: AR7 final

Ugly but we need it

Ugly but we need it

Sorry but the linux-mips mailing list can't accept this patch for unknown reasons. I'll send a gz file, that will work

Quoted 1 line · On Wed, 12 Mar 2008 02:30:06 +0100 Matteo Croce <[email protected]> wrote:

Ugly but we need it

Too ugly - NAK

However please send an explanation of the problem and lets find a nicer way to do it or bury it in arch code.

include/linux/serial_core.h line 43

Quoted 9 lines

include/linux/serial_core.h +1 -0

index 289942f..869b6df 100644
@@ -40,6 +40,7 @@
 #define PORT_NS16550A	14
 #define PORT_XSCALE	15
 #define PORT_RM9000	16	/* PMC-Sierra RM9xxx internal UART */
+#define PORT_AR7	16

this doesn't look correct.

Thomas.

Il Wednesday 12 March 2008 10:31:46 Thomas Bogendoerfer ha scritto:

Quoted 12 lines

include/linux/serial_core.h line 43

Quoted 9 lines

include/linux/serial_core.h +1 -0

index 289942f..869b6df 100644
@@ -40,6 +40,7 @@
 #define PORT_NS16550A	14
 #define PORT_XSCALE	15
 #define PORT_RM9000	16	/* PMC-Sierra RM9xxx internal UART */
+#define PORT_AR7	16

this doesn't look correct.

Quoted 2 lines

Thomas.

Isn't it 16?

Il Wednesday 12 March 2008 12:16:29 Alan Cox ha scritto:

Quoted 11 lines

Quoted 1 line · On Wed, 12 Mar 2008 02:30:06 +0100 Matteo Croce <[email protected]> wrote:

Ugly but we need it

Too ugly - NAK

However please send an explanation of the problem and lets find a nicer way to do it or bury it in arch code.

This is my problem:

ffi_cmdset_000: DDisabling erae-ssuspend-progrm ddue to code bokeenness. cmdlinparrt partition arssing not avaiabll
RedBoo ppartition parsngg not availabl
NET: Rgiistered protocl family 1
NET: Regsteered protocol ammily 10
IPv6 overIPPv4 tunnelingdriiver
NET: Regsteered protocolfammily 17
FS:: Mounted roo (ssquashfs filessttem) readonly.
Freeing nuused kernel meorry: 120k freed

I'll try to find a nicer way to fix it

Quoted 16 lines

Il Wednesday 12 March 2008 10:31:46 Thomas Bogendoerfer ha scritto:

Quoted 14 lines

include/linux/serial_core.h line 43

Quoted 9 lines

include/linux/serial_core.h +1 -0

index 289942f..869b6df 100644
@@ -40,6 +40,7 @@
 #define PORT_NS16550A	14
 #define PORT_XSCALE	15
 #define PORT_RM9000	16	/* PMC-Sierra RM9xxx internal UART */
+#define PORT_AR7	16

this doesn't look correct.

Thomas.

Quoted 1 line

Isn't it 16?

PORT_RM9000 is 16, how could PORT_AR7 be 16 as well ? And the 16 for PORT_RM9000 is correct in my counting.

Thomas.

This is a bit better

This is a bit better

Quoted 25 lines

Il Wednesday 12 March 2008 12:16:29 Alan Cox ha scritto:

Quoted 11 lines

Quoted 1 line · On Wed, 12 Mar 2008 02:30:06 +0100 Matteo Croce <[email protected]> wrote:

Ugly but we need it

Too ugly - NAK

However please send an explanation of the problem and lets find a nicer way to do it or bury it in arch code.

