Skip to contentSkip to conversation

arm: socfpga: Question about FPGA/HPS SDRAM Bridge

U-Boot Jian Luo 2 participants 8 messages Sep 1, 2015 – Sep 2, 2015 FPGA manager framework

lore
Share & tools
lore

Jian Luo

author
#0 lore raw

Hi!

I've read about an implementation requirement regarding the usage of FPGA/HPS SDRAM bridge.
(Link and Text attached at the end.)

The 3. step involves writing the APPLYCFG bit in the STATICCFG register while the SDRAM interface is completely IDLE.
IMHO, it's only possible in SPL stage where everything runs in SRAM. FPGA however can not be configured until U-Boot is ready (step 2). So warm reset should be performed after FPGA configuration.

My idea is to patch sdram.c to dynamically write FPGAPORTRST and APPLYCFG based on information in sysmgr_regs->romcodegrp_bootromswstate.

Is any one working on this?

Thanks!

Signature · 65 lines
Best regards,

Jian Luo


----
Link and Text of the note:

https://support.criticallink.com/redmine/projects/mityarm-5cs/wiki/Important_Note_about_FPGAHPS_SDRAM_Bridge

Important Note about FPGA/HPS SDRAM Bridge (2013-12-13)

Altera has recently disclosed an implementation requirement related to 
the use of the FPGA to HPS SDRAM bridges. Until more formal technical 
notes are published at Altera, this page will be maintained to outline 
the understood issue for the benefits of MitySOM-5CSX customers.

Configuration of the FPGA to HPS SDRAM (fpga2sdram) AXI bridges involves 
the following major steps:

1.   First, the FPGA ports on the fpga2sdram peripheral must be placed 
in reset. This is accomplished by writing a zero to the FPGAPORTRST 
register in the SDRAM Controller control group.

2.   Second, the FPGA must be configured with an image that includes the 
configuration of the fpga2sdram ports. The FPGA fabric asserts 
configuration input ports at the input to the fpga2sdram bridges. The 
configuration ports affect such things as the width of the port as well 
as the direction, etc. When the FPGA is not configured, these 
configuration inputs are not defined.

3.   Third, once the configuration inputs are set, the configuration 
must be then latched / applied to the fpga2sdram bridge peripheral. This 
is accomplished by writing a one to the APPLYCFG bit in the STATICCFG 
register in the SDRAM Controller control group. This bit can only be 
written to while the SDRAM DDR interface is guaranteed to be completely 
IDLE (including transfers from the ARM cores, DMAs, etc.).

4.   Finally, the FPGA ports on the fpga2sdram peripheral can be taken 
out of reset based on your configuration. This is accomplished by 
writing ones to the appropriate bits in the FPGAPORTRST register in the 
SDRAM Controller control group.

If these steps are not followed, attempting to use an FPGA port on the 
HPS SDRAM bridge controller may result in critical failure -- the HPS 
subsystem may freeze and effectively lock up the processor.

As a consequence to the note bolded in step 3, the fpga2sdram controller 
cannot be practically configured while most high level operating systems 
are running (linux, windows, android, etc.). Altera has provided the 
capability to set the configuration bit in step three with a macro 
command in their (and Critical Link's) u-Boot port. This is accomplished 
by copying a small routine to on-chip RAM that disables caches and 
asserts the APPLYCFG bit and then returns operation to the typical DDR 
space.

Therefor, if you have an FPGA design that utilizes the fpga2sdram 
controller, you must program the FPGA in u-Boot following an FPGA / 
power on reset situation.

Once the SDRAM controller is properly configured, the FPGA ports may be 
reset and enabled (steps 1 and 4) as often as necessary in order to 
facilitate reloading of the FPGA -- as long as the new fpga2sdram port 
configuration matches the original configuration. This will allow for 
reconfiguration of an FPGA while running linux, if necessary.
#1 lore raw
Quoted 1 line · Jian Luo

Hi!

Hi,

Quoted 10 lines

I've read about an implementation requirement regarding the usage of FPGA/HPS SDRAM bridge.
(Link and Text attached at the end.)

The 3. step involves writing the APPLYCFG bit in the STATICCFG register while the SDRAM interface is completely IDLE.
IMHO, it's only possible in SPL stage where everything runs in SRAM. FPGA however can not be configured until U-Boot is ready (step 2). So warm reset should be performed after FPGA configuration.

