-#if 0
-static void setup_coherent_ht_domain(void)
-{
- static const unsigned int register_values[] = {
- /* Routing Table Node i
- * F0:0x40 i = 0,
- * F0:0x44 i = 1,
- * F0:0x48 i = 2,
- * F0:0x4c i = 3,
- * F0:0x50 i = 4,
- * F0:0x54 i = 5,
- * F0:0x58 i = 6,
- * F0:0x5c i = 7
- * [ 0: 3] Request Route
- * [0] Route to this node
- * [1] Route to Link 0
- * [2] Route to Link 1
- * [3] Route to Link 2
- * [11: 8] Response Route
- * [0] Route to this node
- * [1] Route to Link 0
- * [2] Route to Link 1
- * [3] Route to Link 2
- * [19:16] Broadcast route
- * [0] Route to this node
- * [1] Route to Link 0
- * [2] Route to Link 1
- * [3] Route to Link 2
- */
- PCI_ADDR(0, 0x18, 0, 0x40), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x44), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x48), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x4c), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x50), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x54), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x58), 0xfff0f0f0, 0x00010101,
- PCI_ADDR(0, 0x18, 0, 0x5c), 0xfff0f0f0, 0x00010101,
-
- /* Hypetransport Transaction Control Register
- * F0:0x68
- * [ 0: 0] Disable read byte probe
- * 0 = Probes issues
- * 1 = Probes not issued
- * [ 1: 1] Disable Read Doubleword probe
- * 0 = Probes issued
- * 1 = Probes not issued
- * [ 2: 2] Disable write byte probes
- * 0 = Probes issued
- * 1 = Probes not issued
- * [ 3: 3] Disable Write Doubleword Probes
- * 0 = Probes issued
- * 1 = Probes not issued.
- * [ 4: 4] Disable Memroy Controller Target Start
- * 0 = TgtStart packets are generated
- * 1 = TgtStart packets are not generated.
- * [ 5: 5] CPU1 Enable
- * 0 = Second CPU disabled or not present
- * 1 = Second CPU enabled.
- * [ 6: 6] CPU Request PassPW
- * 0 = CPU requests do not pass posted writes
- * 1 = CPU requests pass posted writes.
- * [ 7: 7] CPU read Respons PassPW
- * 0 = CPU Responses do not pass posted writes
- * 1 = CPU responses pass posted writes.
- * [ 8: 8] Disable Probe Memory Cancel
- * 0 = Probes may generate MemCancels
- * 1 = Probes may not generate MemCancels
- * [ 9: 9] Disable Remote Probe Memory Cancel.
- * 0 = Probes hitting dirty blocks generate memory cancel packets
- * 1 = Only probed caches on the same node as the memory controller
- * generate cancel packets.
- * [10:10] Disable Fill Probe
- * 0 = Probes issued for cache fills
- * 1 = Probes not issued for cache fills.
- * [11:11] Response PassPw
- * 0 = Downstream response PassPW based on original request
- * 1 = Downstream response PassPW set to 1
- * [12:12] Change ISOC to Ordered
- * 0 = Bit 1 of coherent HT RdSz/WrSz command used for iosynchronous prioritization
- * 1 = Bit 1 of coherent HT RdSz/WrSz command used for ordering.
- * [14:13] Buffer Release Priority select
- * 00 = 64
- * 01 = 16
- * 10 = 8
- * 11 = 2
- * [15:15] Limit Coherent HT Configuration Space Range
- * 0 = No coherent HT configuration space restrictions
- * 1 = Limit coherent HT configuration space based on node count
- * [16:16] Local Interrupt Conversion Enable.
- * 0 = ExtInt/NMI interrups unaffected.
- * 1 = ExtInt/NMI broadcat interrupts converted to LINT0/1
- * [17:17] APIC Extended Broadcast Enable.
- * 0 = APIC broadcast is 0F
- * 1 = APIC broadcast is FF
- * [18:18] APIC Extended ID Enable
- * 0 = APIC ID is 4 bits.
- * 1 = APIC ID is 8 bits.
- * [19:19] APIC Extended Spurious Vector Enable
- * 0 = Lower 4 bits of spurious vector are read-only 1111
- * 1 = Lower 4 bits of spurious vecotr are writeable.
- * [20:20] Sequence ID Source Node Enable
- * 0 = Normal operation
- * 1 = Keep SeqID on routed packets for debugging.
- * [22:21] Downstream non-posted request limit
- * 00 = No limit
- * 01 = Limited to 1
- * 10 = Limited to 4
- * 11 = Limited to 8
- * [23:23] RESERVED
- * [25:24] Medium-Priority Bypass Count
- * - Maximum # of times a medium priority access can pass a low
- * priority access before Medium-Priority mode is disabled for one access.
- * [27:26] High-Priority Bypass Count
- * - Maximum # of times a high prioirty access can pass a medium or low
- * priority access before High-prioirty mode is disabled for one access.
