PERFORCE change 31143 for review

Peter Wemm peter at FreeBSD.org
Tue May 13 23:59:36 PDT 2003


http://perforce.freebsd.org/chv.cgi?CH=31143

Change 31143 by peter at peter_hammer on 2003/05/13 23:59:13

	reimplement the lazy critical section masking stuff.

Affected files ...

.. //depot/projects/hammer/sys/amd64/amd64/critical.c#7 edit
.. //depot/projects/hammer/sys/amd64/amd64/exception.S#14 edit
.. //depot/projects/hammer/sys/amd64/amd64/genassym.c#14 edit
.. //depot/projects/hammer/sys/amd64/include/critical.h#6 edit
.. //depot/projects/hammer/sys/amd64/include/pcpu.h#6 edit
.. //depot/projects/hammer/sys/amd64/include/proc.h#6 edit
.. //depot/projects/hammer/sys/amd64/isa/icu_vector.S#4 edit
.. //depot/projects/hammer/sys/amd64/isa/intr_machdep.c#6 edit

Differences ...

==== //depot/projects/hammer/sys/amd64/amd64/critical.c#7 (text+ko) ====

@@ -19,24 +19,148 @@
 #include <machine/critical.h>
 
 /*
- * cpu_critical_fork_exit() - cleanup after fork
+ * XXX this mess to get sched_ithd() and call_fast_unpend()
+ */
+#include <sys/bus.h>
+#include <machine/frame.h>
+#include <machine/segments.h>
+#include <amd64/isa/icu.h>
+#include <amd64/isa/intr_machdep.h>
+
+void amd64_unpend(void);	/* NOTE: not static, called from assembly */
+
+/*
+ * cpu_unpend() -	called from critical_exit() inline after quick
+ *			interrupt-pending check.
  */
 void
-cpu_critical_fork_exit(void)
+cpu_unpend(void)
 {
+	register_t rflags;
 	struct thread *td;
 
 	td = curthread;
-	td->td_critnest = 1;
-	td->td_md.md_savecrit = read_rflags() | PSL_I;
+	rflags = intr_disable();
+	if (PCPU_GET(int_pending)) {
+		++td->td_intr_nesting_level;
+		amd64_unpend();
+		--td->td_intr_nesting_level;
+	}
+	intr_restore(rflags);
+}
+
+/*
+ * cpu_critical_fork_exit() - cleanup after fork
+ *
+ *	For amd64 we do not have to do anything, td_critnest is
+ *	handled by the fork trampoline code.
+ */
+void
+cpu_critical_fork_exit(void)
+{
 }
 
 /*
  * cpu_thread_link() - thread linkup, initialize machine-dependant fields
+ *
+ *	There are currently no machine-dependant fields that require 
+ *	initialization.
  */
 void
 cpu_thread_link(struct thread *td)
 {
+}
+
+/*
+ * Called from cpu_unpend or called from the assembly vector code
+ * to process any interrupts which may have occured while we were in
+ * a critical section.
+ *
+ * 	- interrupts must be disabled
+ *	- td_critnest must be 0
+ *	- td_intr_nesting_level must be incremented by the caller
+ *
+ * NOT STATIC (called from assembly)
+ */
+static __inline u_int
+bsfq(u_long mask)
+{
+        u_long   result;
+   
+        __asm __volatile("bsfq %1,%0" : "=r" (result) : "rm" (mask));
+        return (result);
+} 
+
+void
+amd64_unpend(void)
+{
+	struct clockframe frame;
+
+	frame.cf_cs = SEL_KPL;
+	frame.cf_rip = (register_t)amd64_unpend;
+	frame.cf_rflags = PSL_KERNEL;
+	KASSERT(curthread->td_critnest == 0, ("unpend critnest != 0"));
+	KASSERT((read_rflags() & PSL_I) == 0, ("unpend interrupts enabled1"));
+	curthread->td_critnest = 1;
+	for (;;) {
+		u_int64_t mask;
+		int irq;
+
+		/*
+		 * Fast interrupts have priority
+		 */
+		if ((mask = PCPU_GET(fpending)) != 0) {
+			irq = bsfq(mask);
+			PCPU_SET(fpending, mask & ~(1ul << irq));
+			call_fast_unpend(irq);
+			KASSERT((read_rflags() & PSL_I) == 0,
+			    ("unpend interrupts enabled2 %d", irq));
+			continue;
+		}
 
