Date:   Mon,  4 May 2020 16:49:36 +0200
From:   Alexandre Chartre <alexandre.chartre@...cle.com>
To:     rkrcmar@...hat.com, tglx@...utronix.de, mingo@...hat.com,
        bp@...en8.de, hpa@...or.com, dave.hansen@...ux.intel.com,
        luto@...nel.org, peterz@...radead.org, x86@...nel.org,
        linux-mm@...ck.org, linux-kernel@...r.kernel.org
Cc:     pbonzini@...hat.com, konrad.wilk@...cle.com,
        jan.setjeeilers@...cle.com, liran.alon@...cle.com,
        junaids@...gle.com, graf@...zon.de, rppt@...ux.vnet.ibm.com,
        kuzuno@...il.com, mgross@...ux.intel.com,
        alexandre.chartre@...cle.com
Subject: [RFC v4][PATCH part-1 4/7] mm/asi: Interrupt ASI on interrupt/exception/NMI

If an interrupt/exception/NMI is triggered while using ASI then
ASI is interrupted and the system switches back to the (kernel)
page-table used before entering ASI.

When the interrupt/exception/NMI handler returns then ASI is
resumed by switching back to the ASI page-table.

Signed-off-by: Alexandre Chartre <alexandre.chartre@...cle.com>
---
 arch/x86/entry/calling.h           |  26 +++++-
 arch/x86/entry/entry_64.S          |  22 ++++++
 arch/x86/include/asm/asi.h         | 122 +++++++++++++++++++++++++++++
 arch/x86/include/asm/asi_session.h |   7 ++
 arch/x86/include/asm/mmu_context.h |   3 +-
 arch/x86/kernel/asm-offsets.c      |   5 ++
 arch/x86/mm/asi.c                  |  67 ++++++++++++++--
 7 files changed, 242 insertions(+), 10 deletions(-)

diff --git a/arch/x86/entry/calling.h b/arch/x86/entry/calling.h
index 0789e13ece90..ca23b79adecf 100644
--- a/arch/x86/entry/calling.h
+++ b/arch/x86/entry/calling.h
@@ -6,6 +6,7 @@
 #include <asm/percpu.h>
 #include <asm/asm-offsets.h>
 #include <asm/processor-flags.h>
+#include <asm/asi.h>
 
 /*
 
@@ -172,7 +173,30 @@ For 32-bit we have the following conventions - kernel is built with
 	.endif
 .endm
 
-#ifdef CONFIG_PAGE_TABLE_ISOLATION
+#if defined(CONFIG_ADDRESS_SPACE_ISOLATION)
+
+/*
+ * For now, ASI is not compatible with PTI.
+ */
+
+.macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
+.endm
+
+.macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req
+.endm
+
+.macro SWITCH_TO_USER_CR3_STACK	scratch_reg:req
+.endm
+
+.macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req
+	ASI_INTERRUPT_AND_SAVE_CR3 \scratch_reg \save_reg
+.endm
+
+.macro RESTORE_CR3 scratch_reg:req save_reg:req
+	ASI_RESUME_AND_RESTORE_CR3 \save_reg
+.endm
+
+#elif defined(CONFIG_PAGE_TABLE_ISOLATION)
 
 /*
  * PAGE_TABLE_ISOLATION PGDs are 8k.  Flip bit 12 to switch between the two
diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index 0e9504fabe52..ac47da63a29f 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -573,7 +573,15 @@ SYM_CODE_START(interrupt_entry)
 
 	CALL_enter_from_user_mode
 
+#ifdef CONFIG_ADDRESS_SPACE_ISOLATION
+	jmp	2f
+#endif
 1:
+#ifdef CONFIG_ADDRESS_SPACE_ISOLATION
+	/* Interrupt address space isolation if it is active */
+	ASI_INTERRUPT scratch_reg=%rdi
+2:
+#endif
 	ENTER_IRQ_STACK old_rsp=%rdi save_ret=1
 	/* We entered an interrupt context - irqs are off: */
 	TRACE_IRQS_OFF
@@ -673,6 +681,10 @@ retint_kernel:
 	jnz	1f
 	call	preempt_schedule_irq
 1:
+#endif
+#ifdef CONFIG_ADDRESS_SPACE_ISOLATION
+	ASI_PREPARE_RESUME
+	ASI_RESUME scratch_reg=%rdi
 #endif
 	/*
 	 * The iretq could re-enable interrupts:
@@ -1238,6 +1250,9 @@ SYM_CODE_START_LOCAL(paranoid_entry)
 	 * This is also why CS (stashed in the "iret frame" by the
 	 * hardware at entry) can not be used: this may be a return
 	 * to kernel code, but with a user CR3 value.
+	 *
+	 * If ASI is enabled, this also handles the case where we are
+	 * using an ASI CR3 value.
 	 */
 	SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14
 
