linux_dsm_epyc7002/arch/x86/vdso/vdso2c.h

319 lines
10 KiB
C
Raw Normal View History

x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
/*
* This file is included twice from vdso2c.c. It generates code for 32-bit
* and 64-bit vDSOs. We need both for 64-bit builds, since 32-bit vDSOs
* are built for 32-bit userspace.
*/
/*
* We're writing a section table for a few reasons:
*
* The Go runtime had a couple of bugs: it would read the section
* table to try to figure out how many dynamic symbols there were (it
* shouldn't have looked at the section table at all) and, if there
* were no SHT_SYNDYM section table entry, it would use an
* uninitialized value for the number of symbols. An empty DYNSYM
* table would work, but I see no reason not to write a valid one (and
* keep full performance for old Go programs). This hack is only
* needed on x86_64.
*
* The bug was introduced on 2012-08-31 by:
* https://code.google.com/p/go/source/detail?r=56ea40aac72b
* and was fixed on 2014-06-13 by:
* https://code.google.com/p/go/source/detail?r=fc1cd5e12595
*
* Binutils has issues debugging the vDSO: it reads the section table to
* find SHT_NOTE; it won't look at PT_NOTE for the in-memory vDSO, which
* would break build-id if we removed the section table. Binutils
* also requires that shstrndx != 0. See:
* https://sourceware.org/bugzilla/show_bug.cgi?id=17064
*
* elfutils might not look for PT_NOTE if there is a section table at
* all. I don't know whether this matters for any practical purpose.
*
* For simplicity, rather than hacking up a partial section table, we
* just write a mostly complete one. We omit non-dynamic symbols,
* though, since they're rather large.
*
* Once binutils gets fixed, we might be able to drop this for all but
* the 64-bit vdso, since build-id only works in kernel RPMs, and
* systems that update to new enough kernel RPMs will likely update
* binutils in sync. build-id has never worked for home-built kernel
* RPMs without manual symlinking, and I suspect that no one ever does
* that.
*/
struct BITSFUNC(fake_sections)
{
ELF(Shdr) *table;
unsigned long table_offset;
int count, max_count;
int in_shstrndx;
unsigned long shstr_offset;
const char *shstrtab;
size_t shstrtab_len;
int out_shstrndx;
};
static unsigned int BITSFUNC(find_shname)(struct BITSFUNC(fake_sections) *out,
const char *name)
{
const char *outname = out->shstrtab;
while (outname - out->shstrtab < out->shstrtab_len) {
if (!strcmp(name, outname))
return (outname - out->shstrtab) + out->shstr_offset;
outname += strlen(outname) + 1;
}
if (*name)
printf("Warning: could not find output name \"%s\"\n", name);
return out->shstr_offset + out->shstrtab_len - 1; /* Use a null. */
}
static void BITSFUNC(init_sections)(struct BITSFUNC(fake_sections) *out)
{
if (!out->in_shstrndx)
fail("didn't find the fake shstrndx\n");
memset(out->table, 0, out->max_count * sizeof(ELF(Shdr)));
if (out->max_count < 1)
fail("we need at least two fake output sections\n");
PUT_LE(&out->table[0].sh_type, SHT_NULL);
PUT_LE(&out->table[0].sh_name, BITSFUNC(find_shname)(out, ""));
out->count = 1;
}
static void BITSFUNC(copy_section)(struct BITSFUNC(fake_sections) *out,
int in_idx, const ELF(Shdr) *in,
const char *name)
{
uint64_t flags = GET_LE(&in->sh_flags);
bool copy = flags & SHF_ALLOC &&
(GET_LE(&in->sh_size) ||
(GET_LE(&in->sh_type) != SHT_RELA &&
GET_LE(&in->sh_type) != SHT_REL)) &&
strcmp(name, ".altinstructions") &&
strcmp(name, ".altinstr_replacement");
if (!copy)
return;
if (out->count >= out->max_count)
fail("too many copied sections (max = %d)\n", out->max_count);
if (in_idx == out->in_shstrndx)
out->out_shstrndx = out->count;
out->table[out->count] = *in;
PUT_LE(&out->table[out->count].sh_name,
BITSFUNC(find_shname)(out, name));
/* elfutils requires that a strtab have the correct type. */
if (!strcmp(name, ".fake_shstrtab"))
PUT_LE(&out->table[out->count].sh_type, SHT_STRTAB);
out->count++;
}
static void BITSFUNC(go)(void *addr, size_t len,
FILE *outfile, const char *name)
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
{
int found_load = 0;
unsigned long load_size = -1; /* Work around bogus warning */
unsigned long data_size;
ELF(Ehdr) *hdr = (ELF(Ehdr) *)addr;
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
int i;
unsigned long j;
ELF(Shdr) *symtab_hdr = NULL, *strtab_hdr, *secstrings_hdr,
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
*alt_sec = NULL;
ELF(Dyn) *dyn = 0, *dyn_end = 0;
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
const char *secstrings;