- * [28:28] Enable High Priority CPU Reads
- * 0 = Cpu reads are medium prioirty
- * 1 = Cpu reads are high prioirty
- * [29:29] Disable Low Priority Writes
- * 0 = Non-isochronous writes are low priority
- * 1 = Non-isochronous writes are medium prioirty
- * [30:30] Disable High Priority Isochronous writes
- * 0 = Isochronous writes are high priority
- * 1 = Isochronous writes are medium priority
- * [31:31] Disable Medium Priority Isochronous writes
- * 0 = Isochronous writes are medium are high
- * 1 = With bit 30 set makes Isochrouns writes low priority.
- */
- PCI_ADDR(0, 0x18, 0, 0x68), 0x00800000, 0x0f00840f,
- /* HT Initialization Control Register
- * F0:0x6C ok...
- * [ 0: 0] Routing Table Disable
- * 0 = Packets are routed according to routing tables
- * 1 = Packets are routed according to the default link field
- * [ 1: 1] Request Disable (BSP should clear this)
- * 0 = Request packets may be generated
- * 1 = Request packets may not be generated.
- * [ 3: 2] Default Link (Read-only)
- * 00 = LDT0
- * 01 = LDT1
- * 10 = LDT2
- * 11 = CPU on same node
- * [ 4: 4] Cold Reset
- * - Scratch bit cleared by a cold reset
- * [ 5: 5] BIOS Reset Detect
- * - Scratch bit cleared by a cold reset
- * [ 6: 6] INIT Detect
- * - Scratch bit cleared by a warm or cold reset not by an INIT
- *
- */
- PCI_ADDR(0, 0x18, 0, 0x6C), 0xffffff8c, 0x00000000 | (1 << 6) |(1 << 5)| (1 << 4),
- /* LDTi Capabilities Registers
- * F0:0x80 i = 0,
- * F0:0xA0 i = 1,
- * F0:0xC0 i = 2,
- */
- /* LDTi Link Control Registrs
- * F0:0x84 i = 0,
- * F0:0xA4 i = 1,
- * F0:0xC4 i = 2,
- * [ 1: 1] CRC Flood Enable
- * 0 = Do not generate sync packets on CRC error
- * 1 = Generate sync packets on CRC error
- * [ 2: 2] CRC Start Test (Read-Only)
- * [ 3: 3] CRC Force Frame Error
- * 0 = Do not generate bad CRC
- * 1 = Generate bad CRC
- * [ 4: 4] Link Failure
- * 0 = No link failure detected
- * 1 = Link failure detected
- * [ 5: 5] Initialization Complete
- * 0 = Initialization not complete
- * 1 = Initialization complete
- * [ 6: 6] Receiver off
- * 0 = Recevier on
- * 1 = Receiver off
- * [ 7: 7] Transmitter Off
- * 0 = Transmitter on
- * 1 = Transmitter off
- * [ 9: 8] CRC_Error
- * 00 = No error
- * [0] = 1 Error on byte lane 0
- * [1] = 1 Error on byte lane 1
- * [12:12] Isochrnous Enable (Read-Only)
- * [13:13] HT Stop Tristate Enable
- * 0 = Driven during an LDTSTOP_L
- * 1 = Tristated during and LDTSTOP_L
- * [14:14] Extended CTL Time
- * 0 = CTL is asserted for 16 bit times during link initialization
- * 1 = CTL is asserted for 50us during link initialization
- * [18:16] Max Link Width In (Read-Only?)
- * 000 = 8 bit link
- * 001 = 16bit link
- * [19:19] Doubleword Flow Control in (Read-Only)
- * 0 = This link does not support doubleword flow control
- * 1 = This link supports doubleword flow control
- * [22:20] Max Link Width Out (Read-Only?)
- * 000 = 8 bit link
- * 001 = 16bit link
- * [23:23] Doubleworld Flow Control out (Read-Only)
- * 0 = This link does not support doubleword flow control
- * 1 = This link supports doubleworkd flow control
- * [26:24] Link Width In
- * 000 = Use 8 bits
- * 001 = Use 16 bits
- * 010 = reserved
- * 011 = Use 32 bits
- * 100 = Use 2 bits
- * 101 = Use 4 bits
- * 110 = reserved
- * 111 = Link physically not connected
- * [27:27] Doubleword Flow Control In Enable
- * 0 = Doubleword flow control disabled
- * 1 = Doubleword flow control enabled (Not currently supported)
- * [30:28] Link Width Out
- * 000 = Use 8 bits
- * 001 = Use 16 bits
- * 010 = reserved
- * 011 = Use 32 bits
- * 100 = Use 2 bits
- * 101 = Use 4 bits
- * 110 = reserved
- * 111 = Link physically not connected
- * [31:31] Doubleworld Flow Control Out Enable
- * 0 = Doubleworld flow control disabled
- * 1 = Doubleword flow control enabled (Not currently supported)
- */
- PCI_ADDR(0, 0x18, 0, 0x84), 0x00009c05, 0x11110020,
- /* LDTi Frequency/Revision Registers
- * F0:0x88 i = 0,
- * F0:0xA8 i = 1,
- * F0:0xC8 i = 2,
- * [ 4: 0] Minor Revision
- * Contains the HT Minor revision
- * [ 7: 5] Major Revision
- * Contains the HT Major revision
