On 25/08/2017 13:57, Yu Zhang wrote:
> Provide paging mode switching logic to run access test in 5
> level paging mode if LA57 is detected. Qemu parameter +la57
> should be used to expose this feature, for example:
> ./x86-run ./x86/access.flat -cpu qemu64,+la57
>
> Signed-off-by: Yu Zhang <yu.c.zhang@xxxxxxxxxxxxxxx>
> ---
> x86/access.c | 17 +++++++++++++++--
> x86/cstart64.S | 43 ++++++++++++++++++++++++++++++++++++++++++-
> 2 files changed, 57 insertions(+), 3 deletions(-)
I need this for the test to pass in QEMU. Probably the TLB is helping
you on real hardware! Pushed with this change.
diff --git a/x86/access.c b/x86/access.c
index 640ca8c..c99c5c1 100644
--- a/x86/access.c
+++ b/x86/access.c
@@ -322,7 +322,7 @@ pt_element_t ac_test_alloc_pt(ac_pool_t *pool)
_Bool ac_test_enough_room(ac_pool_t *pool)
{
- return pool->pt_pool_current + 4 * PAGE_SIZE <= pool->pt_pool_size;
+ return pool->pt_pool_current + 5 * PAGE_SIZE <= pool->pt_pool_size;
}
void ac_test_reset_pt_pool(ac_pool_t *pool)
@@ -467,6 +467,7 @@ void __ac_setup_specific_pages(ac_test_t *at, ac_pool_t *pool, u64 pd_page,
{
unsigned long root = read_cr3();
int flags = at->flags;
+ bool skip = true;
if (!ac_test_enough_room(pool))
ac_test_reset_pt_pool(pool);
@@ -476,6 +477,16 @@ void __ac_setup_specific_pages(ac_test_t *at, ac_pool_t *pool, u64 pd_page,
pt_element_t *vroot = va(root & PT_BASE_ADDR_MASK);
unsigned index = PT_INDEX((unsigned long)at->virt, i);
pt_element_t pte = 0;
+
+ /*
+ * Reuse existing page tables along the path to the test code and data
+ * (which is in the bottom 2MB).
+ */
+ if (skip && i >= 2 && index == 0) {
+ goto next;
+ }
+ skip = false;
+
switch (i) {
case 5:
case 4:
@@ -535,6 +546,7 @@ void __ac_setup_specific_pages(ac_test_t *at, ac_pool_t *pool, u64 pd_page,
break;
}
vroot[index] = pte;
+ next:
root = vroot[index];
}
ac_set_expected_status(at);
> diff --git a/x86/access.c b/x86/access.c
> index 0546dbb..56d17a1 100644
> --- a/x86/access.c
> +++ b/x86/access.c
> @@ -15,6 +15,7 @@ typedef unsigned long pt_element_t;
> static int cpuid_7_ebx;
> static int cpuid_7_ecx;
> static int invalid_mask;
> +static int page_table_levels;
>
> #define PT_BASE_ADDR_MASK ((pt_element_t)((((pt_element_t)1 << 40) - 1) & PAGE_MASK))
> #define PT_PSE_BASE_ADDR_MASK (PT_BASE_ADDR_MASK & ~(1ull << 21))
> @@ -107,6 +108,8 @@ enum {
> #define AC_CPU_CR4_SMEP_MASK (1 << AC_CPU_CR4_SMEP_BIT)
> #define AC_CPU_CR4_PKE_MASK (1 << AC_CPU_CR4_PKE_BIT)
>
> +extern void setup_5level_page_table();
> +
> const char *ac_names[] = {
> [AC_PTE_PRESENT_BIT] = "pte.p",
> [AC_PTE_ACCESSED_BIT] = "pte.a",