-	td->td_md.md_savecrit = 0;
+		/*
+		 * Threaded interrupts come next
+		 */
+		if ((mask = PCPU_GET(ipending)) != 0) {
+			irq = bsfq(mask);
+			PCPU_SET(ipending, mask & ~(1ul << irq));
+			sched_ithd((void *)(uintptr_t)irq);
+			KASSERT((read_rflags() & PSL_I) == 0,
+			    ("unpend interrupts enabled3 %d", irq));
+			continue;
+		}
+
+		/*
+		 * Software interrupts and delayed IPIs are last
+		 *
+		 * XXX give the bits #defined names.  see also
+		 * isa/xxx_vector.s
+		 */
+		if ((mask = PCPU_GET(spending)) != 0) {
+			irq = bsfq(mask);
+			PCPU_SET(spending, mask & ~(1ul << irq));
+			switch(irq) {
+			case 0:		/* bit 0 - hardclock */
+				hardclock_process(&frame);
+				break;
+			case 1:		/* bit 1 - statclock */
+				if (profprocs != 0)
+					profclock(&frame);
+				if (pscnt == psdiv)
+					statclock(&frame);
+				break;
+			}
+			KASSERT((read_rflags() & PSL_I) == 0,
+			    ("unpend interrupts enabled4 %d", irq));
+			continue;
+		}
+		break;
+	}
+	/*
+	 * Interrupts are still disabled, we can safely clear int_pending 
+	 * and td_critnest.
+	 */
+	KASSERT((read_rflags() & PSL_I) == 0, ("unpend interrupts enabled5"));
+	PCPU_SET(int_pending, 0);
+	curthread->td_critnest = 0;
 }

==== //depot/projects/hammer/sys/amd64/amd64/exception.S#14 (text+ko) ====

@@ -297,6 +297,9 @@
 	movq	%r12, %rdi		/* function */
 	movq	%rbx, %rsi		/* arg1 */
 	movq	%rsp, %rdx		/* trapframe pointer */
+	movq	PCPU(CURTHREAD),%rbx	/* setup critnest */
+	movl	$1,TD_CRITNEST(%rbx)
+	sti
 	call	fork_exit
 	MEXITCOUNT
 	jmp	doreti			/* Handle any ASTs */

==== //depot/projects/hammer/sys/amd64/amd64/genassym.c#14 (text+ko) ====

@@ -181,6 +181,10 @@
 ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
 ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
 ASSYM(PC_SCRATCH_RSP, offsetof(struct pcpu, pc_scratch_rsp));
+ASSYM(PC_INT_PENDING, offsetof(struct pcpu, pc_int_pending));
+ASSYM(PC_IPENDING, offsetof(struct pcpu, pc_ipending));
+ASSYM(PC_FPENDING, offsetof(struct pcpu, pc_fpending));
+ASSYM(PC_SPENDING, offsetof(struct pcpu, pc_spending));
 
 ASSYM(KCSEL, GSEL(GCODE_SEL, SEL_KPL));
 ASSYM(KDSEL, GSEL(GDATA_SEL, SEL_KPL));

==== //depot/projects/hammer/sys/amd64/include/critical.h#6 (text+ko) ====

@@ -23,6 +23,7 @@
 /*
  * Prototypes - see <arch>/<arch>/critical.c
  */
+void cpu_unpend(void);
 void cpu_critical_fork_exit(void);
 void cpu_thread_link(struct thread *td);
 
@@ -33,15 +34,12 @@
  *
  *	This routine is called from critical_enter() on the 0->1 transition
  *	of td_critnest, prior to it being incremented to 1.
+ *
+ *	If new-style critical section handling we do not have to do anything.
+ *	However, as a side effect any interrupts occuring while td_critnest
+ *	is non-zero will be deferred.
  */
-static __inline void
-cpu_critical_enter(void)
-{
-	struct thread *td;
-
-	td = curthread;
-	td->td_md.md_savecrit = intr_disable();
-}
+#define cpu_critical_enter()
 
 /*
  *	cpu_critical_exit:
@@ -49,14 +47,27 @@
  *	This routine is called from critical_exit() on a 1->0 transition
  *	of td_critnest, after it has been decremented to 0.  We are
  *	exiting the last critical section.
+ *
+ *	Note that the td->critnest (1->0) transition interrupt race against
+ *	our int_pending/unpend() check below is handled by the interrupt
+ *	code for us, so we do not have to do anything fancy.
  */
 static __inline void
 cpu_critical_exit(void)
 {
-	struct thread *td;
-
-	td = curthread;
-	intr_restore(td->td_md.md_savecrit);
+	/*
+	 * We may have to schedule pending interrupts.  Create
+	 * conditions similar to an interrupt context and call
+	 * unpend().
+	 *
+	 * note: we do this even if we are in an interrupt
+	 * nesting level.  Deep nesting is protected by
+	 * critical_*() and if we conditionalized it then we
+	 * would have to check int_pending again whenever
+	 * we decrement td_intr_nesting_level to 0.
+	 */
+	if (PCPU_GET(int_pending))
+		cpu_unpend();
 }
 
