Changeset 216 in xtideuniversalbios for trunk/XTIDE_Universal_BIOS/Src

Timestamp:

Jan 23, 2012, 8:17:19 AM (13 years ago)

Author:

gregli@…

google:author:

gregli@hotmail.com

Message:

Serial port code - improved pointer re-normalization.

File:

• trunk/XTIDE_Universal_BIOS/Src/Device/Serial/SerialCommand.asm (modified) (7 diffs)

trunk/XTIDE_Universal_BIOS/Src/Device/Serial/SerialCommand.asm

@@ -14 +14 @@
 SerialCommand_UART_transmitByte                 EQU     0
 SerialCommand_UART_receiveByte                  EQU     0
-SerialCommand_UART_divisorLow                   EQU     0
+
+;
 ; Values for UART_divisorLow:
 ; 60h = 1200, 30h = 2400, 18h = 4800, 0ch = 9600, 6 = 19200, 3 = 38400, 2 = 57600, 1 = 115200
-
-SerialCommand_UART_divisorLow_startingBaud      EQU   030h
+;
+SerialCommand_UART_divisorLow                   EQU     0
+
+;
 ; We support 4 baud rates, starting here going higher and skipping every other baud rate
 ; Starting with 30h, that means 30h (1200 baud), 0ch (9600 baud), 3 (38400 baud), and 1 (115200 baud)
 ; Note: hardware baud multipliers (2x, 4x) will impact the final baud rate and are not known at this level
+;
+SerialCommand_UART_divisorLow_startingBaud      EQU   030h
 
 SerialCommand_UART_interruptEnable              EQU     1
+
+;
+; UART_divisorHigh is zero for all speeds including and above 1200 baud (which is all we do)
+;
 SerialCommand_UART_divisorHigh                  EQU     1
-; UART_divisorHigh is zero for all speeds including and above 1200 baud
 
 SerialCommand_UART_interruptIdent               EQU     2
@@ -76 +84 @@
 ;  all other commands return success
 ;  including function 0ech which should return drive information, this is handled with the identify functions
+;
         xor     ah,ah           ;  also clears carry
         ret
@@ -130 +139 @@
         push    ax
 
-;       cld     ; Shouldn't be needed. DF has already been cleared (line 24, Int13h.asm)
+%if 0
+        cld     ; Shouldn't be needed. DF has already been cleared (line 24, Int13h.asm)
+%endif
 
 ;----------------------------------------------------------------------
@@ -309 +320 @@
         jnz     SerialCommand_OutputWithParameters_Error
 