- * [11: 8] Link Frequency (Takes effect the next time the link is reconnected)
- * 0000 = 200Mhz
- * 0001 = reserved
- * 0010 = 400Mhz
- * 0011 = reserved
- * 0100 = 600Mhz
- * 0101 = 800Mhz
- * 0110 = 1000Mhz
- * 0111 = reserved
- * 1000 = reserved
- * 1001 = reserved
- * 1010 = reserved
- * 1011 = reserved
- * 1100 = reserved
- * 1101 = reserved
- * 1110 = reserved
- * 1111 = 100 Mhz
- * [15:12] Error (Not currently Implemented)
- * [31:16] Indicates the frequency capabilities of the link
- * [16] = 1 encoding 0000 of freq supported
- * [17] = 1 encoding 0001 of freq supported
- * [18] = 1 encoding 0010 of freq supported
- * [19] = 1 encoding 0011 of freq supported
- * [20] = 1 encoding 0100 of freq supported
- * [21] = 1 encoding 0101 of freq supported
- * [22] = 1 encoding 0110 of freq supported
- * [23] = 1 encoding 0111 of freq supported
- * [24] = 1 encoding 1000 of freq supported
- * [25] = 1 encoding 1001 of freq supported
- * [26] = 1 encoding 1010 of freq supported
- * [27] = 1 encoding 1011 of freq supported
- * [28] = 1 encoding 1100 of freq supported
- * [29] = 1 encoding 1101 of freq supported
- * [30] = 1 encoding 1110 of freq supported
- * [31] = 1 encoding 1111 of freq supported
- */
- PCI_ADDR(0, 0x18, 0, 0x88), 0xfffff0ff, 0x00000200,
- /* LDTi Feature Capability
- * F0:0x8C i = 0,
- * F0:0xAC i = 1,
- * F0:0xCC i = 2,
- */
- /* LDTi Buffer Count Registers
- * F0:0x90 i = 0,
- * F0:0xB0 i = 1,
- * F0:0xD0 i = 2,
- */
- /* LDTi Bus Number Registers
- * F0:0x94 i = 0,
- * F0:0xB4 i = 1,
- * F0:0xD4 i = 2,
- * For NonCoherent HT specifies the bus number downstream (behind the host bridge)
- * [ 0: 7] Primary Bus Number
- * [15: 8] Secondary Bus Number
- * [23:15] Subordiante Bus Number
- * [31:24] reserved
- */
- PCI_ADDR(0, 0x18, 0, 0x94), 0xff000000, 0x00ff0000,
- /* LDTi Type Registers
- * F0:0x98 i = 0,
- * F0:0xB8 i = 1,
- * F0:0xD8 i = 2,
- */
- };
- int i;
- int max;
- print_debug("setting up coherent ht domain....\r\n");
- max = sizeof(register_values)/sizeof(register_values[0]);
- for(i = 0; i < max; i += 3) {
- device_t dev;
- unsigned where;
- unsigned long reg;
-#if 0
- print_debug_hex32(register_values[i]);
- print_debug(" <-");
- print_debug_hex32(register_values[i+2]);
- print_debug("\r\n");
-#endif
- dev = register_values[i] & ~0xff;
- where = register_values[i] & 0xff;
- reg = pci_read_config32(dev, where);
- reg &= register_values[i+1];
- reg |= register_values[i+2];
- pci_write_config32(dev, where, reg);
-#if 0
- reg = pci_read_config32(register_values[i]);
- reg &= register_values[i+1];
- reg |= register_values[i+2] & ~register_values[i+1];
- pci_write_config32(register_values[i], reg);
-#endif
- }
- print_debug("done.\r\n");
-}
-#else
/* coherent hypertransport initialization for AMD64
- * written by Stefan Reinauer <stepan@openbios.info>
- * (c) 2003 by SuSE Linux AG
+ *
+ * written by Stefan Reinauer <stepan@openbios.org>
+ * (c) 2003-2004 by SuSE Linux AG
*
* This code is licensed under GPL.
*/
*
*/
-#if 0
-#include "compat.h"
-#endif
-
#include <device/pci_def.h>
+#include <device/pci_ids.h>
+#include <device/hypertransport_def.h>
#include "arch/romcc_io.h"
-
+#include "amdk8.h"
/* when generating a temporary row configuration we
* don't want broadcast to be enabled for that node.
*/
-#define generate_temp_row(x...) ((generate_row(x)&(~0x0f0000))|0x010000)
+#define generate_temp_row(...) ((generate_row(__VA_ARGS__)&(~0x0f0000))|0x010000)
#define clear_temp_row(x) fill_row(x,7,DEFAULT)
#define enable_bsp_routing() enable_routing(0)
typedef uint8_t u8;
typedef uint32_t u32;
-typedef int8_t bool;
+typedef int bool;
#define TRUE (-1)
#define FALSE (0)
+static u8 link_to_register(int ldt)
+{
+ /*
+ * [ 0: 3] Request Route
+ * [0] Route to this node
+ * [1] Route to Link 0
+ * [2] Route to Link 1
+ * [3] Route to Link 2
+ */
+
+ if (ldt&0x08) return 0x40;
+ if (ldt&0x04) return 0x20;
+ if (ldt&0x02) return 0x00;
+
+ /* we should never get here */
+ print_debug("Unknown Link\n");
+ return 0;
+}
+
+static int link_connection(int src, int dest)
+{
+ /* we generate the needed link information from the rows
+ * by taking the Request Route of the according row.