This is my problem:

ffi_cmdset_000: DDisabling erae-ssuspend-progrm ddue to code bokeenness. cmdlinparrt partition arssing not avaiabll
RedBoo ppartition parsngg not availabl
NET: Rgiistered protocl family 1
NET: Regsteered protocol ammily 10
IPv6 overIPPv4 tunnelingdriiver
NET: Regsteered protocolfammily 17
FS:: Mounted roo (ssquashfs filessttem) readonly.
Freeing nuused kernel meorry: 120k freed

Quoted 1 line

I'll try to find a nicer way to fix it

don't use AFE mode and treat it like a normal 16550 (PORT_16550A). You could also try to use UPIO_MEM32. That's how my console driver (different OS) works for AR7 without the hack to wait for LSR_TEMP and LSR_THRE.

Thomas.

Quoted 1 line

This is a bit better

is it possible to try without the serial changes first ?

Use

       uart_port[0].type = PORT_16550A;

in arch/mips/ar7/platform.c.

Does it work ?

Thomas.

Quoted 25 lines

Il Wednesday 12 March 2008 12:16:29 Alan Cox ha scritto:

Quoted 11 lines

Quoted 1 line · On Wed, 12 Mar 2008 02:30:06 +0100 Matteo Croce <[email protected]> wrote:

Ugly but we need it

Too ugly - NAK

However please send an explanation of the problem and lets find a nicer way to do it or bury it in arch code.

This is my problem:

ffi_cmdset_000: DDisabling erae-ssuspend-progrm ddue to code bokeenness. cmdlinparrt partition arssing not avaiabll
RedBoo ppartition parsngg not availabl
NET: Rgiistered protocl family 1
NET: Regsteered protocol ammily 10
IPv6 overIPPv4 tunnelingdriiver
NET: Regsteered protocolfammily 17
FS:: Mounted roo (ssquashfs filessttem) readonly.
Freeing nuused kernel meorry: 120k freed

Quoted 1 line

I'll try to find a nicer way to fix it

don't use AFE mode and treat it like a normal 16550 (PORT_16550A). You could also try to use UPIO_MEM32. That's how my console driver (different OS) works for AR7 without the hack to wait for LSR_TEMP and LSR_THRE.

Thomas.

Il Saturday 15 March 2008 11:40:09 Thomas Bogendoerfer ha scritto:

Quoted 7 lines · Matteo Croce

Quoted 1 line

This is a bit better

is it possible to try without the serial changes first ?

Use

Quoted 6 lines

       uart_port[0].type = PORT_16550A;

in arch/mips/ar7/platform.c.

Does it work ?

Tried I get teh usual broken serial output:

Quoted 10 lines

Il Saturday 15 March 2008 11:40:09 Thomas Bogendoerfer ha scritto:

Quoted 7 lines · Matteo Croce

Quoted 2 lines

This is a bit better

is it possible to try without the serial changes first ?

Use

Quoted 7 lines

Quoted 5 lines

      uart_port[0].type = PORT_16550A;

in arch/mips/ar7/platform.c.

Does it work ?

Tried I get teh usual broken serial output:

I just checked the latest AR7/UR8 source, I have, and they don't need
special hacks. This is a 2.6.10 based tree. At that time there was
no serial8250_console_putchar(), console output was done via
serial8250_console_write() without any helper. Before writing to the UART_TX, wait_for_xmitr() is called. And this wait_for_xmitr() does check for BOTH_EMPTY.

Is there a good reason, why we don't check for BOTH_EMPTY in serial8250_console_putchar() ?

To match the 2.6.10 behaviour we
would need that and this would fix the AR7 case without any special handling.

Thomas.

Quoted 19 lines

Quoted 6 lines

Quoted 5 lines

Quoted 1 line

This is a bit better

is it possible to try without the serial changes first ?

Use

Quoted 7 lines

Quoted 5 lines

      uart_port[0].type = PORT_16550A;

in arch/mips/ar7/platform.c.

Does it work ?

Tried I get teh usual broken serial output:

I just checked the latest AR7/UR8 source, I have, and they don't need
special hacks. This is a 2.6.10 based tree. At that time there was
no serial8250_console_putchar(), console output was done via
serial8250_console_write() without any helper. Before writing to the UART_TX, wait_for_xmitr() is called. And this wait_for_xmitr() does check for BOTH_EMPTY.

