Hi, I am a newbie to FPGA design and implementation. While installing the PCI express board on LINUX machine, I don't see Demo application driver for DMA read and write for LINUX (Manual talks about windows installation). I want to generate DMA read and write in Linux box. Does anyone have pointer to a sample application, driver or sof file for the same. Let me know if I am wrong in understanding. Thanks in advance. -Vineet

Hello Vineet, Another forum member and I wrote a Linux Device Driver for the scatter gather design that is generated with the Altera PCI Express core wizard. It is now in the kernel tree, in the drivers/staging/ area: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=tree;f=drivers/staging/altpciechdma;h=ef7dd8fe0e082860742da72655892f12fb088114;hb=f3b8436ad9a8ad36b3c9fa1fe030c7f38e5d3d0b

--- Quote Start --- Hello Vineet, Another forum member and I wrote a Linux Device Driver for the scatter gather design that is generated with the Altera PCI Express core wizard. It is now in the kernel tree, in the drivers/staging/ area: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=tree;f=drivers/staging/altpciechdma;h=ef7dd8fe0e082860742da72655892f12fb088114;hb=f3b8436ad9a8ad36b3c9fa1fe030c7f38e5d3d0b --- Quote End --- Do you have a README or test application to test this driver ? Also, it would be good if you can exactly specify which kernel version it has been tested on. Thanks

According to the .c file, the kernel has been tested by the authors on 2.6.20 and 2.6.24. I've compiled it on 2.6.27 and 2.6.28, and now on 2.6.18 (Red Hat 5), but I had to comment out the cdev.h# include. The authors also state that the file is primarily for testing designs. Linux modules are kind of different from normal c programming and compiling, but good references are LDD3 <http://lwn.net/Kernel/LDD3/> and <http://dirac.org/linux/writing/lkmpg/2.6/lkmpg-2.6.0.html>. The Authors frequently refer to LDD3, but I found the 2nd site's compiling instructions to be more useful. So you compile the module using make, insert the module using insmod, and then if the driver and board work then you can view the results by issuing the dmesg command. The driver includes a test function that is ran automatically at insmod, when the driver "probes" the device. If it doesn't work, make sure the BAR array defined in the C file matches the BARs defined in your megafunction. In my case, I'm using an Arria GX Design Example that does not use BAR 4, which is non zero in the unmodified driver code. Easy fix though.

Thanks for the reply. I was able to insmod and rmmod the module. I was trying in latest kernel 2.6.29. I will check with 2.6.24. Thanks for pointers. I also refer to LDD3. Update you if I succeed in running the driver. Thanks again

Altera PCI Express Development Kit: Linux DMA Driver

64 Replies

Altera_Forum
Honored Contributor
16 years ago
As far as I know, the altpciechdma.c is not a char driver,or it is?
I wonder,how could I access this driver from yhe userspace?
Somebody has experience using this driver in Linux?
I cannot see this driver in $/dev/ .
Altera_Forum
Honored Contributor
16 years ago
--- Quote Start ---
As far as I know, the altpciechdma.c is not a char driver,or it is?
I wonder,how could I access this driver from yhe userspace?
Somebody has experience using this driver in Linux?
I cannot see this driver in $/dev/ .
--- Quote End ---

Up to now, I used the driver only for debugging my design. Perfomed some read/write accesses and allocated some memory for DMA writing from FPGA.

I just removed the selftest function and replaced it by my own code (plus some minor changes in the rest of the driver).

For this purpose you just have to compile the modules first (take a look at LDD3 oder at http://www.dirac.org/linux/writing/lkmpg/2.6/lkmpg-2.6.0.html which was quite well for me since I had some version problems). Then you have load the modules into the kernel (insmod), thats it!
Altera_Forum
Honored Contributor
16 years ago
I've also had problems with the compiling of this driver.
"Make" command gives a lot of errors,so I builded this driver using KDelevop.
I loaded him into the kernel using modprobe command.
"lsmod" displays this driver,but he is absent in$/dev/....
I suppose,I have to write the register part by myself.(register_chrdev(200,"altera",&fops.....)...).
By the way this driver is still under development.
Under selftest function,you mean the "probe" function.
Can you post the changes?
Altera_Forum
Honored Contributor
16 years ago
--- Quote Start ---
I've also had problems with the compiling of this driver.
"Make" command gives a lot of errors,so I builded this driver using KDelevop.
I loaded him into the kernel using modprobe command.
"lsmod" displays this driver,but he is absent in$/dev/....
I suppose,I have to write the register part by myself.(register_chrdev(200,"altera",&fops.....)...).
By the way this driver is still under development.
Under selftest function,you mean the "probe" function.
Can you post the changes?
--- Quote End ---

At the end of probe() there is a function call for the dma selftest. If you loaded the module you can use dmesg to view the printk message output.
Altera_Forum
Honored Contributor
16 years ago
Thank's for help.
Altera_Forum
Honored Contributor
16 years ago
Hi

I am using x4 Arria GX Dev kit on Intel Core2Duo x86 64.
My probe function gives me :
......
Enabled MSI interrupting.
Using a 64-bit DMA mask.
IRQ pin# 0 (0=none, 1=INTA#...4=INTD#).
IRQ line# 0.
Could not request IRQ# 251, error -38
altpciechdma: probe of 0000:01:00.0 failed with error -38
....
How do I have to specify the BAR's?
static const unsigned long bar_min_len[APE_BAR_NUM] =
{ 32768, 0, 256, 0, 32768, 0 };

I dont understand,if the WinDriver doesnt let me to take the interrupt line,
which I have already uninstalled,
or I've incorrectly specified the BAR's?
Altera_Forum
Honored Contributor
16 years ago
I followed this thread to test the DMA driver. It worked OK with 32bit Fedora 11. When we moved the card to CentOS 4.3 64bit linux, the dma_test failed. I can't figure out why, please help!

