Re: [PATCH v2] x86, mm: set NX across entire PMD at boot

From: Yinghai Lu
Date: Tue Nov 18 2014 - 02:39:51 EST


On Mon, Nov 17, 2014 at 12:27 PM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> On Sun, Nov 16, 2014 at 3:44 PM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>>
>> So the initial patch to get rid of the X mapping is of course to just
>> extend the area to the PMD. A little bit different to your initial
>> patch, but essentially the same.
>>
>> - unsigned long all_end = PFN_ALIGN(&_end);
>> + unsigned long all_end = roundup((unsigned long) &_end, PMD_SIZE);
>>
>> I'm going to apply your V1 patch with the above roundup()
>> simplification. If a page of that area gets used later on then we are
>> going to split up the PMD anyway.
>
> That's fine by me. Yinghai Lu seems to have potentially better
> solutions, but my head is spinning after reading more of this thread.
> :) I just want to make sure that at the end of the day, there are no
> RW+x mappings.
>

Please check v3 that cleanup highmap in the middle.


Before patch:
---[ High Kernel Mapping ]---
0xffffffff80000000-0xffffffff81000000 16M pmd
0xffffffff81000000-0xffffffff82200000 18M ro PSE GLB x pmd
0xffffffff82200000-0xffffffff82c00000 10M ro PSE GLB NX pmd
0xffffffff82c00000-0xffffffff82e00000 2M RW PSE GLB NX pmd
0xffffffff82e00000-0xffffffff83000000 2M RW GLB NX pte
0xffffffff83000000-0xffffffff83200000 2M RW PSE GLB NX pmd
0xffffffff83200000-0xffffffff83400000 2M RW GLB NX pte
0xffffffff83400000-0xffffffff84200000 14M RW PSE GLB NX pmd
0xffffffff84200000-0xffffffff843a2000 1672K RW GLB NX pte
0xffffffff843a2000-0xffffffff84400000 376K RW GLB x pte
0xffffffff84400000-0xffffffffa0000000 444M pmd

After patch:
---[ High Kernel Mapping ]---
0xffffffff80000000-0xffffffff81000000 16M pmd
0xffffffff81000000-0xffffffff82000000 16M ro PSE GLB x pmd
0xffffffff82000000-0xffffffff82011000 68K ro GLB x pte
0xffffffff82011000-0xffffffff82200000 1980K pte
0xffffffff82200000-0xffffffff82a00000 8M ro PSE GLB NX pmd
0xffffffff82a00000-0xffffffff82a1d000 116K ro GLB NX pte
0xffffffff82a1d000-0xffffffff82c00000 1932K pte
0xffffffff82c00000-0xffffffff82e00000 2M RW PSE GLB NX pmd
0xffffffff82e00000-0xffffffff82e52000 328K RW GLB NX pte
0xffffffff82e52000-0xffffffff83000000 1720K pte
0xffffffff83000000-0xffffffff83200000 2M pmd
0xffffffff83200000-0xffffffff83213000 76K pte
0xffffffff83213000-0xffffffff83400000 1972K RW GLB NX pte
0xffffffff83400000-0xffffffff84200000 14M RW PSE GLB NX pmd
0xffffffff84200000-0xffffffff84383000 1548K RW GLB NX pte
0xffffffff84383000-0xffffffff84400000 500K pte
0xffffffff84400000-0xffffffffa0000000 444M pmd

Thanks

Yinghai
Subject: [PATCH 1/2] x86, 64bit: add pfn_range_is_highmapped()

will use it to support holes in highmap.

Signed-off-by: Yinghai Lu <yinghai@xxxxxxxxxx>

---
arch/x86/include/asm/pgtable_64.h | 2 ++
arch/x86/mm/init_64.c | 22 ++++++++++++++++++++++
arch/x86/mm/pageattr.c | 16 +---------------
3 files changed, 25 insertions(+), 15 deletions(-)

Index: linux-2.6/arch/x86/mm/init_64.c
===================================================================
--- linux-2.6.orig/arch/x86/mm/init_64.c
+++ linux-2.6/arch/x86/mm/init_64.c
@@ -375,6 +375,23 @@ void __init init_extra_mapping_uc(unsign
__init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
}

