower, upper - lower);
mmu_config_range(lower / MiB, (upper - lower) / MiB, DCACHE_OFF);
switch (conf->vop_mode) {
case VOP_MODE_NONE:
return;
case VOP_MODE_AUTO_DETECT:
/* try EDP first, then HDMI */
case VOP_MODE_EDP:
printk(BIOS_DEBUG, "Attempting to setup EDP display.\n");
rkclk_configure_edp();
rkclk_configure_vop_aclk(conf->vop_id, 192 * MHz);
/* grf_edp_ref_clk_sel: from internal 24MHz or 27MHz clock */
write32(&rk3288_grf->soc_con12, RK_SETBITS(1 << 4));
/* select epd signal from vop0 or vop1 */
val = (conf->vop_id == 1) ? RK_SETBITS(1 << 5) :
RK_CLRBITS(1 << 5);
write32(&rk3288_grf->soc_con6, val);
rk_edp_init();
if (rk_edp_get_edid(&edid) == 0) {
detected_mode = VOP_MODE_EDP;
break;
} else {
printk(BIOS_WARNING, "Cannot get EDID from EDP.\n");
if (conf->vop_mode == VOP_MODE_EDP)
return;
}
/* fall thru */
case VOP_MODE_HDMI:
printk(BIOS_DEBUG, "Attempting to setup HDMI display.\n");
rkclk_configure_hdmi();
rkclk_configure_vop_aclk(conf->vop_id, 384 * MHz);
rk_hdmi_init(conf->vop_id);
if (rk_hdmi_get_edid(&edid) == 0) {
detected_mode = VOP_MODE_HDMI;
break;
} else {
printk(BIOS_WARNING, "Cannot get EDID from HDMI.\n");
if (conf->vop_mode == VOP_MODE_HDMI)
return;
}
/* fall thru */
default:
printk(BIOS_WARNING, "Cannot read any edid info, aborting.\n");
return;
}
if (rkclk_configure_vop_dclk(conf->vop_id, edid.mode.pixel_clock * KHz)) {
printk(BIOS_WARNING, "config vop err\n");
return;
}
edid_set_framebuffer_bits_per_pixel(&edid,
conf->framebuffer_bits_per_pixel, 0);
rkvop_mode_set(conf->vop_id, &edid, detected_mode);
rkvop_prepare(conf->vop_id, &edid);
rkvop_enable(conf->vop_id, lcdbase);
switch (detected_mode) {
case VOP_MODE_HDMI:
if (rk_hdmi_enable(&edid)) {
printk(BIOS_WARNING, "hdmi enable err\n");
return;
}
/*
* HACK: if we do remove this delay, HDMI TV may not show
* anythings. So we make an delay here, ensure TV have
* enough time to respond.
*/
mdelay(2000);
break;
case VOP_MODE_EDP:
default:
if (rk_edp_prepare()) {
printk(BIOS_WARNING, "edp prepare err\n");
return;
}
if (rk_edp_enable()) {
printk(BIOS_WARNING, "edp enable err\n");
return;
}
mainboard_power_on_backlight();
break;
}
set_vbe_mode_info_valid(&edid, (uintptr_t)lcdbase);
}
static int scalecuda_resize(AVFilterContext *ctx,
AVFrame *out, AVFrame *in)
{
AVHWFramesContext *in_frames_ctx = (AVHWFramesContext*)in->hw_frames_ctx->data;
CUDAScaleContext *s = ctx->priv;
switch (in_frames_ctx->sw_format) {
case AV_PIX_FMT_YUV420P:
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0], in->width, in->height, in->linesize[0],
out->data[0], out->width, out->height, out->linesize[0],
1);
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0]+in->linesize[0]*in->height, in->width/2, in->height/2, in->linesize[0]/2,
out->data[0]+out->linesize[0]*out->height, out->width/2, out->height/2, out->linesize[0]/2,
1);
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0]+ ALIGN_UP((in->linesize[0]*in->height*5)/4, s->tex_alignment), in->width/2, in->height/2, in->linesize[0]/2,
out->data[0]+(out->linesize[0]*out->height*5)/4, out->width/2, out->height/2, out->linesize[0]/2,
1);
break;
case AV_PIX_FMT_YUV444P:
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0], in->width, in->height, in->linesize[0],
out->data[0], out->width, out->height, out->linesize[0],
1);
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0]+in->linesize[0]*in->height, in->width, in->height, in->linesize[0],
out->data[0]+out->linesize[0]*out->height, out->width, out->height, out->linesize[0],
1);
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0]+in->linesize[0]*in->height*2, in->width, in->height, in->linesize[0],
