Date: Sat, 25 May 1996 03:39:11 -0700 (PDT) From: asami@cs.berkeley.edu (Satoshi Asami) To: bde@zeta.org.au Cc: current@freebsd.org, ccd@stampede.cs.berkeley.edu Subject: More on kernel bcopy Message-ID: <199605251039.DAA21280@silvia.HIP.Berkeley.EDU>
next in thread | raw e-mail | index | archive | help
Bruce,
I tried calling fastmove from bcopy. Unfortunately, it crashed right
after it loaded. I remember you mentioning that the fnsave/frstor
part is only used when we call it from bcopy and an interrupt handler
calls bcopy. Maybe there is a bug in there still.
Here's the diff.
===
Index: support.s
===================================================================
RCS file: /usr/cvs/src/sys/i386/i386/support.s,v
retrieving revision 1.35
diff -u -r1.35 support.s
--- support.s 1996/05/03 21:01:00 1.35
+++ support.s 1996/05/25 09:30:20
@@ -291,6 +291,18 @@
subl %esi,%eax
cmpl %ecx,%eax /* overlapping? */
jb 1f
+#ifdef I586_FAST_BCOPY
+ cmpl $128,%ecx
+ jbe slow_bcopy
+
+ jmp slow_bcopy /* XXX take this out and see it crash */
+
+ call fastmove
+ jmp done_bcopy
+
+ ALIGN_TEXT
+slow_bcopy:
+#endif /* I586_FAST_BCOPY */
shrl $2,%ecx /* copy by 32-bit words */
cld /* nope, copy forwards */
rep
@@ -299,6 +311,9 @@
andl $3,%ecx /* any bytes left? */
rep
movsb
+#ifdef I586_FAST_BCOPY
+done_bcopy:
+#endif /* I586_FAST_BCOPY */
popl %edi
popl %esi
ret
@@ -453,6 +468,16 @@
/* bcopy(%esi, %edi, %ebx) */
3:
movl %ebx,%ecx
+#ifdef I586_FAST_BCOPY
+ cmpl $128,%ecx
+ jbe slow_copyout
+
+ call fastmove
+ jmp done_copyout
+
+ ALIGN_TEXT
+slow_copyout:
+#endif /* I586_FAST_BCOPY */
shrl $2,%ecx
cld
rep
@@ -500,6 +525,16 @@
cmpl $VM_MAXUSER_ADDRESS,%edx
ja copyin_fault
+#ifdef I586_FAST_BCOPY
+ cmpl $128,%ecx
+ jbe slow_copyin
+
+ call fastmove
+ jmp done_copyin
+
+ ALIGN_TEXT
+slow_copyin:
+#endif /* I586_FAST_BCOPY */
movb %cl,%al
shrl $2,%ecx /* copy longword-wise */
cld
@@ -510,6 +545,10 @@
rep
movsb
+#ifdef I586_FAST_BCOPY
+ ALIGN_TEXT
+done_copyin:
+#endif /* I586_FAST_BCOPY */
popl %edi
popl %esi
xorl %eax,%eax
@@ -525,6 +564,206 @@
movl $0,PCB_ONFAULT(%edx)
movl $EFAULT,%eax
ret
+
+#ifdef I586_FAST_BCOPY
+/* fastmove(src, dst, len)
+ src in %esi
+ dst in %edi
+ len in %ecx
+ uses %eax and %edx for tmp. storage
+ */
+/*
+LC0:
+ .ascii "npxproc == curproc\0"
+LC1:
+ .ascii "support.s"
+ */
+ ALIGN_TEXT
+fastmove:
+ cmpl $63,%ecx
+ jbe fastmove_tail
+
+ testl $7,%esi /* check if src addr is multiple of 8 */
+ jnz fastmove_tail
+
+ testl $7,%edi /* check if dst addr is multiple of 8 */
+ jnz fastmove_tail
+
+ pushl %ebp
+ movl %esp,%ebp
+ subl $176,%esp
+
+/* if (intr_nesting_level > 0) */
+ cmpb $0,_intr_nesting_level
+ je L6
+/* save reentrantly */
+ movl %cr0,%edx
+ clts
+ fnsave -176(%ebp)
+ jmp L7
+
+/* else { */
+ ALIGN_TEXT
+L6:
+/* if (npxproc != NULL) { */
+ cmpl $0,_npxproc
+ je L8
+/* assert(npxproc == curproc); */
+/* movl _npxproc,%eax
+ cmpl %eax,_curproc
+ je L6b
+ pushl LC0
+ pushl $599
+ pushl LC1
+ call ___assert
+ addl $12,%esp
+L6b: */
+/* fnsave(&curpcb->pcb_savefpu); */
+ movl _curpcb,%eax
+ fnsave 112(%eax)
+/* npxproc = NULL; */
+ movl $0,_npxproc
+/* } */
+L8:
+/* now we own the FPU. */
+
+/*
+ * The process' FP state is saved in the pcb, but if we get
+ * switched, the cpu_switch() will store our FP state in the
+ * pcb. It should be possible to avoid all the copying for
+ * this, e.g., by setting a flag to tell cpu_switch() to
+ * save the state somewhere else.
