#include <mach/ppc/vm_param.h>
#include <ppc/exception.h>
+
+/*
+ * ml_set_physical() -- turn off DR and (if 64-bit) turn SF on
+ * it is assumed that pf64Bit is already in cr6
+ * ml_set_physical_get_ffs() -- turn DR off, SF on, and get feature flags
+ * ml_set_physical_disabled() -- turn DR and EE off, SF on, get feature flags
+ * ml_set_translation_off() -- turn DR, IR, and EE off, SF on, get feature flags
+ *
+ * Callable only from assembler, these return:
+ * r2 -- new MSR
+ * r11 -- old MSR
+ * r10 -- feature flags (pf64Bit etc, ie SPRG 2)
+ * cr6 -- feature flags 24-27, ie pf64Bit, pf128Byte, and pf32Byte
+ *
+ * Uses r0 and r2. ml_set_translation_off also uses r3 and cr5.
+ */
+
+ .align 4
+ .globl EXT(ml_set_translation_off)
+LEXT(ml_set_translation_off)
+ mfsprg r10,2 // get feature flags
+ li r0,0 ; Clear this
+ mtcrf 0x02,r10 // move pf64Bit etc to cr6
+ ori r0,r0,lo16(MASK(MSR_EE)+MASK(MSR_FP)+MASK(MSR_IR)+MASK(MSR_DR)) // turn off all 4
+ mfmsr r11 // get MSR
+ oris r0,r0,hi16(MASK(MSR_VEC)) // Turn off vector too
+ mtcrf 0x04,r10 // move pfNoMSRir etc to cr5
+ andc r2,r11,r0 // turn off EE, IR, and DR
+ bt++ pf64Bitb,ml_set_physical_64 // skip if 64-bit (only they take the hint)
+ bf pfNoMSRirb,ml_set_physical_32 // skip if we can load MSR directly
+ li r0,loadMSR // Get the MSR setter SC
+ mr r3,r2 // copy new MSR to r2
+ sc // Set it
+ blr
+
+ .align 4
+ .globl EXT(ml_set_physical_disabled)
+
+LEXT(ml_set_physical_disabled)
+ li r0,0 ; Clear
+ mfsprg r10,2 // get feature flags
+ ori r0,r0,lo16(MASK(MSR_EE)) // turn EE and fp off
+ mtcrf 0x02,r10 // move pf64Bit etc to cr6
+ b ml_set_physical_join
+
+ .align 5
+ .globl EXT(ml_set_physical_get_ffs)
+
+LEXT(ml_set_physical_get_ffs)
+ mfsprg r10,2 // get feature flags
+ mtcrf 0x02,r10 // move pf64Bit etc to cr6
+
+ .globl EXT(ml_set_physical)
+LEXT(ml_set_physical)
+
+ li r0,0 // do not turn off interrupts
+
+ml_set_physical_join:
+ oris r0,r0,hi16(MASK(MSR_VEC)) // Always gonna turn of vectors
+ mfmsr r11 // get MSR
+ ori r0,r0,lo16(MASK(MSR_DR)+MASK(MSR_FP)) // always turn off DR and FP bit
+ andc r2,r11,r0 // turn off DR and maybe EE
+ bt++ pf64Bitb,ml_set_physical_64 // skip if 64-bit (only they take the hint)
+ml_set_physical_32:
+ mtmsr r2 // turn off translation
+ isync
+ blr
+
+ml_set_physical_64:
+ li r0,1 // get a 1 to slam into SF
+ rldimi r2,r0,63,MSR_SF_BIT // set SF bit (bit 0)
+ mtmsrd r2 // set 64-bit mode, turn off data relocation
+ isync // synchronize
+ blr
+
+
+/*
+ * ml_restore(old_MSR)
+ *
+ * Callable only from assembler, restores the MSR in r11 saved by ml_set_physical.
+ * We assume cr6 and r11 are as set by ml_set_physical, ie:
+ * cr6 - pf64Bit flag (feature flags 24-27)
+ * r11 - old MSR
+ */
+
+ .align 5
+ .globl EXT(ml_restore)
+
+LEXT(ml_restore)
+ bt++ pf64Bitb,ml_restore_64 // handle 64-bit cpus (only they take the hint)
+ mtmsr r11 // restore a 32-bit MSR
+ isync
+ blr
+
+ml_restore_64:
+ mtmsrd r11 // restore a 64-bit MSR
+ isync
+ blr
+
+
/* PCI config cycle probing
*
* boolean_t ml_probe_read(vm_offset_t paddr, unsigned int *val)
LEXT(ml_probe_read)
mfsprg r9,2 ; Get feature flags
+
+ rlwinm. r0,r9,0,pf64Bitb,pf64Bitb ; Are we on a 64-bit machine?
+ rlwinm r3,r3,0,0,31 ; Clean up for 64-bit machines
+ bne++ mpr64bit ; Go do this the 64-bit way...
+
+mpr32bit: lis r8,hi16(MASK(MSR_VEC)) ; Get the vector flag
mfmsr r0 ; Save the current MSR
- rlwinm r0,r0,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r0,r0,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
+ ori r8,r8,lo16(MASK(MSR_FP)) ; Add the FP flag
+
neg r10,r3 ; Number of bytes to end of page
- rlwinm r2,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions
+ andc r0,r0,r8 ; Clear VEC and FP
rlwinm. r10,r10,0,20,31 ; Clear excess junk and test for page bndry
+ ori r8,r8,lo16(MASK(MSR_EE)|MASK(MSR_IR)|MASK(MSR_DR)) ; Drop EE, IR, and DR
mr r12,r3 ; Save the load address
+ andc r2,r0,r8 ; Clear VEC, FP, and EE
mtcrf 0x04,r9 ; Set the features
cmplwi cr1,r10,4 ; At least 4 bytes left in page?
- rlwinm r2,r2,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Clear translation
beq- mprdoit ; We are right on the boundary...
li r3,0
bltlr- cr1 ; No, just return failure...
mtmsr r2 ; Turn translation back off
isync
- mtspr hid0, r6 ; Restore HID0
- isync
-
lis r10,hi16(EXT(shadow_BAT)+shdDBAT) ; Get shadow address
ori r10,r10,lo16(EXT(shadow_BAT)+shdDBAT) ; Get shadow address
.globl EXT(ml_probe_read_mck)
LEXT(ml_probe_read_mck)
-/* Read physical address
+
+/* PCI config cycle probing - 64-bit
+ *
+ * boolean_t ml_probe_read_64(addr64_t paddr, unsigned int *val)
+ *
+ * Read the memory location at physical address paddr.
+ * This is a part of a device probe, so there is a good chance we will
+ * have a machine check here. So we have to be able to handle that.
+ * We assume that machine checks are enabled both in MSR and HIDs
+ */
+
+; Force a line boundry here
+ .align 6
+ .globl EXT(ml_probe_read_64)
+
+LEXT(ml_probe_read_64)
+
+ mfsprg r9,2 ; Get feature flags
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwinm. r0,r9,0,pf64Bitb,pf64Bitb ; Are we on a 64-bit machine?
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+
+ mr r4,r5 ; Move result to common register
+ beq-- mpr32bit ; Go do this the 32-bit way...
+
+mpr64bit: andi. r0,r3,3 ; Check if we are on a word boundary
+ li r0,0 ; Clear the EE bit (and everything else for that matter)
+ bne-- mprFail ; Boundary not good...