Quoted 4 lines

My idea is to patch sdram.c to dynamically write FPGAPORTRST and APPLYCFG based on information in sysmgr_regs->romcodegrp_bootromswstate.

Is any one working on this?

We do this sort of stuff by running code from cache, see socfpga_sdram_apply_static_cfg() in arch/arm/mach-socfpga/misc.c .

You want to use the "bridge enable" and "bridge disable" commands to enable/disable the bridges between FPGA and HPS in U-Boot.

Hope that helps ;-)

Best regards,
Marek Vasut

author
#2 lore raw

**
Hi Marek,

Quoted 14 lines · Marek Vasut
Quoted 1 line · Jian Luo

Hi!

Hi,

Quoted 10 lines

I've read about an implementation requirement regarding the usage of FPGA/HPS SDRAM bridge.
(Link and Text attached at the end.)

The 3. step involves writing the APPLYCFG bit in the STATICCFG register while the SDRAM interface is completely IDLE.
IMHO, it's only possible in SPL stage where everything runs in SRAM. FPGA however can not be configured until U-Boot is ready (step 2). So warm reset should be performed after FPGA configuration.

Quoted 6 lines
Quoted 4 lines

My idea is to patch sdram.c to dynamically write FPGAPORTRST and APPLYCFG based on information in sysmgr_regs->romcodegrp_bootromswstate.

Is any one working on this?

We do this sort of stuff by running code from cache, see socfpga_sdram_apply_static_cfg() in arch/arm/mach-socfpga/misc.c .

Thanks. It's way better than run bootloader twice!

Quoted 3 lines

You want to use the "bridge enable" and "bridge disable" commands to enable/disable the bridges between FPGA and HPS in U-Boot.

What about calling socfpga_sdram_apply_static_cfg() direct in socfpga_load() in drivers/fpga/socfpga.c to make it generic? socfpga_load() already did step 1 and 2.
Is there any side effect when fpga2sdram bridge is not used?

Quoted 5 lines

Hope that helps ;-)

Best regards,
Marek Vasut

#3 lore raw
Quoted 2 lines · Jian Luo

**
Hi Marek,

Hi!

Quoted 19 lines · Marek Vasut
Quoted 18 lines · Jian Luo
Quoted 1 line

Hi!

Hi,

Quoted 13 lines

I've read about an implementation requirement regarding the usage of FPGA/HPS SDRAM bridge.
(Link and Text attached at the end.)

The 3. step involves writing the APPLYCFG bit in the STATICCFG register while the SDRAM interface is completely IDLE.
IMHO, it's only possible in SPL stage where everything runs in SRAM. FPGA however can not be configured until U-Boot is ready (step 2). So warm reset should be performed after FPGA configuration.

My idea is to patch sdram.c to dynamically write FPGAPORTRST and APPLYCFG based on information in sysmgr_regs->romcodegrp_bootromswstate.

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

Is any one working on this?

We do this sort of stuff by running code from cache, see socfpga_sdram_apply_static_cfg() in arch/arm/mach-socfpga/misc.c .

Thanks. It's way better than run bootloader twice!

Yeah :)

Quoted 5 lines
Quoted 2 lines

You want to use the "bridge enable" and "bridge disable" commands to enable/disable the bridges between FPGA and HPS in U-Boot.

What about calling socfpga_sdram_apply_static_cfg() direct in socfpga_load() in drivers/fpga/socfpga.c to make it generic?

Which code exactly do you refer to ?

Quoted 2 lines

socfpga_load() already did step 1 and 2.
Is there any side effect when fpga2sdram bridge is not used?

The idea is to keep the FPGA programming code and the code which controls the bridges separate. The reason for that is that you can have content in the FPGA which is running completely independent of the HPS software. You don't always need to enable the bridges.

The FPGA2SDRAM bridge (or, more like fpga2hps bridge) is used for stuff like framebuffers implemented in the FPGA, so that the FPGA can fetch data from the SDRAM.

Best regards,
Marek Vasut

author
#4 lore raw

Hi Marek,

Quoted 1 line · Marek Vasut

On Tuesday, September 01, 2015 at 01:49:43 PM, Jian Luo wrote:

----snip----

Quoted 4 lines
Quoted 2 lines

What about calling socfpga_sdram_apply_static_cfg() direct in socfpga_load() in drivers/fpga/socfpga.c to make it generic?

Which code exactly do you refer to ?