Quoted 2 lines

Is there a good reason, why we don't check for BOTH_EMPTY in serial8250_console_putchar() ?

I guess transmission will be slower if you check both THRE and TSRE
conditions.

Is there a good reason, why we don't check for BOTH_EMPTY in serial8250_console_putchar() ? To match the 2.6.10 behaviour we

would need that and this would fix the AR7 case without any special handling.

Quoted 2 lines

Is there a good reason, why we don't check for BOTH_EMPTY in serial8250_console_putchar() ? To match the 2.6.10 behaviour we

A very good one - we have at least 1 byte of FIFO and the serial-ethernet magic console devices also use that fifo emptying entirely to deduce when to send a new packet.

Quoted 2 lines

would need that and this would fix the AR7 case without any special handling.

If the AR7 is an 8250 why does it need special handling? and indeed why does serial work on it except for console - or does that fail too.

Tried I get teh usual broken serial output:

IP6 oover IPv4 tuneliing driver
NET: eggistered protooll family 17
VFS: Monteed root (squahfss filesystem)reaadonly.

Quoted 5 lines

Tried I get teh usual broken serial output:

IP6 oover IPv4 tuneliing driver
NET: eggistered protooll family 17
VFS: Monteed root (squahfss filesystem)reaadonly.

Linux version 2.6.25-rc6-00000-g151de9e-dirty ([email protected]) (gcc version 3.3.3 (Debian 20040320)) #29 Fri Mar 21 02:40:21 CET 2008 console [early0] enabled
CPU revision is: 00018448 (MIPS 4KEc)
TI AR7 (Unknown), ID: 0x5700, Revision: 0x00
Determined physical RAM map:
memory: 02000000 @ 14000000 (usable)
Zone PFN ranges:
Normal 81920 -> 90112
Movable zone start PFN for each node
early_node_map[1] active PFN ranges

    0:    81920 ->    90112

Built 1 zonelists in Zone order, mobility grouping on. Total pages:
8128
Kernel command line: console=ttyS0,115200
Primary instruction cache 16kB, VIPT, 4-way, linesize 16 bytes. Primary data cache 16kB, 4-way, VIPT, no aliases, linesize 16 bytes Synthesized clear page handler (26 instructions). Synthesized copy page handler (46 instructions).
PID hash table entries: 128 (order: 7, 512 bytes)
Dentry cache hash table entries: 4096 (order: 2, 16384 bytes) Inode-cache hash table entries: 2048 (order: 1, 8192 bytes)
Memory: 29700k/32768k available (2075k kernel code, 3068k reserved, 466k data, 136k init, 0k highmem)
Mount-cache hash table entries: 512
net_namespace: 152 bytes
NET: Registered protocol family 16
SCSI subsystem initialized
NET: Registered protocol family 2
Time: MIPS clocksource has been installed.
IP route cache hash table entries: 1024 (order: 0, 4096 bytes) TCP established hash table entries: 1024 (order: 1, 8192 bytes) TCP bind hash table entries: 1024 (order: 0, 4096 bytes)
TCP: Hash tables configured (established 1024 bind 1024)
TCP reno registered
io scheduler noop registered
io scheduler anticipatory registered (default)
io scheduler deadline registered
Serial: 8250/16550 driver $Revision: 1.90 $ 2 ports, IRQ sharing disabled
serial8250: ttyS0 at MMIO 0x8610e03 (irq = 15) is a 16550A console handover: boot [early0] -> real [ttyS0]
serial8250: ttyS1 at MMIO 0x8610f03 (irq = 16) is a 16550A
loop: module loaded
Fixed MDIO Bus: probed
ar7_wdt: failed to unlock WDT disable reg
ar7_wdt: failed to unlock WDT prescale reg
ar7_wdt: failed to unlock WDT change reg
ar7_wdt: timer margin 59 seconds (prescale 65535, change 57180, freq 62500000)
TCP cubic registered
NET: Registered protocol family 1
NET: Registered protocol family 17
RPC: Registered udp transport module.
RPC: Registered tcp transport module.
Root-NFS: No NFS server available, giving up.
VFS: Unable to mount root fs via NFS, trying floppy.
List of all partitions:
No filesystem could mount root, tried:

Kernel panic - not syncing: VFS: Unable to mount root fs on

unknown-block(2,0)
Rebooting in 3 seconds..