uint64_t syms[NSYMS] = {};
struct BITSFUNC(fake_sections) fake_sections = {};
ELF(Phdr) *pt = (ELF(Phdr) *)(addr + GET_LE(&hdr->e_phoff));
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
/* Walk the segment table. */
for (i = 0; i < GET_LE(&hdr->e_phnum); i++) {
if (GET_LE(&pt[i].p_type) == PT_LOAD) {
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
if (found_load)
fail("multiple PT_LOAD segs\n");
if (GET_LE(&pt[i].p_offset) != 0 ||
GET_LE(&pt[i].p_vaddr) != 0)
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
fail("PT_LOAD in wrong place\n");
if (GET_LE(&pt[i].p_memsz) != GET_LE(&pt[i].p_filesz))
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
fail("cannot handle memsz != filesz\n");
load_size = GET_LE(&pt[i].p_memsz);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
found_load = 1;
} else if (GET_LE(&pt[i].p_type) == PT_DYNAMIC) {
dyn = addr + GET_LE(&pt[i].p_offset);
dyn_end = addr + GET_LE(&pt[i].p_offset) +
GET_LE(&pt[i].p_memsz);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
}
if (!found_load)
fail("no PT_LOAD seg\n");
data_size = (load_size + 4095) / 4096 * 4096;
/* Walk the dynamic table */
for (i = 0; dyn + i < dyn_end &&
GET_LE(&dyn[i].d_tag) != DT_NULL; i++) {
typeof(dyn[i].d_tag) tag = GET_LE(&dyn[i].d_tag);
if (tag == DT_REL || tag == DT_RELSZ || tag == DT_RELA ||
tag == DT_RELENT || tag == DT_TEXTREL)
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
fail("vdso image contains dynamic relocations\n");
}
/* Walk the section table */
secstrings_hdr = addr + GET_LE(&hdr->e_shoff) +
GET_LE(&hdr->e_shentsize)*GET_LE(&hdr->e_shstrndx);
secstrings = addr + GET_LE(&secstrings_hdr->sh_offset);
for (i = 0; i < GET_LE(&hdr->e_shnum); i++) {
ELF(Shdr) *sh = addr + GET_LE(&hdr->e_shoff) +
GET_LE(&hdr->e_shentsize) * i;
if (GET_LE(&sh->sh_type) == SHT_SYMTAB)
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
symtab_hdr = sh;
if (!strcmp(secstrings + GET_LE(&sh->sh_name),
".altinstructions"))
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
alt_sec = sh;
}
if (!symtab_hdr)
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
fail("no symbol table\n");
strtab_hdr = addr + GET_LE(&hdr->e_shoff) +
GET_LE(&hdr->e_shentsize) * GET_LE(&symtab_hdr->sh_link);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
/* Walk the symbol table */
for (i = 0;
i < GET_LE(&symtab_hdr->sh_size) / GET_LE(&symtab_hdr->sh_entsize);
i++) {
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
int k;
ELF(Sym) *sym = addr + GET_LE(&symtab_hdr->sh_offset) +
GET_LE(&symtab_hdr->sh_entsize) * i;
const char *name = addr + GET_LE(&strtab_hdr->sh_offset) +
GET_LE(&sym->st_name);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
for (k = 0; k < NSYMS; k++) {
if (!strcmp(name, required_syms[k].name)) {
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
if (syms[k]) {
fail("duplicate symbol %s\n",
required_syms[k].name);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
syms[k] = GET_LE(&sym->st_value);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
}
if (!strcmp(name, "fake_shstrtab")) {
ELF(Shdr) *sh;
fake_sections.in_shstrndx = GET_LE(&sym->st_shndx);
fake_sections.shstrtab = addr + GET_LE(&sym->st_value);
fake_sections.shstrtab_len = GET_LE(&sym->st_size);
sh = addr + GET_LE(&hdr->e_shoff) +
GET_LE(&hdr->e_shentsize) *
fake_sections.in_shstrndx;
fake_sections.shstr_offset = GET_LE(&sym->st_value) -
GET_LE(&sh->sh_addr);
}
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
/* Build the output section table. */
if (!syms[sym_VDSO_FAKE_SECTION_TABLE_START] ||
!syms[sym_VDSO_FAKE_SECTION_TABLE_END])
fail("couldn't find fake section table\n");
if ((syms[sym_VDSO_FAKE_SECTION_TABLE_END] -
syms[sym_VDSO_FAKE_SECTION_TABLE_START]) % sizeof(ELF(Shdr)))
fail("fake section table size isn't a multiple of sizeof(Shdr)\n");
fake_sections.table = addr + syms[sym_VDSO_FAKE_SECTION_TABLE_START];
fake_sections.table_offset = syms[sym_VDSO_FAKE_SECTION_TABLE_START];
fake_sections.max_count = (syms[sym_VDSO_FAKE_SECTION_TABLE_END] -
syms[sym_VDSO_FAKE_SECTION_TABLE_START]) /
sizeof(ELF(Shdr));
BITSFUNC(init_sections)(&fake_sections);
for (i = 0; i < GET_LE(&hdr->e_shnum); i++) {
ELF(Shdr) *sh = addr + GET_LE(&hdr->e_shoff) +
GET_LE(&hdr->e_shentsize) * i;
BITSFUNC(copy_section)(&fake_sections, i, sh,
secstrings + GET_LE(&sh->sh_name));
}
if (!fake_sections.out_shstrndx)
fail("didn't generate shstrndx?!?\n");
PUT_LE(&hdr->e_shoff, fake_sections.table_offset);
PUT_LE(&hdr->e_shentsize, sizeof(ELF(Shdr)));
