Re: FSL DMA engine transfer to PCI memory
Hi Ira, On 01/25/2011 06:29 PM, Ira W. Snyder wrote: On Tue, Jan 25, 2011 at 04:32:02PM +0200, Felix Radensky wrote: Hi Ira, On 01/25/2011 02:18 AM, Ira W. Snyder wrote: On Tue, Jan 25, 2011 at 01:39:39AM +0200, Felix Radensky wrote: Hi Ira, Scott On 01/25/2011 12:26 AM, Ira W. Snyder wrote: On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort-TAbort-MAbort-SERR-PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s64ns, L11us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param =cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination address is wrong. The device_prep_dma_memcpy() routine works in physical addresses only (dma_addr_t type). Your source address looks fine: you're using the result of dma_map_single(), which returns a physical address. Your destination address should be something that comes from struct pci_dev.resource[x].start
Re: FSL DMA engine transfer to PCI memory
On Thu, Jan 27, 2011 at 10:32:19AM +0200, Felix Radensky wrote: Hi Ira, On 01/25/2011 06:29 PM, Ira W. Snyder wrote: On Tue, Jan 25, 2011 at 04:32:02PM +0200, Felix Radensky wrote: Hi Ira, On 01/25/2011 02:18 AM, Ira W. Snyder wrote: On Tue, Jan 25, 2011 at 01:39:39AM +0200, Felix Radensky wrote: Hi Ira, Scott On 01/25/2011 12:26 AM, Ira W. Snyder wrote: On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort-TAbort-MAbort-SERR-PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s64ns, L11us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param =cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination
RE: FSL DMA engine transfer to PCI memory
What was the ppc you used? The 8315E PowerQUIICC II On 85xx/QorIQ-family chips such as P2020, there is no DMA controller inside the PCIe controller itself (or are you talking about bus mastering by the PCIe device[1]? interface is a bit ambiguous), though it was considered part of the PCI controller on 82xx. The DMA engine and PCIe are both on OCeaN, so the traffic does not need to pass through the e500 Coherency Module. My understanding -- for what it's worth, coming from a software person :-) -- is that you should be able to get large transfer chunks using the DMA engine. It might be possible - but the ppc's pcie would need to know the length of the dma (or at least be told that there was more data to arrive) before even starting the pcie transfer. I used 128 bytes per pcie transfer (which the altera slave can handle) but that is longer than you want a burst on the internal (CSB in my case) bus on the ppc. It is also longer than a cache line - so the dma engine's memory reads might induce a cache flush. I suggest getting things working, and then seeing whether the performance is acceptable. The only reason for using dma (instead of pio) is to get long pcie transfers - otherwise it isn't really worth the effort. Transfers are unlikely to take long enough to make it worth taking an interrupt at the end of the dma. My device driver implements read() and write() (and poll() to wait for interrupts). So I do overlap the copy_to/from_user with the next dma. The generic dma controller can't even generate 64bit cycles into the ppc's PCIe engine. Could you elaborate? The pcie is (apparantly) a 64bit interface, to a single 32bit transfer is actually a 64bit one with only 4 byte enables driven. I couldn't see anything that would allow a CSB master to generate two 32bit cycles (since it is a 32bit bus) that the pcie hardware could convert into a single 64bit pcie transfer. The fpga target is likely to have 32bit targets (it could have 64 bit ones, but if you've instantiated a NiosII cpu it wont!) so you get a bus width adapter (which carefully does the cycle with no byte enables driven) as well as the clock crossing bridge. These both make the slave even slower than it would otherwise be! IIRC We managed to get 2us for a read and 500ns for a write cycle. The per byte costs are relatively small in comparison. David ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
Re: FSL DMA engine transfer to PCI memory