I mean something like this.

drivers/fpga/socfpga.c +0 -0
index 4448250..d0f67f6 100644
@@ -300,5 +300,12 @@ int socfpga_load(Altera_desc *desc, const void 

*rbf_data, size_t rbf_size)

                 return status;

         /* Ensure the FPGA entering user mode */
-       return fpgamgr_program_poll_usermode();
+       status = fpgamgr_program_poll_usermode();
+       if (status)

+ return status;
+
+ /* Latch FPGA2SDRAM bridge configuration */
+ socfpga_sdram_apply_static_cfg();
+
+ return 0;
}

Quoted 8 lines
Quoted 3 lines

socfpga_load() already did step 1 and 2.
Is there any side effect when fpga2sdram bridge is not used?

The idea is to keep the FPGA programming code and the code which controls the bridges separate. The reason for that is that you can have content in the FPGA which is running completely independent of the HPS software. You don't always need to enable the bridges.

No, it's the other way around here.
socfpga_load() runs eventually when command "fpga load" is called. So when I explicitly use "fpga load" command, the old FPGA program is expected to stop running, right?

And I don't want to enable the bridge here, just to latch the SDRAM configuration.

Quoted 7 lines

The FPGA2SDRAM bridge (or, more like fpga2hps bridge) is used for stuff like framebuffers implemented in the FPGA, so that the FPGA can fetch data from the SDRAM.

Best regards,
Marek Vasut

Best regards,

Jian Luo

#5 lore raw
Quoted 1 line · Jian Luo

Hi Marek,

Hi!

Quoted 39 lines · Marek Vasut
Quoted 1 line · Jian Luo

----snip----

Quoted 4 lines
Quoted 2 lines

What about calling socfpga_sdram_apply_static_cfg() direct in socfpga_load() in drivers/fpga/socfpga.c to make it generic?

Which code exactly do you refer to ?

I mean something like this.

drivers/fpga/socfpga.c +8 -1
index 4448250..d0f67f6 100644
@@ -300,5 +300,12 @@ int socfpga_load(Altera_desc *desc, const void
 *rbf_data, size_t rbf_size)
                 return status;
 
         /* Ensure the FPGA entering user mode */
-       return fpgamgr_program_poll_usermode();
+       status = fpgamgr_program_poll_usermode();
+       if (status)
+               return status;
+
+       /* Latch FPGA2SDRAM bridge configuration */
+       socfpga_sdram_apply_static_cfg();
+
+       return 0;
  }
Quoted 7 lines
Quoted 2 lines

socfpga_load() already did step 1 and 2.
Is there any side effect when fpga2sdram bridge is not used?

The idea is to keep the FPGA programming code and the code which controls the bridges separate. The reason for that is that you can have content in the FPGA which is running completely independent of the HPS software. You don't always need to enable the bridges.

Quoted 4 lines

No, it's the other way around here.
socfpga_load() runs eventually when command "fpga load" is called. So when I explicitly use "fpga load" command, the old FPGA program is expected to stop running, right?

Right

Quoted 2 lines

And I don't want to enable the bridge here, just to latch the SDRAM configuration.

So what you want to do is "bridge disable ; fpga load ... ; bridge enable", isn't that the correct sequence ? I might be missing something though, sorry.

Quoted 10 lines
Quoted 6 lines

The FPGA2SDRAM bridge (or, more like fpga2hps bridge) is used for stuff like framebuffers implemented in the FPGA, so that the FPGA can fetch data from the SDRAM.

Best regards,
Marek Vasut

Best regards,

Jian Luo

Best regards,
Marek Vasut

author
#6 lore raw

Hi Marek,

Quoted 36 lines · Marek Vasut
Quoted 1 line · Jian Luo

Hi Marek,

Hi!

Quoted 31 lines · Marek Vasut
Quoted 1 line · Jian Luo

----snip----

Quoted 4 lines
Quoted 2 lines

What about calling socfpga_sdram_apply_static_cfg() direct in socfpga_load() in drivers/fpga/socfpga.c to make it generic?

Which code exactly do you refer to ?

I mean something like this.

drivers/fpga/socfpga.c +8 -1
index 4448250..d0f67f6 100644
@@ -300,5 +300,12 @@ int socfpga_load(Altera_desc *desc, const void
 *rbf_data, size_t rbf_size)
                 return status;
 
         /* Ensure the FPGA entering user mode */
-       return fpgamgr_program_poll_usermode();
+       status = fpgamgr_program_poll_usermode();
+       if (status)
+               return status;
+
+       /* Latch FPGA2SDRAM bridge configuration */
+       socfpga_sdram_apply_static_cfg();
+
+       return 0;
  }
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 2 lines

socfpga_load() already did step 1 and 2.
Is there any side effect when fpga2sdram bridge is not used?

