Minor improvements to the compat vdso and sigpage

There's no need to allocate the compat vDSO and signal pages using GFP_ATOMIC allocations, so use GFP_KERNEL instead.

Signed-off-by Will Deacon <[email protected]>

---
 arch/arm64/kernel/vdso.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

arch/arm64/kernel/vdso.c +2 -2

index cee5d04ea9ad..2d057a4dc787 100644
@@ -299,7 +299,7 @@ static int aarch32_alloc_kuser_vdso_page(void)
 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 		return 0;
 
-	vdso_page = get_zeroed_page(GFP_ATOMIC);
+	vdso_page = get_zeroed_page(GFP_KERNEL);
 	if (!vdso_page)
 		return -ENOMEM;
 

@@ -316,7 +316,7 @@ static int aarch32_alloc_sigpage(void)
 	int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
 	unsigned long sigpage;
 
-	sigpage = get_zeroed_page(GFP_ATOMIC);
+	sigpage = get_zeroed_page(GFP_KERNEL);
 	if (!sigpage)
 		return -ENOMEM;
 

Quoted 4 lines · Will Deacon

There's no need to allocate the compat vDSO and signal pages using GFP_ATOMIC allocations, so use GFP_KERNEL instead.

Signed-off-by Will Deacon <[email protected]>

Nit: We could explain why GFP_KERNEL is sufficient in the commit message.

flush_dcache_page() ensures that the 'PG_dcache_clean' flag for its 'page' argument is clear so that cache maintenance will be performed if the page is mapped into userspace with execute permissions.

Newly allocated pages have this flag clear, so there is no need to call flush_dcache_page() for the compat vdso or signal pages. Remove the redundant calls.

Signed-off-by Will Deacon <[email protected]>

---
 arch/arm64/kernel/vdso.c | 2 --
 1 file changed, 2 deletions(-)

arch/arm64/kernel/vdso.c +0 -2

index 2d057a4dc787..421411981dc3 100644
@@ -306,7 +306,6 @@ static int aarch32_alloc_kuser_vdso_page(void)
 	memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
 	       kuser_sz);
 	aarch32_vectors_page = virt_to_page(vdso_page);
-	flush_dcache_page(aarch32_vectors_page);
 	return 0;
 }
 

@@ -322,7 +321,6 @@ static int aarch32_alloc_sigpage(void)
 
 	memcpy((void *)sigpage, __aarch32_sigret_code_start, sigret_sz);
 	aarch32_sig_page = virt_to_page(sigpage);
-	flush_dcache_page(aarch32_sig_page);
 	return 0;
 }
 

For compatability with 32-bit Arm, allow the compat signal page to be remapped via mremap().

Signed-off-by Will Deacon <[email protected]>

---
 arch/arm64/kernel/vdso.c | 9 +++++++++
 1 file changed, 9 insertions(+)

arch/arm64/kernel/vdso.c +9 -0

index 421411981dc3..16bf0b46fb70 100644
@@ -271,6 +271,14 @@ enum aarch32_map {
 static struct page *aarch32_vectors_page __ro_after_init;
 static struct page *aarch32_sig_page __ro_after_init;
 
+static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
+				  struct vm_area_struct *new_vma)
+{
+	current->mm->context.sigpage = (void *)new_vma->vm_start;
+
+	return 0;
+}
+
 static struct vm_special_mapping aarch32_vdso_maps[] = {
 	[AA32_MAP_VECTORS] = {
 		.name	= "[vectors]", /* ABI */

@@ -279,6 +287,7 @@ static struct vm_special_mapping aarch32_vdso_maps[] = {
 	[AA32_MAP_SIGPAGE] = {
 		.name	= "[sigpage]", /* ABI */
 		.pages	= &aarch32_sig_page,
+		.mremap	= aarch32_sigpage_mremap,
 	},
 	[AA32_MAP_VVAR] = {
 		.name = "[vvar]",

We can avoid the expensive ISB instruction after reading the counter in the vDSO gettime functions by creating a fake address hazard against a dummy stack read, just like we do inside the kernel.