On Wed, 26 Jan 2011 10:18:01 + David Laight david.lai...@aculab.com wrote: What was the ppc you used? The 8315E PowerQUIICC II Ah. The interconnect between the DMA engine and PCIe is different on 83xx. The DMA engine and PCIe are both on OCeaN, so the traffic does not need to pass through the e500 Coherency Module. My understanding -- for what it's worth, coming from a software person :-) -- is that you should be able to get large transfer chunks using the DMA engine. It might be possible - but the ppc's pcie would need to know the length of the dma (or at least be told that there was more data to arrive) before even starting the pcie transfer. On 85xx/QorIQ, I believe the connection between the DMA engine and the PCIe controller allows the data to arrive in suitably large chunks. I suggest getting things working, and then seeing whether the performance is acceptable. The only reason for using dma (instead of pio) is to get long pcie transfers - otherwise it isn't really worth the effort. Transfers are unlikely to take long enough to make it worth taking an interrupt at the end of the dma. But in the absence of specific knowledge about this specific chip, implementing it and testing is a good way of determining whether you get those large PCIe transactions on this particular hardware. And even if the transfers aren't particularly fast, if the total transfer size (not the size of the chunks that go on the bus) is large enough, it could be worth freeing up the core to do something else. It could also avoid running the data through the core's caches, or be a transfer from one PCIe device to another, etc. Don't be too quick to say don't bother. :-) The generic dma controller can't even generate 64bit cycles into the ppc's PCIe engine. Could you elaborate? The pcie is (apparantly) a 64bit interface, to a single 32bit transfer is actually a 64bit one with only 4 byte enables driven. My understanding is that PCIe is an aggregation of one or more serial links, over which packets are sent. I'm not sure to what extent it makes sense to call it a 64-bit interface, other than addressing. I couldn't see anything that would allow a CSB master to generate two 32bit cycles (since it is a 32bit bus) that the pcie hardware could convert into a single 64bit pcie transfer. Again, that's an 83xx thing, 85xx/QorIQ is different. Though from the 8315 manual it looks like the CSB can do 64-bit data (but not addresses). -Scott ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
RE: FSL DMA engine transfer to PCI memory
I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. You'll need to use the dma engine that is part of the PCIe interface in order to get large PCIe transfers. I think everything else will still generate single 32bit PCIe transfers - which are (if your measurements match mine) exceptionally lethargic - the ISA bus is faster! That does work provided you remember to give the dma controller physical addresses and byteswap absolutely everything. (Oh, and I didn't get single word transfers to work - they locked the dma controller - not a problem since they are faster by PIO.) Note that the PPC Linux (Linux in general??) doesn't have a 'virtual to physical' function that works for all addresses, you'll need to remember the physical address of the PCIe slave and use malloc'ed memory for the descriptors (on which virt_to_phys() actually works). I don't think there is a standard device driver for the PCIe dma, I couldn't even find any header files that were vaugely relevent except in the uboot sources. I certainly wrote some code that just assumes it is on the right hardware! These are the relevant bits of code Global initialisation: /* Enable the read/write dma controllers */ csb_ctrl = in_le32(pex-pex_csb_ctrl); csb_ctrl |= PEX_CSB_CTRL_WDMAE | PEX_CSB_CTRL_RDMAE; out_le32(pex-pex_csb_ctrl, csb_ctrl); /* We don't rely on the dma polling the descriptor, I have NFI * whether the default of 0 means 'never poll' or 'poll very quickly'. * Set a large slow value for sanity. */ out_le32(pex-pex_dms_dstmr, ~0u); Transfer setup: /* We only support aligned writes - caller must verify */ dma_ctrl = PDMAD_CTRL_VALID; dma_ctrl |= PDMAD_CTRL_SNOOP_CSB; dma_ctrl |= PDMAD_CTRL_1ST_BYTES | PDMAD_CTRL_LAST_BYTES; dma_ctrl |= PDMAD_CTRL_NEXT_VALID; dma_ctrl |= len (PDMAD_CTRL_LEN_SHIFT - 2); /* Fill in DMA descriptor */ st_le32(desc-pdmad_ctrl, dma_ctrl); /* We MUST clear the status - otherwise the xfer will be skipped */ st_le32(desc-pdmad_stat, 0); st_le32(desc-pdmad_src_address, src_phys); st_le32(desc-pdmad_dst_address, dst_phys); st_le32(desc-pdmad_next_desc, 0); /* Clear old status */ st_le32(pex_dma-pex_dma_stat, in_le32(pex_dma-pex_dma_stat)); /* Give descriptor address to dma engine */ st_le32(pex_dma-pex_dma_addr, virt_to_phys(desc)); /* Wait for all above memory cycles, then start xfer */ iosync(); st_le32(pex_dma-pex_dma_ctrl, PEX_DMA_CTRL_START | PEX_DMA_CTRL_SNOOP); Poll for completion: /* Wait for transfer to complete/fail */ do { desc_stat = ld_le32(desc-pdmad_stat); } while (!(desc_stat PDMAD_STAT_DONE)); status = ld_le32(pex_dma-pex_dma_stat); if (status == (PEX_DMA_STAT_DSCPL | PEX_DMA_STAT_CHCPL) desc_stat == PDMAD_STAT_DONE) /* Transfer ok */ return 0; /* Transfer failed */ Oh, since I couldn't find it in the documentation, the first word of the dma descriptor is 'ctrl' and the last 'next_desc'. David ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
Re: FSL DMA engine transfer to PCI memory
Hi Ira, On 01/25/2011 02:18 AM, Ira W. Snyder wrote: On Tue, Jan 25, 2011 at 01:39:39AM +0200, Felix Radensky wrote: Hi Ira, Scott On 01/25/2011 12:26 AM, Ira W. Snyder wrote: On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort-TAbort-MAbort-SERR-PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s64ns, L11us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param =cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination address is wrong. The device_prep_dma_memcpy() routine works in physical addresses only (dma_addr_t type). Your source address looks fine: you're using the result of dma_map_single(), which returns a physical address. Your destination address should be something that comes from struct pci_dev.resource[x].start + offset if necessary. In your lspci output above, that will be 0xc000. Another possible problem: AFAIK you must use the _ONSTACK() variants from include/linux/completion.h for struct
FW: FSL DMA engine transfer to PCI memory
memcpy_toio() works fine, the data is written correctly. After DMA, the correct data appears at offsets 0xC, 0x1C, 0x2C, etc. of the destination buffer. I have 12 bytes of junk, 4 bytes of correct data, then again 12 bytes of junk and so on. Does your Avalon MM slave decode the 4 byte enables? The slave always sees 2 Avalon MM writes, one of which will have no byte enables asserted. (Actually it is a 64bit Avalon transfer - and a bus width adapter gets inserted for you.) Internal M9K memory blocks get this right... David ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
Re: FSL DMA engine transfer to PCI memory
On Tue, Jan 25, 2011 at 04:32:02PM +0200, Felix Radensky wrote: Hi Ira, On 01/25/2011 02:18 AM, Ira W. Snyder wrote: On Tue, Jan 25, 2011 at 01:39:39AM +0200, Felix Radensky wrote: Hi Ira, Scott On 01/25/2011 12:26 AM, Ira W. Snyder wrote: On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort-TAbort-MAbort-SERR-PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s64ns, L11us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param =cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination address is wrong. The device_prep_dma_memcpy() routine works in physical addresses only (dma_addr_t type). Your source address looks fine: you're using the result of dma_map_single(), which returns a
RE: FSL DMA engine transfer to PCI memory
custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. This sounds like your FPGA doesn't handle burst mode accesses correctly. A logic analyzer will help you prove it. He is doing PCIe, not PCI. A PCIe transfers is an HDLC packet pair, one containing the request, the other the response. In order to get any significant throughput the hdlc packet(s) have to contain all the data (eg 128 bytes). On the ppc we used that means you have to use the dma controller inside the PCIe interface block. The generic dma controller can't even generate 64bit cycles into the ppc's PCIe engine. David ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
