Altera PCI Express Development Kit: Linux DMA Driver

--- Quote Start ---

Your bar_min_len should be alright. I kept mine the same length (as in {32768,0,256,0,0,0}) as it was before, but I've tried both ways and I'm successful in either case. Have your tried running the driver against the reference design provided with the board? That design was compiled in 7.2, which works for me on my computer, running 32 bit Red Hat. In fact, 7.2 & 9.0 are the only software versions I can get to work... lspci won't show the board using 8.x.

--- Quote End ---

I am not sure what you mean by reference design. I am using following:

www.altera.com/literature/an/an456.pdf

With windows driver it works. Also, in Linux with 8.x, I can see the board using lspci.

Thanks again Rob.

--- Quote Start ---

Oh, I've never even seen that before, but I think AN456 is the same as what I'm talking about, which came on a disc (windoze) with the AGX development kit. Furthermore, when you create a new PCIe megafunction, Quartus generates a chaining dma example design in <mf_name>_examples/chaining_dma. I use that project file and set the device and pin-outs according to these reference designs. Also, I can only get it to work by using data/descriptor, and in the aforementioned verisions. Are you using one *.pof from the application note, or are you compiling and building your own?

--- Quote End ---

I using default and haven't tried building my own

Alright:

I downloaded the application note and tried out all 4 of the .sof files provided for Arria GX...

x1 Legacy: no device in lspci

x4 Legacy: device shows, driver hangs with EPLAST=44 on the 1st memory transfer.

x1 ST: no device in lspci

x4 ST: no device in lspci

So I think I have more or less duplicated your environment with the x4 legacy example. The quartus settings files for these projects indicate that the application note was built using Quartus 8.0 sp1. You can use any version to program the device, but in less you recompile the project AND the megafunction, the design remains 8.0, which is error-prone according to the linux driver authors. The best advice I can give you is:

1. Try the 7.2 High Performance design that is on the CD that comes with the AGX development board. For me the linux driver's memory test passes with that example.

2. Instantiate a PCIe megafunction in quartus, open the project it builds in <mf_name>_examples/chainingdma, set the device and pins, and use that design. This is the route I take, but it only works using versions 7.2 or 9.0.

--- Quote Start ---

Alright:

I downloaded the application note and tried out all 4 of the .sof files provided for Arria GX...

x1 Legacy: no device in lspci

x4 Legacy: device shows, driver hangs with EPLAST=44 on the 1st memory transfer.

x1 ST: no device in lspci

x4 ST: no device in lspci

So I think I have more or less duplicated your environment with the x4 legacy example. The quartus settings files for these projects indicate that the application note was built using Quartus 8.0 sp1. You can use any version to program the device, but in less you recompile the project AND the megafunction, the design remains 8.0, which is error-prone according to the linux driver authors. The best advice I can give you is:

1. Try the 7.2 High Performance design that is on the CD that comes with the AGX development board. For me the linux driver's memory test passes with that example.

2. Instantiate a PCIe megafunction in quartus, open the project it builds in <mf_name>_examples/chainingdma, set the device and pins, and use that design. This is the route I take, but it only works using versions 7.2 or 9.0.

--- Quote End ---

THanks Rob. I will get back to you with updates

Rob, Is following line is redundant in driver ?

irq_line = dev->irq;

I mean why is this needed when pci_read_config_byte is called to get the irq_line

--- Quote Start ---

Yes, it looks like irq_line is assigned two different values, first from reading the appropriate configuration byte, then with the line above. Probably left over from testing the driver. I would say that "irq_line = dev->irq;" is the more important line in the driver, since the variable discards what it receives from the configuration byte and uses this value. So you could delete the pci_read_config_byte line and the printk statement that prints its value. But it provides a nice little bit of debugging info and it works just fine.

--- Quote End ---

Well it didn't make any difference though. In mean time, i have tested it in 32-bit Linux. Log is as below (it is with Quartus 8.0/8.1/9.0):

=======================================

altpciechdma init(), built at Mar 20 2009 22:14:41

probe(dev = 0xf79c5400, pciid = 0xf8b77c20)

Testing probe function

probe() ape = 0xf63196c0

sizeof(struct ape_chdma_table) = 4096.

table_virt = 0xfffffffff3e57000, table_bus = 0x 33e57000.

ACPI: PCI Interrupt 0000:06:00.0[A] -> GSI 16 (level, low) -> IRQ 16

PCI: Setting latency timer of device 0000:06:00.0 to 64

Enabled MSI interrupting.

