1. Updated CpuFlushTlb() according to IA-64 programmer's guide

2. Updated Thunk code to preserve EFLAGS/RFLAGS across real mode code invocation
3. Updated comments in x86Thunk.c

git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@248 6f19259b-4bc3-4df7-8a09-765794883524

diff --git a/MdePkg/Library/BaseLib/BaseLib.msa b/MdePkg/Library/BaseLib/BaseLib.msa
index 7719c3af01b52f23259cf08b588b3d8a5574ceb1..0bba18d0bc4884ec37b3568bec950e8cd49c27bd 100644 (file)
--- a/MdePkg/Library/BaseLib/BaseLib.msa
+++ b/MdePkg/Library/BaseLib/BaseLib.msa
@@ -265,6 +265,7 @@ WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
     </Arch>\r
     <Arch ArchType="IPF">\r
       <Filename>Math64.c</Filename>\r
+      <Filename>Ipf/PalCallStatic.s</Filename>\r
       <Filename>Ipf/setjmp.s</Filename>\r
       <Filename>Ipf/SwitchStack.s</Filename>\r
       <Filename>Ipf/Unaligned.c</Filename>\r

diff --git a/MdePkg/Library/BaseLib/Ia32/Thunk16.asm b/MdePkg/Library/BaseLib/Ia32/Thunk16.asm
index 89d4f679b6bd9ede99cd65512d0f9c397503e25c..454b8905bc3eb876f949d3e5ac80bb99e57b9f54 100644 (file)
--- a/MdePkg/Library/BaseLib/Ia32/Thunk16.asm
+++ b/MdePkg/Library/BaseLib/Ia32/Thunk16.asm
@@ -141,6 +141,12 @@ _ToUserCode PROC
     call    @Base                       ; push eip\r
 @Base:\r
     pop     bp                          ; ebp <- offset @Base\r
+    DB      67h                         ; address size override\r
+    push    [esp + sizeof (IA32_REGS) + 2]\r
+    lea     eax, [esi + (offset @RealMode - offset @Base)]\r
+    push    eax\r
+    retf\r
+@RealMode:\r
     mov     cs:[esi + (offset SavedSs - offset @Base)], edx\r
     mov     cs:[esi + (offset SavedEsp - offset @Base)], bx\r
     DB      66h\r
@@ -208,7 +214,9 @@ InternalAsmThunk16  PROC    USES    ebp ebx esi edi ds  es  fs  gs
     push    10h\r
     pop     ecx                         ; ecx <- selector for data segments\r
     lgdt    fword ptr [edx + (offset _16Gdtr - offset SavedCr0)]\r
+    pushfd\r
     call    fword ptr [edx + (offset _EntryPoint - offset SavedCr0)]\r
+    popfd\r
     lidt    fword ptr [esp + 36]        ; restore protected mode IDTR\r
     lea     eax, [ebp - sizeof (IA32_REGS)]\r
     ret\r

diff --git a/MdePkg/Library/BaseLib/Ipf/CpuFlushTlb.s b/MdePkg/Library/BaseLib/Ipf/CpuFlushTlb.s
index 0afd87e88b2da928da44001d32318add24bf14b5..608830496c99674aa801f4735eddb91327e52066 100644 (file)
--- a/MdePkg/Library/BaseLib/Ipf/CpuFlushTlb.s
+++ b/MdePkg/Library/BaseLib/Ipf/CpuFlushTlb.s
@@ -17,13 +17,41 @@
 .auto\r
 .text\r
 \r
+.global PalCallStatic\r
+.type   PalCallStatic, @function\r
+\r
 .proc   CpuFlushTlb\r
 .type   CpuFlushTlb, @function\r
 CpuFlushTlb::\r
-        mov                 r8  = ip\r
-        mov                 r9  = -1\r
-        dep.z               r10 = -1, 61, 3\r
-        and                 r8  = r8, r10\r
-        ptc.l               r8, r9\r
+        alloc               loc0 = ar.pfs, 0, 2, 5, 0\r
+        mov                 out0 = 0\r
+        mov                 out1 = 6\r
+        mov                 out2 = 0\r
+        mov                 out3 = 0\r
+        mov                 out4 = 0\r
+        mov                 loc1 = b0\r
+        br.call.sptk        b0  = PalCallStatic\r
+        rsm                 1 << 14                 // Disable interrupts\r
+        mov                 ar.pfs = loc0\r
+        extr.u              r14 = r10, 32, 32       // r14 <- count1\r
+        extr.u              r15 = r11, 32, 32       // r15 <- stride1\r
+        extr.u              r10 = r10, 0, 32        // r10 <- count2\r
+        mov                 loc0 = psr\r
+        extr.u              r11 = r11, 0, 32        // r11 <- stride2\r
+        br.cond.sptk        LoopPredicate\r
+LoopOuter:\r
+        mov                 ar.lc = r10             // LC <- count2\r
+        mov                 ar.ec = r0              // EC <- 0\r
+Loop:\r
+        ptc.e               r9\r
+        add                 r9 = r11, r9            // r9 += stride2\r
+        br.ctop.sptk        Loop\r
+        add                 r9 = r15, r9            // r9 += stride1\r
+LoopPredicate:\r
+        cmp.ne              p6, p7 = r0, r14        // count1 == 0?\r
+        add                 r14 = -1, r14\r
+(p6)    br.cond.sptk        LoopOuter\r
+        mov                 psr.l = loc0\r
+        mov                 b0  = loc1\r
         br.ret.sptk.many    b0\r
 .endp\r