+ mfmsr r11 ; Get the MSR
+ mtmsrd r0,1 ; Set the EE bit only (do not care about RI)
+ rlwinm r11,r11,0,MSR_EE_BIT,MSR_EE_BIT ; Isolate just the EE bit
+ mfmsr r10 ; Refresh our view of the MSR (VMX/FP may have changed)
+ or r12,r10,r11 ; Turn on EE if on before we turned it off
+ ori r0,r0,lo16(MASK(MSR_IR)|MASK(MSR_DR)) ; Get the IR and DR bits
+ li r2,1 ; Get a 1
+ sldi r2,r2,63 ; Get the 64-bit bit
+ andc r10,r10,r0 ; Clear IR and DR
+ or r10,r10,r2 ; Set 64-bit
+
+ li r0,1 ; Get a 1
+ mtmsrd r10 ; Translation and EE off, 64-bit on
+ isync
+
+ sldi r0,r0,32+8 ; Get the right bit to inhibit caching
+
+ mfspr r8,hid4 ; Get HID4
+ or r2,r8,r0 ; Set bit to make real accesses cache-inhibited
+ sync ; Sync up
+ mtspr hid4,r2 ; Make real accesses cache-inhibited
+ isync ; Toss prefetches
+
+ lis r7,0xE000 ; Get the unlikeliest ESID possible
+ srdi r7,r7,1 ; Make 0x7FFFFFFFF0000000
+ slbie r7 ; Make sure the ERAT is cleared
+
+ sync
+ isync
+
+ eieio ; Make sure of all previous accesses
+
+ lwz r11,0(r3) ; Get it and maybe machine check here
+
+ eieio ; Make sure of ordering again
+ sync ; Get caught up yet again
+ isync ; Do not go further till we are here
+
+ sync ; Sync up
+ mtspr hid4,r8 ; Make real accesses not cache-inhibited
+ isync ; Toss prefetches
+
+ lis r7,0xE000 ; Get the unlikeliest ESID possible
+ srdi r7,r7,1 ; Make 0x7FFFFFFFF0000000
+ slbie r7 ; Make sure the ERAT is cleared
+
+ mtmsrd r12 ; Restore entry MSR
+ isync
+
+ stw r11,0(r4) ; Pass back the result
+ li r3,1 ; Indicate success
+ blr ; Leave...
+
+mprFail: li r3,0 ; Set failure
+ blr ; Leave...
+
+; Force a line boundry here. This means we will be able to check addresses better
+ .align 6
+ .globl EXT(ml_probe_read_mck_64)
+LEXT(ml_probe_read_mck_64)
+
+
+/* Read physical address byte
*
* unsigned int ml_phys_read_byte(vm_offset_t paddr)
+ * unsigned int ml_phys_read_byte_64(addr64_t paddr)
*
* Read the byte at physical address paddr. Memory should not be cache inhibited.
*/
; Force a line boundry here
+
.align 5
+ .globl EXT(ml_phys_read_byte_64)
+
+LEXT(ml_phys_read_byte_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ b ml_phys_read_byte_join
+
.globl EXT(ml_phys_read_byte)
LEXT(ml_phys_read_byte)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_read_byte_join: ; r3 = address to read (reg64_t)
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
+
+ lbz r3,0(r3) ; Get the byte
+ b rdwrpost ; Clean up and leave...
- mfmsr r10 ; Save the current MSR
- rlwinm r10,r10,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r10,r10,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r4,r10,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions
- rlwinm r4,r4,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation
- mtmsr r4 ; Translation and all off
- isync ; Toss prefetch
+/* Read physical address half word
+ *
+ * unsigned int ml_phys_read_half(vm_offset_t paddr)
+ * unsigned int ml_phys_read_half_64(addr64_t paddr)
+ *
+ * Read the half word at physical address paddr. Memory should not be cache inhibited.
+ */
- lbz r3,0(r3) ; Get the byte
- sync
+; Force a line boundry here
- mtmsr r10 ; Restore translation and rupts
- isync
- blr
+ .align 5
+ .globl EXT(ml_phys_read_half_64)
+
+LEXT(ml_phys_read_half_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ b ml_phys_read_half_join
+
+ .globl EXT(ml_phys_read_half)
-/* Read physical address
+LEXT(ml_phys_read_half)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_read_half_join: ; r3 = address to read (reg64_t)
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
+
+ lhz r3,0(r3) ; Get the half word
+ b rdwrpost ; Clean up and leave...
+
+
+/* Read physical address word
*
* unsigned int ml_phys_read(vm_offset_t paddr)
+ * unsigned int ml_phys_read_64(addr64_t paddr)
+ * unsigned int ml_phys_read_word(vm_offset_t paddr)
+ * unsigned int ml_phys_read_word_64(addr64_t paddr)
*
* Read the word at physical address paddr. Memory should not be cache inhibited.
*/
; Force a line boundry here
+
.align 5
+ .globl EXT(ml_phys_read_64)
+ .globl EXT(ml_phys_read_word_64)
+
+LEXT(ml_phys_read_64)
+LEXT(ml_phys_read_word_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ b ml_phys_read_word_join
+
.globl EXT(ml_phys_read)
+ .globl EXT(ml_phys_read_word)
LEXT(ml_phys_read)
+LEXT(ml_phys_read_word)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_read_word_join: ; r3 = address to read (reg64_t)
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
+
+ lwz r3,0(r3) ; Get the word
+ b rdwrpost ; Clean up and leave...
- mfmsr r0 ; Save the current MSR
- rlwinm r0,r0,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r0,r0,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r4,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions
- rlwinm r4,r4,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation
- mtmsr r4 ; Translation and all off
- isync ; Toss prefetch
+/* Read physical address double word
+ *
+ * unsigned long long ml_phys_read_double(vm_offset_t paddr)
+ * unsigned long long ml_phys_read_double_64(addr64_t paddr)
+ *
+ * Read the double word at physical address paddr. Memory should not be cache inhibited.
+ */
+
+; Force a line boundry here
+
+ .align 5
+ .globl EXT(ml_phys_read_double_64)
+
+LEXT(ml_phys_read_double_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ b ml_phys_read_double_join
+
+ .globl EXT(ml_phys_read_double)
+
+LEXT(ml_phys_read_double)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_read_double_join: ; r3 = address to read (reg64_t)
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
- lwz r3,0(r3) ; Get the word
- sync
+ lwz r4,4(r3) ; Get the low word
+ lwz r3,0(r3) ; Get the high word
+ b rdwrpost ; Clean up and leave...
- mtmsr r0 ; Restore translation and rupts
- isync
- blr
/* Write physical address byte
*
* void ml_phys_write_byte(vm_offset_t paddr, unsigned int data)
+ * void ml_phys_write_byte_64(addr64_t paddr, unsigned int data)
*
* Write the byte at physical address paddr. Memory should not be cache inhibited.
*/
-; Force a line boundry here
.align 5
+ .globl EXT(ml_phys_write_byte_64)
+
+LEXT(ml_phys_write_byte_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ mr r4,r5 ; Copy over the data
+ b ml_phys_write_byte_join
+
.globl EXT(ml_phys_write_byte)
LEXT(ml_phys_write_byte)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_write_byte_join: ; r3 = address to write (reg64_t), r4 = data
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
+
+ stb r4,0(r3) ; Set the byte
+ b rdwrpost ; Clean up and leave...
- mfmsr r0 ; Save the current MSR
- rlwinm r0,r0,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r0,r0,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r5,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions
- rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation
- mtmsr r5 ; Translation and all off
- isync ; Toss prefetch
+/* Write physical address half word
+ *
+ * void ml_phys_write_half(vm_offset_t paddr, unsigned int data)
+ * void ml_phys_write_half_64(addr64_t paddr, unsigned int data)
+ *
+ * Write the half word at physical address paddr. Memory should not be cache inhibited.
+ */
+
+ .align 5
+ .globl EXT(ml_phys_write_half_64)
+
+LEXT(ml_phys_write_half_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ mr r4,r5 ; Copy over the data
+ b ml_phys_write_half_join
+
+ .globl EXT(ml_phys_write_half)
+
+LEXT(ml_phys_write_half)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_write_half_join: ; r3 = address to write (reg64_t), r4 = data
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
- stb r4,0(r3) ; Set the byte
- sync
+ sth r4,0(r3) ; Set the half word
+ b rdwrpost ; Clean up and leave...
- mtmsr r0 ; Restore translation and rupts
- isync
- blr
-/* Write physical address
+/* Write physical address word
*
* void ml_phys_write(vm_offset_t paddr, unsigned int data)
+ * void ml_phys_write_64(addr64_t paddr, unsigned int data)
+ * void ml_phys_write_word(vm_offset_t paddr, unsigned int data)
+ * void ml_phys_write_word_64(addr64_t paddr, unsigned int data)
*
* Write the word at physical address paddr. Memory should not be cache inhibited.
*/
-; Force a line boundry here
.align 5
+ .globl EXT(ml_phys_write_64)
+ .globl EXT(ml_phys_write_word_64)
+
+LEXT(ml_phys_write_64)
+LEXT(ml_phys_write_word_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ mr r4,r5 ; Copy over the data
+ b ml_phys_write_word_join
+
.globl EXT(ml_phys_write)
+ .globl EXT(ml_phys_write_word)
LEXT(ml_phys_write)
+LEXT(ml_phys_write_word)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_write_word_join: ; r3 = address to write (reg64_t), r4 = data
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
+
+ stw r4,0(r3) ; Set the word
+ b rdwrpost ; Clean up and leave...