> @@ -467,11 +470,12 @@ void __ac_setup_specific_pages(ac_test_t *at, ac_pool_t *pool, u64 pd_page,
> ac_test_reset_pt_pool(pool);
>
> at->ptep = 0;
> - for (int i = 4; i >= 1 && (i >= 2 || !F(AC_PDE_PSE)); --i) {
> + for (int i = page_table_levels; i >= 1 && (i >= 2 || !F(AC_PDE_PSE)); --i) {
> pt_element_t *vroot = va(root & PT_BASE_ADDR_MASK);
> unsigned index = PT_INDEX((unsigned long)at->virt, i);
> pt_element_t pte = 0;
> switch (i) {
> + case 5:
> case 4:
> case 3:
> pte = pd_page ? pd_page : ac_test_alloc_pt(pool);
> @@ -552,7 +556,7 @@ static void dump_mapping(ac_test_t *at)
> int i;
>
> printf("Dump mapping: address: %p\n", at->virt);
> - for (i = 4; i >= 1 && (i >= 2 || !F(AC_PDE_PSE)); --i) {
> + for (i = page_table_levels ; i >= 1 && (i >= 2 || !F(AC_PDE_PSE)); --i) {
> pt_element_t *vroot = va(root & PT_BASE_ADDR_MASK);
> unsigned index = PT_INDEX((unsigned long)at->virt, i);
> pt_element_t pte = vroot[index];
> @@ -986,6 +990,15 @@ int main()
> cpuid_7_ecx = cpuid(7).c;
>
> printf("starting test\n\n");
> + page_table_levels = 4;
> r = ac_test_run();
> +
> + if (cpuid_7_ecx & (1 << 16)) {
> + page_table_levels = 5;
> + setup_5level_page_table();
> + printf("starting 5-level paging test.\n\n");
> + r = ac_test_run();
> + }
> +
> return r ? 0 : 1;
> }
> diff --git a/x86/cstart64.S b/x86/cstart64.S
> index 4c26fb2..8e4a1f3 100644
> --- a/x86/cstart64.S
> +++ b/x86/cstart64.S
> @@ -45,6 +45,10 @@ ptl4:
> .quad ptl3 + 7
>
> .align 4096
> +ptl5:
> + .quad ptl4 + 7
> +
> +.align 4096
>
> gdt64_desc:
> .word gdt64_end - gdt64 - 1
> @@ -91,6 +95,8 @@ tss_end:
>
> mb_boot_info: .quad 0
>
> +pt_root: .quad ptl4
> +
> .section .init
>
> .code32
> @@ -119,14 +125,36 @@ start:
> call prepare_64
> jmpl $8, $start64
>
> +switch_to_5level:
> + /* Disable CR4.PCIDE */
> + mov %cr4, %eax
> + btr $17, %eax
> + mov %eax, %cr4
> +
> + mov %cr0, %eax
> + btr $31, %eax
> + mov %eax, %cr0
> +
> + mov $ptl5, %eax
> + mov %eax, pt_root
> +
> + /* Enable CR4.LA57 */
> + mov %cr4, %eax
> + bts $12, %eax
> + mov %eax, %cr4
> +
> + call enter_long_mode
> + jmpl $8, $lvl5
> +
> prepare_64:
> lgdt gdt64_desc
>
> +enter_long_mode:
> mov %cr4, %eax
> bts $5, %eax // pae
> mov %eax, %cr4
>
> - mov $ptl4, %eax
> + mov pt_root, %eax
> mov %eax, %cr3
>
> efer = 0xc0000080
> @@ -211,6 +239,19 @@ start64:
> mov %eax, %edi
> call exit
>
> +.globl setup_5level_page_table
> +setup_5level_page_table:
> + /* Check if 5-level paging has already enabled */
> + mov %cr4, %rax
> + test $12, %eax
> + jnz lvl5
> +
> + pushq $32
> + pushq $switch_to_5level
> + lretq
> +lvl5:
> + retq
> +
> idt_descr:
> .word 16 * 256 - 1
> .quad boot_idt
>
