Skip to contentSkip to conversation

phosphor-ipmi-flash: Update over eSPI interface

DISCUSSION Harry Sung1 3 participants 4 messages Sep 11, 2019 – Sep 12, 2019

lore
Share & tools
lore
#0 lore raw
Quoted 10 lines · Oskar Senft
Quoted 4 lines

Hi Harry

What's the behavior on eSPI? I assume you still have the aspeed-lpc-ctrl

enabled, right?

Quoted 3 lines

Thanks
Oskar.

Hi Oskar,
Yes, I still enabled the aspeed-lpc-ctrl in my build. Because phosphor-ipmi-flash has some mandatory actions on /dev/aspeed-lpc-ctrl before flash (settings for HICR5, HICR7 and HICR8) even though these settings are meaningless for eSPI.

Currently, I set ESPI084 (source address) and ESPI088 (target address) registers manually because linux seems not have a driver can help us to set ESPI084 and ESPI088.

Due to the limitation of AST2500, we can only write 256 bytes in one write operation (write shared memory). Based on the test result, it takes about 30 mins to transfer a 32MB image over eSPI.

Thanks,
Harry

Quoted 44 lines
Quoted 9 lines · On Mon, Sep 9, 2019 at 4:41 AM Harry Sung1 <[email protected]> wrote:
Quoted 7 lines

Hi Patrick,

I found “phosphor-ipmi-flash” have not support flash over eSPI yet.

Quoted 7 lines
Quoted 7 lines

May I ask if you have any plans to support flash over eSPI?

I have done a simple test about shared memory between host and BMC :

The shared memory is work after I set ESPI084 (source address) and ESPI088

(target address) registers.

Quoted 27 lines
Quoted 27 lines

But it has an limitation that only 256 bytes are available on each page (4KB).

For example, if host address starts to write from 0xFE0B0000 (BMC reserved enough memory already)

Writable area are:

0xFE0B0000 ~ 0xFE0B00FF

0xFE0B1000 ~ 0xFE0B10FF

0xFE0B2000 ~ 0xFE0B20FF

0xFE0B3000 ~ 0xFE0B30FF

…

…

…

Quoted 7 lines
Quoted 2 lines
Quoted 2 lines

Thanks,
Harry

Harry, currently there's no plan to implement it as I have no method of testing it, However, it should prove fairly straightforward to add another option to the transport mechanism list. Please let me know if you run into any blockers.

Hi Patrick,
Got it. The better way to set eSPI register is setting them by the driver, right?
For quick validation, I am going to use the " ipmilpc" interface and set necessary eSPI registers manually.

Thanks,
Harry

#1 lore raw
Quoted 12 lines · On Wed, Sep 11, 2019 at 1:59 AM Harry Sung1 <[email protected]> wrote:
Quoted 11 lines · Oskar Senft
Quoted 4 lines

Hi Harry

What's the behavior on eSPI? I assume you still have the aspeed-lpc-ctrl

enabled, right?

Quoted 3 lines

Thanks
Oskar.

Quoted 7 lines

Hi Oskar,
Yes, I still enabled the aspeed-lpc-ctrl in my build. Because phosphor-ipmi-flash has some mandatory actions on /dev/aspeed-lpc-ctrl before flash (settings for HICR5, HICR7 and HICR8) even though these settings are meaningless for eSPI.

Currently, I set ESPI084 (source address) and ESPI088 (target address) registers manually because linux seems not have a driver can help us to set ESPI084 and ESPI088.

Due to the limitation of AST2500, we can only write 256 bytes in one write operation (write shared memory). Based on the test result, it takes about 30 mins to transfer a 32MB image over eSPI.

:( wow, that's unfortunately rather slow.

Quoted 50 lines

Thanks,
Harry

Quoted 47 lines
Quoted 9 lines · On Mon, Sep 9, 2019 at 4:41 AM Harry Sung1 <[email protected]> wrote:
Quoted 7 lines

Hi Patrick,

I found “phosphor-ipmi-flash” have not support flash over eSPI yet.

Quoted 7 lines
Quoted 7 lines

May I ask if you have any plans to support flash over eSPI?

I have done a simple test about shared memory between host and BMC :

The shared memory is work after I set ESPI084 (source address) and ESPI088

(target address) registers.

Quoted 29 lines
Quoted 29 lines

But it has an limitation that only 256 bytes are available on each page (4KB).

For example, if host address starts to write from 0xFE0B0000 (BMC reserved enough memory already)

Writable area are:

0xFE0B0000 ~ 0xFE0B00FF

0xFE0B1000 ~ 0xFE0B10FF

0xFE0B2000 ~ 0xFE0B20FF

0xFE0B3000 ~ 0xFE0B30FF

…

…

…

Thanks,
Harry

Quoted 8 lines
Quoted 4 lines

Harry, currently there's no plan to implement it as I have no method of testing it, However, it should prove fairly straightforward to add another option to the transport mechanism list. Please let me know if you run into any blockers.

Hi Patrick,
Got it. The better way to set eSPI register is setting them by the driver, right?
For quick validation, I am going to use the " ipmilpc" interface and set necessary eSPI registers manually.

