Implement memcpy_far and checksum_far, and replace _fl variants.

The "flat" mode variants work in real mode, but will not work in
    protected mode.  So, replace with versions that take explicit
    segments.

7 files changed, 48 insertions, 35 deletions

diff --git a/src/biosvar.h b/src/biosvar.h
index 4d3b3f0..28f956f 100644
--- a/src/biosvar.h
+++ b/src/biosvar.h
@@ -258,8 +258,12 @@ get_ebda_ptr()
  * Global variables
  ****************************************************************/
 
-#define GET_GLOBAL(var) \
-    GET_VAR(CS, (var))
+#define GLOBAL_SEGREG CS
+static inline u16 get_global_seg() {
+    return GET_SEG(GLOBAL_SEGREG);
+}
+#define GET_GLOBAL(var)                         \
+    GET_VAR(GLOBAL_SEGREG, (var))
 #define SET_GLOBAL(var, val) do {                                       \
         extern void __force_link_error__set_global_only_in_32bit();     \
         if (MODE16)                                                     \
diff --git a/src/boot.c b/src/boot.c
index 142544b..1f0647a 100644
--- a/src/boot.c
+++ b/src/boot.c
@@ -40,7 +40,7 @@ printf_bootdev(u16 bootdev)
     if (type == 4 && description_fl != 0) {
         char description[33];
         /* first 32 bytes are significant */
-        memcpy_fl(MAKE_FLATPTR(GET_SEG(SS), description), description_fl, 32);
+        memcpy(description, description_fl, 32);
         /* terminate string */
         description[32] = 0;
         printf(" [%.s]", description);
diff --git a/src/disk.c b/src/disk.c
index 248c839..03d2d38 100644
--- a/src/disk.c
+++ b/src/disk.c
@@ -41,12 +41,12 @@ __disk_stub(struct bregs *regs, int lineno, const char *fname)
     __disk_stub((regs), __LINE__, __func__)
 
 static int
-__send_disk_op(struct disk_op_s *op_p, u16 op_s)
+__send_disk_op(struct disk_op_s *op_far, u16 op_seg)
 {
     struct disk_op_s dop;
-    memcpy_fl(MAKE_FLATPTR(GET_SEG(SS), &dop)
-              , MAKE_FLATPTR(op_s, op_p)
-              , sizeof(dop));
+    memcpy_far(GET_SEG(SS), &dop
+               , op_seg, op_far
+               , sizeof(dop));
 
     dprintf(DEBUG_HDL_13, "disk_op d=%d lba=%d buf=%p count=%d cmd=%d\n"
             , dop.driveid, (u32)dop.lba, dop.buf_fl
@@ -486,9 +486,9 @@ disk_1348(struct bregs *regs, u8 device)
     SET_EBDA2(ebda_seg, dpte.reserved, 0);
     SET_EBDA2(ebda_seg, dpte.revision, 0x11);
 
-    u8 *p = MAKE_FLATPTR(ebda_seg
-                        , offsetof(struct extended_bios_data_area_s, dpte));
-    SET_EBDA2(ebda_seg, dpte.checksum, -checksum(p, 15));
+    u8 sum = checksum_far(
+        ebda_seg, (void*)offsetof(struct extended_bios_data_area_s, dpte), 15);
+    SET_EBDA2(ebda_seg, dpte.checksum, -sum);
 
     if (size < 66) {
         disk_ret(regs, DISK_RET_SUCCESS);
@@ -522,7 +522,7 @@ disk_1348(struct bregs *regs, u8 device)
 
     SET_INT13DPT(regs, device_path, slave);
 
-    SET_INT13DPT(regs, checksum, -checksum(MAKE_FLATPTR(regs->ds, 30), 35));
+    SET_INT13DPT(regs, checksum, -checksum_far(regs->ds, (void*)30, 35));
 
     disk_ret(regs, DISK_RET_SUCCESS);
 }
diff --git a/src/pcibios.c b/src/pcibios.c
index 950871c..cfcb25e 100644
--- a/src/pcibios.c
+++ b/src/pcibios.c
@@ -144,13 +144,13 @@ handle_1ab10e(struct bregs *regs)
     }
 
     // Get dest buffer.
-    u16 d = (GET_FARVAR(regs->es, *(u16*)(regs->di+2)) + 0);
-    u16 destseg = GET_FARVAR(regs->es, *(u16*)(regs->di+4));
+    void *d_far = (void*)(GET_FARVAR(regs->es, *(u16*)(regs->di+2)) + 0);
+    u16 d_seg = GET_FARVAR(regs->es, *(u16*)(regs->di+4));
 