I don't see any problems with using PORT_16550A. What I'm still wondering how your kernel could work at all, since there is a missing case for setting up the TLB refill handler. Something like the patch below.

And most of the AR7 device driver code will not work for big endian. The log above is from a big endian AR7 system, where I needed to disable CPMAC to get it booting that far.

Thomas.

arch/mips/mm/tlbex.c +0 -0

index 3a93d4c..382738c 100644

Quoted 5 lines

Tried I get teh usual broken serial output:

IP6 oover IPv4 tuneliing driver
NET: eggistered protooll family 17
VFS: Monteed root (squahfss filesystem)reaadonly.

Linux version 2.6.25-rc6-00000-g151de9e-dirty ([email protected]) (gcc version 3.3.3 (Debian 20040320)) #29 Fri Mar 21 02:40:21 CET 2008 console [early0] enabled
CPU revision is: 00018448 (MIPS 4KEc)
TI AR7 (Unknown), ID: 0x5700, Revision: 0x00
Determined physical RAM map:
memory: 02000000 @ 14000000 (usable)
Zone PFN ranges:
Normal 81920 -> 90112
Movable zone start PFN for each node
early_node_map[1] active PFN ranges

    0:    81920 ->    90112

Built 1 zonelists in Zone order, mobility grouping on. Total pages:
8128
Kernel command line: console=ttyS0,115200
Primary instruction cache 16kB, VIPT, 4-way, linesize 16 bytes. Primary data cache 16kB, 4-way, VIPT, no aliases, linesize 16 bytes Synthesized clear page handler (26 instructions). Synthesized copy page handler (46 instructions).
PID hash table entries: 128 (order: 7, 512 bytes)
Dentry cache hash table entries: 4096 (order: 2, 16384 bytes) Inode-cache hash table entries: 2048 (order: 1, 8192 bytes)
Memory: 29700k/32768k available (2075k kernel code, 3068k reserved, 466k data, 136k init, 0k highmem)
Mount-cache hash table entries: 512
net_namespace: 152 bytes
NET: Registered protocol family 16
SCSI subsystem initialized
NET: Registered protocol family 2
Time: MIPS clocksource has been installed.
IP route cache hash table entries: 1024 (order: 0, 4096 bytes) TCP established hash table entries: 1024 (order: 1, 8192 bytes) TCP bind hash table entries: 1024 (order: 0, 4096 bytes)
TCP: Hash tables configured (established 1024 bind 1024)
TCP reno registered
io scheduler noop registered
io scheduler anticipatory registered (default)
io scheduler deadline registered
Serial: 8250/16550 driver $Revision: 1.90 $ 2 ports, IRQ sharing disabled
serial8250: ttyS0 at MMIO 0x8610e03 (irq = 15) is a 16550A console handover: boot [early0] -> real [ttyS0]
serial8250: ttyS1 at MMIO 0x8610f03 (irq = 16) is a 16550A
loop: module loaded
Fixed MDIO Bus: probed
ar7_wdt: failed to unlock WDT disable reg
ar7_wdt: failed to unlock WDT prescale reg
ar7_wdt: failed to unlock WDT change reg
ar7_wdt: timer margin 59 seconds (prescale 65535, change 57180, freq 62500000)
TCP cubic registered
NET: Registered protocol family 1
NET: Registered protocol family 17
RPC: Registered udp transport module.
RPC: Registered tcp transport module.
Root-NFS: No NFS server available, giving up.
VFS: Unable to mount root fs via NFS, trying floppy.
List of all partitions:
No filesystem could mount root, tried:

Kernel panic - not syncing: VFS: Unable to mount root fs on

unknown-block(2,0)
Rebooting in 3 seconds..