 #else /* !__GNUC__ */

==== //depot/projects/hammer/sys/amd64/include/pcpu.h#6 (text+ko) ====

@@ -40,7 +40,12 @@
  */
 #define	PCPU_MD_FIELDS							\
 	struct	pcpu *pc_prvspace;	/* Self-reference */		\
-	register_t pc_scratch_rsp;	/* User %rsp in syscall */
+	register_t pc_scratch_rsp;	/* User %rsp in syscall */	\
+	u_int64_t pc_int_pending;	/* master int pending flag */	\
+	u_int64_t pc_ipending;		/* pending slow interrupts */	\
+	u_int64_t pc_fpending;		/* pending fast interrupts */	\
+	u_int64_t pc_spending		/* pending soft interrupts */
+
 
 #if defined(lint)
  

==== //depot/projects/hammer/sys/amd64/include/proc.h#6 (text+ko) ====

@@ -41,7 +41,7 @@
  * Machine-dependent part of the proc structure for AMD64.
  */
 struct mdthread {
-	register_t	md_savecrit;
+	int __dummy__;
 };
 
 struct mdproc {

==== //depot/projects/hammer/sys/amd64/isa/icu_vector.S#4 (text+ko) ====

@@ -4,6 +4,7 @@
  */
 
 #define	IRQ_BIT(irq_num)	(1 << ((irq_num) % 8))
+#define	IRQ_LBIT(irq_num)	(1 << (irq_num))
 #define	IRQ_BYTE(irq_num)	((irq_num) >> 3)
 
 #define	ENABLE_ICU1							\
@@ -15,12 +16,11 @@
 	outb	%al,$IO_ICU2 ;	/* but do second icu first ... */	\
 	outb	%al,$IO_ICU1	/* ... then first icu */
 
-
 /*
  * Macros for interrupt interrupt entry, call to handler, and exit.
  */
 
-#define	FAST_INTR(irq_num, vec_name, enable_icus)			\
+#define	FAST_INTR(irq_num, vec_name, icu, enable_icus)			\
 	.text ;								\
 	SUPERALIGN_TEXT ;						\
 IDTVEC(vec_name) ;							\
@@ -44,8 +44,19 @@
 	movq	%r14,TF_R14(%rsp) ;					\
 	movq	%r15,TF_R15(%rsp) ;					\
 	FAKE_MCOUNT((12)*4(%rsp)) ;					\
-	call	critical_enter ;					\
 	movq	PCPU(CURTHREAD),%rbx ;					\
+	cmpl	$0,TD_CRITNEST(%rbx) ;					\
+	je	1f ;							\
+	movq	$1,PCPU(INT_PENDING) ;					\
+	orq	$IRQ_LBIT(irq_num),PCPU(FPENDING) ;			\
+	movb	imen + IRQ_BYTE(irq_num),%al ;				\
+	orb	$IRQ_BIT(irq_num),%al ;					\
+	movb	%al,imen + IRQ_BYTE(irq_num) ;				\
+	outb	%al,$icu+ICU_IMR_OFFSET ;				\
+	enable_icus ;							\
+	jmp	10f ;							\
+1: ;									\
+	incl	TD_CRITNEST(%rbx) ;					\
 	incl	TD_INTR_NESTING_LEVEL(%rbx) ;				\
 	movq	intr_unit + (irq_num) * 8, %rdi ;			\
 	call	*intr_handler + (irq_num) * 8 ;	/* do the work ASAP */	\
@@ -53,11 +64,47 @@
 	incl	cnt+V_INTR ;	/* book-keeping can wait */		\
 	movq	intr_countp + (irq_num) * 8,%rax ;			\
 	incq	(%rax) ;						\
+	decl	TD_CRITNEST(%rbx) ;					\
+	cmpq	$0,PCPU(INT_PENDING) ;					\
+	je	2f ;							\
+	call	amd64_unpend ;						\
+2: ;									\
 	decl	TD_INTR_NESTING_LEVEL(%rbx) ;				\
-	call	critical_exit ;						\
+10: ;									\
 	MEXITCOUNT ;							\
 	jmp	doreti
 