RevisionID is supported by this driver 0x1

VendorID is supported by this driver 0x1172

DeviceID is supported by this driver 0xe001

Could not set 64-bit DMA mask.

Using a 32-bit DMA mask.

IRQ pin# 1 (0=none, 1=INTA#...4=INTD#).

IRQ line# 11.

IRQ line post assigning# 216.

Succesfully requested IRQ# 216 with dev_id 0xf63196c0

BAR0 0xf0000000-0xf0ffffff flags 0x0000120c

BAR2 0xb7200000-0xb72003ff flags 0x00000200

BAR[0] mapped at 0xf8b60000 with length 32768(/16777216).

BAR[2] mapped at 0xf8b58000 with length 256(/1024).

bar_tests(), PAGE_SIZE = 0x1000

write_header = 0xf8b58000.

read_header = 0xf8b58010.

&write_header->w3 (intiates DMA) = 0xf8b5800c

&read_header->w3 (initiates DMA) = 0xf8b5801c

ape->table_virt = 0xf3e57000.

Allocated cache-coherent DMA buffer (virtual address = 0xfffffffff3e70000, bus address = 0x0000000033e70000).

writing read header 0x000400ff to 0xf8b58010

writing higher bus address 0x00000000 to 0xf8b58014

writing lower bus address 0x33e57000 to 0xf8b58018

Flush posted writes

Start DMA read

writing (to w3 initiate DMA) 0x000000fe to 0xf8b5801c

EPLAST(w3 register ORed) = 64222

POLL FOR READ:

EPLAST = 64222, n = 254

EPLAST = 9, n = 254

EPLAST = 20, n = 254

EPLAST = 31, n = 254

EPLAST = 42, n = 254

EPLAST = 52, n = 254

EPLAST = 63, n = 254

EPLAST = 73, n = 254

EPLAST = 84, n = 254

EPLAST = 94, n = 254

EPLAST = 105, n = 254

EPLAST = 116, n = 254

EPLAST = 127, n = 254

EPLAST = 137, n = 254

EPLAST = 148, n = 254

EPLAST = 159, n = 254

EPLAST = 169, n = 254

EPLAST = 180, n = 254

EPLAST = 190, n = 254

EPLAST = 201, n = 254

EPLAST = 211, n = 254

EPLAST = 222, n = 254

EPLAST = 232, n = 254

EPLAST = 243, n = 254

EPLAST = 254, n = 254

DONE# IRQs during transfer: 2

Start DMA write

POLL FOR WRITE:

EPLAST = 64222, n = 254

EPLAST = 9, n = 254

EPLAST = 21, n = 254

EPLAST = 32, n = 254

EPLAST = 43, n = 254

EPLAST = 55, n = 254

EPLAST = 66, n = 254

EPLAST = 77, n = 254

EPLAST = 88, n = 254

EPLAST = 100, n = 254

EPLAST = 111, n = 254

EPLAST = 122, n = 254

EPLAST = 133, n = 254

EPLAST = 145, n = 254

EPLAST = 156, n = 254

EPLAST = 167, n = 254

EPLAST = 178, n = 254

EPLAST = 189, n = 254

EPLAST = 201, n = 254

EPLAST = 212, n = 254

EPLAST = 223, n = 254

EPLAST = 234, n = 254

EPLAST = 245, n = 254

EPLAST = 254, n = 254

DONE# IRQs during transfer: 1

[f3e70000] = 0xf3e70000 != [f3e72000] = 0x00000000 ?!

[f3e70004] = 0xf3e70004 != [f3e72004] = 0x00000000 ?!

[f3e70008] = 0xf3e70008 != [f3e72008] = 0x00000000 ?!

[f3e7000c] = 0xf3e7000c != [f3e7200c] = 0x00000000 ?!

[f3e70010] = 0xf3e70010 != [f3e72010] = 0x00000000 ?!

[f3e70014] = 0xf3e70014 != [f3e72014] = 0x00000000 ?!

[f3e70018] = 0xf3e70018 != [f3e72018] = 0x00000000 ?!

[f3e7001c] = 0xf3e7001c != [f3e7201c] = 0x00000000 ?!

[f3e70020] = 0xf3e70020 != [f3e72020] = 0x00000000 ?!

---more errors follow! not printed---

DMA loop back test FAILED.

bar_tests() end, result -1

probe() successful.

altpciechdma 0