[PATCH] x86/hyperv: don't use percpu areas for pcpu_flush/pcpu_flush_ex structures

[PATCH] x86/hyperv: don't use percpu areas for pcpu_flush/pcpu_flush_ex structures

[Vitaly Kuznetsov]

hv_do_hypercall() does virt_to_phys() translation and with some configs
(CONFIG_SLAB) this doesn't work for percpu areas, we pass wrong memory to
hypervisor and get #GP. We could use working slow_virt_to_phys() instead
but doing so kills the performance.

Move pcpu_flush/pcpu_flush_ex structures out of percpu areas and
allocate memory on first call. The additional level of indirection gives
us a small performance penalty, in future we may consider introducing
hypercall functions which avoid virt_to_phys() conversion and cache
physical addresses of pcpu_flush/pcpu_flush_ex structures somewhere.

Reported-by: Simon Xiao <sixiao@microsoft.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
---
- Please cosider this for 4.14.
---
 arch/x86/hyperv/mmu.c | 34 ++++++++++++++++++++++++++++------
 1 file changed, 28 insertions(+), 6 deletions(-)

diff --git a/arch/x86/hyperv/mmu.c b/arch/x86/hyperv/mmu.c
index 123d9b501108..b8957309bdc4 100644
--- a/arch/x86/hyperv/mmu.c
+++ b/arch/x86/hyperv/mmu.c
@@ -36,9 +36,9 @@ struct hv_flush_pcpu_ex {
 /* Each gva in gva_list encodes up to 4096 pages to flush */
 #define HV_TLB_FLUSH_UNIT (4096 * PAGE_SIZE)
 
-static struct hv_flush_pcpu __percpu *pcpu_flush;
+static struct hv_flush_pcpu __percpu **pcpu_flush;
 
-static struct hv_flush_pcpu_ex __percpu *pcpu_flush_ex;
+static struct hv_flush_pcpu_ex __percpu **pcpu_flush_ex;
 
 /*
  * Fills in gva_list starting from offset. Returns the number of items added.
@@ -102,6 +102,7 @@ static void hyperv_flush_tlb_others(const struct cpumask *cpus,
 				    const struct flush_tlb_info *info)
 {
 	int cpu, vcpu, gva_n, max_gvas;
+	struct hv_flush_pcpu **flush_pcpu;
 	struct hv_flush_pcpu *flush;
 	u64 status = U64_MAX;
 	unsigned long flags;
@@ -116,7 +117,17 @@ static void hyperv_flush_tlb_others(const struct cpumask *cpus,
 
 	local_irq_save(flags);
 
-	flush = this_cpu_ptr(pcpu_flush);
+	flush_pcpu = this_cpu_ptr(pcpu_flush);
+
+	if (unlikely(!*flush_pcpu))
+		*flush_pcpu = page_address(alloc_page(GFP_ATOMIC));
+
+	flush = *flush_pcpu;
+
+	if (unlikely(!flush)) {
+		local_irq_restore(flags);
+		goto do_native;
+	}
 
 	if (info->mm) {
 		flush->address_space = virt_to_phys(info->mm->pgd);
@@ -173,6 +184,7 @@ static void hyperv_flush_tlb_others_ex(const struct cpumask *cpus,
 				       const struct flush_tlb_info *info)
 {
 	int nr_bank = 0, max_gvas, gva_n;
+	struct hv_flush_pcpu_ex **flush_pcpu;
 	struct hv_flush_pcpu_ex *flush;
 	u64 status = U64_MAX;
 	unsigned long flags;
@@ -187,7 +199,17 @@ static void hyperv_flush_tlb_others_ex(const struct cpumask *cpus,
 
 	local_irq_save(flags);
 
-	flush = this_cpu_ptr(pcpu_flush_ex);
+	flush_pcpu = this_cpu_ptr(pcpu_flush_ex);
+
+	if (unlikely(!*flush_pcpu))
+		*flush_pcpu = page_address(alloc_page(GFP_ATOMIC));
+
+	flush = *flush_pcpu;
+
+	if (unlikely(!flush)) {
+		local_irq_restore(flags);
+		goto do_native;
+	}
 
 	if (info->mm) {
 		flush->address_space = virt_to_phys(info->mm->pgd);
@@ -266,7 +288,7 @@ void hyper_alloc_mmu(void)
 		return;
 
 	if (!(ms_hyperv.hints & HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
-		pcpu_flush = __alloc_percpu(PAGE_SIZE, PAGE_SIZE);
+		pcpu_flush = alloc_percpu(struct hv_flush_pcpu *);
 	else
-		pcpu_flush_ex = __alloc_percpu(PAGE_SIZE, PAGE_SIZE);
+		pcpu_flush_ex = alloc_percpu(struct hv_flush_pcpu_ex *);
 }
-- 
2.13.6

[tip:x86/urgent] x86/hyperv: Don't use percpu areas for pcpu_flush/pcpu_flush_ex structures

[tip-bot for Vitaly Kuznetsov]

Commit-ID:  60d73a7c96601434dfdb56d5b9167ff3b850d8d7
Gitweb:     https://git.kernel.org/tip/60d73a7c96601434dfdb56d5b9167ff3b850d8d7
Author:     Vitaly Kuznetsov <vkuznets@redhat.com>
AuthorDate: Thu, 5 Oct 2017 13:39:24 +0200
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Tue, 10 Oct 2017 12:54:56 +0200

x86/hyperv: Don't use percpu areas for pcpu_flush/pcpu_flush_ex structures

hv_do_hypercall() does virt_to_phys() translation and with some configs
(CONFIG_SLAB) this doesn't work for percpu areas, we pass wrong memory to
hypervisor and get #GP. We could use working slow_virt_to_phys() instead
but doing so kills the performance.

