PCI 2.1 VPD support
cover PCI 2.1 VPD support
PCI 2.1 specifies a way to provide VPD in the expansion ROM. This patch series exposes that in the same way as PCI 2.2 VPD.
I do not have any production devices with VPD in ROM so I reflashed a NIC with an image including it. This code should be tested against some production ROMs since I may have misinterpreted the specification both when reading and writing!
There are two remaining aspects of this that I'm not quite happy about:
1. I understand that the expansion ROM may share a decoder with BAR 0, which makes expansion ROM access dangerous when a driver is loaded. The sysfs "rom" attribute must be specifically read-enabled by writing to it, which I assume is intended to protect against this. Perhaps pci_vpd_pci21_read() should test pdev->rom_attr_enabled?
2. PCI resource allocation may fail during pci_scan_device() and therefore I could not insert the call to pci_vpd_pci21_init() there. Instead I added it to pci_create_sysfs_dev_files() - but I don't really think this function should be probing. Is there a better place to add the call?
Ben.
Ben Hutchings (2):
PCI: ROM access changes to support PCI 2.1 VPD
PCI: Expose PCI 2.1 VPD through sysfs
drivers/pci/pci-sysfs.c | 2 +- drivers/pci/pci.h | 1 + drivers/pci/rom.c | 245 +++++++++++++++++++++++++++++++++++++++-------- 3 files changed, 206 insertions(+), 42 deletions(-)
Signature · 2 lines
Ben Hutchings, Senior Software Engineer, Solarflare Communications Not speaking for my employer; that's the marketing department's job.
Quoted 7 lines · Ben Hutchings
PCI 2.1 specifies a way to provide VPD in the expansion ROM. This patch series exposes that in the same way as PCI 2.2 VPD.
I do not have any production devices with VPD in ROM so I reflashed a NIC with an image including it. This code should be tested against some production ROMs since I may have misinterpreted the specification both when reading and writing!
I'll have to dig around to see if I can find any here for testing.
Quoted 7 lines
There are two remaining aspects of this that I'm not quite happy about:
1. I understand that the expansion ROM may share a decoder with BAR 0, which makes expansion ROM access dangerous when a driver is loaded. The sysfs "rom" attribute must be specifically read-enabled by writing to it, which I assume is intended to protect against this. Perhaps
pci_vpd_pci21_read()should testpdev->rom_attr_enabled?
Right, that can get a little ugly... In the case of 2.1 VPD, it does make sense to require that the ROM be enabled before poking at it, and is definitely safer.
Quoted 5 lines
2. PCI resource allocation may fail during
pci_scan_device()and therefore I could not insert the call topci_vpd_pci21_init()there. Instead I added it topci_create_sysfs_dev_files()- but I don't really think this function should be probing. Is there a better place to add the call?
Well, one thing we could do is try to either use the allocated ROM or assign it later on, then cache the VPD data. That way we wouldn't have to worry about accessing it with the rom enabled later on. But that would eat up memory.
You could also just make the pci_vpd_pci21_init a device_initcall so it can fill in the dev->vpd before the PCI sysfs late_initcall. Or we could just stuff it into pci_init in the same loop that calls the final fixups.
Other than that, things look pretty good to me. I'd definitely like to get this tested on at least one PCI 2.1 device w/VPD before pushing though. You know what they say about untested code. :)
Thanks,
Jesse