Skip to contentSkip to conversation

i386: CS5535 chip support - cpu

PATCH Jordan Crouse 3 participants 4 messages Dec 7, 2005 – Dec 10, 2005

lore
Share & tools
lore
#0 lore raw

Greetings Ben -

Quoted 2 lines

- verifies the existence of the CS5535 by checking the DIVIL signature
- configures UART1 as a NS16550A

I'm confused by all this. Not so much your code, which is correct, but your reason for doing all of this in the first place. On a somewhat sane BIOS, it should have already set up all your descriptors and GPIO pins long before the kernel booted. Not that you should trust blindly the BIOS in all things, but as long as this is handled for you, you might as well take advantage of it.

Now, if your particular board has its own very good reasons for handling this, then thats fine, but I don't think this should be accepted as CS5535 support, because it stands a fairly good chance of causing trouble over
the larger set of Geode devices. I'll certainly listen to arguments
to the contrary, though.

Quoted 1 line

- (optionally) enables UART2 and configures it as a NS16550A

This is similar to code that has previously been submitted, and is attached below. Please review and comment.

Quoted 1 line

- (optionally) enables the SMBus/I2C interface

Is there any reason why you can't use the PCI devices instead? Granted, the SMBUS and GPIO are part of a multi-function device (which has a higher annoyance level), but it still seems better then passing around the global variables that you read from the MSR. Also, PCI has the advantage of being a much cleaner and well traveled path, which is always nice.

Regards,
Jordan

Signature · 18 lines
Jordan Crouse
Senior Linux Engineer
AMD - Personal Connectivity Solutions Group
<www.amd.com/embeddedprocessors>


GEODE:  Support second UART on CS5535

The CS5535 has a muxed serial port that can either be used to drive GPIO pins
or a second 16550 UART.  This code enables the UART via a command line option.

Signed-off-by: Jordan Crouse <[email protected]>
---

 drivers/serial/Kconfig       |   11 ++++++
 drivers/serial/Makefile      |    1 +
 drivers/serial/cs5535_uart.c |   83 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 95 insertions(+), 0 deletions(-)
drivers/serial/Kconfig +11 -0
index ad47c1b..0b1bb39 100644
@@ -616,6 +616,17 @@ config SERIAL_AU1X00_CONSOLE
 	  If you have an Alchemy AU1X00 processor (MIPS based) and you want
 	  to use a console on a serial port, say Y.  Otherwise, say N.
 
+config SERIAL_GEODE_UART2
+	bool "Enable AMD CS5535 UART2 as a serial port"
+	depends on MGEODE_LX
+	default y
+	select SERIAL_CORE
+	help
+	  Select this to allow the user to select the secondary CS5535 UART
+	  as a 16550 serial port instead of the default DDC interface. The
+	  UART2 can be selected by specifying geodeuart2 on the command
+	  line.
+
 config SERIAL_CORE
 	tristate
 
drivers/serial/Makefile +1 -0
index d7c7c71..16dee50 100644
@@ -57,3 +57,4 @@ obj-$(CONFIG_SERIAL_JSM) += jsm/
 obj-$(CONFIG_SERIAL_TXX9) += serial_txx9.o
 obj-$(CONFIG_SERIAL_VR41XX) += vr41xx_siu.o
 obj-$(CONFIG_SERIAL_SGI_IOC4) += ioc4_serial.o
+obj-$(CONFIG_SERIAL_GEODE_UART2) += cs5535_uart.o
drivers/serial/cs5535_uart.c new file +83 -0

83 lines not on this page · raw message

#1 lore raw

Hi Jordan,

Sorry it took so long for me to respond.

Quoted 11 lines · Jordan Crouse

Greetings Ben -

Quoted 2 lines

- verifies the existence of the CS5535 by checking the DIVIL signature
- configures UART1 as a NS16550A

I'm confused by all this. Not so much your code, which is correct, but your reason for doing all of this in the first place. On a somewhat sane BIOS, it should have already set up all your descriptors and GPIO pins long before the kernel booted. Not that you should trust blindly the BIOS in all things, but as long as this is handled for you, you might as well take advantage of it.

I think your phrase "somewhat sane BIOS" sums up the problem nicely. I threw in the UART1 config because UART2 has problems. Paranoia. I'm not sure that the UART1 config is needed, so I'm willing to scrap it, or wrap it in a config option.

If you are interested in my experience with UART2, see the kernel buzilla report #5677.

Quoted 5 lines

Now, if your particular board has its own very good reasons for handling this, then thats fine, but I don't think this should be accepted as CS5535 support, because it stands a fairly good chance of causing trouble over
the larger set of Geode devices. I'll certainly listen to arguments
to the contrary, though.

I'm curious as to what trouble this could cause?

Quoted 4 lines
Quoted 1 line

- (optionally) enables UART2 and configures it as a NS16550A

This is similar to code that has previously been submitted, and is attached below. Please review and comment.