- mfmsr r0 ; Save the current MSR
- rlwinm r0,r0,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r0,r0,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r5,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions
- rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation
- mtmsr r5 ; Translation and all off
- isync ; Toss prefetch
+/* Write physical address double word
+ *
+ * void ml_phys_write_double(vm_offset_t paddr, unsigned long long data)
+ * void ml_phys_write_double_64(addr64_t paddr, unsigned long long data)
+ *
+ * Write the double word at physical address paddr. Memory should not be cache inhibited.
+ */
+
+ .align 5
+ .globl EXT(ml_phys_write_double_64)
+
+LEXT(ml_phys_write_double_64)
+
+ rlwinm r3,r3,0,1,0 ; Copy low 32 bits to top 32
+ rlwimi r3,r4,0,0,31 ; Insert low part of 64-bit address in bottom 32 bits
+ mr r4,r5 ; Copy over the high data
+ mr r5,r6 ; Copy over the low data
+ b ml_phys_write_double_join
+
+ .globl EXT(ml_phys_write_double)
+
+LEXT(ml_phys_write_double)
+ rlwinm r3,r3,0,0,31 ; truncate address to 32-bits
+ml_phys_write_double_join: ; r3 = address to write (reg64_t), r4,r5 = data (long long)
+ mflr r11 ; Save the return
+ bl rdwrpre ; Get set up, translation/interrupts off, 64-bit on, etc.
+
+ stw r4,0(r3) ; Set the high word
+ stw r5,4(r3) ; Set the low word
+ b rdwrpost ; Clean up and leave...
+
+
+ .align 5
+
+rdwrpre: mfsprg r12,2 ; Get feature flags
+ lis r8,hi16(MASK(MSR_VEC)) ; Get the vector flag
+ mfmsr r10 ; Save the MSR
+ ori r8,r8,lo16(MASK(MSR_FP)) ; Add the FP flag
+ mtcrf 0x02,r12 ; move pf64Bit
+ andc r10,r10,r8 ; Clear VEC and FP
+ ori r9,r8,lo16(MASK(MSR_EE)|MASK(MSR_IR)|MASK(MSR_DR)) ; Drop EE, DR, and IR
+ li r2,1 ; Prepare for 64 bit
+ andc r9,r10,r9 ; Clear VEC, FP, DR, and EE
+ bf-- pf64Bitb,rdwrpre32 ; Join 32-bit code...
+
+ srdi r7,r3,31 ; Get a 1 if address is in I/O memory
+ rldimi r9,r2,63,MSR_SF_BIT ; set SF bit (bit 0)
+ cmpldi cr7,r7,1 ; Is source in I/O memory?
+ mtmsrd r9 ; set 64-bit mode, turn off EE, DR, and IR
+ isync ; synchronize
+
+ sldi r0,r2,32+8 ; Get the right bit to turn off caching
+
+ bnelr++ cr7 ; We are not in the I/O area, all ready...
+
+ mfspr r8,hid4 ; Get HID4
+ or r2,r8,r0 ; Set bit to make real accesses cache-inhibited
+ sync ; Sync up
+ mtspr hid4,r2 ; Make real accesses cache-inhibited
+ isync ; Toss prefetches
+
+ lis r7,0xE000 ; Get the unlikeliest ESID possible
+ srdi r7,r7,1 ; Make 0x7FFFFFFFF0000000
+ slbie r7 ; Make sure the ERAT is cleared
- stw r4,0(r3) ; Set the word
sync
+ isync
+ blr ; Finally, all ready...
+
+ .align 5
+
+rdwrpre32: rlwimi r9,r10,0,MSR_IR_BIT,MSR_IR_BIT ; Leave the IR bit unchanged
+ mtmsr r9 ; Drop EE, DR, and leave IR unchanged
+ isync
+ blr ; All set up, leave...
+
+ .align 5
+
+rdwrpost: mtlr r11 ; Restore the return
+ bt++ pf64Bitb,rdwrpost64 ; Join 64-bit code...
+
+ mtmsr r10 ; Restore entry MSR (sans FP and VEC)
+ isync
+ blr ; Leave...
+
+rdwrpost64: bne++ cr7,rdwrpcok ; Skip enabling real mode caching if we did not change it...
+
+ sync ; Sync up
+ mtspr hid4,r8 ; Make real accesses not cache-inhibited
+ isync ; Toss prefetches
- mtmsr r0 ; Restore translation and rupts
+ lis r7,0xE000 ; Get the unlikeliest ESID possible
+ srdi r7,r7,1 ; Make 0x7FFFFFFFF0000000
+ slbie r7 ; Make sure the ERAT is cleared
+
+rdwrpcok: mtmsrd r10 ; Restore entry MSR (sans FP and VEC)
isync
- blr
+ blr ; Leave...
/* set interrupts enabled or disabled
LEXT(ml_set_interrupts_enabled)
- mfsprg r7,0
- lwz r4,PP_INTS_ENABLED(r7)
- mr. r4,r4
- beq- EXT(fake_set_interrupts_enabled)
+ andi. r4,r3,1 ; Are we turning interruptions on?
+ lis r0,hi16(MASK(MSR_VEC)) ; Get vector enable
mfmsr r5 ; Get the current MSR
- rlwinm r5,r5,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- mr r4,r3 ; Save the old value
- rlwinm r5,r5,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
+ ori r0,r0,lo16(MASK(MSR_EE)|MASK(MSR_FP)) ; Get float enable and EE enable
rlwinm r3,r5,17,31,31 ; Set return value
- rlwimi r5,r4,15,16,16 ; Insert new EE bit
- andi. r8,r5,lo16(MASK(MSR_EE)) ; Interruptions
- bne CheckPreemption
-NoPreemption:
- mtmsr r5 ; Slam enablement
+ andc r5,r5,r0 ; Force VEC and FP off
+ bne CheckPreemption ; Interrupts going on, check ASTs...
+
+ mtmsr r5 ; Slam diable (always going disabled here)
+ isync ; Need this because FP/Vec might go off
blr
+ .align 5
+
CheckPreemption:
- lwz r8,PP_NEED_AST(r7)
- li r6,AST_URGENT
- lwz r8,0(r8)
- lwz r7,PP_PREEMPT_CNT(r7)
- lis r0,HIGH_ADDR(DoPreemptCall)
- and. r8,r8,r6
- ori r0,r0,LOW_ADDR(DoPreemptCall)
- beq+ NoPreemption
- cmpi cr0, r7, 0
+ mfsprg r7,0
+ ori r5,r5,lo16(MASK(MSR_EE)) ; Turn on the enable
+ lwz r8,PP_NEED_AST(r7) ; Get pointer to AST flags
+ mfsprg r9,1 ; Get current activation
+ li r6,AST_URGENT ; Get the type we will preempt for
+ lwz r7,ACT_PREEMPT_CNT(r9) ; Get preemption count
+ lwz r8,0(r8) ; Get AST flags
+ lis r0,hi16(DoPreemptCall) ; High part of Preempt FW call
+ cmpwi cr1,r7,0 ; Are preemptions masked off?
+ and. r8,r8,r6 ; Are we urgent?
+ crorc cr1_eq,cr0_eq,cr1_eq ; Remember if preemptions are masked or not urgent
+ ori r0,r0,lo16(DoPreemptCall) ; Bottome of FW call
+
mtmsr r5 ; Restore the MSR now, before we can preempt
- bnelr+ ; Return if no premption
+ isync ; Need this because FP/Vec might go off
+
+ beqlr++ cr1 ; Return if no premption...
sc ; Preempt
blr
-
/* Emulate a decremeter exception
*
* void machine_clock_assist(void)
mfsprg r7,0
lwz r4,PP_INTS_ENABLED(r7)
mr. r4,r4
- beq- EXT(CreateFakeDEC)
- blr
+ bnelr+ cr0
+ b EXT(CreateFakeDEC)
/* Set machine into idle power-saving mode.
