On 05/29/2015 03:58 PM, Andy Lutomirski wrote:
> Test a couple of special cases in 32-bit kernels for entries from
> vm86 mode. This will OOPS both old kernels due to a bug and and
> 4.1-rc5 due to a regression I introduced, and it should make sure
> that the SYSENTER-from-vm86-mode hack in the kernel keeps working.
>
> Tests: 394838c96013 x86/asm/entry/32: Fix user_mode() misuses
> Tests: 7ba554b5ac69 x86/asm/entry/32: Really make user_mode() work correctly for VM86 mode
> Signed-off-by: Andy Lutomirski <[email protected]>
> ---
>
> Ingo, Shuah: I think this should go in through -tip. (In general, I think
> new x86 tests that don't have interesting interactions with the kselftest
> infrastructure should go in through -tip, especially tests such as this one
> that are related to recent regressions.)
No problem going through tip. Could you please make sure
"make kselftest" run from top level and
tools/testing/selftests/kselftest_install.sh
don't break?
Once the above are verified, here is my
Acked-by: Shuah Khan <[email protected]>
Thanks,
-- Shuah
>
> tools/testing/selftests/x86/Makefile | 6 +-
> tools/testing/selftests/x86/entry_from_vm86.c | 114 ++++++++++++++++++++++++++
> 2 files changed, 118 insertions(+), 2 deletions(-)
> create mode 100644 tools/testing/selftests/x86/entry_from_vm86.c
>
> diff --git a/tools/testing/selftests/x86/Makefile b/tools/testing/selftests/x86/Makefile
> index 5bdb781163d1..9b0d8baf2934 100644
> --- a/tools/testing/selftests/x86/Makefile
> +++ b/tools/testing/selftests/x86/Makefile
> @@ -5,8 +5,10 @@ include ../lib.mk
> .PHONY: all all_32 all_64 warn_32bit_failure clean
>
> TARGETS_C_BOTHBITS := sigreturn single_step_syscall
> +TARGETS_C_32BIT_ONLY := entry_from_vm86
>
> -BINARIES_32 := $(TARGETS_C_BOTHBITS:%=%_32)
> +TARGETS_C_32BIT_ALL := $(TARGETS_C_BOTHBITS) $(TARGETS_C_32BIT_ONLY)
> +BINARIES_32 := $(TARGETS_C_32BIT_ALL:%=%_32)
> BINARIES_64 := $(TARGETS_C_BOTHBITS:%=%_64)
>
> CFLAGS := -O2 -g -std=gnu99 -pthread -Wall
> @@ -32,7 +34,7 @@ all_64: $(BINARIES_64)
> clean:
> $(RM) $(BINARIES_32) $(BINARIES_64)
>
> -$(TARGETS_C_BOTHBITS:%=%_32): %_32: %.c
> +$(TARGETS_C_32BIT_ALL:%=%_32): %_32: %.c
> $(CC) -m32 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl
>
> $(TARGETS_C_BOTHBITS:%=%_64): %_64: %.c
> diff --git a/tools/testing/selftests/x86/entry_from_vm86.c b/tools/testing/selftests/x86/entry_from_vm86.c
> new file mode 100644
> index 000000000000..5c38a187677b
> --- /dev/null
> +++ b/tools/testing/selftests/x86/entry_from_vm86.c
> @@ -0,0 +1,114 @@
> +/*
> + * entry_from_vm86.c - tests kernel entries from vm86 mode
> + * Copyright (c) 2014-2015 Andrew Lutomirski
> + *
> + * This exercises a few paths that need to special-case vm86 mode.
> + *
> + * GPL v2.
