Re: [PATCH v1 7/7] fpga: dfl: pci: Add generic OFS PCI PID

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On Tue, 22 Feb 2022, Tom Rix wrote:


On 2/21/22 7:11 PM, Zhang, Tianfei wrote:

-----Original Message-----
From: Tom Rix <trix@xxxxxxxxxx>
Sent: Tuesday, February 22, 2022 2:10 AM
To: matthew.gerlach@xxxxxxxxxxxxxxx
Cc: Zhang, Tianfei <tianfei.zhang@xxxxxxxxx>; Wu, Hao <hao.wu@xxxxxxxxx>;
mdf@xxxxxxxxxx; Xu, Yilun <yilun.xu@xxxxxxxxx>; linux-fpga@xxxxxxxxxxxxxxx;
linux-doc@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; corbet@xxxxxxx
Subject: Re: [PATCH v1 7/7] fpga: dfl: pci: Add generic OFS PCI PID


On 2/21/22 9:50 AM, matthew.gerlach@xxxxxxxxxxxxxxx wrote:

On Fri, 18 Feb 2022, Tom Rix wrote:

On 2/18/22 1:03 AM, Zhang, Tianfei wrote:
-----Original Message-----
From: Tom Rix <trix@xxxxxxxxxx>
Sent: Wednesday, February 16, 2022 12:16 AM
To: Zhang, Tianfei <tianfei.zhang@xxxxxxxxx>; Wu, Hao
<hao.wu@xxxxxxxxx>; mdf@xxxxxxxxxx; Xu, Yilun <yilun.xu@xxxxxxxxx>;
linux-fpga@xxxxxxxxxxxxxxx; linux-doc@xxxxxxxxxxxxxxx;
linux-kernel@xxxxxxxxxxxxxxx
Cc: corbet@xxxxxxx; Matthew Gerlach
<matthew.gerlach@xxxxxxxxxxxxxxx>
Subject: Re: [PATCH v1 7/7] fpga: dfl: pci: Add generic OFS PCI PID


On 2/14/22 3:26 AM, Tianfei zhang wrote:
From: Matthew Gerlach <matthew.gerlach@xxxxxxxxxxxxxxx>

Add the PCI product id for an Open FPGA Stack PCI card.
Is there a URL to the card ?
This PCIe Device IDs have registered by Intel.
A URL is useful to introduce the board, Is there one ?
Signed-off-by: Matthew Gerlach <matthew.gerlach@xxxxxxxxxxxxxxx>
Signed-off-by: Tianfei Zhang <tianfei.zhang@xxxxxxxxx>
---
    drivers/fpga/dfl-pci.c | 4 ++++
    1 file changed, 4 insertions(+)

diff --git a/drivers/fpga/dfl-pci.c b/drivers/fpga/dfl-pci.c index
83b604d6dbe6..cb2fbf3eb918 100644
--- a/drivers/fpga/dfl-pci.c
+++ b/drivers/fpga/dfl-pci.c
@@ -76,12 +76,14 @@ static void cci_pci_free_irq(struct pci_dev
*pcidev)
    #define PCIE_DEVICE_ID_INTEL_PAC_D5005        0x0B2B
    #define PCIE_DEVICE_ID_SILICOM_PAC_N5010    0x1000
    #define PCIE_DEVICE_ID_SILICOM_PAC_N5011    0x1001
+#define PCIE_DEVICE_ID_INTEL_OFS        0xbcce
INTEL_OFS is a generic name, pci id's map to specific cards

Is there a more specific name for this card ?
I think using INTEL_OFS is better, because INTEL_OFS is the Generic
development platform can support multiple cards which using OFS
specification, like Intel PAC N6000 card.
I would prefer something like PCIE_DEVICE_ID_INTEL_PAC_N6000 because
it follows an existing pattern.  Make it easy on a developer, they
will look at their board or box, see X and try to find something
similar in the driver source.

To use OSF_ * the name needs a suffix to differentiate it from future
cards that will also use ofs.

If this really is a generic id please explain in the doc patch how
every future board with use this single id and how a driver could
work around a hw problem in a specific board with a pci id covering
multiple boards.

Tom
Hi Tom,

The intent is to have a generic device id that can be used with many
different boards.  Currently, we have FPGA implementations for 3
different boards using this generic id.  We may need a better name for
device id than OFS.  More precisely this generic device id means a PCI
function that is described by a Device Feature List (DFL).  How about
PCIE_DEVICE_ID_INTEL_DFL?