*
*
*/
-
; Force a line boundry here
.align 5
.globl EXT(machine_idle_ppc)
LEXT(machine_idle_ppc)
- mfmsr r3 ; Get the current MSR
- rlwinm r3,r3,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r3,r3,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r5,r3,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions
+ lis r0,hi16(MASK(MSR_VEC)) ; Get the vector flag
+ mfmsr r3 ; Save the MSR
+ ori r0,r0,lo16(MASK(MSR_FP)) ; Add the FP flag
+ andc r3,r3,r0 ; Clear VEC and FP
+ ori r0,r0,lo16(MASK(MSR_EE)) ; Drop EE also
+ andc r5,r3,r0 ; Clear VEC, FP, DR, and EE
+
mtmsr r5 ; Hold up interruptions for now
isync ; May have messed with fp/vec
mfsprg r12,0 ; Get the per_proc_info
- mfspr r6,hid0 ; Get the current power-saving mode
mfsprg r11,2 ; Get CPU specific features
- rlwinm r6,r6,0,sleep+1,doze-1 ; Clear all possible power-saving modes (not DPM though)
+ mfspr r6,hid0 ; Get the current power-saving mode
mtcrf 0xC7,r11 ; Get the facility flags
lis r4,hi16(napm) ; Assume we can nap
mftbu r8 ; Get the upper one again
cmplw r9,r8 ; Did the top tick?
bne- yesnap ; Yeah, need to get it again...
+
stw r8,napStamp(r12) ; Set high order time stamp
stw r7,napStamp+4(r12) ; Set low order nap stamp
- rlwinm. r7,r11,0,pfNoL2PFNapb,pfNoL2PFNapb ; Turn off L2 Prefetch before nap?
+ rlwinm. r0,r11,0,pfAltivecb,pfAltivecb ; Do we have altivec?
+ beq- minovec ; No...
+ dssall ; Stop the streams before we nap/doze
+ sync
+ lwz r8,napStamp(r12) ; Reload high order time stamp
+clearpipe:
+ cmplw r8,r8
+ bne- clearpipe
+ isync
+minovec:
+
+ rlwinm. r7,r11,0,pfNoL2PFNapb,pfNoL2PFNapb ; Turn off L2 Prefetch before nap?
beq miL2PFok
mfspr r7,msscr0 ; Get currect MSSCR0 value
- rlwinm r7,r7,0,0,l2pfes-1 ; Dissable L2 Prefetch
+ rlwinm r7,r7,0,0,l2pfes-1 ; Disable L2 Prefetch
mtspr msscr0,r7 ; Updates MSSCR0 value
sync
isync
miL2PFok:
- rlwinm. r7,r11,0,pfSlowNapb,pfSlowNapb ; Should nap at slow speed?
+ rlwinm. r7,r11,0,pfSlowNapb,pfSlowNapb ; Should nap at slow speed?
beq minoslownap
mfspr r7,hid1 ; Get current HID1 value
- oris r7,r7,hi16(hid1psm) ; Select PLL1
+ oris r7,r7,hi16(hid1psm) ; Select PLL1
mtspr hid1,r7 ; Update HID1 value
minoslownap:
; is taken and set everything up to return directly to machine_idle_ret.
; So, make sure everything we need there is already set up...
;
+
+ lis r10,hi16(dozem|napm|sleepm) ; Mask of power management bits
+
+ bf-- pf64Bitb,mipNSF1 ; skip if 32-bit...
+
+ sldi r4,r4,32 ; Position the flags
+ sldi r10,r10,32 ; Position the masks
+
+
+mipNSF1: andc r6,r6,r10 ; Clean up the old power bits
+
ori r7,r5,lo16(MASK(MSR_EE)) ; Flip on EE
or r6,r6,r4 ; Set nap or doze
oris r5,r7,hi16(MASK(MSR_POW)) ; Turn on power management in next MSR
+
+ sync
mtspr hid0,r6 ; Set up the HID for nap/doze
+ mfspr r6,hid0 ; Yes, this is silly, keep it here
+ mfspr r6,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r6,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r6,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r6,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r6,hid0 ; Yes, this is a duplicate, keep it here
isync ; Make sure it is set
+
mtmsr r7 ; Enable for interrupts
- rlwinm. r11,r11,0,pfAltivecb,pfAltivecb ; Do we have altivec?
- beq- minovec ; No...
- dssall ; Stop the streams before we nap/doze
-minovec: sync ; Make sure queues are clear
- mtmsr r5 ; Nap or doze
+ bf-- pf64Bitb,mipowloop ; skip if 32-bit...
+
+ li r3,0x10 ; Fancy nap threashold is 0x10 ticks
+ mftb r8 ; Get the low half of the time base
+ mfdec r4 ; Get the decrementer ticks
+ cmplw r4,r3 ; Less than threashold?
+ blt mipowloop
+
+ mtdec r3 ; Load decrimenter with threshold
+ isync ; and make sure,
+ mfdec r3 ; really sure, it gets there
+
+ rlwinm r6,r5,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear out the EE bit
+ sync ; Make sure queues are clear
+ mtmsr r6 ; Set MSR with EE off but POW on
isync ; Make sure this takes before we proceed
- b minovec ; loop if POW does not take
+
+ mftb r9 ; Get the low half of the time base
+ sub r9,r9,r8 ; Get the number of ticks spent waiting
+ sub r4,r4,r9 ; Adjust the decrementer value
+
+ mtdec r4 ; Load decrimenter with the rest of the timeout
+ isync ; and make sure,
+ mfdec r4 ; really sure, it gets there
+
+mipowloop:
+ sync ; Make sure queues are clear
+ mtmsr r5 ; Nap or doze, MSR with POW and EE set
+ isync ; Make sure this takes before we proceed
+ b mipowloop ; loop if POW does not take
+
;
; Note that the interrupt handler will turn off the nap/doze bits in the hid.
; Also remember that the interrupt handler will force return to here whenever
b deadsleep ; Die the death of 1000 joys...
#endif
+ mfsprg r12,0 ; Get the per_proc_info
mfspr r4,hid0 ; Get the current power-saving mode
eqv r10,r10,r10 ; Get all foxes
mfsprg r11,2 ; Get CPU specific features
- rlwinm. r5,r11,0,pfNoL2PFNapb,pfNoL2PFNapb ; Turn off L2 Prefetch before sleep?
+ rlwinm. r5,r11,0,pfNoL2PFNapb,pfNoL2PFNapb ; Turn off L2 Prefetch before sleep?
beq mpsL2PFok
mfspr r5,msscr0 ; Get currect MSSCR0 value
- rlwinm r5,r5,0,0,l2pfes-1 ; Dissable L2 Prefetch
+ rlwinm r5,r5,0,0,l2pfes-1 ; Disable L2 Prefetch
mtspr msscr0,r5 ; Updates MSSCR0 value
sync
isync
mpsL2PFok:
+ rlwinm. r5,r11,0,pf64Bitb,pf64Bitb ; PM bits are shifted on 64bit systems.
+ bne mpsPF64bit
+
+ rlwinm r4,r4,0,sleep+1,doze-1 ; Clear all possible power-saving modes (not DPM though)
+ oris r4,r4,hi16(sleepm) ; Set sleep
+ b mpsClearDEC
+
+mpsPF64bit:
+ lis r5, hi16(dozem|napm|sleepm) ; Clear all possible power-saving modes (not DPM though)
+ sldi r5, r5, 32
+ andc r4, r4, r5
+ lis r5, hi16(napm) ; Set sleep
+ sldi r5, r5, 32
+ or r4, r4, r5
+
+mpsClearDEC:
mfmsr r5 ; Get the current MSR
rlwinm r10,r10,0,1,31 ; Make 0x7FFFFFFF
- rlwinm r4,r4,0,sleep+1,doze-1 ; Clear all possible power-saving modes (not DPM though)
mtdec r10 ; Load decrimenter with 0x7FFFFFFF
isync ; and make sure,
mfdec r9 ; really sure, it gets there
mtcrf 0x07,r11 ; Get the cache flags, etc
- oris r4,r4,hi16(sleepm) ; Set sleep
rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Turn off translation
;
; Note that we need translation off before we set the HID to sleep. Otherwise
ori r3,r5,lo16(MASK(MSR_EE)) ; Flip on EE
sync
mtspr hid0,r4 ; Set up the HID to sleep
+ mfspr r4,hid0 ; Yes, this is silly, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
mtmsr r3 ; Enable for interrupts to drain decrimenter
mfmsr r5 ; Get the current MSR
oris r5,r5,hi16(MASK(MSR_POW)) ; Turn on power management in next MSR
; Leave EE off because power goes off shortly
-
-slSleepNow: sync ; Sync it all up
+ mfsprg r12,0 ; Get the per_proc_info
+ li r10,PP_CPU_FLAGS
+ lhz r11,PP_CPU_FLAGS(r12) ; Get the flags
+ ori r11,r11,SleepState ; Marked SleepState
+ sth r11,PP_CPU_FLAGS(r12) ; Set the flags
+ dcbf r10,r12
+
+ mfsprg r11,2 ; Get CPU specific features
+ rlwinm. r0,r11,0,pf64Bitb,pf64Bitb ; Test for 64 bit processor
+ eqv r4,r4,r4 ; Get all foxes
+ rlwinm r4,r4,0,1,31 ; Make 0x7FFFFFFF
+ beq slSleepNow ; skip if 32-bit...