+ */
+
+ return generate_row(src, dest, CONFIG_MAX_CPUS) & 0x0f;
+}
+
static void disable_probes(void)
{
/* disable read/write/fill probes for uniprocessor setup
u32 val;
- print_debug("Disabling read/write/fill probes for UP... ");
+ print_spew("Disabling read/write/fill probes for UP... ");
val=pci_read_config32(NODE_HT(0), 0x68);
- val |= 0x0000040f;
+ val |= (1<<10)|(1<<9)|(1<<8)|(1<<4)|(1<<3)|(1<<2)|(1<<1)|(1 << 0);
pci_write_config32(NODE_HT(0), 0x68, val);
- print_debug("done.\r\n");
+ print_spew("done.\r\n");
}
*
*/
- /* Enable routing table for BSP */
- print_debug("Enabling routing table for node ");
- print_debug_hex32(node);
+ /* Enable routing table */
+ print_spew("Enabling routing table for node ");
+ print_spew_hex8(node);
val=pci_read_config32(NODE_HT(node), 0x6c);
- val |= (1 << 6) | (1 << 5) | (1 << 4);
-#if 0
val &= ~((1<<1)|(1<<0));
-#else
- /* Don't enable requests here as the indicated processor starts booting */
- val &= ~(1<<0);
-#endif
pci_write_config32(NODE_HT(node), 0x6c, val);
- print_debug(" done.\r\n");
+ print_spew(" done.\r\n");
}
-#if MAX_CPUS > 1
+#if CONFIG_MAX_CPUS > 1
static void rename_temp_node(u8 node)
{
- u32 val;
+ uint32_t val;
- print_debug("Renaming current temp node to ");
- print_debug_hex32(node);
+ print_spew("Renaming current temporary node to ");
+ print_spew_hex8(node);
val=pci_read_config32(NODE_HT(7), 0x60);
val &= (~7); /* clear low bits. */
val |= node; /* new node */
pci_write_config32(NODE_HT(7), 0x60, val);
- print_debug(" done.\r\n");
-
-
+ print_spew(" done.\r\n");
}
static bool check_connection(u8 src, u8 dest, u8 link)
{
- /* this function does 2 things:
- * 1) detect whether the coherent HT link is connected
- * 2) verify that the coherent hypertransport link
- * is established and actually working by reading the
- * remote node's vendor/device id
- */
-
-#define UP 0x00
-#define ACROSS 0x20
-#define DOWN 0x40
-
+ /* See if we have a valid connection to dest */
u32 val;
- /* 1) */
- val=pci_read_config32(NODE_HT(src), 0x98+link);
+ /* Detect if the coherent HT link is connected. */
+ val = pci_read_config32(NODE_HT(src), 0x98+link);
if ( (val&0x17) != 0x03)
return 0;
- /* 2) */
- val=pci_read_config32(NODE_HT(dest),0);
+ /* Verify that the coherent hypertransport link is
+ * established and actually working by reading the
+ * remode node's vendor/device id
+ */
+ val = pci_read_config32(NODE_HT(dest),0);
if(val != 0x11001022)
return 0;
return 1;
}
-static unsigned int generate_row(u8 node, u8 row, u8 maxnodes)
+static unsigned read_freq_cap(device_t dev, unsigned pos)
{
- /* Routing Table Node i
- *
- * F0: 0x40, 0x44, 0x48, 0x4c, 0x50, 0x54, 0x58, 0x5c
- * i: 0, 1, 2, 3, 4, 5, 6, 7
- *
- * [ 0: 3] Request Route
- * [0] Route to this node
- * [1] Route to Link 0
- * [2] Route to Link 1
- * [3] Route to Link 2
- * [11: 8] Response Route
- * [0] Route to this node
- * [1] Route to Link 0
- * [2] Route to Link 1
- * [3] Route to Link 2
- * [19:16] Broadcast route
- * [0] Route to this node
- * [1] Route to Link 0
- * [2] Route to Link 1
- * [3] Route to Link 2
- */
+ /* Handle bugs in valid hypertransport frequency reporting */
+ unsigned freq_cap;
+ uint32_t id;
- u32 ret=DEFAULT;
+ freq_cap = pci_read_config16(dev, pos);
+ freq_cap &= ~(1 << HT_FREQ_VENDOR); /* Ignore Vendor HT frequencies */
- static const unsigned int rows_2p[2][2] = {
- { 0x00030101, 0x00010404 },
- { 0x00010404, 0x00030101 }
- };
+ id = pci_read_config32(dev, 0);
- static const unsigned int rows_4p[4][4] = {
- { 0x00070101, 0x00010404, 0x00050202, 0x00010402 },
- { 0x00010808, 0x000b0101, 0x00010802, 0x00090202 },
- { 0x00090202, 0x00010802, 0x000b0101, 0x00010808 },
- { 0x00010402, 0x00050202, 0x00010404, 0x00070101 }
- };
-
- if (!(node>=maxnodes || row>=maxnodes)) {
- if (maxnodes==2)
