[dpdk-dev] [PATCH] pci: fix check uio bind

[Tan, Jianfeng]

Hi Gaëtan,

> -----Original Message-----
> From: Gaëtan Rivet [mailto:gaetan.rivet at 6wind.com]
> Sent: Saturday, October 21, 2017 4:08 AM
> To: Tan, Jianfeng
> Cc: dev at dpdk.org; santosh.shukla at caviumnetworks.com;
> jerin.jacob at caviumnetworks.com; Burakov, Anatoly
> Subject: Re: [PATCH] pci: fix check uio bind
> 
> On Sat, Oct 21, 2017 at 12:47:14AM +0800, Tan, Jianfeng wrote:
> > Hi Gaëtan,
> >
> >
> > On 10/19/2017 7:42 PM, Gaëtan Rivet wrote:
> > >Hi Jianfeng,
> > >
> > >On Thu, Oct 19, 2017 at 11:18:29AM +0000, Jianfeng Tan wrote:
> > >>When checking if any devices bound to uio, we did not exclud
> > >>those which are blacklisted (or in the case that a whitelist
> > >>is specified).
> > >>
> > >>This patch fixes it by only checking whitelisted devices.
> > >>
> > >>Fixes: 815c7deaed2d ("pci: get IOMMU class on Linux")
> > >>
> > >>Signed-off-by: Jianfeng Tan <jianfeng.tan at intel.com>
> > >>---
> > >>  lib/librte_eal/linuxapp/eal/eal_pci.c | 18 ++++++++++++++++++
> > >>  1 file changed, 18 insertions(+)
> > >>
> > >>diff --git a/lib/librte_eal/linuxapp/eal/eal_pci.c
> b/lib/librte_eal/linuxapp/eal/eal_pci.c
> > >>index b4dbf95..2b23d67 100644
> > >>--- a/lib/librte_eal/linuxapp/eal/eal_pci.c
> > >>+++ b/lib/librte_eal/linuxapp/eal/eal_pci.c
> > >>@@ -516,8 +516,26 @@ static inline int
> > >>  pci_one_device_bound_uio(void)
> > >>  {
> > >>  	struct rte_pci_device *dev = NULL;
> > >>+	struct rte_devargs *devargs;
> > >>+	int check_all = 1;
> > >>+	int need_check;
> > >>+
> > >>+	if (rte_pci_bus.bus.conf.scan_mode == RTE_BUS_SCAN_WHITELIST)
> > >>+		check_all = 0;
> > >>  	FOREACH_DEVICE_ON_PCIBUS(dev) {
> > >>+		devargs = dev->device.devargs;
> > >>+
> > >>+		need_check = 0;
> > >>+		if (check_all)
> > >Unless I'm mistaken, you will check blacklisted devices as well here.
> >
> > Thank you for pointing out this.
> >
> > I was referring to rte_pci_probe(), which also only check "probe_all" and
> > (devargs && RTE_DEV_WHITELISTED); but turns out it double checks the
> > blacklisted devices in rte_pci_probe_one_driver().
> >
> > I'll fix it.
> >
> > >The condition should be something like:
> > >
> > >if (check_all && devargs == NULL)
> >
> > >Which means that both ifs can be refactored as
> > >
> > >if ((check_all ^ (devargs != NULL)) == 0)
> > >         continue;
> > >
> > >Removing need_check. But it can be hard to read.
> >
> > Yes, I prefer to make it easy to understand. Please let me know if you are
> > OK with below code (remove check_all):
> >
> >         FOREACH_DEVICE_ON_PCIBUS(dev) {
> >                 devargs = dev->device.devargs;
> >
> >                 need_check = 0;
> >                 switch (rte_pci_bus.bus.conf.scan_mode) {
> >                 case RTE_BUS_SCAN_UNDEFINED:
> >                         need_check = 1;
> >                         break;
> >                 case RTE_BUS_SCAN_WHITELIST:
> >                         if (devargs && devargs->policy ==
> > RTE_DEV_WHITELISTED)
> >                                 need_check = 1;
> >                         break;
> >                 case RTE_BUS_SCAN_BLACKLIST:
> >                         if (!devargs || devargs->policy !=
> > RTE_DEV_BLACKLISTED)
> >                                 need_check = 1;
> >                         break;
> >                 }
> >
> >                 if (!need_check)
> >                         continue;
> > ...
> 
> I like the switch, two remarks however:
> 
> 1. The SCAN_UNDEFINED basically means blacklist mode for the PCI bus.
>    This is the reason probe_all was set by testing for WHITELIST
>    mode: either of the other too would thus trigger the blacklist
>    behavior.
> 
>    Thus, I think you could write a fallthrough case for UNDEFINED, that
>    would go into the BLACKLIST mode.
> 
> 2. For pointers in general I would test against NULL instead of using
>    the unary '!'.
>    I think it is the general policy in DPDK to always explicitly check
>    against the constant value, but I personally think that for booleans
>    like need_check the "not" operator is ok.
>    So I will only highlight the !devargs :)

Make sense! Will send out a new version as per your above suggestions.

Thanks,
Jianfeng