@@ -1313,6 +1328,13 @@ SYM_CODE_START_LOCAL(error_entry)
 
 .Lerror_entry_done_lfence:
 	FENCE_SWAPGS_KERNEL_ENTRY
+#ifdef CONFIG_ADDRESS_SPACE_ISOLATION
+	/*
+	 * Interrupt address space isolation if it is active. This will restore
+	 * the original kernel CR3.
+	 */
+	ASI_INTERRUPT scratch_reg=%rdi
+#endif
 .Lerror_entry_done:
 	ret
 
diff --git a/arch/x86/include/asm/asi.h b/arch/x86/include/asm/asi.h
index bcfb68e8e392..d240954b2f85 100644
--- a/arch/x86/include/asm/asi.h
+++ b/arch/x86/include/asm/asi.h
@@ -108,6 +108,128 @@ extern void asi_set_pagetable(struct asi *asi, pgd_t *pagetable);
 extern int asi_enter(struct asi *asi);
 extern void asi_exit(struct asi *asi);
 
+#else  /* __ASSEMBLY__ */
+
+#include <asm/alternative-asm.h>
+#include <asm/asm-offsets.h>
+#include <asm/cpufeatures.h>
+#include <asm/percpu.h>
+#include <asm/processor-flags.h>
+
+#define THIS_ASI_SESSION_asi		\
+	PER_CPU_VAR(cpu_tlbstate + TLB_STATE_asi)
+#define THIS_ASI_SESSION_isolation_cr3	\
+	PER_CPU_VAR(cpu_tlbstate + TLB_STATE_asi_isolation_cr3)
+#define THIS_ASI_SESSION_original_cr3	\
+	PER_CPU_VAR(cpu_tlbstate + TLB_STATE_asi_original_cr3)
+#define THIS_ASI_SESSION_idepth	\
+	PER_CPU_VAR(cpu_tlbstate + TLB_STATE_asi_idepth)
+
+.macro SET_NOFLUSH_BIT	reg:req
+	bts	$X86_CR3_PCID_NOFLUSH_BIT, \reg
+.endm
+
+/*
+ * Switch CR3 to the original kernel CR3 value. This is used when exiting
+ * interrupting ASI.
+ */
+.macro ASI_SWITCH_TO_KERNEL_CR3 scratch_reg:req
+	/*
+	 * KERNEL pages can always resume with NOFLUSH as we do
+	 * explicit flushes.
+	 */
+	movq	THIS_ASI_SESSION_original_cr3, \scratch_reg
+	ALTERNATIVE "", "SET_NOFLUSH_BIT \scratch_reg", X86_FEATURE_PCID
+	movq	\scratch_reg, %cr3
+.endm
+
+/*
+ * Interrupt ASI, when there's an interrupt or exception while we
+ * were running with ASI.
+ */
+.macro ASI_INTERRUPT scratch_reg:req
+	movq	THIS_ASI_SESSION_asi, \scratch_reg
+	testq	\scratch_reg, \scratch_reg
+	jz	.Lasi_interrupt_done_\@
+	incl	THIS_ASI_SESSION_idepth
+	cmp	$1, THIS_ASI_SESSION_idepth
+	jne	.Lasi_interrupt_done_\@
+	ASI_SWITCH_TO_KERNEL_CR3 \scratch_reg
+.Lasi_interrupt_done_\@:
+.endm
+
+.macro ASI_PREPARE_RESUME
+	call	asi_prepare_resume
+.endm
+
+/*
+ * Resume ASI, after it was interrupted by an interrupt or an exception.
+ */
+.macro ASI_RESUME scratch_reg:req
+	movq	THIS_ASI_SESSION_asi, \scratch_reg
+	testq	\scratch_reg, \scratch_reg
+	jz	.Lasi_resume_done_\@
+	decl	THIS_ASI_SESSION_idepth
+	jnz	.Lasi_resume_done_\@
+	movq	THIS_ASI_SESSION_isolation_cr3, \scratch_reg
+	mov	\scratch_reg, %cr3
+.Lasi_resume_done_\@:
+.endm
+
+/*
+ * Interrupt ASI, special processing when ASI is interrupted by a NMI
+ * or a paranoid interrupt/exception.
+ */
+.macro ASI_INTERRUPT_AND_SAVE_CR3 scratch_reg:req save_reg:req
+	movq	%cr3, \save_reg
+	/*
+	 * Test the ASI PCID bits. If set, then an ASI page table
+	 * is active. If clear, CR3 already has the kernel page table
+	 * active.
+	 */
+	bt	$ASI_PGTABLE_BIT, \save_reg
+	jnc	.Ldone_\@
+	incl	THIS_ASI_SESSION_idepth
+	ASI_SWITCH_TO_KERNEL_CR3 \scratch_reg
+.Ldone_\@:
+.endm
+
+/*
+ * Resume ASI, special processing when ASI is resumed from a NMI
+ * or a paranoid interrupt/exception.
+ */
+.macro ASI_RESUME_AND_RESTORE_CR3 save_reg:req
+
+	ALTERNATIVE "jmp .Lwrite_cr3_\@", "", X86_FEATURE_PCID
+
+	bt	$ASI_PGTABLE_BIT, \save_reg
+	jnc	.Lrestore_kernel_cr3_\@
+
+	/*
+	 * Restore ASI CR3. We need to update TLB flushing
+	 * information.
+	 */
+	movq	THIS_ASI_SESSION_asi, %rdi
+	movq	\save_reg, %rsi
+	call	asi_update_flush
+	movq	%rax, THIS_ASI_SESSION_isolation_cr3
+	decl	THIS_ASI_SESSION_idepth
+	movq	%rax, %cr3
+	jmp	.Ldone_\@
+
+.Lrestore_kernel_cr3_\@:
+	/*
+	 * Restore kernel CR3. KERNEL pages can always resume
+	 * with NOFLUSH as we do explicit flushes.
+	 */
+	SET_NOFLUSH_BIT \save_reg
+
+.Lwrite_cr3_\@:
+	movq	\save_reg, %cr3
+
+.Ldone_\@:
+.endm
+
 #endif	/* __ASSEMBLY__ */
 