- ret=rows_2p[node][row];
- if (maxnodes==4)
- ret=rows_4p[node][row];
+ /* AMD 8131 Errata 48 */
+ if (id == (PCI_VENDOR_ID_AMD | (PCI_DEVICE_ID_AMD_8131_PCIX << 16))) {
+ freq_cap &= ~(1 << HT_FREQ_800Mhz);
+ }
+ /* AMD 8151 Errata 23 */
+ if (id == (PCI_VENDOR_ID_AMD | (PCI_DEVICE_ID_AMD_8151_SYSCTRL << 16))) {
+ freq_cap &= ~(1 << HT_FREQ_800Mhz);
}
+ /* AMD K8 Unsupported 1Ghz? */
+ if (id == (PCI_VENDOR_ID_AMD | (0x1100 << 16))) {
+ freq_cap &= ~(1 << HT_FREQ_1000Mhz);
+ }
+ return freq_cap;
+}
-#if 0
- printk_spew("generating entry n=%d, r=%d, max=%d - row=%x\n",
- node,row,maxnodes,ret);
-#endif
+static int optimize_connection(device_t node1, uint8_t link1, device_t node2, uint8_t link2)
+{
+ static const uint8_t link_width_to_pow2[]= { 3, 4, 0, 5, 1, 2, 0, 0 };
+ static const uint8_t pow2_to_link_width[] = { 0x7, 4, 5, 0, 1, 3 };
+ uint16_t freq_cap1, freq_cap2, freq_cap, freq_mask;
+ uint8_t width_cap1, width_cap2, width_cap, width, old_width, ln_width1, ln_width2;
+ uint8_t freq, old_freq;
+ int needs_reset;
+ /* Set link width and frequency */
+
+ /* Initially assume everything is already optimized and I don't need a reset */
+ needs_reset = 0;
+
+ /* Get the frequency capabilities */
+ freq_cap1 = read_freq_cap(node1, link1 + PCI_HT_CAP_HOST_FREQ_CAP);
+ freq_cap2 = read_freq_cap(node2, link2 + PCI_HT_CAP_HOST_FREQ_CAP);
+
+ /* Calculate the highest possible frequency */
+ freq = log2(freq_cap1 & freq_cap2);
+
+ /* See if I am changing the link freqency */
+ old_freq = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_FREQ);
+ needs_reset |= old_freq != freq;
+ old_freq = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_FREQ);
+ needs_reset |= old_freq != freq;
+
+ /* Set the Calulcated link frequency */
+ pci_write_config8(node1, link1 + PCI_HT_CAP_HOST_FREQ, freq);
+ pci_write_config8(node2, link2 + PCI_HT_CAP_HOST_FREQ, freq);
+
+ /* Get the width capabilities */
+ width_cap1 = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_WIDTH);
+ width_cap2 = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH);
+
+ /* Calculate node1's input width */
+ ln_width1 = link_width_to_pow2[width_cap1 & 7];
+ ln_width2 = link_width_to_pow2[(width_cap2 >> 4) & 7];
+ if (ln_width1 > ln_width2) {
+ ln_width1 = ln_width2;
+ }
+ width = pow2_to_link_width[ln_width1];
+ /* Calculate node1's output width */
+ ln_width1 = link_width_to_pow2[(width_cap1 >> 4) & 7];
+ ln_width2 = link_width_to_pow2[width_cap2 & 7];
+ if (ln_width1 > ln_width2) {
+ ln_width1 = ln_width2;
+ }
+ width |= pow2_to_link_width[ln_width1] << 4;
- return ret;
+ /* See if I am changing node1's width */
+ old_width = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_WIDTH + 1);
+ needs_reset |= old_width != width;
+
+ /* Set node1's widths */
+ pci_write_config8(node1, link1 + PCI_HT_CAP_HOST_WIDTH + 1, width);
+
+ /* Calculate node2's width */
+ width = ((width & 0x70) >> 4) | ((width & 0x7) << 4);
+
+ /* See if I am changing node2's width */
+ old_width = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH + 1);
+ needs_reset |= old_width != width;
+
+ /* Set node2's widths */
+ pci_write_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH + 1, width);
+
+ return needs_reset;
}
static void fill_row(u8 node, u8 row, u32 value)
{
-#if 0
- print_debug("fill_row: pci_write_config32(");
- print_debug_hex32(NODE_HT(node));
- print_debug_char(',');
- print_debug_hex32(0x40 + (row << 2));
- print_debug_char(',');
- print_debug_hex32(value);
- print_debug(")\r\n");
-#endif
pci_write_config32(NODE_HT(node), 0x40+(row<<2), value);
}
static void setup_row(u8 source, u8 dest, u8 cpus)
{
-#if 0
- printk_spew("setting up link from node %d to %d (%d cpus)\r\n",
- source, dest, cpus);
-#endif
-
fill_row(source,dest,generate_row(source,dest,cpus));