     // Memcpy pir table slots to dest buffer.
-    memcpy_fl(MAKE_FLATPTR(destseg, d)
-              , MAKE_FLATPTR(SEG_BIOS, pirtable_g->slots)
-              , pirsize);
+    memcpy_far(d_seg, d_far
+               , get_global_seg(), pirtable_g->slots
+               , pirsize);
 
     // XXX - bochs bios sets bx to (1 << 9) | (1 << 11)
     regs->bx = GET_GLOBAL(pirtable_g->exclusive_irqs);
diff --git a/src/system.c b/src/system.c
index 5f39a98..798dc25 100644
--- a/src/system.c
+++ b/src/system.c
@@ -279,9 +279,9 @@ handle_15e820(struct bregs *regs)
         return;
     }
 
-    memcpy_fl(MAKE_FLATPTR(regs->es, regs->di)
-              , MAKE_FLATPTR(SEG_BIOS, &e820_list[regs->bx])
-              , sizeof(e820_list[0]));
+    memcpy_far(regs->es, (void*)(regs->di+0)
+               , get_global_seg(), &e820_list[regs->bx]
+               , sizeof(e820_list[0]));
     if (regs->bx == count-1)
         regs->ebx = 0;
     else
diff --git a/src/util.c b/src/util.c
index 0a60e17..66b3343 100644
--- a/src/util.c
+++ b/src/util.c
@@ -99,15 +99,22 @@ stack_hop(u32 eax, u32 edx, u32 ecx, void *func)
 
 // Sum the bytes in the specified area.
 u8
-checksum(u8 *buf_fl, u32 len)
+checksum_far(u16 buf_seg, u8 *buf_far, u32 len)
 {
+    SET_SEG(ES, buf_seg);
     u32 i;
     u8 sum = 0;
     for (i=0; i<len; i++)
-        sum += GET_FLATPTR(buf_fl[i]);
+        sum += GET_VAR(ES, buf_far[i]);
     return sum;
 }
 
+u8
+checksum(u8 *buf, u32 len)
+{
+    return checksum_far(GET_SEG(SS), buf, len);
+}
+
 void *
 memset(void *s, int c, size_t n)
 {
@@ -116,19 +123,19 @@ memset(void *s, int c, size_t n)
     return s;
 }
 
-void *
-memcpy_fl(void *d_fl, const void *s_fl, size_t len)
+inline void
+memcpy_far(u16 d_seg, void *d_far, u16 s_seg, const void *s_far, size_t len)
 {
-    u8 *d = d_fl;
-    u8 *s = (u8*)s_fl;
-
-    while (len--) {
-        SET_FLATPTR(*d, GET_FLATPTR(*s));
-        d++;
-        s++;
-    }
-
-    return d_fl;
+    SET_SEG(ES, d_seg);
+    u16 bkup_ds;
+    asm volatile(
+        "movw %%ds, %w0\n"
+        "movw %w4, %%ds\n"
+        "rep movsb (%%si),%%es:(%%di)\n"
+        "movw %w0, %%ds\n"
+        : "=&r"(bkup_ds), "+c"(len), "+S"(s_far), "+D"(d_far)
+        : "r"(s_seg)
+        : "cc", "memory");
 }
 
 void *
diff --git a/src/util.h b/src/util.h
index 78812cb..1531ffd 100644
--- a/src/util.h
+++ b/src/util.h
@@ -67,10 +67,12 @@ static inline u64 rdtscll(void)
 
 // util.c
 inline u32 stack_hop(u32 eax, u32 edx, u32 ecx, void *func);
-u8 checksum(u8 *buf_fl, u32 len);
+u8 checksum_far(u16 buf_seg, u8 *buf_far, u32 len);
+u8 checksum(u8 *buf, u32 len);
 void *memset(void *s, int c, size_t n);
 void *memcpy(void *d1, const void *s1, size_t len);
-void *memcpy_fl(void *d_fl, const void *s_fl, size_t len);
+inline void memcpy_far(u16 d_seg, void *d_far
+                       , u16 s_seg, const void *s_far, size_t len);
 void *memmove(void *d, const void *s, size_t len);
 struct bregs;
 inline void call16(struct bregs *callregs);