Here is the dmesg:

altpciechdma init(), built at Oct 13 2009 13:54:17
probe(dev = 0xffff81015fb33800, pciid = 0xffffffff88661b80)
probe() ape = 0xffff81015c9ac7c0
sizeof(struct ape_chdma_table) = 4096.
table_virt = 0xffff810140e35000, table_bus = 0x 140e35000.
PCI: Enabling device 0000:06:00.0 (0000 -> 0002)
PCI: Setting latency timer of device 0000:06:00.0 to 64
Enabled MSI interrupting.
Using a 64-bit DMA mask.
IRQ pin# 0 (0=none, 1=INTA#...4=INTD#).
IRQ line# 0.
Succesfully requested IRQ# 194 with dev_id 0xffff81015c9ac7c0
BAR0 0xc0000000-0xcfffffff flags 0x00000200
BAR1 0xdcf80000-0xdcfbffff flags 0x00000200
BAR2 0xdcfc0000-0xdcffffff flags 0x00000200
BAR[0] mapped at 0xffffc20010370000 with length 32768(/268435456).
BAR[2] mapped at 0xffffc2000002a000 with length 256(/262144).
bar_tests(), PAGE_SIZE = 0x1000
write_header = 0xffffc2000002a000.
read_header = 0xffffc2000002a010.
&write_header->w3 = 0xffffc2000002a00c
&read_header->w3 = 0xffffc2000002a01c
ape->table_virt = 0xffff810140e35000.
Allocated cache-coherent DMA buffer (virtual address = 0xffff810140784000, bus address = 0x0000000140784000).
writing 0x000400ff to 0xffffc2000002a010
writing 0x00000001 to 0xffffc2000002a014
writing 0x40e35000 to 0xffffc2000002a018
Flush posted writes
Start DMA read
writing 0x000000fe to 0xffffc2000002a01c
EPLAST = 64222
POLL FOR READ:
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254

..............

Start DMA write
POLL FOR WRITE:
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254
EPLAST = 64222, n = 254

................

[ffff810140784000] = 0x40784000 != [ffff810140786000] = 0x40786000 ?!
[ffff810140784004] = 0x40784004 != [ffff810140786004] = 0x40786004 ?!
[ffff810140784008] = 0x40784008 != [ffff810140786008] = 0x40786008 ?!
[ffff81014078400c] = 0x4078400c != [ffff81014078600c] = 0x4078600c ?!
[ffff810140784010] = 0x40784010 != [ffff810140786010] = 0x40786010 ?!
[ffff810140784014] = 0x40784014 != [ffff810140786014] = 0x40786014 ?!
[ffff810140784018] = 0x40784018 != [ffff810140786018] = 0x40786018 ?!
[ffff81014078401c] = 0x4078401c != [ffff81014078601c] = 0x4078601c ?!
[ffff810140784020] = 0x40784020 != [ffff810140786020] = 0x40786020 ?!
---more errors follow! not printed---
DMA loop back test FAILED.
bar_tests() end, result -1
probe() successful.
Altera_Forum
Honored Contributor
16 years ago
Well,I decided to give the altpcichdma driver manually the interrupt line.
From the /proc/interrupts,give the driver a free INT.In mine case INT#7.
Write dev->irq_line=7.
irq_line=7.
rc=request_int(7,...);
Dont let the driver to automatically detect the INT.
In that case I've passed the DMA test successfuly.

BAR[0] mapped at 0xffffc20010370000 with length 32768(/268435456).
BAR[2] mapped at 0xffffc2000002a000 with length 256(/262144).

Specify the bar struct={268435456,0,262144};
Altera_Forum
Honored Contributor
16 years ago
Is there any way to watch the DMA transfer ArriaGX<->PCIe,or even better to save it in a file?
Maybe there are specified progs?
Altera_Forum
Honored Contributor
16 years ago
In altpciechdma, the EP address is set to 4096. Can anyone help explaining why 4096? and if 46bit Linux will have different address?

Thanks

Forum Discussion

Altera PCI Express Development Kit: Linux DMA Driver

64 Replies

Recent Discussions

DK-DEV-AGI027RBES missing device selection

SyncE clock as the reference for DDR, HBM, and FPGA control interfaces in the Agilex M-Series design

DK-DEV-AGI027RES Install Package

Cyclone V GT Dev Kit - new units fail to boot from FPP

Why 390.625MHz clock to F-Tile Reference fequency