}
static void setup_temp_row(u8 source, u8 dest, u8 cpus)
{
-#if 0
- printk_spew("setting up temp. link from node %d to %d (%d cpus)\r\n",
- source, dest, cpus);
-#endif
-
fill_row(source,7,generate_temp_row(source,dest,cpus));
}
};
uint8_t row;
int i;
-#if 1
- print_debug("setup_remote_node\r\n");
-#endif
+
+ print_spew("setup_remote_node: ");
for(row=0; row<cpus; row++)
setup_remote_row(node, row, cpus);
uint32_t value;
uint8_t reg;
reg = pci_reg[i];
- print_debug("copying reg: ");
- print_debug_hex8(reg);
- print_debug("\r\n");
value = pci_read_config32(NODE_MP(0), reg);
pci_write_config32(NODE_MP(7), reg, value);
}
-#if 1
- print_debug("setup_remote_done\r\n");
-#endif
+ print_spew("done\r\n");
}
#endif
-#if MAX_CPUS > 2
+#if CONFIG_MAX_CPUS > 2
static void setup_temp_node(u8 node, u8 cpus)
{
u8 row;
}
#endif
-static u8 setup_uniprocessor(void)
+static void setup_uniprocessor(void)
{
- print_debug("Enabling UP settings\r\n");
+ print_spew("Enabling UP settings\r\n");
disable_probes();
- return 1;
}
-#if MAX_CPUS > 1
-static u8 setup_smp(void)
+struct setup_smp_result {
+ int cpus;
+ int needs_reset;
+};
+
+#if CONFIG_MAX_CPUS > 1
+static struct setup_smp_result setup_smp(void)
{
- u8 cpus=2;
+ struct setup_smp_result result;
+ result.cpus = 2;
+ result.needs_reset = 0;
- print_debug("Enabling SMP settings\r\n");
+ print_spew("Enabling SMP settings\r\n");
- setup_row(0,0,cpus);
+ setup_row(0, 0, result.cpus);
/* Setup and check a temporary connection to node 1 */
- setup_temp_row(0,1,cpus);
+ setup_temp_row(0, 1, result.cpus);
- if (!check_connection(0, 7, ACROSS)) { // Link: ACROSS
+ if (!check_connection(0, 7, link_to_register(link_connection(0,1)))) {
print_debug("No connection to Node 1.\r\n");
clear_temp_row(0); /* delete temp connection */
setup_uniprocessor(); /* and get up working */
- return 1;
+ result.cpus = 1;
+ return result;
}
/* We found 2 nodes so far */
- setup_node(0, cpus); /* Node 1 is there. Setup Node 0 correctly */
- setup_remote_node(1, cpus); /* Setup the routes on the remote node */
- enable_routing(1); /* Enable routing on Node 1 */
- rename_temp_node(1); /* Rename Node 7 to Node 1 */
+ setup_node(0, result.cpus); /* Node 1 is there. Setup Node 0 correctly */
+ setup_remote_node(1, result.cpus); /* Setup the routes on the remote node */
+ rename_temp_node(1); /* Rename Node 7 to Node 1 */
+ enable_routing(1); /* Enable routing on Node 1 */
+
clear_temp_row(0); /* delete temporary connection */
-#if MAX_CPUS > 2
- cpus=4;
+ result.needs_reset =
+ optimize_connection(NODE_HT(0), 0x80 + link_to_register(link_connection(0,1)),
+ NODE_HT(1), 0x80 + link_to_register(link_connection(1,0)) );
+
+#if CONFIG_MAX_CPUS > 2
+ result.cpus=4;
/* Setup and check temporary connection from Node 0 to Node 2 */
- setup_temp_row(0,2,cpus);
+ setup_temp_row(0,2, result.cpus);
- if (!check_connection(0, 7, UP)) { // Link: UP
+ if (!check_connection(0, 7, link_to_register(link_connection(0,2))) ) {
print_debug("No connection to Node 2.\r\n");
clear_temp_row(0); /* delete temp connection */
- // detect_mp_capability(2); /* and get 2p working */
- return 2;
+ result.cpus = 2;
+ return result;
}
/* We found 3 nodes so far. Now setup a temporary
* connection from node 0 to node 3 via node 1
*/
- setup_temp_row(0,1,cpus); /* temp. link between nodes 0 and 1 */
- setup_temp_row(1,3,cpus); /* temp. link between nodes 1 and 3 */
+ setup_temp_row(0,1, result.cpus); /* temp. link between nodes 0 and 1 */
+ setup_temp_row(1,3, result.cpus); /* temp. link between nodes 1 and 3 */
- if (!check_connection(0, 7, UP)) { // Link: UP
+ if (!check_connection(1, 7, link_to_register(link_connection(1,3)))) {
print_debug("No connection to Node 3.\r\n");
clear_temp_row(0); /* delete temp connection */
clear_temp_row(1); /* delete temp connection */