+/* three holes at most*/
+#define NR_RANGE 4
+static struct range pfn_highmapped[NR_RANGE];
+static int nr_pfn_highmapped;
+
+int pfn_range_is_highmapped(unsigned long start_pfn, unsigned long end_pfn)
+{
+ int i;
+
+ for (i = 0; i < nr_pfn_highmapped; i++)
+ if ((start_pfn >= pfn_highmapped[i].start) &&
+ (end_pfn <= pfn_highmapped[i].end))
+ return 1;
+
+ return 0;
+}
+
/*
* The head.S code sets up the kernel high mapping:
*
@@ -409,6 +426,11 @@ void __init cleanup_highmap(void)
if (vaddr < (unsigned long) _text || vaddr > end)
set_pmd(pmd, __pmd(0));
}
+
+ nr_pfn_highmapped = add_range(pfn_highmapped, NR_RANGE,
+ nr_pfn_highmapped,
+ __pa_symbol(_text) >> PAGE_SHIFT,
+ __pa_symbol(roundup(_brk_end, PMD_SIZE)) >> PAGE_SHIFT);
}

static unsigned long __meminit
Index: linux-2.6/arch/x86/mm/pageattr.c
===================================================================
--- linux-2.6.orig/arch/x86/mm/pageattr.c
+++ linux-2.6/arch/x86/mm/pageattr.c
@@ -91,20 +91,6 @@ void arch_report_meminfo(struct seq_file
static inline void split_page_count(int level) { }
#endif

-#ifdef CONFIG_X86_64
-
-static inline unsigned long highmap_start_pfn(void)
-{
- return __pa_symbol(_text) >> PAGE_SHIFT;
-}
-
-static inline unsigned long highmap_end_pfn(void)
-{
- return __pa_symbol(roundup(_brk_end, PMD_SIZE)) >> PAGE_SHIFT;
-}
-
-#endif
-
#ifdef CONFIG_DEBUG_PAGEALLOC
# define debug_pagealloc 1
#else
@@ -1242,7 +1228,7 @@ static int cpa_process_alias(struct cpa_
* to touch the high mapped kernel as well:
*/
if (!within(vaddr, (unsigned long)_text, _brk_end) &&
- within(cpa->pfn, highmap_start_pfn(), highmap_end_pfn())) {
+ pfn_range_is_highmapped(cpa->pfn, 1)) {
unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) +
__START_KERNEL_map - phys_base;
alias_cpa = *cpa;
Index: linux-2.6/arch/x86/include/asm/pgtable_64.h
===================================================================
--- linux-2.6.orig/arch/x86/include/asm/pgtable_64.h
+++ linux-2.6/arch/x86/include/asm/pgtable_64.h
@@ -167,6 +167,8 @@ static inline int pgd_large(pgd_t pgd) {
extern int kern_addr_valid(unsigned long addr);
extern void cleanup_highmap(void);

+int pfn_range_is_highmapped(unsigned long start_pfn, unsigned long end_pfn);
+
#define HAVE_ARCH_UNMAPPED_AREA
#define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN

Subject: [PATCH 2/2] x86, 64bit: cleanup highmap late for not needed range

1. should use _brk_end instead of &_end in mark_rodata_ro().
_brk_end can move up to &_end, i.e. to __brk_limit. It's safe to
use _brk_end when mark_rodata_ro() is called because extend_brk()
is gone already at that point.
2. add cleanup_highmap_late for range for initmem, around rodata, and
[_brk_end, pmd_end]

Kernel Layout:

[ 0.000000] .text: [0x01000000-0x0200d608]
[ 0.000000] .rodata: [0x02200000-0x02a1cfff]
[ 0.000000] .data: [0x02c00000-0x02e50e7f]
[ 0.000000] .init: [0x02e52000-0x03212fff]
[ 0.000000] .bss: [0x03221000-0x0437bfff]
[ 0.000000] .brk: [0x0437c000-0x043a1fff]

Actually used brk:
[ 0.272959] memblock_reserve: [0x0000000437c000-0x00000004382fff] flags 0x0 BRK

Before patch:
---[ High Kernel Mapping ]---
0xffffffff80000000-0xffffffff81000000 16M pmd
0xffffffff81000000-0xffffffff82200000 18M ro PSE GLB x pmd
0xffffffff82200000-0xffffffff82c00000 10M ro PSE GLB NX pmd
0xffffffff82c00000-0xffffffff82e00000 2M RW PSE GLB NX pmd
0xffffffff82e00000-0xffffffff83000000 2M RW GLB NX pte
0xffffffff83000000-0xffffffff83200000 2M RW PSE GLB NX pmd
0xffffffff83200000-0xffffffff83400000 2M RW GLB NX pte
0xffffffff83400000-0xffffffff84200000 14M RW PSE GLB NX pmd
0xffffffff84200000-0xffffffff843a2000 1672K RW GLB NX pte
0xffffffff843a2000-0xffffffff84400000 376K RW GLB x pte
0xffffffff84400000-0xffffffffa0000000 444M pmd