I don't see any problems with using PORT_16550A. What I'm still wondering how your kernel could work at all, since there is a missing case for setting up the TLB refill handler. Something like the patch below.

And most of the AR7 device driver code will not work for big endian. The log above is from a big endian AR7 system, where I needed to disable CPMAC to get it booting that far.

Thomas.

arch/mips/mm/tlbex.c +0 -0

index 3a93d4c..382738c 100644

Sorry but the linux-mips mailing list can't accept this patch for unknown reasons. I'll send a gz file, that will work

Hi Matteo,

Review

Quoted 3 lines · Matteo Croce

(quoted diff) +3 -0

+	switch (dev->divisor) {
+	case vlynq_div_auto:
+		/* Only try locally supplied clock, others cause problems */

i have a platform (AR7VWi - Leonardo Board) which has an external supplied clock from an ACX111 so the div_auto autoprobing will not work.

I put the vlynq_div_external code in front of this which should simply listen on the vlynq if there is a clock and use it. This solved one of the issues for me. Can you elaborate on the above comment where caused problems? I have seen multiple implementations of this, all of them did autoprobing first by listening on the remote clock first.

Review

Quoted 13 lines

(quoted diff) +13 -0

+		vlynq_reg_write(dev->remote->control, 0);
+		for (i = vlynq_ldiv2; i <= vlynq_ldiv8; i++) {
+			vlynq_reg_write(dev->local->control,
+					VLYNQ_CTRL_CLOCK_INT |
+					VLYNQ_CTRL_CLOCK_DIV(i - vlynq_ldiv1));
+			if (vlynq_linked(dev)) {
+				printk(KERN_DEBUG
+				       "%s: using local clock divisor %d\n",
+				       dev->dev.bus_id, i - vlynq_ldiv1 + 1);
+				dev->divisor = i;
+				return 0;
+			}
+		}

Hi Matteo,

Review

Quoted 3 lines · Matteo Croce

(quoted diff) +3 -0

+	switch (dev->divisor) {
+	case vlynq_div_auto:
+		/* Only try locally supplied clock, others cause problems */

i have a platform (AR7VWi - Leonardo Board) which has an external supplied clock from an ACX111 so the div_auto autoprobing will not work.

I put the vlynq_div_external code in front of this which should simply listen on the vlynq if there is a clock and use it. This solved one of the issues for me. Can you elaborate on the above comment where caused problems? I have seen multiple implementations of this, all of them did autoprobing first by listening on the remote clock first.

Review

Quoted 13 lines

(quoted diff) +13 -0

+		vlynq_reg_write(dev->remote->control, 0);
+		for (i = vlynq_ldiv2; i <= vlynq_ldiv8; i++) {
+			vlynq_reg_write(dev->local->control,
+					VLYNQ_CTRL_CLOCK_INT |
+					VLYNQ_CTRL_CLOCK_DIV(i - vlynq_ldiv1));
+			if (vlynq_linked(dev)) {
+				printk(KERN_DEBUG
+				       "%s: using local clock divisor %d\n",
+				       dev->dev.bus_id, i - vlynq_ldiv1 + 1);
+				dev->divisor = i;
+				return 0;
+			}
+		}

What i found in the TI code is that FIRST the local->control needs to get set before issueing a remote access so shouldnd the

	vlynq_reg_write(dev->remote ... )

move behind the dev->local ? I mean the logic is to set a local clock divisor - try to access something on the remote end and see if the link got up ?!?