- //detect_mp_capability(2); /* and get 2p working */
- return 2;
+ result.cpus = 2;
+ return result;
}
/* We found 4 nodes so far. Now setup all nodes for 4p */
- setup_node(0, cpus); /* The first 2 nodes are configured */
- setup_node(1, cpus); /* already. Just configure them for 4p */
+ setup_node(0, result.cpus); /* The first 2 nodes are configured */
+ setup_node(1, result.cpus); /* already. Just configure them for 4p */
- setup_temp_row(0,2,cpus);
- setup_temp_node(2,cpus);
- enable_routing(7);
- rename_temp_node(2);
-
- setup_temp_row(0,1,cpus);
- setup_temp_row(1,3,cpus);
- setup_temp_node(3,cpus);
- enable_routing(3);
- rename_temp_node(3);
+ setup_temp_row(0,2, result.cpus);
+ setup_temp_node(2, result.cpus);
+ rename_temp_node(2);
+ enable_routing(2);
+
+ setup_temp_row(0,1, result.cpus);
+ setup_temp_row(1,3, result.cpus);
+ setup_temp_node(3, result.cpus);
+ rename_temp_node(3);
+ enable_routing(3); /* enable routing on node 3 (temp.) */
clear_temp_row(0);
clear_temp_row(1);
clear_temp_row(2);
clear_temp_row(3);
-#endif
- print_debug_hex32(cpus);
+ /* optimize physical connections - by LYH */
+ result.needs_reset = optimize_connection(
+ NODE_HT(0), 0x80 + link_to_register(link_connection(0,2)),
+ NODE_HT(2), 0x80 + link_to_register(link_connection(2,0)) );
+
+ result.needs_reset = optimize_connection(
+ NODE_HT(1), 0x80 + link_to_register(link_connection(1,3)),
+ NODE_HT(3), 0x80 + link_to_register(link_connection(3,1)) );
+
+ result.needs_reset = optimize_connection(
+ NODE_HT(2), 0x80 + link_to_register(link_connection(2,3)),
+ NODE_HT(3), 0x80 + link_to_register(link_connection(3,2)) );
+
+#endif /* CONFIG_MAX_CPUS > 2 */
+
+ print_debug_hex8(result.cpus);
print_debug(" nodes initialized.\r\n");
- return cpus;
+ return result;
}
#endif
-#if MAX_CPUS > 1
-static unsigned detect_mp_capabilities(unsigned cpus)
+#if CONFIG_MAX_CPUS > 1
+static unsigned verify_mp_capabilities(unsigned cpus)
{
unsigned node, row, mask;
bool mp_cap=TRUE;
-#if 1
- print_debug("detect_mp_capabilities: ");
- print_debug_hex32(cpus);
- print_debug("\r\n");
-#endif
- if (cpus>2)
- mask=0x04; /* BigMPCap */
- else
+ if (cpus > 2) {
+ mask=0x06; /* BigMPCap */
+ } else {
mask=0x02; /* MPCap */
+ }
for (node=0; node<cpus; node++) {
- if (!(pci_read_config32(NODE_MC(node), 0xe8) & mask))
- mp_cap=FALSE;
+ if ((pci_read_config32(NODE_MC(node), 0xe8) & mask) != mask) {
+ mp_cap = FALSE;
+ }
}
- if (mp_cap)
+ if (mp_cap) {
return cpus;
+ }
/* one of our cpus is not mp capable */
- print_debug("One of the CPUs is not MP capable. Going back to UP\r\n");
+ print_err("One of the CPUs is not MP capable. Going back to UP\r\n");
- for (node=cpus; node>0; node--)
- for (row=cpus; row>0; row--)
- fill_row(NODE_HT(node-1), row-1, DEFAULT);
-
- return setup_uniprocessor();
+ for (node = cpus; node > 0; node--) {
+ for (row = cpus; row > 0; row--) {
+ fill_row(NODE_HT(node-1), row-1, DEFAULT);
+ }
+ }
+ setup_uniprocessor();
+ return 1;
}
#endif
-/* this is a shrunken cpuid. */
-
-static unsigned int cpuid(unsigned int op)
-{
- unsigned int ret;
-
- asm volatile ( "cpuid" : "=a" (ret) : "a" (op));
-
- return ret;
-}
-
static void coherent_ht_finalize(unsigned cpus)
{
- int node;
+ unsigned node;
bool rev_a0;
/* set up cpu count and node count and enable Limit
#if 1
print_debug("coherent_ht_finalize\r\n");
#endif
- rev_a0=((cpuid(1)&0xffff)==0x0f10);
+ rev_a0 = is_cpu_rev_a0();
+ for (node = 0; node < cpus; node++) {
+ device_t dev;
+ uint32_t val;
+ dev = NODE_HT(node);
- for (node=0; node<cpus; node++) {
- u32 val;
- val=pci_read_config32(NODE_HT(node), 0x60);
- val &= 0x000F0070;
+ /* Set the Total CPU and Node count in the system */
+ val = pci_read_config32(dev, 0x60);
+ val &= (~0x000F0070);
val |= ((cpus-1)<<16)|((cpus-1)<<4);