out->data[0]+out->linesize[0]*out->height*2, out->width, out->height, out->linesize[0],
1);
break;
case AV_PIX_FMT_NV12:
call_resize_kernel(s, s->cu_func_uchar, s->cu_tex_uchar, 1,
in->data[0], in->width, in->height, in->linesize[0],
out->data[0], out->width, out->height, out->linesize[0],
1);
call_resize_kernel(s, s->cu_func_uchar2, s->cu_tex_uchar2, 2,
in->data[1], in->width/2, in->height/2, in->linesize[1],
out->data[0] + out->linesize[0] * ((out->height + 31) & ~0x1f), out->width/2, out->height/2, out->linesize[1]/2,
1);
break;
case AV_PIX_FMT_P010LE:
call_resize_kernel(s, s->cu_func_ushort, s->cu_tex_ushort, 1,
in->data[0], in->width, in->height, in->linesize[0]/2,
out->data[0], out->width, out->height, out->linesize[0]/2,
2);
call_resize_kernel(s, s->cu_func_ushort2, s->cu_tex_ushort2, 2,
in->data[1], in->width / 2, in->height / 2, in->linesize[1]/2,
out->data[0] + out->linesize[0] * ((out->height + 31) & ~0x1f), out->width / 2, out->height / 2, out->linesize[1] / 4,
2);
break;
case AV_PIX_FMT_P016LE:
call_resize_kernel(s, s->cu_func_ushort, s->cu_tex_ushort, 1,
in->data[0], in->width, in->height, in->linesize[0] / 2,
out->data[0], out->width, out->height, out->linesize[0] / 2,
2);
call_resize_kernel(s, s->cu_func_ushort2, s->cu_tex_ushort2, 2,
in->data[1], in->width / 2, in->height / 2, in->linesize[1] / 2,
out->data[0] + out->linesize[0] * ((out->height + 31) & ~0x1f), out->width / 2, out->height / 2, out->linesize[1] / 4,
2);
break;
default:
return AVERROR_BUG;
}
return 0;
}
/*************************************************
* HOST DRIVERS
*************************************************/
void hostGPUDRV(CUfunction drvfun, int N, int nrhs, hostdrv_pars_t *prhs) {
unsigned int maxthreads = MAXTHREADS_STREAM;
int nstreams = iDivUp(N, maxthreads*BLOCK_DIM1D);
CUresult err = CUDA_SUCCESS;
for (int str = 0; str < nstreams; str++) {
int offset = str * maxthreads * BLOCK_DIM1D;
int size = 0;
if (str == (nstreams - 1))
size = N - str * maxthreads * BLOCK_DIM1D;
else
size = maxthreads * BLOCK_DIM1D;
int gridx = iDivUp(size, BLOCK_DIM1D); // number of x blocks
// setup execution parameters
if (CUDA_SUCCESS != (err = cuFuncSetBlockShape(drvfun, BLOCK_DIM1D, 1, 1))) {
mexErrMsgTxt("Error in cuFuncSetBlockShape");
}
if (CUDA_SUCCESS != cuFuncSetSharedSize(drvfun, 0)) {
mexErrMsgTxt("Error in cuFuncSetSharedSize");
}
// add parameters
int poffset = 0;
// CUDA kernels interface
// N: number of elements
// offset: used for streams
ALIGN_UP(poffset, __alignof(size));
if (CUDA_SUCCESS != cuParamSeti(drvfun, poffset, size)) {
mexErrMsgTxt("Error in cuParamSeti");
}
poffset += sizeof(size);
ALIGN_UP(poffset, __alignof(offset));
if (CUDA_SUCCESS != cuParamSeti(drvfun, poffset, offset)) {
mexErrMsgTxt("Error in cuParamSeti");
}
poffset += sizeof(offset);
for (int p=0;p<nrhs;p++) {
ALIGN_UP(poffset, prhs[p].align);
if (CUDA_SUCCESS
!= cuParamSetv(drvfun, poffset, prhs[p].par, prhs[p].psize)) {
mexErrMsgTxt("Error in cuParamSetv");
}
poffset += prhs[p].psize;
}
if (CUDA_SUCCESS != cuParamSetSize(drvfun, poffset)) {
mexErrMsgTxt("Error in cuParamSetSize");
}
err = cuLaunchGridAsync(drvfun, gridx, 1, 0);
if (CUDA_SUCCESS != err) {
mexErrMsgTxt("Error running kernel");
}
}
}
internal_function
new_heap (size_t size, size_t top_pad)
{
size_t pagesize = GLRO (dl_pagesize);
char *p1, *p2;
unsigned long ul;
heap_info *h;
if (size + top_pad < HEAP_MIN_SIZE)
size = HEAP_MIN_SIZE;
else if (size + top_pad <= HEAP_MAX_SIZE)
size += top_pad;
else if (size > HEAP_MAX_SIZE)
return 0;
else
size = HEAP_MAX_SIZE;
size = ALIGN_UP (size, pagesize);
/* A memory region aligned to a multiple of HEAP_MAX_SIZE is needed.