Flo

Hi Matteo,

Review

Quoted 3 lines · Matteo Croce

(quoted diff) +3 -0

+	switch (dev->divisor) {
+	case vlynq_div_auto:
+		/* Only try locally supplied clock, others cause problems */

i have a platform (AR7VWi - Leonardo Board) which has an external supplied clock from an ACX111 so the div_auto autoprobing will not work.

I put the vlynq_div_external code in front of this which should simply listen on the vlynq if there is a clock and use it. This solved one of the issues for me. Can you elaborate on the above comment where caused problems? I have seen multiple implementations of this, all of them did autoprobing first by listening on the remote clock first.

Review

Quoted 13 lines

(quoted diff) +13 -0

+		vlynq_reg_write(dev->remote->control, 0);
+		for (i = vlynq_ldiv2; i <= vlynq_ldiv8; i++) {
+			vlynq_reg_write(dev->local->control,
+					VLYNQ_CTRL_CLOCK_INT |
+					VLYNQ_CTRL_CLOCK_DIV(i - vlynq_ldiv1));
+			if (vlynq_linked(dev)) {
+				printk(KERN_DEBUG
+				       "%s: using local clock divisor %d\n",
+				       dev->dev.bus_id, i - vlynq_ldiv1 + 1);
+				dev->divisor = i;
+				return 0;
+			}
+		}

What i found in the TI code is that FIRST the local->control needs to get set before issueing a remote access so shouldnd the

	vlynq_reg_write(dev->remote ... )

move behind the dev->local ? I mean the logic is to set a local clock divisor - try to access something on the remote end and see if the link got up ?!?

Flo

Works fine for my AR7 which has an interlan clock.

Quoted 3 lines

Works fine for my AR7 which has an interlan clock.

Its doesnt for me with an external clock - thats what i mean - Auto probing should first try to listen for an external clock before letting clocks run against each other. This is the hunk of a patch on top of yours ...

(quoted diff) +9 -1

@@ -371,12 +371,20 @@ static int __vlynq_enable_device(struct 
 
        switch (dev->divisor) {
        case vlynq_div_auto:
-               /* Only try locally supplied clock, others cause problems */
+       
+               vlynq_reg_write(dev->local->control, 0);
                vlynq_reg_write(dev->remote->control, 0);
+               if (vlynq_linked(dev)) {
+                       printk(KERN_DEBUG "%s: using external clock\n",
+                              dev->dev.bus_id);
+                       return 0;
+               }
+
                for (i = vlynq_ldiv2; i <= vlynq_ldiv8; i++) {
                        vlynq_reg_write(dev->local->control,
                                        VLYNQ_CTRL_CLOCK_INT |
                                        VLYNQ_CTRL_CLOCK_DIV(i - vlynq_ldiv1));
+                       vlynq_reg_write(dev->remote->control, 0);
                        if (vlynq_linked(dev)) {
                                printk(KERN_DEBUG
                                       "%s: using local clock divisor %d\n",

Flo

Quoted 23 lines

(quoted diff) +9 -1

@@ -371,12 +371,20 @@ static int __vlynq_enable_device(struct 
 
        switch (dev->divisor) {
        case vlynq_div_auto:
-               /* Only try locally supplied clock, others cause problems */
+       
+               vlynq_reg_write(dev->local->control, 0);
                vlynq_reg_write(dev->remote->control, 0);
+               if (vlynq_linked(dev)) {
+                       printk(KERN_DEBUG "%s: using external clock\n",
+                              dev->dev.bus_id);
+                       return 0;
+               }
+
                for (i = vlynq_ldiv2; i <= vlynq_ldiv8; i++) {
                        vlynq_reg_write(dev->local->control,
                                        VLYNQ_CTRL_CLOCK_INT |
                                        VLYNQ_CTRL_CLOCK_DIV(i - vlynq_ldiv1));
+                       vlynq_reg_write(dev->remote->control, 0);
                        if (vlynq_linked(dev)) {
                                printk(KERN_DEBUG
                                       "%s: using local clock divisor %d\n",

Quoted 1 line

Flo

isn't this what I do in my last patch?