+ li r3, 0x4000 ; Cause decrimenter to roll over soon
+ mtdec r3 ; Load decrimenter with 0x00004000
+ isync ; and make sure,
+ mfdec r3 ; really sure, it gets there
+
+slSleepNow:
+ sync ; Sync it all up
mtmsr r5 ; Do sleep with interruptions enabled
isync ; Take a pill
+ mtdec r4 ; Load decrimenter with 0x7FFFFFFF
+ isync ; and make sure,
+ mfdec r3 ; really sure, it gets there
b slSleepNow ; Go back to sleep if we wake up...
mfsprg r11,2 ; Get CPU specific features
mfmsr r7 ; Get the current MSR
- rlwinm r4,r9,0,dpm+1,doze-1 ; Clear all possible power-saving modes (also disable DPM)
rlwinm r7,r7,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
rlwinm r7,r7,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
rlwimi r11,r11,pfLClckb+1,31,31 ; Move pfLClck to another position (to keep from using non-volatile CRs)
rlwinm r5,r7,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Turn off translation
rlwinm r5,r5,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions
mtcrf 0x87,r11 ; Get the feature flags
+ lis r10,hi16(dozem|napm|sleepm|dpmm) ; Mask of power management bits
+ bf-- pf64Bitb,cIniNSF1 ; Skip if 32-bit...
+
+ sldi r10,r10,32 ; Position the masks
+
+cIniNSF1: andc r4,r9,r10 ; Clean up the old power bits
mtspr hid0,r4 ; Set up the HID
+ mfspr r4,hid0 ; Yes, this is silly, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r4,hid0 ; Yes, this is a duplicate, keep it here
bt pfNoMSRirb,ciNoMSR ; No MSR...
dssall ; Stop streams
sync
-cinoDSS: lis r5,hi16(EXT(tlb_system_lock)) ; Get the TLBIE lock
+cinoDSS: li r5,tlbieLock ; Get the TLBIE lock
li r0,128 ; Get number of TLB entries
- ori r5,r5,lo16(EXT(tlb_system_lock)) ; Grab up the bottom part
li r6,0 ; Start at 0
+ bf-- pf64Bitb,citlbhang ; Skip if 32-bit...
+ li r0,1024 ; Get the number of TLB entries
citlbhang: lwarx r2,0,r5 ; Get the TLBIE lock
mr. r2,r2 ; Is it locked?
sync
isync
-cinoSMP: stw r2,0(r5) ; Unlock TLBIE lock
+ bf-- pf64Bitb,cinoSMP ; Skip if 32-bit...
+ ptesync ; Wait for quiet again
+ sync
+
+cinoSMP: stw r2,tlbieLock(0) ; Unlock TLBIE lock
+
+ bt++ pf64Bitb,cin64 ; Skip if 64-bit...
- cror cr0_eq,pfL1ib,pfL1db ; Check for either I- or D-cache
- bf- cr0_eq,cinoL1 ; No level 1 to flush...
rlwinm. r0,r9,0,ice,dce ; Were either of the level 1s on?
beq- cinoL1 ; No, no need to flush...
- bf pfL1fab,ciswdl1 ; If no hw flush assist, go do by software...
+ rlwinm. r0,r11,0,pfL1fab,pfL1fab ; do we have L1 flush assist?
+ beq ciswdl1 ; If no hw flush assist, go do by software...
mfspr r8,msscr0 ; Get the memory system control register
oris r8,r8,hi16(dl1hwfm) ; Turn on the hardware flush request
;
; Flush and disable the level 2
;
- bf pfL2b,cinol2 ; No level 2 cache to flush
+ mfsprg r10,2 ; need to check 2 features we did not put in CR
+ rlwinm. r0,r10,0,pfL2b,pfL2b ; do we have L2?
+ beq cinol2 ; No level 2 cache to flush
mfspr r8,l2cr ; Get the L2CR
lwz r3,pfl2cr(r12) ; Get the L2CR value
bne- ciinvdl2 ; Yes, just invalidate and get PLL synced...
ciflushl2:
- bf pfL2fab,ciswfl2 ; Flush not in hardware...
+ rlwinm. r0,r10,0,pfL2fab,pfL2fab ; hardware-assisted L2 flush?
+ beq ciswfl2 ; Flush not in hardware...
mr r10,r8 ; Take a copy now
addi r10,r10,32 ; Next line
bdnz ciswfldl2a ; Do the lot...
-ciinvdl2: rlwinm r8,r3,0,l2e+1,31 ; Use the saved L2CR and clear the enable bit
+ciinvdl2: rlwinm r8,r3,0,l2e+1,31 ; Clear the enable bit
b cinla ; Branch to next line...
.align 5
sync
isync
ciinvdl2a: mfspr r2,l2cr ; Get the L2CR
- bf pfL2ib,ciinvdl2b ; Flush not in hardware...
+ mfsprg r0,2 ; need to check a feature in "non-volatile" set
+ rlwinm. r0,r0,0,pfL2ib,pfL2ib ; flush in HW?
+ beq ciinvdl2b ; Flush not in hardware...
rlwinm. r2,r2,0,l2i,l2i ; Is the invalidate still going?
bne+ ciinvdl2a ; Assume so, this will take a looong time...
sync
rlwinm. r10,r10,0,l3hwf,l3hwf ; Is the flush over?
bne+ cihwfl3 ; Nope, keep going...
-ciinvdl3: rlwinm r8,r3,0,l3e+1,31 ; Use saved L3CR value and clear the enable bit
+ciinvdl3: rlwinm r8,r3,0,l3e+1,31 ; Clear the enable bit
sync ; Make sure of life, liberty, and justice
mtspr l3cr,r8 ; Disable L3
sync
mtspr l3cr,r3 ; Enable it as desired
sync
cinol3:
- bf pfL2b,cinol2a ; No level 2 cache to enable
+ mfsprg r0,2 ; need to check a feature in "non-volatile" set
+ rlwinm. r0,r0,0,pfL2b,pfL2b ; is there an L2 cache?
+ beq cinol2a ; No level 2 cache to enable
lwz r3,pfl2cr(r12) ; Get the L2CR value
cmplwi r3, 0 ; Should the L2 be all the way off?
- beq cinol2a : Yes, done with L2
+ beq cinol2a : Yes, done with L2
mtspr l2cr,r3 ; Enable it as desired
sync
blr ; Return...
+;
+; Handle 64-bit architecture
+; This processor can not run without caches, so we just push everything out
+; and flush. It will be relativily clean afterwards
+;
+
+ .align 5
+
+cin64:
+ mfspr r10,hid1 ; Save hid1
+ mfspr r4,hid4 ; Save hid4
+ mr r12,r10 ; Really save hid1
+ mr r11,r4 ; Get a working copy of hid4
+
+ li r0,0 ; Get a 0
+ eqv r2,r2,r2 ; Get all foxes
+
+ rldimi r10,r0,55,7 ; Clear I$ prefetch bits (7:8)
+
+ isync
+ mtspr hid1,r10 ; Stick it
+ mtspr hid1,r10 ; Stick it again
+ isync
+
+ rldimi r11,r2,38,25 ; Disable D$ prefetch (25:25)
+
+ sync
+ mtspr hid4,r11 ; Stick it
+ isync
+
+ li r3,8 ; Set bit 28+32
+ sldi r3,r3,32 ; Make it bit 28
+ or r3,r3,r11 ; Turn on the flash invalidate L1D$
+
+ oris r5,r11,0x0600 ; Set disable L1D$ bits
+ sync
+ mtspr hid4,r3 ; Invalidate
+ isync
+
+ mtspr hid4,r5 ; Un-invalidate and disable L1D$
+ isync
+
+ lis r8,GUSModeReg ; Get the GUS mode ring address
+ mfsprg r0,2 ; Get the feature flags
+ ori r8,r8,0x8000 ; Set to read data
+ rlwinm. r0,r0,pfSCOMFixUpb+1,31,31 ; Set shift if we need a fix me up
+
+ sync
+
+ mtspr scomc,r8 ; Request the GUS mode
+ mfspr r11,scomd ; Get the GUS mode
+ mfspr r8,scomc ; Get back the status (we just ignore it)
+ sync
+ isync
+
+ sld r11,r11,r0 ; Fix up if needed
+
+ ori r6,r11,lo16(GUSMdmapen) ; Set the bit that means direct L2 cache address
+ lis r8,GUSModeReg ; Get GUS mode register address
+
+ sync
+
+ mtspr scomd,r6 ; Set that we want direct L2 mode
+ mtspr scomc,r8 ; Tell GUS we want direct L2 mode
+ mfspr r3,scomc ; Get back the status
+ sync
+ isync
+
+ li r3,0 ; Clear start point
+
+cflushlp: lis r6,0x0040 ; Pick 4MB line as our target
+ or r6,r6,r3 ; Put in the line offset
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+ addis r6,r6,8 ; Roll bit 42:44
+ lwz r5,0(r6) ; Load a line
+
+ addi r3,r3,128 ; Next line
+ andis. r5,r3,8 ; Have we done enough?