> + */
> +
> +#define _GNU_SOURCE
> +
> +#include <assert.h>
> +#include <stdlib.h>
> +#include <sys/syscall.h>
> +#include <sys/signal.h>
> +#include <sys/ucontext.h>
> +#include <unistd.h>
> +#include <stdio.h>
> +#include <string.h>
> +#include <inttypes.h>
> +#include <sys/mman.h>
> +#include <err.h>
> +#include <stddef.h>
> +#include <stdbool.h>
> +#include <errno.h>
> +#include <sys/vm86.h>
> +
> +static unsigned long load_addr = 0x10000;
> +static int nerrs = 0;
> +
> +asm (
> + ".pushsection .rodata\n\t"
> + ".type vmcode_bound, @object\n\t"
> + "vmcode:\n\t"
> + "vmcode_bound:\n\t"
> + ".code16\n\t"
> + "bound %ax, (2048)\n\t"
> + "int3\n\t"
> + "vmcode_sysenter:\n\t"
> + "sysenter\n\t"
> + ".size vmcode, . - vmcode\n\t"
> + "end_vmcode:\n\t"
> + ".code32\n\t"
> + ".popsection"
> + );
> +
> +extern unsigned char vmcode[], end_vmcode[];
> +extern unsigned char vmcode_bound[], vmcode_sysenter[];
> +
> +static void do_test(struct vm86plus_struct *v86, unsigned long eip,
> + const char *text)
> +{
> + long ret;
> +
> + printf("[RUN]\t%s from vm86 mode\n", text);
> + v86->regs.eip = eip;
> + ret = vm86(VM86_ENTER, v86);
> +
> + if (ret == -1 && errno == ENOSYS) {
> + printf("[SKIP]\tvm86 not supported\n");
> + return;
> + }
> +
> + if (VM86_TYPE(ret) == VM86_INTx) {
> + char trapname[32];
> + int trapno = VM86_ARG(ret);
> + if (trapno == 13)
> + strcpy(trapname, "GP");
> + else if (trapno == 5)
> + strcpy(trapname, "BR");
> + else if (trapno == 14)
> + strcpy(trapname, "PF");
> + else
> + sprintf(trapname, "%d", trapno);
> +
> + printf("[OK]\tExited vm86 mode due to #%s\n", trapname);
> + } else if (VM86_TYPE(ret) == VM86_UNKNOWN) {
> + printf("[OK]\tExited vm86 mode due to unhandled GP fault\n");
> + } else {
> + printf("[OK]\tExited vm86 mode due to type %ld, arg %ld\n",
> + VM86_TYPE(ret), VM86_ARG(ret));
> + }
> +}
> +
> +int main(void)
> +{
> + struct vm86plus_struct v86;
> + unsigned char *addr = mmap((void *)load_addr, 4096,
> + PROT_READ | PROT_WRITE | PROT_EXEC,
> + MAP_ANONYMOUS | MAP_PRIVATE, -1,0);
> + if (addr != (unsigned char *)load_addr)
> + err(1, "mmap");
> +
> + memcpy(addr, vmcode, end_vmcode - vmcode);
> + addr[2048] = 2;
> + addr[2050] = 3;
> +
> + memset(&v86, 0, sizeof(v86));
> +
> + v86.regs.cs = load_addr / 16;
> + v86.regs.ss = load_addr / 16;
> + v86.regs.ds = load_addr / 16;
> + v86.regs.es = load_addr / 16;
> +
> + assert((v86.regs.cs & 3) == 0); /* Looks like RPL = 0 */
> +
> + /* #BR -- should deliver SIG??? */
> + do_test(&v86, vmcode_bound - vmcode, "#BR");
> +
> + /* SYSENTER -- should cause #GP or #UD depending on CPU */
> + do_test(&v86, vmcode_sysenter - vmcode, "SYSENTER");
> +
> + return (nerrs == 0 ? 0 : 1);
> +}
>
--
Shuah Khan
Sr. Linux Kernel Developer
Open Source Innovation Group
Samsung Research America (Silicon Valley)
[email protected] | (970) 217-8978
On Mon, Jun 1, 2015 at 9:02 AM, Shuah Khan <[email protected]> wrote:
> On 05/29/2015 03:58 PM, Andy Lutomirski wrote:
>> Test a couple of special cases in 32-bit kernels for entries from
>> vm86 mode. This will OOPS both old kernels due to a bug and and
>> 4.1-rc5 due to a regression I introduced, and it should make sure
>> that the SYSENTER-from-vm86-mode hack in the kernel keeps working.
>>
>> Tests: 394838c96013 x86/asm/entry/32: Fix user_mode() misuses
>> Tests: 7ba554b5ac69 x86/asm/entry/32: Really make user_mode() work correctly for VM86 mode
>> Signed-off-by: Andy Lutomirski <[email protected]>
>> ---
>>
>> Ingo, Shuah: I think this should go in through -tip. (In general, I think
>> new x86 tests that don't have interesting interactions with the kselftest
>> infrastructure should go in through -tip, especially tests such as this one
>> that are related to recent regressions.)
>
> No problem going through tip. Could you please make sure
>
> "make kselftest" run from top level and
> tools/testing/selftests/kselftest_install.sh
>
> don't break?
Both 'make kselftest -j12 TARGETS=x86' and
../../../tools/testing/selftests/kselftest_install.sh work for me.