With a DFL device id, the functionality of the PF/VF is determined by
the contents of the DFL.  Each Device Feature Header (DFH) in the DFL
has a revision field that can be used identify "broken" hw, or new
functionality added to a feature.  Additionally, since the DFL is
typically used in a FPGA, the broken hardware, can and should be fixed
in most cases.
How is lspci supposed to work ?
There is an example for one card using IOFS and DFL.

# lspci | grep acc
b1:00.0 Processing accelerators: Intel Corporation Device bcce (rev 01)
b1:00.1 Processing accelerators: Intel Corporation Device bcce
b1:00.2 Processing accelerators: Intel Corporation Device bcce
b1:00.3 Processing accelerators: Red Hat, Inc. Virtio network device
b1:00.4 Processing accelerators: Intel Corporation Device bcce

Note: There 5 PFs in this card, it exports the management functions via PF0(b1:00.0), Other PFs like b1:00.1, b1:00.2, b1:00.4, are using for testing, which depends on RTL designer or project requirement. The PF3 instance a VirtIO net device for example, will bind with virtio-net driver
presenting itself as a network interface to the OS.

Hi Tom,

These are very good questions, and the answers will be addressed in the documentation associated with a v2 submission of this patch.


What I mean there is heterogeneous set of cards in one machine, how do you tell which card is which ?

If the PCI PID/VID is generic, indicating only that there is one or more DFL, then some other mechanism must be used to differentiate the cards. One could use unique PCI sub-PID/sub-VIDs to differentiate specific implementations. One could also use some register in BAR space to help identify the card, or one could use PCI Vital Product Data (VPD) to provide detailed information about the running FPGA design on the card.


Or in a datacenter where the machines are all remote and admin has to flash just the n6000's ?

This problem exists with the N3000 cards. Depending on the FPGA configuration, the line side of the card could be very different (e.g. 4x10Gb or 2x2x25Gb). The network operator must make sure to update a particular N3000 card with the correct FPGA image type. In the case of the N3000 there is a register exposed through sysfs containing the "Bitstream ID" which contains the line side configuration of the FPGA.


How could she find just the n6000's with lspci ?

If you only wanted to use lspci to determine the card, then differentiating PCI sub-VID/sub-PID could be used or VPD could be used.


How would the driver know ?

The dfl-pci driver is fairly generic in that it doesn't really care about the PCI PID/VID because all it really does enumerate the DFLs. It is the individual dfl drivers that may need to know hw differences/bugs for that component IP.


Tom


A dfl set can change with fw updates and in theory different boards could have
the same set.

Tom

Matthew
Tom

    /* VF Device */
    #define PCIE_DEVICE_ID_VF_INT_5_X        0xBCBF
    #define PCIE_DEVICE_ID_VF_INT_6_X        0xBCC1
    #define PCIE_DEVICE_ID_VF_DSC_1_X        0x09C5
    #define PCIE_DEVICE_ID_INTEL_PAC_D5005_VF    0x0B2C
+#define PCIE_DEVICE_ID_INTEL_OFS_VF        0xbccf

    static struct pci_device_id cci_pcie_id_tbl[] = {
        {PCI_DEVICE(PCI_VENDOR_ID_INTEL,
PCIE_DEVICE_ID_PF_INT_5_X),},
@@
-95,6 +97,8 @@ static struct pci_device_id cci_pcie_id_tbl[] = {
        {PCI_DEVICE(PCI_VENDOR_ID_INTEL,
PCIE_DEVICE_ID_INTEL_PAC_D5005_VF),},
{PCI_DEVICE(PCI_VENDOR_ID_SILICOM_DENMARK,
PCIE_DEVICE_ID_SILICOM_PAC_N5010),},
{PCI_DEVICE(PCI_VENDOR_ID_SILICOM_DENMARK,
PCIE_DEVICE_ID_SILICOM_PAC_N5011),},
+    {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCIE_DEVICE_ID_INTEL_OFS),},
+    {PCI_DEVICE(PCI_VENDOR_ID_INTEL,
PCIE_DEVICE_ID_INTEL_OFS_VF),},
        {0,}
    };
    MODULE_DEVICE_TABLE(pci, cci_pcie_id_tbl);



[Index of Archives]     [LM Sensors]     [Linux Sound]     [ALSA Users]     [ALSA Devel]     [Linux Audio Users]     [Linux Media]     [Kernel]     [Gimp]     [Yosemite News]     [Linux Media]

  Powered by Linux