+ beq++ cflushlp ; Not yet...
+
+ sync
+
+ lis r6,0x0040 ; Pick 4MB line as our target
+
+cflushx: dcbf 0,r6 ; Flush line and invalidate
+ addi r6,r6,128 ; Next line
+ andis. r5,r6,0x0080 ; Have we done enough?
+ beq++ cflushx ; Keep on flushing...
+
+ mr r3,r10 ; Copy current hid1
+ rldimi r3,r2,54,9 ; Set force icbi match mode
+
+ li r6,0 ; Set start if ICBI range
+ isync
+ mtspr hid1,r3 ; Stick it
+ mtspr hid1,r3 ; Stick it again
+ isync
+
+cflicbi: icbi 0,r6 ; Kill I$
+ addi r6,r6,128 ; Next line
+ andis. r5,r6,1 ; Have we done them all?
+ beq++ cflicbi ; Not yet...
+
+ lis r8,GUSModeReg ; Get GUS mode register address
+
+ sync
+
+ mtspr scomd,r11 ; Set that we do not want direct mode
+ mtspr scomc,r8 ; Tell GUS we do not want direct mode
+ mfspr r3,scomc ; Get back the status
+ sync
+ isync
+
+ isync
+ mtspr hid0,r9 ; Restore entry hid0
+ mfspr r9,hid0 ; Yes, this is silly, keep it here
+ mfspr r9,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r9,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r9,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r9,hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r9,hid0 ; Yes, this is a duplicate, keep it here
+ isync
+
+ isync
+ mtspr hid1,r12 ; Restore entry hid1
+ mtspr hid1,r12 ; Stick it again
+ isync
+
+ sync
+ mtspr hid4,r4 ; Restore entry hid4
+ isync
+
+ sync
+ mtmsr r7 ; Restore MSR to entry
+ isync
+ blr ; Return...
+
+
+
/* Disables all caches
*
* void cacheDisable(void)
cdNoAlt: sync
+ btlr pf64Bitb ; No way to disable a 64-bit machine...
+
mfspr r5,hid0 ; Get the hid
rlwinm r5,r5,0,dce+1,ice-1 ; Clear the I- and D- cache enables
mtspr hid0,r5 ; Turn off dem caches
sync
- bf pfL2b,cdNoL2 ; Skip if no L2...
-
+ rlwinm. r0,r11,0,pfL2b,pfL2b ; is there an L2?
+ beq cdNoL2 ; Skip if no L2...
+
mfspr r5,l2cr ; Get the L2
rlwinm r5,r5,0,l2e+1,31 ; Turn off enable bit
b cinlcc ; Jump back up and turn off cache...
cdNoL2:
-
+
bf pfL3b,cdNoL3 ; Skip down if no L3...
mfspr r5,l3cr ; Get the L3
/* Initialize processor thermal monitoring
* void ml_thrm_init(void)
*
- * Build initial TAU registers and start them all going.
- * We ca not do this at initial start up because we need to have the processor frequency first.
- * And just why is this in assembler when it does not have to be?? Cause I am just too
- * lazy to open up a "C" file, thats why.
+ * Obsolete, deprecated and will be removed.
*/
; Force a line boundry here
.globl EXT(ml_thrm_init)
LEXT(ml_thrm_init)
-
- mfsprg r12,0 ; Get the per_proc blok
- lis r11,hi16(EXT(gPEClockFrequencyInfo)) ; Get top of processor information
- mfsprg r10,2 ; Get CPU specific features
- ori r11,r11,lo16(EXT(gPEClockFrequencyInfo)) ; Get bottom of processor information
- mtcrf 0x40,r10 ; Get the installed features
-
- li r3,lo16(thrmtidm|thrmvm) ; Set for lower-than thermal event at 0 degrees
- bflr pfThermalb ; No thermal monitoring on this cpu
- mtspr thrm1,r3 ; Do it
-
- lwz r3,thrmthrottleTemp(r12) ; Get our throttle temprature
- rlwinm r3,r3,31-thrmthre,thrmthrs,thrmthre ; Position it
- ori r3,r3,lo16(thrmvm) ; Set for higher-than event
- mtspr thrm2,r3 ; Set it
-
- lis r4,hi16(1000000) ; Top of million
-;
-; Note: some CPU manuals say this is processor clocks, some say bus rate. The latter
-; makes more sense because otherwise we can not get over about 400MHz.
-#if 0
- lwz r3,PECFIcpurate(r11) ; Get the processor speed
-#else
- lwz r3,PECFIbusrate(r11) ; Get the bus speed
-#endif
- ori r4,r4,lo16(1000000) ; Bottom of million
- lis r7,hi16(thrmsitvm>>1) ; Get top of highest possible value
- divwu r3,r3,r4 ; Get number of cycles per microseconds
- ori r7,r7,lo16(thrmsitvm>>1) ; Get the bottom of the highest possible value
- addi r3,r3,1 ; Insure we have enough
- mulli r3,r3,20 ; Get 20 microseconds worth of cycles
- cmplw r3,r7 ; Check against max
- ble+ smallenuf ; It is ok...
- mr r3,r7 ; Saturate
-
-smallenuf: rlwinm r3,r3,31-thrmsitve,thrmsitvs,thrmsitve ; Position
- ori r3,r3,lo16(thrmem) ; Enable with at least 20micro sec sample
- stw r3,thrm3val(r12) ; Save this in case we need it later
- mtspr thrm3,r3 ; Do it
blr
-
/* Set thermal monitor bounds
* void ml_thrm_set(unsigned int low, unsigned int high)
*
- * Set TAU to interrupt below low and above high. A value of
- * zero disables interruptions in that direction.
+ * Obsolete, deprecated and will be removed.
*/
; Force a line boundry here
.globl EXT(ml_thrm_set)
LEXT(ml_thrm_set)
-
- mfmsr r0 ; Get the MSR
- rlwinm r0,r0,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r0,r0,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r6,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear EE bit
- mtmsr r6
- isync
-
- mfsprg r12,0 ; Get the per_proc blok
-
- rlwinm. r6,r3,31-thrmthre,thrmthrs,thrmthre ; Position it and see if enabled
- mfsprg r9,2 ; Get CPU specific features
- stw r3,thrmlowTemp(r12) ; Set the low temprature
- mtcrf 0x40,r9 ; See if we can thermal this machine
- rlwinm r9,r9,(((31-thrmtie)+(pfThermIntb+1))&31),thrmtie,thrmtie ; Set interrupt enable if this machine can handle it
- bf pfThermalb,tsetcant ; No can do...
- beq tsetlowo ; We are setting the low off...
- ori r6,r6,lo16(thrmtidm|thrmvm) ; Set the lower-than and valid bit
- or r6,r6,r9 ; Set interruption request if supported
-
-tsetlowo: mtspr thrm1,r6 ; Cram the register
-
- rlwinm. r6,r4,31-thrmthre,thrmthrs,thrmthre ; Position it and see if enabled
- stw r4,thrmhighTemp(r12) ; Set the high temprature
- beq tsethigho ; We are setting the high off...
- ori r6,r6,lo16(thrmvm) ; Set valid bit
- or r6,r6,r9 ; Set interruption request if supported
-
-tsethigho: mtspr thrm2,r6 ; Cram the register
-
-tsetcant: mtmsr r0 ; Reenable interruptions
- blr ; Leave...