Signed-off-by Will Deacon <[email protected]>

---
 arch/arm64/include/asm/arch_timer.h        | 21 ---------------------
 arch/arm64/include/asm/barrier.h           | 19 +++++++++++++++++++
 arch/arm64/include/asm/vdso/gettimeofday.h |  6 +-----
 3 files changed, 20 insertions(+), 26 deletions(-)

arch/arm64/include/asm/arch_timer.h +0 -21

index 9f0ec21d6327..88d20f04c64a 100644
@@ -165,25 +165,6 @@ static inline void arch_timer_set_cntkctl(u32 cntkctl)
 	isb();
 }
 
-/*
- * Ensure that reads of the counter are treated the same as memory reads
- * for the purposes of ordering by subsequent memory barriers.
- *
- * This insanity brought to you by speculative system register reads,
- * out-of-order memory accesses, sequence locks and Thomas Gleixner.
- *
- * http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631195.html
- */
-#define arch_counter_enforce_ordering(val) do {				\
-	u64 tmp, _val = (val);						\
-									\
-	asm volatile(							\
-	"	eor	%0, %1, %1\n"					\
-	"	add	%0, sp, %0\n"					\
-	"	ldr	xzr, [%0]"					\
-	: "=r" (tmp) : "r" (_val));					\
-} while (0)
-
 static __always_inline u64 __arch_counter_get_cntpct_stable(void)
 {
 	u64 cnt;

@@ -224,8 +205,6 @@ static __always_inline u64 __arch_counter_get_cntvct(void)
 	return cnt;
 }
 
-#undef arch_counter_enforce_ordering
-
 static inline int arch_timer_arch_init(void)
 {
 	return 0;

arch/arm64/include/asm/barrier.h +19 -0

index c3009b0e5239..37d891af8ea5 100644
@@ -70,6 +70,25 @@ static inline unsigned long array_index_mask_nospec(unsigned long idx,
 	return mask;
 }
 
+/*
+ * Ensure that reads of the counter are treated the same as memory reads
+ * for the purposes of ordering by subsequent memory barriers.
+ *
+ * This insanity brought to you by speculative system register reads,
+ * out-of-order memory accesses, sequence locks and Thomas Gleixner.
+ *
+ * http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631195.html
+ */
+#define arch_counter_enforce_ordering(val) do {				\
+	u64 tmp, _val = (val);						\
+									\
+	asm volatile(							\
+	"	eor	%0, %1, %1\n"					\
+	"	add	%0, sp, %0\n"					\
+	"	ldr	xzr, [%0]"					\
+	: "=r" (tmp) : "r" (_val));					\
+} while (0)
+
 #define __smp_mb()	dmb(ish)
 #define __smp_rmb()	dmb(ishld)
 #define __smp_wmb()	dmb(ishst)

arch/arm64/include/asm/vdso/gettimeofday.h +1 -5

index 631ab1281633..4b4c0dac0e14 100644
@@ -83,11 +83,7 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode,
 	 */
 	isb();
 	asm volatile("mrs %0, cntvct_el0" : "=r" (res) :: "memory");
-	/*
-	 * This isb() is required to prevent that the seq lock is
-	 * speculated.#
-	 */
-	isb();
+	arch_counter_enforce_ordering(res);
 
 	return res;
 }

Commit 9c698bff66ab ("ARM: ensure the signal page contains defined contents") poisoned the unused portions of the signal page for 32-bit Arm.

Implement the same poisoning for the compat signal page on arm64 rather than using __GFP_ZERO.

Signed-off-by Will Deacon <[email protected]>

---
 arch/arm64/kernel/vdso.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

arch/arm64/kernel/vdso.c +6 -3

index 16bf0b46fb70..159b72a646ab 100644
@@ -318,17 +318,20 @@ static int aarch32_alloc_kuser_vdso_page(void)
 	return 0;
 }
 
+#define COMPAT_SIGPAGE_POISON_WORD	0xe7fddef1
 static int aarch32_alloc_sigpage(void)
 {
 	extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
 	int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
-	unsigned long sigpage;
+	__le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
+	void *sigpage;
 
-	sigpage = get_zeroed_page(GFP_KERNEL);
+	sigpage = (void *)__get_free_page(GFP_KERNEL);
 	if (!sigpage)
 		return -ENOMEM;
 
-	memcpy((void *)sigpage, __aarch32_sigret_code_start, sigret_sz);
+	memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
+	memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
 	aarch32_sig_page = virt_to_page(sigpage);
 	return 0;
 }

Here are a few small improvements to the compat vDSO and sigpage implementation for arm64 which I cooked up when comparing the code with the native 32-bit counterpart.

Cheers,

Will

[...]