Skip to contentSkip to conversation

Paravirt cpu batching.patch

PATCH In review Zachary Amsden 2 participants 2 messages Dec 12, 2006

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 2/5 Paravirt cpu batching.patch

The VMI ROM has a mode where hypercalls can be queued and batched. This turns out to be a significant win during context switch, but must be done at a specific point before side effects to CPU state are visible to subsequent instructions. This is similar to the MMU batching hooks already provided. The same hooks could be used by the Xen backend to implement a context switch multicall.

To explain a bit more about lazy modes in the paravirt patches, basically, the idea is that only one of lazy CPU or MMU mode can be active at any given time. Lazy MMU mode is similar to this lazy CPU mode, and allows for batching of multiple PTE updates (say, inside a remap loop), but to avoid keeping some kind of state machine about when to flush cpu or mmu updates, we just allow one or the other to be active. Although there is no real reason a more comprehensive scheme could not be implemented, there is also no demonstrated need for this extra complexity.

arch/i386/kernel/paravirt.c +1 -0
diff -r 320f0d4d2280 arch/i386/kernel/paravirt.c
@@ -545,6 +545,7 @@ struct paravirt_ops paravirt_ops = {
 	.apic_write_atomic = native_apic_write_atomic,
 	.apic_read = native_apic_read,
 #endif
+	.set_lazy_mode = (void *)native_nop,
 
 	.flush_tlb_user = native_flush_tlb,
 	.flush_tlb_kernel = native_flush_tlb_global,
arch/i386/kernel/process.c +31 -16

59 lines not on this page · raw message

include/asm-generic/pgtable.h +13 -0
diff -r 320f0d4d2280 include/asm-generic/pgtable.h
@@ -183,6 +183,19 @@ static inline void ptep_set_wrprotect(st
 #endif
 
 /*
+ * A facility to provide batching of the reload of page tables with the
+ * actual context switch code for paravirtualized guests.  By convention,
+ * only one of the lazy modes (CPU, MMU) should be active at any given
+ * time, entry should never be nested, and entry and exits should always
+ * be paired.  This is for sanity of maintaining and reasoning about the
+ * kernel code.
+ */
+#ifndef __HAVE_ARCH_ENTER_LAZY_CPU_MODE
+#define arch_enter_lazy_cpu_mode()	do {} while (0)
+#define arch_leave_lazy_cpu_mode()	do {} while (0)
+#endif
+
+/*
  * When walking page tables, get the address of the next boundary,
  * or the end address of the range if that comes earlier.  Although no
  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
include/asm-i386/paravirt.h +15 -0
diff -r 320f0d4d2280 include/asm-i386/paravirt.h
@@ -146,6 +146,8 @@ struct paravirt_ops
 	void (fastcall *pmd_clear)(pmd_t *pmdp);
 #endif
 
+	void (fastcall *set_lazy_mode)(int mode);
+
 	/* These two are jmp to, not actually called. */
 	void (fastcall *irq_enable_sysexit)(void);
 	void (fastcall *iret)(void);
@@ -386,6 +388,19 @@ static inline void pmd_clear(pmd_t *pmdp
 }
 #endif
 
+/* Lazy mode for batching updates / context switch */
+#define PARAVIRT_LAZY_NONE 0
+#define PARAVIRT_LAZY_MMU  1
+#define PARAVIRT_LAZY_CPU  2
+
+#define  __HAVE_ARCH_ENTER_LAZY_CPU_MODE
+#define arch_enter_lazy_cpu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_CPU)
+#define arch_leave_lazy_cpu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_NONE)
+
+#define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
+#define arch_enter_lazy_mmu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_MMU)
+#define arch_leave_lazy_mmu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_NONE)
+
 /* These all sit in the .parainstructions section to tell us what to patch. */
 struct paravirt_patch {
 	u8 *instr; 		/* original instructions */
kernel/sched.c +7 -0
diff -r 320f0d4d2280 kernel/sched.c
@@ -1833,6 +1833,13 @@ context_switch(struct rq *rq, struct tas
 	struct mm_struct *mm = next->mm;
 	struct mm_struct *oldmm = prev->active_mm;
 
+	/*
+	 * For paravirt, this is coupled with an exit in switch_to to
+	 * combine the page table reload and the switch backend into
+	 * one hypercall.
+	 */
+	arch_enter_lazy_cpu_mode();
+
 	if (!mm) {
 		next->active_mm = oldmm;
 		atomic_inc(&oldmm->mm_count);
#1 lore raw
Quoted 21 lines · * Zachary Amsden ([email protected]) wrote:

The VMI ROM has a mode where hypercalls can be queued and batched. This turns out to be a significant win during context switch, but must be done at a specific point before side effects to CPU state are visible to subsequent instructions. This is similar to the MMU batching hooks already provided. The same hooks could be used by the Xen backend to implement a context switch multicall.

To explain a bit more about lazy modes in the paravirt patches, basically, the idea is that only one of lazy CPU or MMU mode can be active at any given time. Lazy MMU mode is similar to this lazy CPU mode, and allows for batching of multiple PTE updates (say, inside a remap loop), but to avoid keeping some kind of state machine about when to flush cpu or mmu updates, we just allow one or the other to be active. Although there is no real reason a more comprehensive scheme could not be implemented, there is also no demonstrated need for this extra complexity.

arch/i386/kernel/process.c struct task_struct fastcall * __switch_t · line 671

Quoted 7 lines
arch/i386/kernel/process.c +4 -0
@@ -665,6 +665,37 @@ struct task_struct fastcall * __switch_t
 	load_TLS(next, cpu);
 
 	/*
+	 * Restore IOPL if needed.
+	 */
+	if (unlikely(prev->iopl != next->iopl))
+		set_iopl_mask(next->iopl);

Small sidenote that this bit undoes a recent change from Chuck Ebbert, which killed iopl_mask update via hypervisor.

thanks,
-chris