+ blr
/* Read processor temprature
* unsigned int ml_read_temp(void)
*
+ * Obsolete, deprecated and will be removed.
*/
; Force a line boundry here
.globl EXT(ml_read_temp)
LEXT(ml_read_temp)
-
- mfmsr r9 ; Save the MSR
- li r5,15 ; Starting point for ranging (start at 15 so we do not overflow)
- rlwinm r9,r9,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r9,r9,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r8,r9,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions
- mfsprg r7,2 ; Get CPU specific features
- mtmsr r8 ; Do not allow interruptions
- mtcrf 0x40,r7 ; See if we can thermal this machine
- bf pfThermalb,thrmcant ; No can do...
-
- mfspr r11,thrm1 ; Save thrm1
-
-thrmrange: rlwinm r4,r5,31-thrmthre,thrmthrs,thrmthre ; Position it
- ori r4,r4,lo16(thrmtidm|thrmvm) ; Flip on the valid bit and make comparision for less than
-
- mtspr thrm1,r4 ; Set the test value
-
-thrmreada: mfspr r3,thrm1 ; Get the thermal register back
- rlwinm. r0,r3,0,thrmtiv,thrmtiv ; Has it settled yet?
- beq+ thrmreada ; Nope...
-
- rlwinm. r0,r3,0,thrmtin,thrmtin ; Are we still under the threshold?
- bne thrmsearch ; No, we went over...
-
- addi r5,r5,16 ; Start by trying every 16 degrees
- cmplwi r5,127 ; Have we hit the max?
- blt- thrmrange ; Got some more to do...
-
-thrmsearch: rlwinm r4,r5,31-thrmthre,thrmthrs,thrmthre ; Position it
- ori r4,r4,lo16(thrmtidm|thrmvm) ; Flip on the valid bit and make comparision for less than
-
- mtspr thrm1,r4 ; Set the test value
-
-thrmread: mfspr r3,thrm1 ; Get the thermal register back
- rlwinm. r0,r3,0,thrmtiv,thrmtiv ; Has it settled yet?
- beq+ thrmread ; Nope...
-
- rlwinm. r0,r3,0,thrmtin,thrmtin ; Are we still under the threshold?
- beq thrmdone ; No, we hit it...
- addic. r5,r5,-1 ; Go down a degree
- bge+ thrmsearch ; Try again (until we are below freezing)...
-
-thrmdone: addi r3,r5,1 ; Return the temprature (bump it up to make it correct)
- mtspr thrm1,r11 ; Restore the thermal register
- mtmsr r9 ; Re-enable interruptions
- blr ; Leave...
-
-thrmcant: eqv r3,r3,r3 ; Return bogus temprature because we can not read it
- mtmsr r9 ; Re-enable interruptions
- blr ; Leave...
+ li r3,-1
+ blr
/* Throttle processor speed up or down
* unsigned int ml_throttle(unsigned int step)
* Returns old speed and sets new. Both step and return are values from 0 to
* 255 that define number of throttle steps, 0 being off and "ictcfim" is max * 2.
*
+ * Obsolete, deprecated and will be removed.
*/
; Force a line boundry here
.globl EXT(ml_throttle)
LEXT(ml_throttle)
-
- mfmsr r9 ; Save the MSR
- rlwinm r9,r9,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r9,r9,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r8,r9,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions
- cmplwi r3,lo16(ictcfim>>1) ; See if we are going too far
- mtmsr r8 ; Do not allow interruptions
- isync
- ble+ throtok ; Throttle value is ok...
- li r3,lo16(ictcfim>>1) ; Set max
-
-throtok: rlwinm. r4,r3,1,ictcfib,ictcfie ; Set the throttle
- beq throtoff ; Skip if we are turning it off...
- ori r4,r4,lo16(thrmvm) ; Turn on the valid bit
-
-throtoff: mfspr r3,ictc ; Get the old throttle
- mtspr ictc,r4 ; Set the new
- rlwinm r3,r3,31,1,31 ; Shift throttle value over
- mtmsr r9 ; Restore interruptions
- blr ; Return...
+ li r3,0
+ blr
/*
** ml_get_timebase()
LEXT(ml_get_timebase)
loop:
- mftbu r4
- mftb r5
- mftbu r6
- cmpw r6, r4
- bne- loop
+ mftbu r4
+ mftb r5
+ mftbu r6
+ cmpw r6, r4
+ bne- loop
+
+ stw r4, 0(r3)
+ stw r5, 4(r3)
+
+ blr
- stw r4, 0(r3)
- stw r5, 4(r3)
+/*
+ * unsigned int cpu_number(void)
+ *
+ * Returns the current cpu number.
+ */
+
+ .align 5
+ .globl EXT(cpu_number)
+
+LEXT(cpu_number)
+ mfsprg r4,0 ; Get per-proc block
+ lhz r3,PP_CPU_NUMBER(r4) ; Get CPU number
+ blr ; Return...
- blr
/*
- * The routine that implements cpu_number.
+ * void set_machine_current_act(thread_act_t)
+ *
+ * Set the current activation
*/
+ .align 5
+ .globl EXT(set_machine_current_act)
- .align 5
- .globl EXT(cpu_number)
-
-LEXT(cpu_number)
-
- mfmsr r9 /* Save the old MSR */
- rlwinm r9,r9,0,MSR_FP_BIT+1,MSR_FP_BIT-1 ; Force floating point off
- rlwinm r9,r9,0,MSR_VEC_BIT+1,MSR_VEC_BIT-1 ; Force vectors off
- rlwinm r8,r9,0,17,15 /* Clear interruptions */
- mtmsr r8 /* Interrupts off */
- isync
- mfsprg r7,0 /* Get per-proc block */
- lhz r3,PP_CPU_NUMBER(r7) /* Get CPU number */
- mtmsr r9 /* Restore interruptions to entry */
- blr /* Return... */
+LEXT(set_machine_current_act)
+
+ mtsprg 1,r3 ; Set spr1 with the active thread
+ blr ; Return...
+
+/*
+ * thread_t current_act(void)
+ * thread_t current_thread(void)
+ *
+ *
+ * Return the current thread for outside components.
+ */
+ .align 5
+ .globl EXT(current_act)
+ .globl EXT(current_thread)
+
+LEXT(current_act)
+LEXT(current_thread)
+
+ mfsprg r3,1
+ blr
+
+ .align 5
+ .globl EXT(clock_get_uptime)
+LEXT(clock_get_uptime)
+1: mftbu r9
+ mftb r0
+ mftbu r11
+ cmpw r11,r9
+ bne- 1b
+ stw r0,4(r3)
+ stw r9,0(r3)
+ blr
+
+
+ .align 5
+ .globl EXT(mach_absolute_time)
+LEXT(mach_absolute_time)
+1: mftbu r3
+ mftb r4
+ mftbu r0
+ cmpw r0,r3
+ bne- 1b
+ blr
/*
** ml_sense_nmi()
.globl EXT(ml_set_processor_speed)
LEXT(ml_set_processor_speed)
- mfsprg r5, 0 ; Get the per_proc_info
+ mflr r0 ; Save the link register
+ stwu r1, -(FM_ALIGN(4*4)+FM_SIZE)(r1) ; Make some space on the stack
+ stw r28, FM_ARG0+0x00(r1) ; Save a register
+ stw r29, FM_ARG0+0x04(r1) ; Save a register
+ stw r30, FM_ARG0+0x08(r1) ; Save a register
+ stw r31, FM_ARG0+0x0C(r1) ; Save a register
+ stw r0, (FM_ALIGN(4*4)+FM_SIZE+FM_LR_SAVE)(r1) ; Save the return
+
+ mfsprg r31, 0 ; Get the per_proc_info
+
+ lwz r30, pfPowerModes(r31) ; Get the supported power modes
+
+ rlwinm. r0, r30, 0, pmDualPLLb, pmDualPLLb ; Is DualPLL supported?
+ bne spsDPLL
+
+ rlwinm. r0, r30, 0, pmDFSb, pmDFSb ; Is DFS supported?
+ bne spsDFS
+
+ rlwinm. r0, r30, 0, pmPowerTuneb, pmPowerTuneb ; Is PowerTune supported?