The idea is to keep the FPGA programming code and the code which controls the bridges separate. The reason for that is that you

can have

Quoted 1 line
Quoted 1 line
Quoted 1 line

content in the FPGA which is running completely independent of

the HPS

Quoted 7 lines
Quoted 6 lines
Quoted 1 line

software. You don't always need to enable the bridges.

No, it's the other way around here.
socfpga_load() runs eventually when command "fpga load" is called. So when I explicitly use "fpga load" command, the old FPGA program is expected to stop running, right?

Quoted 6 lines

Right

Quoted 2 lines

And I don't want to enable the bridge here, just to latch the SDRAM configuration.

So what you want to do is "bridge disable ; fpga load ... ; bridge

enable",

Quoted 1 line

isn't that the correct sequence ? I might be missing something

though, sorry.

No, you are right. Sorry, I didn't think it through.

My original thought was, that "fpga load" should not depend on "bridge e/d" regardless of how fpga2sdram is configured. Now I think the dependency is a bad idea.
Sorry for the confusion.

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

The FPGA2SDRAM bridge (or, more like fpga2hps bridge) is used for

stuff

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines

like framebuffers implemented in the FPGA, so that the FPGA can fetch data from the SDRAM.

Best regards,
Marek Vasut

Quoted 6 lines
Quoted 3 lines

Best regards,

Jian Luo

Best regards,
Marek Vasut

Best regards,

Jian Luo

#7 lore raw
Quoted 1 line · Jian Luo

Hi Marek,

Hi!

Quoted 70 lines · Marek Vasut
Quoted 37 lines · Jian Luo
Quoted 1 line

Hi Marek,

Hi!

Quoted 32 lines · Marek Vasut
Quoted 1 line · Jian Luo

----snip----

Quoted 4 lines
Quoted 2 lines

What about calling socfpga_sdram_apply_static_cfg() direct in socfpga_load() in drivers/fpga/socfpga.c to make it generic?

Which code exactly do you refer to ?

I mean something like this.

drivers/fpga/socfpga.c +8 -1
index 4448250..d0f67f6 100644
@@ -300,5 +300,12 @@ int socfpga_load(Altera_desc *desc, const void
 *rbf_data, size_t rbf_size)
 
                 return status;
 
         /* Ensure the FPGA entering user mode */
 
-       return fpgamgr_program_poll_usermode();
+       status = fpgamgr_program_poll_usermode();
+       if (status)
+               return status;
+
+       /* Latch FPGA2SDRAM bridge configuration */
+       socfpga_sdram_apply_static_cfg();
+
+       return 0;
Quoted 7 lines
Quoted 7 lines
  }
Quoted 5 lines
Quoted 2 lines

socfpga_load() already did step 1 and 2.
Is there any side effect when fpga2sdram bridge is not used?

The idea is to keep the FPGA programming code and the code which controls the bridges separate. The reason for that is that you

can have

Quoted 1 line
Quoted 1 line
Quoted 1 line

content in the FPGA which is running completely independent of

the HPS

Quoted 7 lines
Quoted 6 lines
Quoted 1 line

software. You don't always need to enable the bridges.

No, it's the other way around here.
socfpga_load() runs eventually when command "fpga load" is called. So when I explicitly use "fpga load" command, the old FPGA program is expected to stop running, right?

Quoted 6 lines

Right

Quoted 2 lines

And I don't want to enable the bridge here, just to latch the SDRAM configuration.

So what you want to do is "bridge disable ; fpga load ... ; bridge

enable",

Quoted 1 line

isn't that the correct sequence ? I might be missing something

Quoted 7 lines

though, sorry.

No, you are right. Sorry, I didn't think it through.

My original thought was, that "fpga load" should not depend on "bridge e/d" regardless of how fpga2sdram is configured. Now I think the dependency is a bad idea.

Yeah, you should be able to control the bridges between the FPGA and the FPGA loading independently for the particular case where the FPGA runs independently of the HPS.

Quoted 1 line

Sorry for the confusion.

Well, I'm glad it got sorted out :)