+/*
+ * Restart a fast interrupt that was held up by a critical section.
+ * This routine is called from unpend().  unpend() ensures we are
+ * in a critical section and deals with the interrupt nesting level
+ * for us.  If we previously masked the irq, we have to unmask it.
+ *
+ * We have a choice.  We can regenerate the irq using the 'int'
+ * instruction or we can create a dummy frame and call the interrupt
+ * handler directly.  I've chosen to use the dummy-frame method.
+ */
+#define	FAST_UNPEND(irq_num, vec_name, icu)				\
+	.text ;								\
+	SUPERALIGN_TEXT ;						\
+IDTVEC(vec_name) ;							\
+	pushfq ;		/* rflags */				\
+	mov	%cs,%ax ;						\
+	pushq	%rax ;		/* cs */				\
+	pushq	24(%rsp) ;	/* original caller rip */		\
+	subq	$TF_RIP,%rsp ;	/* skip rest including tf_err etc */	\
+	movq	intr_unit + (irq_num) * 8, %rdi ;			\
+	call	*intr_handler + (irq_num) * 8 ;	/* do the work ASAP */	\
+	incl	cnt+V_INTR ;	/* book-keeping can wait */		\
+	movq	intr_countp + (irq_num) * 8,%rax ;			\
+	incq	(%rax) ;						\
+	movb	imen + IRQ_BYTE(irq_num),%al ;				\
+	andb	$~IRQ_BIT(irq_num),%al ;				\
+	movb	%al,imen + IRQ_BYTE(irq_num) ;				\
+	outb	%al,$icu+ICU_IMR_OFFSET	;				\
+	addq	$TF_RSP,%rsp ; /* dump frame */				\
+	ret
+
 /* 
  * Slow, threaded interrupts.
  *
@@ -75,7 +122,8 @@
 	testb	$SEL_RPL_MASK,TF_CS(%rsp) ; /* come from kernel? */	\
 	jz	1f ;		/* Yes, dont swapgs again */		\
 	swapgs ;							\
-1:	movq	%rdi,TF_RDI(%rsp) ;					\
+1: ;									\
+	movq	%rdi,TF_RDI(%rsp) ;					\
 	movq	%rsi,TF_RSI(%rsp) ;					\
 	movq	%rdx,TF_RDX(%rsp) ;					\
 	movq	%rcx,TF_RCX(%rsp) ;					\
@@ -97,33 +145,44 @@
 	outb	%al,$icu+ICU_IMR_OFFSET ;				\
 	enable_icus ;							\
 	movq	PCPU(CURTHREAD),%rbx ;					\
+	cmpl	$0,TD_CRITNEST(%rbx) ;					\
+	je	1f ;							\
+	movq	$1,PCPU(INT_PENDING) ;					\
+	orq	$IRQ_LBIT(irq_num),PCPU(IPENDING) ;			\
+	jmp	10f ;							\
+1: ;									\
 	incl	TD_INTR_NESTING_LEVEL(%rbx) ;				\
 	FAKE_MCOUNT(13*4(%rsp)) ;	/* XXX late to avoid double count */ \
+	cmpq	$0,PCPU(INT_PENDING) ;					\
+	je	9f ;							\
+	call	amd64_unpend ;						\
+9: ;									\
 	movq	$irq_num, %rdi;	/* pass the IRQ */			\
 	call	sched_ithd ;						\
 	decl	TD_INTR_NESTING_LEVEL(%rbx) ;				\
+10: ;									\
 	MEXITCOUNT ;							\
 	/* We could usually avoid the following jmp by inlining some of */ \
 	/* doreti, but it's probably better to use less cache. */	\
 	jmp	doreti
 