 #endif	/* CONFIG_ADDRESS_SPACE_ISOLATION */
diff --git a/arch/x86/include/asm/asi_session.h b/arch/x86/include/asm/asi_session.h
index 9d39c936a4ee..85968f7e8f32 100644
--- a/arch/x86/include/asm/asi_session.h
+++ b/arch/x86/include/asm/asi_session.h
@@ -10,6 +10,13 @@ struct asi_session {
 	struct asi		*asi;		/* ASI for this session */
 	unsigned long		isolation_cr3;	/* cr3 when ASI is active */
 	unsigned long		original_cr3;	/* cr3 before entering ASI */
+	/*
+	 * The interrupt depth (idepth) tracks interrupt (actually
+	 * interrupt/exception/NMI) nesting. ASI is interrupted on
+	 * the first interrupt, and it is resumed when that interrupt
+	 * handler returns.
+	 */
+	unsigned int		idepth;		/* interrupt depth */
 };
 
 #endif	/* CONFIG_ADDRESS_SPACE_ISOLATION */
diff --git a/arch/x86/include/asm/mmu_context.h b/arch/x86/include/asm/mmu_context.h
index 9b03bad00b81..b8c81e7b197a 100644
--- a/arch/x86/include/asm/mmu_context.h
+++ b/arch/x86/include/asm/mmu_context.h
@@ -243,7 +243,8 @@ static inline unsigned long __get_current_cr3_fast(void)
 	 * field of the ASI session.
 	 */
 	if (IS_ENABLED(CONFIG_ADDRESS_SPACE_ISOLATION) &&
-	    this_cpu_read(cpu_asi_session.asi)) {
+	    this_cpu_read(cpu_asi_session.asi) &&
+	    !this_cpu_read(cpu_asi_session.idepth)) {
 		cr3 = this_cpu_read(cpu_asi_session.isolation_cr3);
 		/* CR3 read never returns with the NOFLUSH bit */
 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
index 3ca07ad552ae..4c08a688b4b9 100644
--- a/arch/x86/kernel/asm-offsets.c
+++ b/arch/x86/kernel/asm-offsets.c
@@ -94,6 +94,11 @@ static void __used common(void)
 
 	/* TLB state for the entry code */
 	OFFSET(TLB_STATE_user_pcid_flush_mask, tlb_state, user_pcid_flush_mask);
+	OFFSET(TLB_STATE_asi, tlb_state, asi_session.asi);
+	OFFSET(TLB_STATE_asi_isolation_cr3, tlb_state,
+	       asi_session.isolation_cr3);
+	OFFSET(TLB_STATE_asi_original_cr3, tlb_state, asi_session.original_cr3);
+	OFFSET(TLB_STATE_asi_idepth, tlb_state, asi_session.idepth);
 