I don't know as much about the eSPI variation of this. ipmilpc uses whatever LPC memory shared option is available (in coordination with the host+bmc). If eSPI doesn't use the aspeed-lpc-ctrl driver for what it needs, then perhaps a new option should be added ipmiespi?

Quoted 3 lines

Thanks,
Harry

#2 lore raw

Hi Harry

I've done some experiments with the USB virtual NIC on the AST2500 and found that to work rather nicely.

We're currently investigating in my team to use that interface as the primary method for transferring data between the host and the BMC. From what I can tell, this seems to be the fastest, most secure method. The advantage also is that it doesn't need any low-level HW / memory access on the host. However, the host still needs to have the USB NIC on its side supported (driver) and configured (IP address). For our environment (Linux), this is easy to achieve.

It should be possible to update the phosphor-ipmi-flash BMC and host side implementation to use a USB NIC for data transfer. However, we haven't investigated those details yet.

Other methods for data transfer (LPC, PCIe, eSPI, SuperI/O) all seem to open up a large security hole in the AST2500.

Oskar.

Quoted 106 lines · Patrick Venture
Quoted 7 lines · On Wed, Sep 11, 2019 at 1:59 AM Harry Sung1 <[email protected]> wrote:
Quoted 7 lines · Oskar Senft
Quoted 4 lines

Hi Harry

What's the behavior on eSPI? I assume you still have the

aspeed-lpc-ctrl

Quoted 7 lines
Quoted 4 lines

enabled, right?

Quoted 3 lines

Thanks
Oskar.

Hi Oskar,
Yes, I still enabled the aspeed-lpc-ctrl in my build. Because

phosphor-ipmi-flash has some mandatory actions on /dev/aspeed-lpc-ctrl before flash (settings for HICR5, HICR7 and HICR8) even though these settings are meaningless for eSPI.

Quoted 2 lines

Currently, I set ESPI084 (source address) and ESPI088 (target address)

registers manually because linux seems not have a driver can help us to set ESPI084 and ESPI088.

Quoted 2 lines

Due to the limitation of AST2500, we can only write 256 bytes in one

write operation (write shared memory).

Quoted 1 line

Based on the test result, it takes about 30 mins to transfer a 32MB

image over eSPI.

:( wow, that's unfortunately rather slow.

Quoted 5 lines

Thanks,
Harry

Quoted 2 lines
Quoted 2 lines

On Mon, Sep 9, 2019 at 4:41 AM Harry Sung1 <[email protected]>

Quoted 7 lines · wrote:
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Hi Patrick,

I found “phosphor-ipmi-flash” have not support flash over eSPI yet.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

May I ask if you have any plans to support flash over eSPI?

I have done a simple test about shared memory between host and BMC :

The shared memory is work after I set ESPI084 (source address) and

ESPI088

Quoted 3 lines
Quoted 3 lines

(target address) registers.

Quoted 2 lines
Quoted 2 lines

But it has an limitation that only 256 bytes are available on each

page (4KB).

Quoted 25 lines
Quoted 25 lines
Quoted 25 lines
Quoted 25 lines

For example, if host address starts to write from 0xFE0B0000 (BMC reserved enough memory already)

Writable area are:

0xFE0B0000 ~ 0xFE0B00FF

0xFE0B1000 ~ 0xFE0B10FF

0xFE0B2000 ~ 0xFE0B20FF

0xFE0B3000 ~ 0xFE0B30FF

…

…

…

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines

Thanks,
Harry

Harry, currently there's no plan to implement it as I have no method

of testing

Quoted 1 line
Quoted 1 line

it, However, it should prove fairly straightforward to add another

option to

Quoted 5 lines
Quoted 2 lines

the transport mechanism list. Please let me know if you run into any blockers.

Hi Patrick,
Got it. The better way to set eSPI register is setting them by the

driver, right?

Quoted 1 line

For quick validation, I am going to use the " ipmilpc" interface and set

necessary eSPI registers manually.

I don't know as much about the eSPI variation of this. ipmilpc uses whatever LPC memory shared option is available (in coordination with the host+bmc). If eSPI doesn't use the aspeed-lpc-ctrl driver for what it needs, then perhaps a new option should be added ipmiespi?

Quoted 3 lines

Thanks,
Harry

#3 lore raw

Here some more details on how the USB virtual NIC works:
Sources

-

https://events.static.linuxfound.org/sites/events/files/slides/USB%20Gadget%20Configfs%20API_0.pdf
-

https://developer.toradex.com/knowledge-base/usb-device-mode-(linux)

Build Configuration

linux/arch/arm/boot/dts/aspeed-bmc-[machine].dts

+&vhub {

+ status = "okay";

+};

+

gbmc/[...]/recipes-kernel/linux/linux-aspeed/[machine].cfg

+# Enable virtual USB NIC

+CONFIG_USB_CONFIGFS_ECM=y

+CONFIG_USB_CONFIGFS_ECM_SUBSET=y

BMC Runtime Configuration

See attached usb_network.sh. This needs to be executed at startup. Obviously, you'll need to replace the vendor and product ID as well as the strings with something different.