 MCOUNT_LABEL(bintr)
-	FAST_INTR(0,fastintr0, ENABLE_ICU1)
-	FAST_INTR(1,fastintr1, ENABLE_ICU1)
-	FAST_INTR(2,fastintr2, ENABLE_ICU1)
-	FAST_INTR(3,fastintr3, ENABLE_ICU1)
-	FAST_INTR(4,fastintr4, ENABLE_ICU1)
-	FAST_INTR(5,fastintr5, ENABLE_ICU1)
-	FAST_INTR(6,fastintr6, ENABLE_ICU1)
-	FAST_INTR(7,fastintr7, ENABLE_ICU1)
-	FAST_INTR(8,fastintr8, ENABLE_ICU1_AND_2)
-	FAST_INTR(9,fastintr9, ENABLE_ICU1_AND_2)
-	FAST_INTR(10,fastintr10, ENABLE_ICU1_AND_2)
-	FAST_INTR(11,fastintr11, ENABLE_ICU1_AND_2)
-	FAST_INTR(12,fastintr12, ENABLE_ICU1_AND_2)
-	FAST_INTR(13,fastintr13, ENABLE_ICU1_AND_2)
-	FAST_INTR(14,fastintr14, ENABLE_ICU1_AND_2)
-	FAST_INTR(15,fastintr15, ENABLE_ICU1_AND_2)
+	FAST_INTR(0,fastintr0, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(1,fastintr1, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(2,fastintr2, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(3,fastintr3, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(4,fastintr4, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(5,fastintr5, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(6,fastintr6, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(7,fastintr7, IO_ICU1, ENABLE_ICU1)
+	FAST_INTR(8,fastintr8, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(9,fastintr9, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(10,fastintr10, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(11,fastintr11, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(12,fastintr12, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(13,fastintr13, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(14,fastintr14, IO_ICU2, ENABLE_ICU1_AND_2)
+	FAST_INTR(15,fastintr15, IO_ICU2, ENABLE_ICU1_AND_2)
 
 #define	CLKINTR_PENDING	movl $1,CNAME(clkintr_pending)
 /* Threaded interrupts */
@@ -144,5 +203,21 @@
 	INTR(14,intr14, IO_ICU2, ENABLE_ICU1_AND_2,)
 	INTR(15,intr15, IO_ICU2, ENABLE_ICU1_AND_2,)
 
+	FAST_UNPEND(0,fastunpend0, IO_ICU1)
+	FAST_UNPEND(1,fastunpend1, IO_ICU1)
+	FAST_UNPEND(2,fastunpend2, IO_ICU1)
+	FAST_UNPEND(3,fastunpend3, IO_ICU1)
+	FAST_UNPEND(4,fastunpend4, IO_ICU1)
+	FAST_UNPEND(5,fastunpend5, IO_ICU1)
+	FAST_UNPEND(6,fastunpend6, IO_ICU1)
+	FAST_UNPEND(7,fastunpend7, IO_ICU1)
+	FAST_UNPEND(8,fastunpend8, IO_ICU2)
+	FAST_UNPEND(9,fastunpend9, IO_ICU2)
+	FAST_UNPEND(10,fastunpend10, IO_ICU2)
+	FAST_UNPEND(11,fastunpend11, IO_ICU2)
+	FAST_UNPEND(12,fastunpend12, IO_ICU2)
+	FAST_UNPEND(13,fastunpend13, IO_ICU2)
+	FAST_UNPEND(14,fastunpend14, IO_ICU2)
+	FAST_UNPEND(15,fastunpend15, IO_ICU2)
+
 MCOUNT_LABEL(eintr)
-

==== //depot/projects/hammer/sys/amd64/isa/intr_machdep.c#6 (text+ko) ====

@@ -87,6 +87,17 @@
 	IDTVEC(fastintr14), IDTVEC(fastintr15),
 };
 
+static unpendhand_t *fastunpend[ICU_LEN] = {
+	IDTVEC(fastunpend0), IDTVEC(fastunpend1),
+	IDTVEC(fastunpend2), IDTVEC(fastunpend3),
+	IDTVEC(fastunpend4), IDTVEC(fastunpend5),
+	IDTVEC(fastunpend6), IDTVEC(fastunpend7),
+	IDTVEC(fastunpend8), IDTVEC(fastunpend9),
+	IDTVEC(fastunpend10), IDTVEC(fastunpend11),
+	IDTVEC(fastunpend12), IDTVEC(fastunpend13),
+	IDTVEC(fastunpend14), IDTVEC(fastunpend15),
+};
+
 static inthand_t *slowintr[ICU_LEN] = {
 	IDTVEC(intr0), IDTVEC(intr1), IDTVEC(intr2), IDTVEC(intr3),
 	IDTVEC(intr4), IDTVEC(intr5), IDTVEC(intr6), IDTVEC(intr7),
@@ -521,3 +532,10 @@
 
 	return (ithread_remove_handler(cookie));
 }
+
+void
+call_fast_unpend(int irq)
+{
+
+	fastunpend[irq]();
+}


More information about the p4-projects mailing list