diff --git a/MdePkg/Library/BaseLib/Ipf/PalCallStatic.s b/MdePkg/Library/BaseLib/Ipf/PalCallStatic.s
new file mode 100644 (file)
index 0000000..7fa68e1
--- /dev/null
+++ b/MdePkg/Library/BaseLib/Ipf/PalCallStatic.s
@@ -0,0 +1,47 @@
+/// @file\r
+///   Contains an implementation of CallPalProcStatic on Itanium-based\r
+///   architecture.\r
+///\r
+/// Copyright (c) 2006, Intel Corporation\r
+/// All rights reserved. This program and the accompanying materials\r
+/// are licensed and made available under the terms and conditions of the BSD License\r
+/// which accompanies this distribution.  The full text of the license may be found at\r
+/// http://opensource.org/licenses/bsd-license.php\r
+///\r
+/// THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r
+/// WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r
+///\r
+/// Module Name:  PalCallStatic.s\r
+///\r
+///\r
+\r
+.auto\r
+.text\r
+\r
+.proc   PalCallStatic\r
+.type   PalCallStatic, @function\r
+.regstk 5, 0, 0, 0\r
+PalCallStatic::\r
+        cmp.ne              p6, p7 = r0, in0\r
+        mov                 r31 = in4\r
+        mov                 r8  = ip\r
+(p6)    mov                 in0 = ar.k5\r
+        add                 r8  = (PalProcReturn - PalCallStatic), r8\r
+        mov                 in4 = b0\r
+        mov                 r30 = in3\r
+        mov                 r29 = in2\r
+        mov                 b7  = in0\r
+        mov                 in3 = psr\r
+        rsm                 1 << 14                 // Disable interrupts\r
+        mov                 r28 = in1\r
+        mov                 in0 = 256\r
+        mov                 b0  = r8\r
+        br.cond.sptk        b7\r
+PalProcReturn:\r
+        mov                 psr.l = in3\r
+        cmp.eq              p6, p7 = in0, in1       // in1 == PAL_COPY_PAL?\r
+(p6)    cmp.eq              p6, p7 = r0, r8         // Status == Success?\r
+(p6)    mov                 ar.k5 = in2\r
+        mov                 b0  = in4\r
+        br.ret.sptk.many    b0\r
+.endp   PalCallStatic\r

diff --git a/MdePkg/Library/BaseLib/X64/Thunk16.asm b/MdePkg/Library/BaseLib/X64/Thunk16.asm
index 15aaa80090774af1e8a94a03ed398283bde3819a..3121867ebb41e4f886d8df7a924fcc48de0d6e19 100644 (file)
--- a/MdePkg/Library/BaseLib/X64/Thunk16.asm
+++ b/MdePkg/Library/BaseLib/X64/Thunk16.asm
@@ -149,6 +149,11 @@ _ToUserCode PROC
     call    @Base                       ; push eip\r
 @Base:\r
     pop     bp                          ; ebp <- offset @Base\r
+    push    [esp + sizeof (IA32_REGS) + 2]\r
+    lea     eax, [rsi + (offset @RealMode - offset @Base)]\r
+    push    rax\r
+    retf\r
+@RealMode:\r
     DB      2eh                         ; cs:\r
     mov     [rsi + (offset SavedSs - offset @Base)], edi\r
     DB      2eh                         ; cs:\r
@@ -222,7 +227,9 @@ InternalAsmThunk16  PROC    USES    rbp rbx rsi rdi
     push    10h\r
     pop     rdx                         ; rdx <- selector for data segments\r
     lgdt    fword ptr [rcx + (offset _16Gdtr - offset SavedCr4)]\r
+    pushfq\r
     call    fword ptr [rcx + (offset _EntryPoint - offset SavedCr4)]\r
+    popfq\r
     lidt    fword ptr [rsp + 38h]       ; restore protected mode IDTR\r
     lea     eax, [rbp - sizeof (IA32_REGS)]\r
     pop     gs\r

diff --git a/MdePkg/Library/BaseLib/x86Thunk.c b/MdePkg/Library/BaseLib/x86Thunk.c
index d4e37ae59e6671a6cc060f3d527fd8da00ccd8cc..9e69255181fd4a62da0d490fa6a2487524adbb2c 100644 (file)
--- a/MdePkg/Library/BaseLib/x86Thunk.c
+++ b/MdePkg/Library/BaseLib/x86Thunk.c
@@ -95,6 +95,10 @@ AsmGetThunk16Properties (
   ASSERT (ExtraStackSize != NULL);\r
 \r
   *RealModeBufferSize = m16Size;\r
+\r
+  //\r
+  // Extra 4 bytes for return address, and another 4 bytes for mode transition\r
+  //\r
   *ExtraStackSize = sizeof (IA32_DWORD_REGS) + 8;\r
 }\r
 \r