Network configuration needs to go into /etc/systemd/network. See attached 00-bmc-usb0.network.

Host Runtime Configuration

As soon as the BMC is booted, the host should see the BMC as an additional USB hub.

The last command on the BMC will cause an actual USB device to be visible to the host. If it does not get auto-loaded, load the cdc_ether driver manually. Once loaded, this adds a "usb0" network interface on the host that can be configured like any other Ethernet device:

ifconfig usb0 169.254.254.1 netmask 255.255.255.0 up

From here on you can then execute SSH / SCP from the host to the local BMC. However, for phosphor-ipmi-flash, it might be better to implement a new TCP-based method right in phosphor-ipmi-flash both on the BMC and the host side. The important bit is that whatever method you use, it must only stage the image to /tmp where phosphor-ipmi-flash-bios-verify.target can then pick it up for verification. You certainly don't want to have root-level access from the host to the BMC as that would allow the host to take ownership of the BMC.

Oskar.

Quoted 133 lines · Oskar Senft

Hi Harry

I've done some experiments with the USB virtual NIC on the AST2500 and found that to work rather nicely.

We're currently investigating in my team to use that interface as the primary method for transferring data between the host and the BMC. From what I can tell, this seems to be the fastest, most secure method. The advantage also is that it doesn't need any low-level HW / memory access on the host. However, the host still needs to have the USB NIC on its side supported (driver) and configured (IP address). For our environment (Linux), this is easy to achieve.

It should be possible to update the phosphor-ipmi-flash BMC and host side implementation to use a USB NIC for data transfer. However, we haven't investigated those details yet.

Other methods for data transfer (LPC, PCIe, eSPI, SuperI/O) all seem to open up a large security hole in the AST2500.

Oskar.

Quoted 107 lines · Patrick Venture
Quoted 7 lines · On Wed, Sep 11, 2019 at 1:59 AM Harry Sung1 <[email protected]> wrote:
Quoted 7 lines · Oskar Senft
Quoted 4 lines

Hi Harry

What's the behavior on eSPI? I assume you still have the

aspeed-lpc-ctrl

Quoted 7 lines
Quoted 4 lines

enabled, right?

Quoted 3 lines

Thanks
Oskar.

Hi Oskar,
Yes, I still enabled the aspeed-lpc-ctrl in my build. Because

phosphor-ipmi-flash has some mandatory actions on /dev/aspeed-lpc-ctrl before flash (settings for HICR5, HICR7 and HICR8) even though these settings are meaningless for eSPI.

Quoted 2 lines

Currently, I set ESPI084 (source address) and ESPI088 (target address)

registers manually because linux seems not have a driver can help us to set ESPI084 and ESPI088.

Quoted 2 lines

Due to the limitation of AST2500, we can only write 256 bytes in one

write operation (write shared memory).

Quoted 1 line

Based on the test result, it takes about 30 mins to transfer a 32MB

image over eSPI.

:( wow, that's unfortunately rather slow.

Quoted 5 lines

Thanks,
Harry

Quoted 2 lines
Quoted 2 lines

On Mon, Sep 9, 2019 at 4:41 AM Harry Sung1 <[email protected]>

Quoted 5 lines · wrote:
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

Hi Patrick,

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

I found “phosphor-ipmi-flash” have not support flash over eSPI yet.

May I ask if you have any plans to support flash over eSPI?

I have done a simple test about shared memory between host and BMC

:

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

The shared memory is work after I set ESPI084 (source address) and

ESPI088

Quoted 3 lines
Quoted 3 lines

(target address) registers.

Quoted 2 lines
Quoted 2 lines

But it has an limitation that only 256 bytes are available on each

page (4KB).

Quoted 25 lines
Quoted 25 lines
Quoted 25 lines
Quoted 25 lines

For example, if host address starts to write from 0xFE0B0000 (BMC reserved enough memory already)

Writable area are:

0xFE0B0000 ~ 0xFE0B00FF

0xFE0B1000 ~ 0xFE0B10FF

0xFE0B2000 ~ 0xFE0B20FF

0xFE0B3000 ~ 0xFE0B30FF

…

…

…

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines

Thanks,
Harry

Harry, currently there's no plan to implement it as I have no method

of testing

Quoted 1 line
Quoted 1 line

it, However, it should prove fairly straightforward to add another

option to

Quoted 5 lines
Quoted 2 lines

the transport mechanism list. Please let me know if you run into any blockers.

Hi Patrick,
Got it. The better way to set eSPI register is setting them by the

driver, right?

Quoted 1 line

For quick validation, I am going to use the " ipmilpc" interface and

set necessary eSPI registers manually.

I don't know as much about the eSPI variation of this. ipmilpc uses whatever LPC memory shared option is available (in coordination with the host+bmc). If eSPI doesn't use the aspeed-lpc-ctrl driver for what it needs, then perhaps a new option should be added ipmiespi?

Quoted 3 lines

Thanks,
Harry