[dpdk-dev] [PATCH v2] windows/netuio: add Windows NetUIO kernel driver
Narcisa Ana Maria Vasile
navasile at linux.microsoft.com
Wed Sep 9 20:53:53 CEST 2020
On Mon, Aug 24, 2020 at 11:53:44PM +0300, Dmitry Kozlyuk wrote:
> On Thu, 20 Aug 2020 15:23:55 -0700, Narcisa Ana Maria Vasile wrote:
> > From: Narcisa Vasile <navasile at microsoft.com>
> >
> > The Windows NetUIO kernel driver allows the DPDK userspace
> > application to directly access the hardware.
> >
> > Cc: Harini Ramakrishnan <Harini.Ramakrishnan at microsoft.com>
> > Cc: Omar Cardona <ocardona at microsoft.com>
> > Signed-off-by: Narcisa Vasile <navasile at microsoft.com>
> > ---
>
> Major questions:
>
> 1. Does NetUIO still need to allocate and map a contiguous memory segment
> now, when DPDK has user-mode memory management?
>
> 2. IOCTLs require to specify PCI address on each call. This is very
> inconvenient for DPDK consumers and also seems to serve no purpose.
>
> 3. There is a need to document driver's design, preferably in commit message,
> specifically:
>
> 3.1) DMA remapping capability in INF (AFAIK, vendors are notified);
> 3.2) manual BAR probing instead of using resource lists;
> 3.3) reason to use EvtIoInCallerContext and IO queues;
> 3.4) IOCTL format.
>
> Also, I agree with everything Ranjit has noted already.
>
> General suggestions to cleanup the code a bit. We can do it later if you
> wish. This also brings up the question, which code style should Windows
> kernel code for DPDK follow (off-topic for now).
>
> * Remove boilerplate code and comments generated by VS wizard.
> * Place `_Use_decl_annotations` on definitions to make them simpler.
> * Limit line length, try using shorted variable names (e.g.
> "netuio_contextdata" may be "context" or "ctx" with no loss).
>
Thank you for you thorough review!
> More specific comments inline.
>
> [snip]
> > +[Strings]
> > +SPSVCINST_ASSOCSERVICE= 0x00000002
> > +Intel = "Intel"
> > +Broadcom = "Broadcom Corporation"
>
> IHVs are supposed to add this gradually.
>
Thanks! I've modified the inf to remove all the vendor specific strings
but left the sections to avoid errors when generating the binary.
> > +ClassName = "Windows UIO"
> > +DiskName = "DPDK netUIO Installation Disk"
> > +netuio.DeviceDesc = "netuio Device"
> > +netuio.SVCDESC = "netuio Service"
> > +
> > + // Build symbolic link name as <netuio_symbolic_link>_BDF (bus/device/func)
> > + CHAR symbolic_link[64] = { 0 };
> > + sprintf_s(symbolic_link, sizeof(symbolic_link), "%s_%04d%02d%02d",
> > + NETUIO_DEVICE_SYMBOLIC_LINK_ANSI, netuio_contextdata->addr.bus_num,
> > + netuio_contextdata->addr.dev_num, netuio_contextdata->addr.func_num);
> > +
> > + ANSI_STRING ansi_symbolic_link;
> > + RtlInitAnsiString(&ansi_symbolic_link, symbolic_link);
> > +
> > + status = RtlAnsiStringToUnicodeString(&netuio_symbolic_link, &ansi_symbolic_link, TRUE);
> > + if (!NT_SUCCESS(status))
> > + return status;
>
> Why not use Unicode directly?
>
It looks like either way, a cast will be needed (for example, if I use wchar for symbolic_link and then
RtlInitUnicodeString()). I've left it as is, but let me know if there's an elegant solution that I didn't see.
> > +
> > + status = WdfDeviceCreateSymbolicLink(device, &netuio_symbolic_link);
> > +
> > + RtlFreeUnicodeString(&netuio_symbolic_link);
> > +
> > + return status;
More information about the dev
mailing list