Your patch looks like it would also solve the UART2 problem. What I still don't understand is why UART2 works fine in DOS and in linux 2.6.13-rc6-mm1, but not in the current kernel - in short, why is this needed?
My guess (not based on any data) is that a BIOS call (initiated by Linux) uses the DDC, but doesn't restore the GPIO to its original state. So, again, a BIOS problem.

Anyway, I recommend adding sanity checks on MSR_DIVIL_GLD_CAP and MSR_LBAR_GPIO before writing to the IO port. The code as-is may cause problems on a non-cs5535 board.

Quoted 7 lines
Quoted 1 line

- (optionally) enables the SMBus/I2C interface

Is there any reason why you can't use the PCI devices instead? Granted, the SMBUS and GPIO are part of a multi-function device (which has a higher annoyance level), but it still seems better then passing around the global variables that you read from the MSR. Also, PCI has the advantage of being a much cleaner and well traveled path, which is always nice.

I agree that PCI would work much cleaner.
I plan to revisit the GPIO & SMB drivers and get them to use PCI, but that may not happen for a while.

Quoted 2 lines

Regards,
Jordan

Thanks for your input.
Ben

#2 lore raw
Quoted 6 lines · Ben Gardner

I think your phrase "somewhat sane BIOS" sums up the problem nicely. I threw in the UART1 config because UART2 has problems. Paranoia. I'm not sure that the UART1 config is needed, so I'm willing to scrap it, or wrap it in a config option.

Quoted 6 lines

If you are interested in my experience with UART2, see the kernel buzilla report #5677.

My guess (not based on any data) is that a BIOS call (initiated by Linux) uses the DDC, but doesn't restore the GPIO to its original state. So, again, a BIOS problem.

Reading BZ 5677, I get some perspective on your issue now. I would agree that the data sounds like something is subverting the GPIO pins for its own use, and definately DDC is the logical conclusion.

Ok, so now with a bit of perspective on the issue, then I think that adding code to win back the serial port is useful, but it should be wrapped in a config or command line option, so that users can decide for themselves. After all, DDC is useful too (if you're an X user).

Quoted 7 lines
Quoted 5 lines

Now, if your particular board has its own very good reasons for handling this, then thats fine, but I don't think this should be accepted as CS5535 support, because it stands a fairly good chance of causing trouble over
the larger set of Geode devices. I'll certainly listen to arguments
to the contrary, though.

I'm curious as to what trouble this could cause?

I'm mainly worried about resource conflicts and such, especially with the first serial port. I'm less worried about the second port (as you could see from my code), but on the particular platform where that code was most useful, we had a fixed set of features so we knew we be more carefree. Certainly, on the big development platforms with a full BIOS, the code would be far more worriesome.

Quoted 4 lines

Your patch looks like it would also solve the UART2 problem. What I still don't understand is why UART2 works fine in DOS and in linux 2.6.13-rc6-mm1, but not in the current kernel - in short, why is this needed?

Our internal BIOS defaults to DDC on the second CS5535/CS5536 serial port, which normally isn't any big deal, because the first port is available, and if worse comes to worse, I can always switch over to LPC.

But on platforms like the Thin Client RDK, which is more like a production platform, there aren't any physical serial ports populated. So in order to get access, our systems guys made a little dongle that plugged into the VGA port and split out the DDC lines and turned them into a serial connection. And this code was born to turn off the default DDC and enable the serial port.

Quoted 3 lines

Anyway, I recommend adding sanity checks on MSR_DIVIL_GLD_CAP and MSR_LBAR_GPIO before writing to the IO port. The code as-is may cause problems on a non-cs5535 board.

Thats a great point. Actually, we should check for both CS5535 and CS5536, since the code will work on both.. :)

Quoted 1 line

Thanks for your input.

Thanks for your interest.
Jordan

Signature · 4 lines
Jordan Crouse
Senior Linux Engineer
AMD - Personal Connectivity Solutions Group
<www.amd.com/embeddedprocessors>
#3 lore raw
In reply to Jordan Crouse

drivers/serial/cs5535_uart.c line 36

Quoted 5 lines · Jordan Crouse
drivers/serial/cs5535_uart.c new file +5 -0
+/* Note - this does not check to see if the CS5535 actually exists */
+
+#define LO(b)  (((1 << b) << 16) | 0x0000)
+#define HI(b)  ((0x0000 << 16) | (1 << b))
+

What kind of ninja mutant macros are these?

drivers/serial/cs5535_uart.c line 48

Quoted 5 lines
drivers/serial/cs5535_uart.c new file +5 -0
+static u32 intab[16] __initdata = {
+	0x20,HI(3), 0x24,LO(3), 0x28,LO(3),
+	0x2C,LO(3), 0x34,HI(3), 0x38,LO(3),
+     0x40,LO(3), 0x44,LO(3)
+};

Can't you just use values here, without the macros?

Pavel

Signature · 1 line
Thanks, Sharp!