+ bne spsPowerTune
+
+ b spsDone ; No supported power modes
- cmpli cr0, r3, 0 ; Turn off BTIC before low speed
- beq sps1
- mfspr r4, hid0 ; Get the current hid0 value
- rlwinm r4, r4, 0, btic+1, btic-1 ; Clear the BTIC bit
+spsDPLL:
+ cmpli cr0, r3, 0 ; Turn off BTIC before low speed
+ beq spsDPLL1
+ mfspr r4, hid0 ; Get the current hid0 value
+ rlwinm r4, r4, 0, btic+1, btic-1 ; Clear the BTIC bit
sync
- mtspr hid0, r4 ; Set the new hid0 value
+ mtspr hid0, r4 ; Set the new hid0 value
isync
sync
-sps1:
- mfspr r4, hid1 ; Get the current PLL settings
- rlwimi r4, r3, 31-hid1ps, hid1ps, hid1ps ; Copy the PLL Select bit
- stw r4, pfHID1(r5) ; Save the new hid1 value
- mtspr hid1, r4 ; Select desired PLL
+spsDPLL1:
+ mfspr r4, hid1 ; Get the current PLL settings
+ rlwimi r4, r3, 31-hid1ps, hid1ps, hid1ps ; Copy the PLL Select bit
+ stw r4, pfHID1(r31) ; Save the new hid1 value
+ mtspr hid1, r4 ; Select desired PLL
- cmpli cr0, r3, 0 ; Restore BTIC after high speed
- bne sps2
- lwz r4, pfHID0(r5) ; Load the hid0 value
+ cmpli cr0, r3, 0 ; Restore BTIC after high speed
+ bne spsDone
+ lwz r4, pfHID0(r31) ; Load the hid0 value
sync
- mtspr hid0, r4 ; Set the hid0 value
+ mtspr hid0, r4 ; Set the hid0 value
isync
sync
+ b spsDone
-sps2:
+spsDFS:
+ cmplwi r3, 0 ; full speed?
+ mfspr r3, hid1 ; Get the current HID1
+ rlwinm r3, r3, 0, hid1dfs1+1, hid1dfs0-1 ; assume full speed, clear dfs bits
+ beq spsDFS1
+ oris r3, r3, hi16(hid1dfs1m) ; slow, set half speed dfs1 bit
+
+spsDFS1:
+ stw r3, pfHID1(r31) ; Save the new hid1 value
+ sync
+ mtspr hid1, r3 ; Set the new HID1
+ sync
+ isync
+ b spsDone
+
+spsPowerTune:
+ rlwinm r28, r3, 31-dnap, dnap, dnap ; Shift the 1 bit to the dnap+32 bit
+ rlwinm r3, r3, 2, 29, 29 ; Shift the 1 to a 4 and mask
+ addi r3, r3, pfPowerTune0 ; Add in the pfPowerTune0 offset
+ lwzx r29, r31, r3 ; Load the PowerTune number 0 or 1
+
+ sldi r28, r28, 32 ; Shift to the top half
+ ld r3, pfHID0(r31) ; Load the saved hid0 value
+ and r28, r28, r3 ; Save the dnap bit
+ lis r4, hi16(dnapm) ; Make a mask for the dnap bit
+ sldi r4, r4, 32 ; Shift to the top half
+ andc r3, r3, r4 ; Clear the dnap bit
+ or r28, r28, r3 ; Insert the dnap bit as needed for later
+
+ sync
+ mtspr hid0, r3 ; Turn off dnap in hid0
+ mfspr r3, hid0 ; Yes, this is silly, keep it here
+ mfspr r3, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r3, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r3, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r3, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r3, hid0 ; Yes, this is a duplicate, keep it here
+ isync ; Make sure it is set
+
+ lis r3, hi16(PowerTuneControlReg) ; Write zero to the PCR
+ ori r3, r3, lo16(PowerTuneControlReg)
+ li r4, 0
+ li r5, 0
+ bl _ml_scom_write
+
+ lis r3, hi16(PowerTuneControlReg) ; Write the PowerTune value to the PCR
+ ori r3, r3, lo16(PowerTuneControlReg)
+ li r4, 0
+ mr r5, r29
+ bl _ml_scom_write
+
+ rlwinm r29, r29, 13-6, 6, 7 ; Move to PSR speed location and isolate the requested speed
+spsPowerTuneLoop:
+ lis r3, hi16(PowerTuneStatusReg) ; Read the status from the PSR
+ ori r3, r3, lo16(PowerTuneStatusReg)
+ li r4, 0
+ bl _ml_scom_read
+ srdi r5, r5, 32
+ rlwinm r0, r5, 0, 6, 7 ; Isolate the current speed
+ rlwimi r0, r5, 0, 2, 2 ; Copy in the change in progress bit
+ cmpw r0, r29 ; Compare the requested and current speeds
+ beq spsPowerTuneDone
+ rlwinm. r0, r5, 0, 3, 3
+ beq spsPowerTuneLoop
+
+spsPowerTuneDone:
+ sync
+ mtspr hid0, r28 ; Turn on dnap in hid0 if needed
+ mfspr r28, hid0 ; Yes, this is silly, keep it here
+ mfspr r28, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r28, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r28, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r28, hid0 ; Yes, this is a duplicate, keep it here
+ mfspr r28, hid0 ; Yes, this is a duplicate, keep it here
+ isync ; Make sure it is set
+
+ b spsDone
+
+spsDone:
+ lwz r0, (FM_ALIGN(4*4)+FM_SIZE+FM_LR_SAVE)(r1) ; Get the return
+ lwz r28, FM_ARG0+0x00(r1) ; Restore a register
+ lwz r29, FM_ARG0+0x04(r1) ; Restore a register
+ lwz r30, FM_ARG0+0x08(r1) ; Restore a register
+ lwz r31, FM_ARG0+0x0C(r1) ; Restore a register
+ lwz r1, FM_BACKPTR(r1) ; Pop the stack
+ mtlr r0
blr
/*
.globl EXT(ml_set_processor_voltage)
LEXT(ml_set_processor_voltage)
- mfspr r4, hid2 ; Get HID2 value
- rlwimi r4, r3, 31-hid2vmin, hid2vmin, hid2vmin ; Insert the voltage mode bit
- mtspr hid2, r4 ; Set the voltage mode
- sync ; Make sure it is done
+ mfsprg r5, 0 ; Get the per_proc_info
+
+ lwz r6, pfPowerModes(r5) ; Get the supported power modes
+
+ rlwinm. r0, r6, 0, pmDPLLVminb, pmDPLLVminb ; Is DPLL Vmin supported
+ beq spvDone
+
+ mfspr r4, hid2 ; Get HID2 value
+ rlwimi r4, r3, 31-hid2vmin, hid2vmin, hid2vmin ; Insert the voltage mode bit
+ mtspr hid2, r4 ; Set the voltage mode
+ sync ; Make sure it is done
+
+spvDone:
blr
+
+
+;
+; unsigned int ml_scom_write(unsigned int reg, unsigned long long data)
+; 64-bit machines only
+; returns status
+;
+
+ .align 5
+ .globl EXT(ml_scom_write)
+
+LEXT(ml_scom_write)
+
+ rldicr r3,r3,8,47 ; Align register it correctly
+ rldimi r5,r4,32,0 ; Merge the high part of data
+ sync ; Clean up everything
+
+ mtspr scomd,r5 ; Stick in the data
+ mtspr scomc,r3 ; Set write to register
+ sync
+ isync
+
+ mfspr r3,scomc ; Read back status
+ blr ; leave....
+
+;
+; unsigned int ml_read_scom(unsigned int reg, unsigned long long *data)
+; 64-bit machines only
+; returns status
+; ASM Callers: data (r4) can be zero and the 64 bit data will be returned in r5
+;
+
+ .align 5
+ .globl EXT(ml_scom_read)
+
+LEXT(ml_scom_read)
+
+ mfsprg r0,2 ; Get the feature flags
+ rldicr r3,r3,8,47 ; Align register it correctly
+ rlwinm r0,r0,pfSCOMFixUpb+1,31,31 ; Set shift if we need a fix me up
+
+ ori r3,r3,0x8000 ; Set to read data
+ sync
+
+ mtspr scomc,r3 ; Request the register
+ mfspr r5,scomd ; Get the register contents
+ mfspr r3,scomc ; Get back the status
+ sync
+ isync
+
+ sld r5,r5,r0 ; Fix up if needed
+
+ cmplwi r4, 0 ; If data pointer is null, just return
+ beqlr ; the received data in r5
+ std r5,0(r4) ; Pass back the received data
+ blr ; Leave...