No swap space needs to be reserved for the following large
mapping (on Linux, this is the case for all non-writable mappings
anyway). */
p2 = MAP_FAILED;
if (aligned_heap_area)
{
p2 = (char *) MMAP (aligned_heap_area, HEAP_MAX_SIZE, PROT_NONE,
MAP_NORESERVE);
aligned_heap_area = NULL;
if (p2 != MAP_FAILED && ((unsigned long) p2 & (HEAP_MAX_SIZE - 1)))
{
__munmap (p2, HEAP_MAX_SIZE);
p2 = MAP_FAILED;
}
}
if (p2 == MAP_FAILED)
{
p1 = (char *) MMAP (0, HEAP_MAX_SIZE << 1, PROT_NONE, MAP_NORESERVE);
if (p1 != MAP_FAILED)
{
p2 = (char *) (((unsigned long) p1 + (HEAP_MAX_SIZE - 1))
& ~(HEAP_MAX_SIZE - 1));
ul = p2 - p1;
if (ul)
__munmap (p1, ul);
else
aligned_heap_area = p2 + HEAP_MAX_SIZE;
__munmap (p2 + HEAP_MAX_SIZE, HEAP_MAX_SIZE - ul);
}
else
{
/* Try to take the chance that an allocation of only HEAP_MAX_SIZE
is already aligned. */
p2 = (char *) MMAP (0, HEAP_MAX_SIZE, PROT_NONE, MAP_NORESERVE);
if (p2 == MAP_FAILED)
return 0;
if ((unsigned long) p2 & (HEAP_MAX_SIZE - 1))
{
__munmap (p2, HEAP_MAX_SIZE);
return 0;
}
}
}
if (__mprotect (p2, size, PROT_READ | PROT_WRITE) != 0)
{
__munmap (p2, HEAP_MAX_SIZE);
return 0;
}
h = (heap_info *) p2;
h->size = size;
h->mprotect_size = size;
LIBC_PROBE (memory_heap_new, 2, h, h->size);
return h;
}
int main(int argc, char *argv[]) {
int quality = 75;
if (argc == 2) {
quality = atoi(argv[1]);
if (quality < 10 || quality > 100) {
fprintf(stderr, "invalid quality. must be between 10 and 100\n");
return 1;
}
}
bcm_host_init();
uint32_t screen = 0;
DISPMANX_DISPLAY_HANDLE_T display = vc_dispmanx_display_open(screen);
DISPMANX_MODEINFO_T info;
int ret = vc_dispmanx_display_get_info(display, &info);
assert(ret == 0);
/* DispmanX expects buffer rows to be aligned to a 32 bit boundary */
int pitch = ALIGN_UP(2 * info.width, 32);
uint32_t vc_image_ptr;
VC_IMAGE_TYPE_T type = VC_IMAGE_RGB565;
DISPMANX_RESOURCE_HANDLE_T resource = vc_dispmanx_resource_create(
type, info.width, info.height, &vc_image_ptr
);
VC_IMAGE_TRANSFORM_T transform = 0;
vc_dispmanx_snapshot(display, resource, transform);
VC_RECT_T rect;
vc_dispmanx_rect_set(&rect, 0, 0, info.width, info.height);
unsigned char *image = malloc(pitch * info.height);
assert(image);
vc_dispmanx_resource_read_data(resource, &rect, image, info.width * 2);
ret = vc_dispmanx_resource_delete(resource);
assert(ret == 0);
ret = vc_dispmanx_display_close(display);
assert(ret == 0);
struct jpeg_compress_struct cinfo;
struct jpeg_error_mgr jerr;
JSAMPROW row_pointer[1];
cinfo.err = jpeg_std_error(&jerr);
jpeg_create_compress(&cinfo);
jpeg_stdio_dest(&cinfo, stdout);
cinfo.image_width = info.width;
cinfo.image_height = info.height;
cinfo.input_components = 3;
cinfo.in_color_space = JCS_RGB;
jpeg_set_defaults(&cinfo);