(The .. thing is annoying -- would it make sense to fix it to work
from the kernel tree root?)
--Andy
>
> Once the above are verified, here is my
>
> Acked-by: Shuah Khan <[email protected]>
>
> Thanks,
> -- Shuah
>
>>
>> tools/testing/selftests/x86/Makefile | 6 +-
>> tools/testing/selftests/x86/entry_from_vm86.c | 114 ++++++++++++++++++++++++++
>> 2 files changed, 118 insertions(+), 2 deletions(-)
>> create mode 100644 tools/testing/selftests/x86/entry_from_vm86.c
>>
>> diff --git a/tools/testing/selftests/x86/Makefile b/tools/testing/selftests/x86/Makefile
>> index 5bdb781163d1..9b0d8baf2934 100644
>> --- a/tools/testing/selftests/x86/Makefile
>> +++ b/tools/testing/selftests/x86/Makefile
>> @@ -5,8 +5,10 @@ include ../lib.mk
>> .PHONY: all all_32 all_64 warn_32bit_failure clean
>>
>> TARGETS_C_BOTHBITS := sigreturn single_step_syscall
>> +TARGETS_C_32BIT_ONLY := entry_from_vm86
>>
>> -BINARIES_32 := $(TARGETS_C_BOTHBITS:%=%_32)
>> +TARGETS_C_32BIT_ALL := $(TARGETS_C_BOTHBITS) $(TARGETS_C_32BIT_ONLY)
>> +BINARIES_32 := $(TARGETS_C_32BIT_ALL:%=%_32)
>> BINARIES_64 := $(TARGETS_C_BOTHBITS:%=%_64)
>>
>> CFLAGS := -O2 -g -std=gnu99 -pthread -Wall
>> @@ -32,7 +34,7 @@ all_64: $(BINARIES_64)
>> clean:
>> $(RM) $(BINARIES_32) $(BINARIES_64)
>>
>> -$(TARGETS_C_BOTHBITS:%=%_32): %_32: %.c
>> +$(TARGETS_C_32BIT_ALL:%=%_32): %_32: %.c
>> $(CC) -m32 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl
>>
>> $(TARGETS_C_BOTHBITS:%=%_64): %_64: %.c
>> diff --git a/tools/testing/selftests/x86/entry_from_vm86.c b/tools/testing/selftests/x86/entry_from_vm86.c
>> new file mode 100644
>> index 000000000000..5c38a187677b
>> --- /dev/null
>> +++ b/tools/testing/selftests/x86/entry_from_vm86.c
>> @@ -0,0 +1,114 @@
>> +/*
>> + * entry_from_vm86.c - tests kernel entries from vm86 mode
>> + * Copyright (c) 2014-2015 Andrew Lutomirski
>> + *
>> + * This exercises a few paths that need to special-case vm86 mode.
>> + *
>> + * GPL v2.
>> + */
>> +
>> +#define _GNU_SOURCE
>> +
>> +#include <assert.h>
>> +#include <stdlib.h>
>> +#include <sys/syscall.h>
>> +#include <sys/signal.h>
>> +#include <sys/ucontext.h>
>> +#include <unistd.h>
>> +#include <stdio.h>
>> +#include <string.h>
>> +#include <inttypes.h>
>> +#include <sys/mman.h>
>> +#include <err.h>
>> +#include <stddef.h>
>> +#include <stdbool.h>
>> +#include <errno.h>
>> +#include <sys/vm86.h>
>> +
>> +static unsigned long load_addr = 0x10000;
>> +static int nerrs = 0;
>> +
>> +asm (
>> + ".pushsection .rodata\n\t"
>> + ".type vmcode_bound, @object\n\t"
>> + "vmcode:\n\t"
>> + "vmcode_bound:\n\t"
>> + ".code16\n\t"
>> + "bound %ax, (2048)\n\t"
>> + "int3\n\t"
>> + "vmcode_sysenter:\n\t"
>> + "sysenter\n\t"
>> + ".size vmcode, . - vmcode\n\t"
>> + "end_vmcode:\n\t"
>> + ".code32\n\t"
>> + ".popsection"
>> + );
>> +
>> +extern unsigned char vmcode[], end_vmcode[];
>> +extern unsigned char vmcode_bound[], vmcode_sysenter[];
>> +
>> +static void do_test(struct vm86plus_struct *v86, unsigned long eip,
>> + const char *text)
>> +{
>> + long ret;
>> +
>> + printf("[RUN]\t%s from vm86 mode\n", text);
>> + v86->regs.eip = eip;
>> + ret = vm86(VM86_ENTER, v86);
>> +
>> + if (ret == -1 && errno == ENOSYS) {
>> + printf("[SKIP]\tvm86 not supported\n");