 	/* Layout info for cpu_entry_area */
 	OFFSET(CPU_ENTRY_AREA_entry_stack, cpu_entry_area, entry_stack_page);
diff --git a/arch/x86/mm/asi.c b/arch/x86/mm/asi.c
index cf0d122a3c72..c91ba82a095b 100644
--- a/arch/x86/mm/asi.c
+++ b/arch/x86/mm/asi.c
@@ -68,7 +68,7 @@ EXPORT_SYMBOL(asi_set_pagetable);
  * Return an updated ASI CR3 value which specified if TLB needs to
  * be flushed or not.
  */
-static unsigned long asi_update_flush(struct asi *asi, unsigned long asi_cr3)
+unsigned long asi_update_flush(struct asi *asi, unsigned long asi_cr3)
 {
 	struct asi_tlb_pgtable *tlb_pgtable;
 	struct asi_tlb_state *tlb_state;
@@ -90,7 +90,24 @@ static unsigned long asi_update_flush(struct asi *asi, unsigned long asi_cr3)
 	return asi_cr3;
 }
 
-static void asi_switch_to_asi_cr3(struct asi *asi)
+
+/*
+ * Switch to the ASI pagetable.
+ *
+ * If schedule is ASI_SWITCH_NOW, then immediately switch to the ASI
+ * pagetable by updating the CR3 register with the ASI CR3 value.
+ * Otherwise, if schedule is ASI_SWITCH_ON_RESUME, prepare everything
+ * for switching to ASI pagetable but do not update the CR3 register
+ * yet. This will be done by the next ASI_RESUME call.
+ */
+
+enum asi_switch_schedule {
+	ASI_SWITCH_NOW,
+	ASI_SWITCH_ON_RESUME,
+};
+
+static void asi_switch_to_asi_cr3(struct asi *asi,
+				  enum asi_switch_schedule schedule)
 {
 	unsigned long original_cr3, asi_cr3;
 	struct asi_session *asi_session;
@@ -114,8 +131,16 @@ static void asi_switch_to_asi_cr3(struct asi *asi)
 	asi_session->original_cr3 = original_cr3;
 	asi_session->isolation_cr3 = asi_cr3;
 
-	/* Update CR3 to immediately enter ASI */
-	native_write_cr3(asi_cr3);
+	if (schedule == ASI_SWITCH_ON_RESUME) {
+		/*
+		 * Defer the CR3 update the next ASI resume by setting
+		 * the interrupt depth to 1.
+		 */
+		asi_session->idepth = 1;
+	} else {
+		/* Update CR3 to immediately enter ASI */
+		native_write_cr3(asi_cr3);
+	}
 }
 
 static void asi_switch_to_kernel_cr3(struct asi *asi)
@@ -132,6 +157,7 @@ static void asi_switch_to_kernel_cr3(struct asi *asi)
 
 	asi_session = &get_cpu_var(cpu_asi_session);
 	asi_session->asi = NULL;
+	asi_session->idepth = 0;
 }
 
 int asi_enter(struct asi *asi)
@@ -153,7 +179,7 @@ int asi_enter(struct asi *asi)
 	}
 
 	local_irq_save(flags);
-	asi_switch_to_asi_cr3(asi);
+	asi_switch_to_asi_cr3(asi, ASI_SWITCH_NOW);
 	local_irq_restore(flags);
 
 	return 0;
@@ -162,8 +188,10 @@ EXPORT_SYMBOL(asi_enter);
 
 void asi_exit(struct asi *asi)
 {
+	struct asi_session *asi_session;
 	struct asi *current_asi;
 	unsigned long flags;
+	int idepth;
 
 	current_asi = this_cpu_read(cpu_asi_session.asi);
 	if (!current_asi) {
@@ -173,8 +201,31 @@ void asi_exit(struct asi *asi)
 
 	WARN_ON(current_asi != asi);
 
-	local_irq_save(flags);
-	asi_switch_to_kernel_cr3(asi);
-	local_irq_restore(flags);
+	idepth = this_cpu_read(cpu_asi_session.idepth);
+	if (!idepth) {
+		local_irq_save(flags);
+		asi_switch_to_kernel_cr3(asi);
+		local_irq_restore(flags);
+	} else {
+		/*
+		 * ASI was interrupted so we already switched back
+		 * to the back to the kernel page table and we just
+		 * need to clear the ASI session.
+		 */
+		asi_session = &get_cpu_var(cpu_asi_session);
+		asi_session->asi = NULL;
+		asi_session->idepth = 0;
+	}
 }
 EXPORT_SYMBOL(asi_exit);
+
+void asi_prepare_resume(void)
+{
+	struct asi_session *asi_session;
+
+	asi_session = &get_cpu_var(cpu_asi_session);
+	if (!asi_session->asi || asi_session->idepth > 1)
+		return;
+
+	asi_switch_to_asi_cr3(asi_session->asi, ASI_SWITCH_ON_RESUME);
+}
-- 
2.18.2

Powered by blists - more mailing lists