Re: FSL DMA engine transfer to PCI memory
On Tue, 25 Jan 2011 16:34:49 + David Laight david.lai...@aculab.com wrote: custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. This sounds like your FPGA doesn't handle burst mode accesses correctly. A logic analyzer will help you prove it. He is doing PCIe, not PCI. A PCIe transfers is an HDLC packet pair, one containing the request, the other the response. In order to get any significant throughput the hdlc packet(s) have to contain all the data (eg 128 bytes). On the ppc we used that means you have to use the dma controller inside the PCIe interface block. What was the ppc you used? On 85xx/QorIQ-family chips such as P2020, there is no DMA controller inside the PCIe controller itself (or are you talking about bus mastering by the PCIe device[1]? interface is a bit ambiguous), though it was considered part of the PCI controller on 82xx. The DMA engine and PCIe are both on OCeaN, so the traffic does not need to pass through the e500 Coherency Module. My understanding -- for what it's worth, coming from a software person :-) -- is that you should be able to get large transfer chunks using the DMA engine. I suggest getting things working, and then seeing whether the performance is acceptable. The generic dma controller can't even generate 64bit cycles into the ppc's PCIe engine. Could you elaborate? -Scott [1] To the original poster, is there any reason you're not doing bus mastering from the PCIe device, assuming you control the content of the FPGA? ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
FSL DMA engine transfer to PCI memory
Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort- TAbort- MAbort- SERR- PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s 64ns, L1 1us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param = cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Felix. ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
Re: FSL DMA engine transfer to PCI memory
On Mon, 24 Jan 2011 23:47:22 +0200 Felix Radensky fe...@embedded-sol.com wrote: static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; Why are you casting a virtual address to dma_addr_t? The destination address is PCI memory address returned by pci_ioremap_bar(). You need the physical address. -Scott ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
Re: FSL DMA engine transfer to PCI memory
On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort- TAbort- MAbort- SERR- PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s 64ns, L1 1us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param = cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination address is wrong. The device_prep_dma_memcpy() routine works in physical addresses only (dma_addr_t type). Your source address looks fine: you're using the result of dma_map_single(), which returns a physical address. Your destination address should be something that comes from struct pci_dev.resource[x].start + offset if necessary. In your lspci output above, that will be 0xc000. Another possible problem: AFAIK you must use the _ONSTACK() variants from include/linux/completion.h for struct completion which are on the stack. Hope it helps, Ira ___ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org
Re: FSL DMA engine transfer to PCI memory
Hi Ira, Scott On 01/25/2011 12:26 AM, Ira W. Snyder wrote: On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort-TAbort-MAbort-SERR-PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s64ns, L11us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param =cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination address is wrong. The device_prep_dma_memcpy() routine works in physical addresses only (dma_addr_t type). Your source address looks fine: you're using the result of dma_map_single(), which returns a physical address. Your destination address should be something that comes from struct pci_dev.resource[x].start + offset if necessary. In your lspci output above, that will be 0xc000. Another possible problem: AFAIK you must use the _ONSTACK() variants from include/linux/completion.h for struct completion which are on the stack. Hope it helps, Ira Thanks for your help. I'm now passing the result of pci_resource_start(pdev, 0) as destination address, and destination buffer changes