>> + return;
>> + }
>> +
>> + if (VM86_TYPE(ret) == VM86_INTx) {
>> + char trapname[32];
>> + int trapno = VM86_ARG(ret);
>> + if (trapno == 13)
>> + strcpy(trapname, "GP");
>> + else if (trapno == 5)
>> + strcpy(trapname, "BR");
>> + else if (trapno == 14)
>> + strcpy(trapname, "PF");
>> + else
>> + sprintf(trapname, "%d", trapno);
>> +
>> + printf("[OK]\tExited vm86 mode due to #%s\n", trapname);
>> + } else if (VM86_TYPE(ret) == VM86_UNKNOWN) {
>> + printf("[OK]\tExited vm86 mode due to unhandled GP fault\n");
>> + } else {
>> + printf("[OK]\tExited vm86 mode due to type %ld, arg %ld\n",
>> + VM86_TYPE(ret), VM86_ARG(ret));
>> + }
>> +}
>> +
>> +int main(void)
>> +{
>> + struct vm86plus_struct v86;
>> + unsigned char *addr = mmap((void *)load_addr, 4096,
>> + PROT_READ | PROT_WRITE | PROT_EXEC,
>> + MAP_ANONYMOUS | MAP_PRIVATE, -1,0);
>> + if (addr != (unsigned char *)load_addr)
>> + err(1, "mmap");
>> +
>> + memcpy(addr, vmcode, end_vmcode - vmcode);
>> + addr[2048] = 2;
>> + addr[2050] = 3;
>> +
>> + memset(&v86, 0, sizeof(v86));
>> +
>> + v86.regs.cs = load_addr / 16;
>> + v86.regs.ss = load_addr / 16;
>> + v86.regs.ds = load_addr / 16;
>> + v86.regs.es = load_addr / 16;
>> +
>> + assert((v86.regs.cs & 3) == 0); /* Looks like RPL = 0 */
>> +
>> + /* #BR -- should deliver SIG??? */
>> + do_test(&v86, vmcode_bound - vmcode, "#BR");
>> +
>> + /* SYSENTER -- should cause #GP or #UD depending on CPU */
>> + do_test(&v86, vmcode_sysenter - vmcode, "SYSENTER");
>> +
>> + return (nerrs == 0 ? 0 : 1);
>> +}
>>
>
>
> --
> Shuah Khan
> Sr. Linux Kernel Developer
> Open Source Innovation Group
> Samsung Research America (Silicon Valley)
> [email protected] | (970) 217-8978
--
Andy Lutomirski
AMA Capital Management, LLC
On 06/01/2015 04:54 PM, Andy Lutomirski wrote:
> On Mon, Jun 1, 2015 at 9:02 AM, Shuah Khan <[email protected]> wrote:
>> On 05/29/2015 03:58 PM, Andy Lutomirski wrote:
>>> Test a couple of special cases in 32-bit kernels for entries from
>>> vm86 mode. This will OOPS both old kernels due to a bug and and
>>> 4.1-rc5 due to a regression I introduced, and it should make sure
>>> that the SYSENTER-from-vm86-mode hack in the kernel keeps working.
>>>
>>> Tests: 394838c96013 x86/asm/entry/32: Fix user_mode() misuses
>>> Tests: 7ba554b5ac69 x86/asm/entry/32: Really make user_mode() work correctly for VM86 mode
>>> Signed-off-by: Andy Lutomirski <[email protected]>
>>> ---
>>>
>>> Ingo, Shuah: I think this should go in through -tip. (In general, I think
>>> new x86 tests that don't have interesting interactions with the kselftest
>>> infrastructure should go in through -tip, especially tests such as this one
>>> that are related to recent regressions.)
>>
>> No problem going through tip. Could you please make sure
>>
>> "make kselftest" run from top level and
>> tools/testing/selftests/kselftest_install.sh
>>
>> don't break?
>
> Both 'make kselftest -j12 TARGETS=x86' and
> ../../../tools/testing/selftests/kselftest_install.sh work for me.
> (The .. thing is annoying -- would it make sense to fix it to work
> from the kernel tree root?)
>
What is annoying about it and how would it get solved by making it
run from the kernel tree root?