+ */
+/* tmp = curpcb->pcb_savefpu; */
+ pushl %edi
+ pushl %esi
+ pushl %ecx
+ leal -176(%ebp),%edi
+ movl _curpcb,%esi
+ addl $112,%esi
+ cld
+ movl $44,%ecx
+ rep
+ movsl
+ popl %ecx
+ popl %esi
+ popl %edi
+/* stop_emulating(); */
+ clts
+/* npxproc = curproc; */
+ movl _curproc,%eax
+ movl %eax,_npxproc
+/* } */
+L7:
+4:
+ pushl %ecx
+ cmpl $1792,%ecx
+ jbe 2f
+ movl $1792,%ecx
+2:
+ subl %ecx,0(%esp)
+ cmpl $256,%ecx
+ jb 5f
+ pushl %esi
+ pushl %ecx
+ .align 4,0x90
+3:
+ movl 0(%esi),%eax
+ movl 32(%esi),%eax
+ movl 64(%esi),%eax
+ movl 96(%esi),%eax
+ movl 128(%esi),%eax
+ movl 160(%esi),%eax
+ movl 192(%esi),%eax
+ movl 224(%esi),%eax
+ addl $256,%esi
+ subl $256,%ecx
+ cmpl $256,%ecx
+ jae 3b
+ popl %ecx
+ popl %esi
+5:
+ ALIGN_TEXT
+fastmove_loop:
+ fildq 0(%esi)
+ fildq 8(%esi)
+ fildq 16(%esi)
+ fildq 24(%esi)
+ fildq 32(%esi)
+ fildq 40(%esi)
+ fildq 48(%esi)
+ fildq 56(%esi)
+ fistpq 56(%edi)
+ fistpq 48(%edi)
+ fistpq 40(%edi)
+ fistpq 32(%edi)
+ fistpq 24(%edi)
+ fistpq 16(%edi)
+ fistpq 8(%edi)
+ fistpq 0(%edi)
+ addl $-64,%ecx
+ addl $64,%esi
+ addl $64,%edi
+ cmpl $63,%ecx
+ ja fastmove_loop
+ popl %eax
+ addl %eax,%ecx
+ cmpl $64,%ecx
+ jae 4b
+
+/* if (intr_nesting_level > 0) */
+
+ cmpb $0,_intr_nesting_level
+ je L9
+
+/* Restore reentrantly. */
+ frstor -176(%ebp)
+ movl %edx,%cr0
+ jmp L10
+
+/* else { */
+ ALIGN_TEXT
+L9:
+/* curpcb->pcb_savefpu = tmp; */
+ pushl %edi
+ pushl %esi
+ pushl %ecx
+ movl _curpcb,%edi
+ addl $112,%edi
+ leal -176(%ebp),%esi
+ cld
+ movl $44,%ecx
+ rep
+ movsl
+ popl %ecx
+ popl %esi
+ popl %edi
+
+/* start_emulating(); */
+ smsw %ax
+ orb $8,%al
+ lmsw %ax
+/* npxproc = NULL; */
+ movl $0,_npxproc
+/* } */
+L10:
+ movl %ebp,%esp
+ popl %ebp
+
+ ALIGN_TEXT
+fastmove_tail:
+ movb %cl,%al
+ shrl $2,%ecx /* copy longword-wise */
+ cld
+ rep
+ movsl
+ movb %al,%cl
+ andb $3,%cl /* copy remaining bytes */
+ rep
+ movsb
+
+ ret
+#endif /* I586_FAST_BCOPY */
/*
* fu{byte,sword,word} : fetch a byte (sword, word) from user memory
Index: trap.c
===================================================================
RCS file: /usr/cvs/src/sys/i386/i386/trap.c,v
retrieving revision 1.76
diff -u -r1.76 trap.c
--- trap.c 1996/05/18 03:36:19 1.76
+++ trap.c 1996/05/18 11:23:39
@@ -319,6 +319,14 @@
(void) trap_pfault(&frame, FALSE);
return;
+ case T_DNA:
+#if NNPX > 0
+ /* if a transparent fault (due to context switch "late") */
+ if (npxdna())
+ return;
+#endif /* NNPX > 0 */
+ break;
+
case T_PROTFLT: /* general protection fault */
case T_SEGNPFLT: /* segment not present fault */
/*
===
As it is, it will check the copy size but won't call fastmove. When
you take out this line:
jmp slow_bcopy /* XXX take this out and see it crash */
it should crash with fireworks on the screen right after the kernel
load.
Satoshi
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605251039.DAA21280>