PUT_LE(&hdr->e_shnum, fake_sections.count);
PUT_LE(&hdr->e_shstrndx, fake_sections.out_shstrndx);
/* Validate mapping addresses. */
for (i = 0; i < sizeof(special_pages) / sizeof(special_pages[0]); i++) {
if (!syms[i])
continue; /* The mapping isn't used; ignore it. */
if (syms[i] % 4096)
fail("%s must be a multiple of 4096\n",
required_syms[i].name);
if (syms[i] < data_size)
fail("%s must be after the text mapping\n",
required_syms[i].name);
if (syms[sym_end_mapping] < syms[i] + 4096)
fail("%s overruns end_mapping\n",
required_syms[i].name);
}
if (syms[sym_end_mapping] % 4096)
fail("end_mapping must be a multiple of 4096\n");
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
if (!name) {
fwrite(addr, load_size, 1, outfile);
return;
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
fprintf(outfile, "/* AUTOMATICALLY GENERATED -- DO NOT EDIT */\n\n");
fprintf(outfile, "#include <linux/linkage.h>\n");
fprintf(outfile, "#include <asm/page_types.h>\n");
fprintf(outfile, "#include <asm/vdso.h>\n");
fprintf(outfile, "\n");
fprintf(outfile,
"static unsigned char raw_data[%lu] __page_aligned_data = {",
data_size);
for (j = 0; j < load_size; j++) {
if (j % 10 == 0)
fprintf(outfile, "\n\t");
fprintf(outfile, "0x%02X, ", (int)((unsigned char *)addr)[j]);
}
fprintf(outfile, "\n};\n\n");
fprintf(outfile, "static struct page *pages[%lu];\n\n",
data_size / 4096);
fprintf(outfile, "const struct vdso_image %s = {\n", name);
fprintf(outfile, "\t.data = raw_data,\n");
fprintf(outfile, "\t.size = %lu,\n", data_size);
fprintf(outfile, "\t.text_mapping = {\n");
fprintf(outfile, "\t\t.name = \"[vdso]\",\n");
fprintf(outfile, "\t\t.pages = pages,\n");
fprintf(outfile, "\t},\n");
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
if (alt_sec) {
fprintf(outfile, "\t.alt = %lu,\n",
(unsigned long)GET_LE(&alt_sec->sh_offset));
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
fprintf(outfile, "\t.alt_len = %lu,\n",
(unsigned long)GET_LE(&alt_sec->sh_size));
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
for (i = 0; i < NSYMS; i++) {
if (required_syms[i].export && syms[i])
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
fprintf(outfile, "\t.sym_%s = 0x%" PRIx64 ",\n",
required_syms[i].name, syms[i]);
x86, vdso: Reimplement vdso.so preparation in build-time C Currently, vdso.so files are prepared and analyzed by a combination of objcopy, nm, some linker script tricks, and some simple ELF parsers in the kernel. Replace all of that with plain C code that runs at build time. All five vdso images now generate .c files that are compiled and linked in to the kernel image. This should cause only one userspace-visible change: the loaded vDSO images are stripped more heavily than they used to be. Everything outside the loadable segment is dropped. In particular, this causes the section table and section name strings to be missing. This should be fine: real dynamic loaders don't load or inspect these tables anyway. The result is roughly equivalent to eu-strip's --strip-sections option. The purpose of this change is to enable the vvar and hpet mappings to be moved to the page following the vDSO load segment. Currently, it is possible for the section table to extend into the page after the load segment, so, if we map it, it risks overlapping the vvar or hpet page. This happens whenever the load segment is just under a multiple of PAGE_SIZE. The only real subtlety here is that the old code had a C file with inline assembler that did 'call VDSO32_vsyscall' and a linker script that defined 'VDSO32_vsyscall = __kernel_vsyscall'. This most likely worked by accident: the linker script entry defines a symbol associated with an address as opposed to an alias for the real dynamic symbol __kernel_vsyscall. That caused ld to relocate the reference at link time instead of leaving an interposable dynamic relocation. Since the VDSO32_vsyscall hack is no longer needed, I now use 'call __kernel_vsyscall', and I added -Bsymbolic to make it work. vdso2c will generate an error and abort the build if the resulting image contains any dynamic relocations, so we won't silently generate bad vdso images. (Dynamic relocations are a problem because nothing will even attempt to relocate the vdso.) Signed-off-by: Andy Lutomirski <luto@amacapital.net> Link: http://lkml.kernel.org/r/2c4fcf45524162a34d87fdda1eb046b2a5cecee7.1399317206.git.luto@amacapital.net Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2014-05-06 02:19:34 +07:00
}
fprintf(outfile, "};\n");
}