Re: FSL DMA engine transfer to PCI memory
On Tue, Jan 25, 2011 at 01:39:39AM +0200, Felix Radensky wrote: Hi Ira, Scott On 01/25/2011 12:26 AM, Ira W. Snyder wrote: On Mon, Jan 24, 2011 at 11:47:22PM +0200, Felix Radensky wrote: Hi, I'm trying to use FSL DMA engine to perform DMA transfer from memory buffer obtained by kmalloc() to PCI memory. This is on custom board based on P2020 running linux-2.6.35. The PCI device is Altera FPGA, connected directly to SoC PCI-E controller. 01:00.0 Unassigned class [ff00]: Altera Corporation Unknown device 0004 (rev 01) Subsystem: Altera Corporation Unknown device 0004 Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast TAbort-TAbort-MAbort-SERR-PERR- Interrupt: pin A routed to IRQ 16 Region 0: Memory at c000 (32-bit, non-prefetchable) [size=128K] Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable- Address: Data: Capabilities: [78] Power Management version 3 Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-) Status: D0 PME-Enable- DSel=0 DScale=0 PME- Capabilities: [80] Express Endpoint IRQ 0 Device: Supported: MaxPayload 256 bytes, PhantFunc 0, ExtTag- Device: Latency L0s64ns, L11us Device: AtnBtn- AtnInd- PwrInd- Device: Errors: Correctable- Non-Fatal- Fatal- Unsupported- Device: RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+ Device: MaxPayload 128 bytes, MaxReadReq 512 bytes Link: Supported Speed 2.5Gb/s, Width x1, ASPM L0s, Port 1 Link: Latency L0s unlimited, L1 unlimited Link: ASPM Disabled RCB 64 bytes CommClk- ExtSynch- Link: Speed 2.5Gb/s, Width x1 Capabilities: [100] Virtual Channel I can successfully writel() to PCI memory via address obtained from pci_ioremap_bar(). Here's my DMA transfer routine static int dma_transfer(struct dma_chan *chan, void *dst, void *src, size_t len) { int rc = 0; dma_addr_t dma_src; dma_addr_t dma_dst; dma_cookie_t cookie; struct completion cmp; enum dma_status status; enum dma_ctrl_flags flags = 0; struct dma_device *dev = chan-device; struct dma_async_tx_descriptor *tx = NULL; unsigned long tmo = msecs_to_jiffies(FPGA_DMA_TIMEOUT_MS); dma_src = dma_map_single(dev-dev, src, len, DMA_TO_DEVICE); if (dma_mapping_error(dev-dev, dma_src)) { printk(KERN_ERR Failed to map src for DMA\n); return -EIO; } dma_dst = (dma_addr_t)dst; flags = DMA_CTRL_ACK | DMA_COMPL_SRC_UNMAP_SINGLE | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT; tx = dev-device_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags); if (!tx) { printk(KERN_ERR %s: Failed to prepare DMA transfer\n, __FUNCTION__); dma_unmap_single(dev-dev, dma_src, len, DMA_TO_DEVICE); return -ENOMEM; } init_completion(cmp); tx-callback = dma_callback; tx-callback_param =cmp; cookie = tx-tx_submit(tx); if (dma_submit_error(cookie)) { printk(KERN_ERR %s: Failed to start DMA transfer\n, __FUNCTION__); return -ENOMEM; } dma_async_issue_pending(chan); tmo = wait_for_completion_timeout(cmp, tmo); status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); if (tmo == 0) { printk(KERN_ERR %s: Transfer timed out\n, __FUNCTION__); rc = -ETIMEDOUT; } else if (status != DMA_SUCCESS) { printk(KERN_ERR %s: Transfer failed: status is %s\n, __FUNCTION__, status == DMA_ERROR ? error : in progress); dev-device_control(chan, DMA_TERMINATE_ALL, 0); rc = -EIO; } return rc; } The destination address is PCI memory address returned by pci_ioremap_bar(). The transfer silently fails, destination buffer doesn't change contents, but no error condition is reported. What am I doing wrong ? Thanks a lot in advance. Your destination address is wrong. The device_prep_dma_memcpy() routine works in physical addresses only (dma_addr_t type). Your source address looks fine: you're using the result of dma_map_single(), which returns a physical address. Your destination address should be something that comes from struct pci_dev.resource[x].start + offset if necessary. In your lspci output above, that will be 0xc000.