-- Shuah
--
Shuah Khan
Sr. Linux Kernel Developer
Open Source Innovation Group
Samsung Research America (Silicon Valley)
[email protected] | (970) 217-8978
On Mon, Jun 1, 2015 at 4:27 PM, Shuah Khan <[email protected]> wrote:
> On 06/01/2015 04:54 PM, Andy Lutomirski wrote:
>> On Mon, Jun 1, 2015 at 9:02 AM, Shuah Khan <[email protected]> wrote:
>>> On 05/29/2015 03:58 PM, Andy Lutomirski wrote:
>>>> Test a couple of special cases in 32-bit kernels for entries from
>>>> vm86 mode. This will OOPS both old kernels due to a bug and and
>>>> 4.1-rc5 due to a regression I introduced, and it should make sure
>>>> that the SYSENTER-from-vm86-mode hack in the kernel keeps working.
>>>>
>>>> Tests: 394838c96013 x86/asm/entry/32: Fix user_mode() misuses
>>>> Tests: 7ba554b5ac69 x86/asm/entry/32: Really make user_mode() work correctly for VM86 mode
>>>> Signed-off-by: Andy Lutomirski <[email protected]>
>>>> ---
>>>>
>>>> Ingo, Shuah: I think this should go in through -tip. (In general, I think
>>>> new x86 tests that don't have interesting interactions with the kselftest
>>>> infrastructure should go in through -tip, especially tests such as this one
>>>> that are related to recent regressions.)
>>>
>>> No problem going through tip. Could you please make sure
>>>
>>> "make kselftest" run from top level and
>>> tools/testing/selftests/kselftest_install.sh
>>>
>>> don't break?
>>
>> Both 'make kselftest -j12 TARGETS=x86' and
>> ../../../tools/testing/selftests/kselftest_install.sh work for me.
>> (The .. thing is annoying -- would it make sense to fix it to work
>> from the kernel tree root?)
>>
>
> What is annoying about it and how would it get solved by making it
> run from the kernel tree root?
>
I ran it from the kernel tree root and it told me to change my
directory. I think it should have changed the directory itself :)
--Andy
On 06/01/2015 05:45 PM, Andy Lutomirski wrote:
> On Mon, Jun 1, 2015 at 4:27 PM, Shuah Khan <[email protected]> wrote:
>> On 06/01/2015 04:54 PM, Andy Lutomirski wrote:
>>> On Mon, Jun 1, 2015 at 9:02 AM, Shuah Khan <[email protected]> wrote:
>>>> On 05/29/2015 03:58 PM, Andy Lutomirski wrote:
>>>>> Test a couple of special cases in 32-bit kernels for entries from
>>>>> vm86 mode. This will OOPS both old kernels due to a bug and and
>>>>> 4.1-rc5 due to a regression I introduced, and it should make sure
>>>>> that the SYSENTER-from-vm86-mode hack in the kernel keeps working.
>>>>>
>>>>> Tests: 394838c96013 x86/asm/entry/32: Fix user_mode() misuses
>>>>> Tests: 7ba554b5ac69 x86/asm/entry/32: Really make user_mode() work correctly for VM86 mode
>>>>> Signed-off-by: Andy Lutomirski <[email protected]>
>>>>> ---
>>>>>
>>>>> Ingo, Shuah: I think this should go in through -tip. (In general, I think
>>>>> new x86 tests that don't have interesting interactions with the kselftest
>>>>> infrastructure should go in through -tip, especially tests such as this one
>>>>> that are related to recent regressions.)
>>>>
>>>> No problem going through tip. Could you please make sure
>>>>
>>>> "make kselftest" run from top level and
>>>> tools/testing/selftests/kselftest_install.sh
>>>>
>>>> don't break?
>>>
>>> Both 'make kselftest -j12 TARGETS=x86' and
>>> ../../../tools/testing/selftests/kselftest_install.sh work for me.
>>> (The .. thing is annoying -- would it make sense to fix it to work
>>> from the kernel tree root?)
>>>
>>
>> What is annoying about it and how would it get solved by making it
>> run from the kernel tree root?
>>
>
> I ran it from the kernel tree root and it told me to change my
> directory. I think it should have changed the directory itself :)
>
ok that makes sense. I can make it do that. :)
thanks,
-- Shuah
--
Shuah Khan
Sr. Linux Kernel Developer
Open Source Innovation Group
Samsung Research America (Silicon Valley)
[email protected] | (970) 217-8978