jpeg_set_quality(&cinfo, quality, TRUE);
jpeg_start_compress(&cinfo, TRUE);
int row_stride = cinfo.image_width * 3;
while (cinfo.next_scanline < cinfo.image_height) {
unsigned char row[row_stride];
unsigned char *dst = &row[0];
uint16_t *src = (uint16_t*)(image + pitch * cinfo.next_scanline);
for (int x = 0; x < cinfo.image_width; x++, src++) {
*dst++ = ((*src & 0xf800) >> 11) << 3;
*dst++ = ((*src & 0x07e0) >> 5) << 2;
*dst++ = ((*src & 0x001f) >> 0) << 3;
}
row_pointer[0] = row;
jpeg_write_scanlines(&cinfo, row_pointer, 1);
}
jpeg_finish_compress(&cinfo);
jpeg_destroy_compress(&cinfo);
return 0;
}
void PackBvmlinuzI386::buildLoader(const Filter *ft)
{
// prepare loader
initLoader(stub_i386_linux_kernel_vmlinuz, sizeof(stub_i386_linux_kernel_vmlinuz));
if (0!=page_offset) { // relocatable kernel
assert(0==ft->id || 0x40==(0xf0 & ft->id)); // others assume fixed buffer address
addLoader("LINUZ000,LINUZ001,LINUZVGA,LINUZ101,LINUZ110",
((0!=config_physical_align) ? "LINUZ120" : "LINUZ130"),
"LINUZ140,LZCUTPOI,LINUZ141",
(ft->id ? "LINUZ145" : ""),
(ph.first_offset_found == 1 ? "LINUZ010" : ""),
NULL);
}
else {
addLoader("LINUZ000,LINUZ001,LINUZVGA,LINUZ005",
ph.first_offset_found == 1 ? "LINUZ010" : "",
(0x40==(0xf0 & ft->id)) ? "LZCKLLT1" : (ft->id ? "LZCALLT1" : ""),
"LBZIMAGE,IDENTSTR",
"+40", // align the stuff to 4 byte boundary
"UPX1HEAD", // 32 byte
"LZCUTPOI",
NULL);
// fake alignment for the start of the decompressor
//linker->defineSymbol("LZCUTPOI", 0x1000);
}
addLoader(getDecompressorSections(), NULL);
if (ft->id)
{
assert(ft->calls > 0);
if (0x40==(0xf0 & ft->id)) {
addLoader("LZCKLLT9", NULL);
}
else {
addLoader("LZCALLT9", NULL);
}
addFilter32(ft->id);
}
if (0!=page_offset) {
addLoader("LINUZ150,IDENTSTR,+40,UPX1HEAD", NULL);
unsigned const l_len = getLoaderSize();
unsigned const c_len = ALIGN_UP(ph.c_len, 4u);
unsigned const e_len = getLoaderSectionStart("LINUZ141") -
getLoaderSectionStart("LINUZ110");
linker->defineSymbol("compressed_length", c_len);
linker->defineSymbol("load_physical_address", physical_start); // FIXME
if (0!=config_physical_align) {
linker->defineSymbol("neg_config_physical_align", 0u - config_physical_align);
}
linker->defineSymbol("neg_length_mov", 0u - ALIGN_UP(c_len + l_len, 4u));
linker->defineSymbol("neg_page_offset", 0u - page_offset);
//linker->defineSymbol("physical_start", physical_start);
linker->defineSymbol("unc_length", ph.u_len);
linker->defineSymbol("dec_offset", ph.overlap_overhead + e_len);
linker->defineSymbol("unc_offset", ph.overlap_overhead + ph.u_len - c_len);
}
else {
addLoader("LINUZ990", NULL);
}
}
/* this is really aimed at the lcd panel. That said, there are two display
* devices on this part and we may someday want to extend it for other boards.