After patch:
---[ High Kernel Mapping ]---
0xffffffff80000000-0xffffffff81000000 16M pmd
0xffffffff81000000-0xffffffff82000000 16M ro PSE GLB x pmd
0xffffffff82000000-0xffffffff82011000 68K ro GLB x pte
0xffffffff82011000-0xffffffff82200000 1980K pte
0xffffffff82200000-0xffffffff82a00000 8M ro PSE GLB NX pmd
0xffffffff82a00000-0xffffffff82a1d000 116K ro GLB NX pte
0xffffffff82a1d000-0xffffffff82c00000 1932K pte
0xffffffff82c00000-0xffffffff82e00000 2M RW PSE GLB NX pmd
0xffffffff82e00000-0xffffffff82e52000 328K RW GLB NX pte
0xffffffff82e52000-0xffffffff83000000 1720K pte
0xffffffff83000000-0xffffffff83200000 2M pmd
0xffffffff83200000-0xffffffff83213000 76K pte
0xffffffff83213000-0xffffffff83400000 1972K RW GLB NX pte
0xffffffff83400000-0xffffffff84200000 14M RW PSE GLB NX pmd
0xffffffff84200000-0xffffffff84383000 1548K RW GLB NX pte
0xffffffff84383000-0xffffffff84400000 500K pte
0xffffffff84400000-0xffffffffa0000000 444M pmd

-v3: remove all not used highmap ranges.

Signed-off-by: Yinghai Lu <yinghai@xxxxxxxxxx>

---
arch/x86/mm/init_64.c | 70 +++++++++++++++++++++++++++++++++++++++++++++++++-
1 file changed, 69 insertions(+), 1 deletion(-)

Index: linux-2.6/arch/x86/mm/init_64.c
===================================================================
--- linux-2.6.orig/arch/x86/mm/init_64.c
+++ linux-2.6/arch/x86/mm/init_64.c
@@ -1098,6 +1098,67 @@ void __init mem_init(void)
}

#ifdef CONFIG_DEBUG_RODATA
+static void remove_highmap_2m(unsigned long addr)
+{
+ pgd_t *pgd;
+ pud_t *pud;
+ pmd_t *pmd;
+
+ pgd = pgd_offset_k(addr);
+ pud = (pud_t *)pgd_page_vaddr(*pgd) + pud_index(addr);
+ pmd = (pmd_t *)pud_page_vaddr(*pud) + pmd_index(addr);
+
+ set_pmd(pmd, __pmd(0));
+}
+
+static void remove_highmap_2m_partial(unsigned long addr, unsigned long end)
+{
+ int i;
+ pgd_t *pgd;
+ pud_t *pud;
+ pmd_t *pmd;
+ pte_t *pte;
+ int start_index = pte_index(addr);
+ int end_index = pte_index(end - 1) + 1;
+
+ set_memory_4k(addr, end_index - start_index);
+
+ pgd = pgd_offset_k(addr);
+ pud = (pud_t *)pgd_page_vaddr(*pgd) + pud_index(addr);
+ pmd = (pmd_t *)pud_page_vaddr(*pud) + pmd_index(addr);
+ pte = (pte_t *)pmd_page_vaddr(*pmd) + start_index;
+
+ for (i = start_index; i < end_index; i++, pte++)
+ set_pte(pte, __pte(0));
+}
+
+static void cleanup_highmap_late(unsigned long start, unsigned long end)
+{
+ unsigned long addr;
+ unsigned long start_2m_aligned = roundup(start, PMD_SIZE);
+ unsigned long end_2m_aligned = rounddown(end, PMD_SIZE);
+
+ start = PFN_ALIGN(start);
+ end &= PAGE_MASK;
+
+ if (start >= end)
+ return;
+
+ if (start < start_2m_aligned && start_2m_aligned <= end)
+ remove_highmap_2m_partial(start, start_2m_aligned);
+
+ for (addr = start_2m_aligned; addr < end_2m_aligned; addr += PMD_SIZE)
+ remove_highmap_2m(addr);
+
+ if (start <= end_2m_aligned && end_2m_aligned < end)
+ remove_highmap_2m_partial(end_2m_aligned, end);
+
+ subtract_range(pfn_highmapped, NR_RANGE,
+ __pa_symbol(start) >> PAGE_SHIFT,
+ __pa_symbol(end) >> PAGE_SHIFT);
+ nr_pfn_highmapped = clean_sort_range(pfn_highmapped, NR_RANGE);
+}
+
const int rodata_test_data = 0xC3;
EXPORT_SYMBOL_GPL(rodata_test_data);

@@ -1146,7 +1207,8 @@ void mark_rodata_ro(void)
unsigned long end = (unsigned long) &__end_rodata_hpage_align;
unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
- unsigned long all_end = PFN_ALIGN(&_end);
+ unsigned long data_start = PFN_ALIGN(&_sdata);
+ unsigned long all_end = PFN_ALIGN(_brk_end);

printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
(end - start) >> 10);
@@ -1160,6 +1222,12 @@ void mark_rodata_ro(void)
*/
set_memory_nx(rodata_start, (all_end - rodata_start) >> PAGE_SHIFT);

+ cleanup_highmap_late(text_end, rodata_start);
+ cleanup_highmap_late(rodata_end, data_start);
+ cleanup_highmap_late(all_end, roundup(_brk_end, PMD_SIZE));
+ cleanup_highmap_late((unsigned long)(&__init_begin),
+ (unsigned long)(&__init_end));
+
rodata_test();

#ifdef CONFIG_CPA_DEBUG