+        pop     ax              ; sector count and command byte
+        dec     al              ; decrememnt sector count
+        push    ax              ; save
+        jz      SerialCommand_OutputWithParameters_ReturnCodeInALCF    ; CF=0 from "cmp ax,bp" returning Zero above
+
+        cli                     ; interrupts back off for ACK byte to host
+                                ; (host could start sending data immediately)
+        out     dx,al           ; ACK with next sector number
+
+;
+; Normalize buffer pointer for next go round, if needed.
+;
+; We need to re-nomrlize the pointer in ES:DI after processing every 7f sectors.  That number could 
+; have been 80 if we knew the offset was on a segment boundary, but this may not be the case.
+;
+; We re-normalize based on the sector count (flags from "dec al" above)...
+;    a) we normalize before the first sector goes out, immediately after sending the command packet (above)
+;    b) on transitions from FF to FE, very rare case for writing 255 or 256 sectors
+;    c) on transitions from 80 to 7F, a large read/write
+;    d) on transitions from 00 to FF, very, very rare case of writing 256 secotrs
+;       We don't need to renormalize in this case, but it isn't worth the memory/effort to not do 
+;       the extra work, and it does no harm.
+;
+; I really don't care much about (d) because I have not seen cases where any OS makes a request
+; for more than 127 sectors.  Back in the day, it appears that some BIOS could not support more than 127
+; sectors, so that may be the practical limit for OS and application developers.  The Extended BIOS
+; function also appear to be capped at 127 sectors.  So although this can support the full 256 sectors 
+; if needed, we are optimized for that 1-127 range.
+;
+; Assume we start with 0000:000f, with 256 sectors to write...
+;    After first packet, 0000:020f
+;    First decrement of AL, transition from 00 to FF: renormalize to 0020:000f (unnecessary)
+;    After second packet, 0020:020f
+;    Second derement of AL, transition from FF to FE: renormalize to 0040:000f
+;    After 7f more packets, 0040:fe0f
+;    Decrement of AL, transition from 80 to 7F: renormalize to 1020:000f
+;    After 7f more packets, 1020:fe0f or 2000:000f if normalized
+;    Decrement of AL, from 1 to 0: exit
+;
+        jge     short .nextSector       ; OF=SF, branch for 1-7e, most common case
+                                        ; (0 kicked out above for return success)
+
+        add     al,2                    ; 7f-ff moves to 81-01
+                                        ; (0-7e kicked out before we get here)
+                                        ; 7f moves to 81 and OF=1, so OF=SF
+                                        ; fe moves to 0 and OF=0, SF=0, so OF=SF 
+                                        ; ff moves to 1 and OF=0, SF=0, so OF=SF
+                                        ; 80-fd moves to 82-ff and OF=0, so OF<>SF
+
+        jl      short .nextSector       ; OF<>SF, branch for all cases but 7f, fe, and ff
+
+%if 0
+        jpo     short .nextSector       ; if we really wanted to avoid normalizing for ff, this
+                                        ; is one way to do it, but it adds more memory and more
+                                        ; cycles for the 7f and fe cases.  IMHO, given that I've
+                                        ; never seen a request for more than 7f, this seems unnecessary.
+%endif
+
+        jmp     short .nextSectorNormalize  ; our two renormalization cases (plus one for ff)
+
+;---------------------------------------------------------------------------
+;
+; Cleanup, error reporting, and exit
+;
+
+;
+; Used in situations where a call is underway, such as with SerialCommand_WaitAndPoll
+;
+ALIGN JUMP_ALIGN
+SerialCommand_OutputWithParameters_ErrorAndPop4Words:
+        add     sp,8
+;;; fall-through
+
+ALIGN JUMP_ALIGN
+SerialCommand_OutputWithParameters_Error:
 ;----------------------------------------------------------------------
 ;
@@ -326 +412 @@
         in      al,dx
         jc      .clearBuffer    ; note CF from shr above
-        jmp     SerialCommand_OutputWithParameters_Error
-
-.clearBufferComplete:
-
-        pop     ax              ; sector count and command byte
-        dec     al              ; decrememnt sector count
-        push    ax              ; save
-        jz      SerialCommand_OutputWithParameters_ReturnCodeInALCF    ; CF clear from .clearBuffer test above
-
-        cli                     ; interrupts back off for ACK byte to host
-                                ; (host could start sending data immediately)
-        out     dx,al           ; ACK with next sector number
-
-;
-; Normalize buffer pointer for next go round, if needed
-;
-        test    di,di
-        jns     short .nextSector
-        jmp     short .nextSectorNormalize
-
-;---------------------------------------------------------------------------
-;
-; Cleanup, error reporting, and exit
-;
-
-;
-; Used in situations where a call is underway, such as with SerialCommand_WaitAndPoll
-;
-ALIGN JUMP_ALIGN
-SerialCommand_OutputWithParameters_ErrorAndPop4Words:
-        add     sp,8
-
-ALIGN JUMP_ALIGN
-SerialCommand_OutputWithParameters_Error:
+
+.clearBufferComplete:           
         stc
         mov     al,1
@@ -365 +419 @@
 ALIGN JUMP_ALIGN
 SerialCommand_OutputWithParameters_ReturnCodeInALCF:
-        sti
+%if 0
+        sti                     ;  all paths here will already have interrupts turned back on
+%endif
         mov     ah,al
 
@@ -696 +752 @@
 .error: 
         stc
-        ; mov       ah,1        ; setting the error code is unnecessary as this path can only be taken during
-                                ; drive detection, and drive detection works off CF and does not check AH
+%if 0
+        mov     ah,1        ; setting the error code is unnecessary as this path can only be taken during
+                            ; drive detection, and drive detection works off CF and does not check AH
+%endif
         ret