*/
void display_startup(device_t dev)
{
struct soc_nvidia_tegra124_config *config = dev->chip_info;
struct display_controller *disp_ctrl = (void *)config->display_controller;
struct pwm_controller *pwm = (void *)TEGRA_PWM_BASE;
struct tegra_dc *dc = &dc_data;
u32 plld_rate;
/* init dc */
dc->base = (void *)TEGRA_ARM_DISPLAYA;
dc->config = config;
config->dc_data = dc;
/* Note dp_init may read EDID and change some config values. */
dp_init(config);
/* should probably just make it all MiB ... in future */
u32 framebuffer_size_mb = config->framebuffer_size / MiB;
u32 framebuffer_base_mb= config->framebuffer_base / MiB;
/* light it all up */
/* This one may have been done in romstage but that's ok for now. */
if (config->panel_vdd_gpio){
gpio_output(config->panel_vdd_gpio, 1);
printk(BIOS_SPEW,"%s: panel_vdd setting gpio %08x to %d\n",
__func__, config->panel_vdd_gpio, 1);
}
udelay(config->vdd_delay_ms * 1000);
if (config->backlight_vdd_gpio){
gpio_output(config->backlight_vdd_gpio, 1);
printk(BIOS_SPEW,"%s: backlight vdd setting gpio %08x to %d\n",
__func__, config->backlight_vdd_gpio, 1);
}
if (config->lvds_shutdown_gpio){
gpio_output(config->lvds_shutdown_gpio, 0);
printk(BIOS_SPEW,"%s: lvds shutdown setting gpio %08x to %d\n",
__func__, config->lvds_shutdown_gpio, 0);
}
if (framebuffer_size_mb == 0){
framebuffer_size_mb = ALIGN_UP(config->xres * config->yres *
(config->framebuffer_bits_per_pixel / 8), MiB)/MiB;
}
if (! framebuffer_base_mb)
framebuffer_base_mb = fb_base_mb();
config->framebuffer_size = framebuffer_size_mb * MiB;
config->framebuffer_base = framebuffer_base_mb * MiB;
mmu_config_range(framebuffer_base_mb, framebuffer_size_mb,
DCACHE_WRITETHROUGH);
printk(BIOS_SPEW, "LCD frame buffer at %dMiB to %dMiB\n", framebuffer_base_mb,
framebuffer_base_mb + framebuffer_size_mb);
/* GPIO magic here if needed to start powering up things. You
* really only want to enable vdd, wait a bit, and then enable
* the panel. However ... the timings in the tegra20 dts make
* no sense to me. I'm pretty sure they're wrong.
* The panel_vdd is done in the romstage, so we need only
* light things up here once we're sure it's all working.
*/
/* The plld is programmed with the assumption of the SHIFT_CLK_DIVIDER
* and PIXEL_CLK_DIVIDER are zero (divide by 1). See the
* update_display_mode() for detail.
*/
plld_rate = clock_display(config->pixel_clock * 2);
if (plld_rate == 0) {
printk(BIOS_ERR, "dc: clock init failed\n");
return;
} else if (plld_rate != config->pixel_clock * 2) {
printk(BIOS_WARNING, "dc: plld rounded to %u\n", plld_rate);
config->pixel_clock = plld_rate / 2;
}
/* Init dc */
if (tegra_dc_init(disp_ctrl)) {
printk(BIOS_ERR, "dc: init failed\n");
return;
}
/* Configure dc mode */
if (update_display_mode(disp_ctrl, config)) {
printk(BIOS_ERR, "dc: failed to configure display mode.\n");
return;
}
/* Enable dp */
dp_enable(dc->out);
/* Init frame buffer */
memset((void *)(framebuffer_base_mb*MiB), 0x00,
framebuffer_size_mb*MiB);
update_window(disp_ctrl, config);
/* Set up Tegra PWM n (where n is specified in config->pwm) to drive the
* panel backlight.
*/
printk(BIOS_SPEW, "%s: enable panel backlight pwm\n", __func__);
WRITEL(((1 << NV_PWM_CSR_ENABLE_SHIFT) |
(220 << NV_PWM_CSR_PULSE_WIDTH_SHIFT) | /* 220/256 */
0x02e), /* frequency divider */
&pwm->pwm[config->pwm].csr);
udelay(config->pwm_to_bl_delay_ms * 1000);
if (config->backlight_en_gpio){
gpio_output(config->backlight_en_gpio, 1);
printk(BIOS_SPEW,"%s: backlight enable setting gpio %08x to %d\n",
__func__, config->backlight_en_gpio, 1);
}
printk(BIOS_INFO, "%s: display init done.\n", __func__);
/* tell depthcharge ...
*/
struct edid edid;
edid.mode.va = config->yres;
edid.mode.ha = config->xres;
edid_set_framebuffer_bits_per_pixel(&edid,
config->framebuffer_bits_per_pixel, 32);
set_vbe_mode_info_valid(&edid, (uintptr_t)(framebuffer_base_mb*MiB));
}