- pci_write_config32(NODE_HT(node),0x60,val);
-
- val=pci_read_config32(NODE_HT(node), 0x68);
- val |= 0x00008000;
- pci_write_config32(NODE_HT(node),0x68,val);
+ pci_write_config32(dev, 0x60, val);
+
+ /* Only respond to real cpu pci configuration cycles
+ * and optimize the HT settings
+ */
+ val=pci_read_config32(dev, 0x68);
+ val &= ~((HTTC_BUF_REL_PRI_MASK << HTTC_BUF_REL_PRI_SHIFT) |
+ (HTTC_MED_PRI_BYP_CNT_MASK << HTTC_MED_PRI_BYP_CNT_SHIFT) |
+ (HTTC_HI_PRI_BYP_CNT_MASK << HTTC_HI_PRI_BYP_CNT_SHIFT));
+ val |= HTTC_LIMIT_CLDT_CFG |
+ (HTTC_BUF_REL_PRI_8 << HTTC_BUF_REL_PRI_SHIFT) |
+ HTTC_RSP_PASS_PW |
+ (3 << HTTC_MED_PRI_BYP_CNT_SHIFT) |
+ (3 << HTTC_HI_PRI_BYP_CNT_SHIFT);
+ pci_write_config32(dev, 0x68, val);
if (rev_a0) {
- pci_write_config32(NODE_HT(node),0x94,0);
- pci_write_config32(NODE_HT(node),0xb4,0);
- pci_write_config32(NODE_HT(node),0xd4,0);
+ print_debug("shit it is an old cup\n");
+ pci_write_config32(dev, 0x94, 0);
+ pci_write_config32(dev, 0xb4, 0);
+ pci_write_config32(dev, 0xd4, 0);
}
}
#if 1
- print_debug("done\n");
+ print_debug("done\r\n");
#endif
}
-static void setup_coherent_ht_domain(void)
+static int apply_cpu_errata_fixes(unsigned cpus, int needs_reset)
+{
+ unsigned node;
+ for(node = 0; node < cpus; node++) {
+ device_t dev;
+ uint32_t cmd;
+ dev = NODE_MC(node);
+ if (is_cpu_pre_c0()) {
+
+ /* Errata 66
+ * Limit the number of downstream posted requests to 1
+ */
+ cmd = pci_read_config32(dev, 0x70);
+ if ((cmd & (3 << 0)) != 2) {
+ cmd &= ~(3<<0);
+ cmd |= (2<<0);
+ pci_write_config32(dev, 0x70, cmd );
+ needs_reset = 1;
+ }
+ cmd = pci_read_config32(dev, 0x7c);
+ if ((cmd & (3 << 4)) != 0) {
+ cmd &= ~(3<<4);
+ cmd |= (0<<4);
+ pci_write_config32(dev, 0x7c, cmd );
+ needs_reset = 1;
+ }
+ /* Clock Power/Timing Low */
+ cmd = pci_read_config32(dev, 0xd4);
+ if (cmd != 0x000D0001) {
+ cmd = 0x000D0001;
+ pci_write_config32(dev, 0xd4, cmd);
+ needs_reset = 1; /* Needed? */
+ }
+
+ }
+ else {
+ uint32_t cmd_ref;
+ /* Errata 98
+ * Set Clk Ramp Hystersis to 7
+ * Clock Power/Timing Low
+ */
+ cmd_ref = 0x04e20707; /* Registered */
+ cmd = pci_read_config32(dev, 0xd4);
+ if(cmd != cmd_ref) {
+ pci_write_config32(dev, 0xd4, cmd_ref );
+ needs_reset = 1; /* Needed? */
+ }
+ }
+ }
+ return needs_reset;
+}
+
+static int optimize_link_read_pointers(unsigned cpus, int needs_reset)
{
- unsigned cpus;
+ unsigned node;
+ for(node = 0; node < cpus; node = node + 1) {
+ device_t f0_dev, f3_dev;
+ uint32_t cmd_ref, cmd;
+ int link;
+ f0_dev = NODE_HT(node);
+ f3_dev = NODE_MC(node);
+ cmd_ref = cmd = pci_read_config32(f3_dev, 0xdc);
+ for(link = 0; link < 3; link = link + 1) {
+ uint32_t link_type;
+ unsigned reg;
+ reg = 0x98 + (link * 0x20);
+ link_type = pci_read_config32(f0_dev, reg);
+ if (link_type & LinkConnected) {
+ cmd &= 0xff << (link *8);
+ /* FIXME this assumes the device on the other
+ * side is an AMD device */
+ cmd |= 0x25 << (link *8);
+ }
+ }
+ if (cmd != cmd_ref) {
+ pci_write_config32(f3_dev, 0xdc, cmd);
+ needs_reset = 1;
+ }
+ }
+ return needs_reset;
+}
+
+static int setup_coherent_ht_domain(void)
+{
+ struct setup_smp_result result;
+ result.cpus = 1;
+ result.needs_reset = 0;
enable_bsp_routing();
-#if MAX_CPUS == 1
- cpus=setup_uniprocessor();
+#if CONFIG_MAX_CPUS == 1
+ setup_uniprocessor();
#else
- cpus=setup_smp();
- cpus=detect_mp_capabilities(cpus);
+ result = setup_smp();
+ result.cpus = verify_mp_capabilities(result.cpus);
#endif
- coherent_ht_finalize(cpus);
-}
+ coherent_ht_finalize(result.cpus);
+ result.needs_reset = apply_cpu_errata_fixes(result.cpus, result.needs_reset);
+
+#if CONFIG_MAX_CPUS > 1 /* Why doesn't this work on the solo? */
+ result.needs_reset = optimize_link_read_pointers(result.cpus, result.needs_reset);
#endif
+
+ return result.needs_reset;
+}