Move pcpu_flush/pcpu_flush_ex structures out of percpu areas and
allocate memory on first call. The additional level of indirection gives
us a small performance penalty, in future we may consider introducing
hypercall functions which avoid virt_to_phys() conversion and cache
physical addresses of pcpu_flush/pcpu_flush_ex structures somewhere.

Reported-by: Simon Xiao <sixiao@microsoft.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Dexuan Cui <decui@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Jork Loeser <Jork.Loeser@microsoft.com>
Cc: K. Y. Srinivasan <kys@microsoft.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: devel@linuxdriverproject.org
Link: http://lkml.kernel.org/r/20171005113924.28021-1-vkuznets@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/hyperv/mmu.c | 34 ++++++++++++++++++++++++++++------
 1 file changed, 28 insertions(+), 6 deletions(-)

diff --git a/arch/x86/hyperv/mmu.c b/arch/x86/hyperv/mmu.c
index 9502d04..f21cebb 100644
--- a/arch/x86/hyperv/mmu.c
+++ b/arch/x86/hyperv/mmu.c
@@ -36,9 +36,9 @@ struct hv_flush_pcpu_ex {
 /* Each gva in gva_list encodes up to 4096 pages to flush */
 #define HV_TLB_FLUSH_UNIT (4096 * PAGE_SIZE)
 
-static struct hv_flush_pcpu __percpu *pcpu_flush;
+static struct hv_flush_pcpu __percpu **pcpu_flush;
 
-static struct hv_flush_pcpu_ex __percpu *pcpu_flush_ex;
+static struct hv_flush_pcpu_ex __percpu **pcpu_flush_ex;
 
 /*
  * Fills in gva_list starting from offset. Returns the number of items added.
@@ -109,6 +109,7 @@ static void hyperv_flush_tlb_others(const struct cpumask *cpus,
 				    const struct flush_tlb_info *info)
 {
 	int cpu, vcpu, gva_n, max_gvas;
+	struct hv_flush_pcpu **flush_pcpu;
 	struct hv_flush_pcpu *flush;
 	u64 status = U64_MAX;
 	unsigned long flags;
@@ -123,7 +124,17 @@ static void hyperv_flush_tlb_others(const struct cpumask *cpus,
 
 	local_irq_save(flags);
 
-	flush = this_cpu_ptr(pcpu_flush);
+	flush_pcpu = this_cpu_ptr(pcpu_flush);
+
+	if (unlikely(!*flush_pcpu))
+		*flush_pcpu = page_address(alloc_page(GFP_ATOMIC));
+
+	flush = *flush_pcpu;
+
+	if (unlikely(!flush)) {
+		local_irq_restore(flags);
+		goto do_native;
+	}
 
 	if (info->mm) {
 		flush->address_space = virt_to_phys(info->mm->pgd);
@@ -180,6 +191,7 @@ static void hyperv_flush_tlb_others_ex(const struct cpumask *cpus,
 				       const struct flush_tlb_info *info)
 {
 	int nr_bank = 0, max_gvas, gva_n;
+	struct hv_flush_pcpu_ex **flush_pcpu;
 	struct hv_flush_pcpu_ex *flush;
 	u64 status = U64_MAX;
 	unsigned long flags;
@@ -194,7 +206,17 @@ static void hyperv_flush_tlb_others_ex(const struct cpumask *cpus,
 
 	local_irq_save(flags);
 
-	flush = this_cpu_ptr(pcpu_flush_ex);
+	flush_pcpu = this_cpu_ptr(pcpu_flush_ex);
+
+	if (unlikely(!*flush_pcpu))
+		*flush_pcpu = page_address(alloc_page(GFP_ATOMIC));
+
+	flush = *flush_pcpu;
+
+	if (unlikely(!flush)) {
+		local_irq_restore(flags);
+		goto do_native;
+	}
 
 	if (info->mm) {
 		flush->address_space = virt_to_phys(info->mm->pgd);
@@ -273,7 +295,7 @@ void hyper_alloc_mmu(void)
 		return;
 
 	if (!(ms_hyperv.hints & HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED))
-		pcpu_flush = __alloc_percpu(PAGE_SIZE, PAGE_SIZE);
+		pcpu_flush = alloc_percpu(struct hv_flush_pcpu *);
 	else
-		pcpu_flush_ex = __alloc_percpu(PAGE_SIZE, PAGE_SIZE);
+		pcpu_flush_ex = alloc_percpu(struct hv_flush_pcpu_ex *);
 }