On Tue, Mar 27, 2018 at 12:28:42PM -0500, Bjorn Helgaas wrote: > On Tue, Mar 27, 2018 at 08:17:26PM +0300, Tal Gilboa wrote: > > On 3/21/2018 10:10 PM, Bjorn Helgaas wrote: > > > On Wed, Mar 21, 2018 at 07:59:21PM +0000, Keller, Jacob E wrote: > > > > > -----Original Message----- > > > > > From: Bjorn Helgaas [mailto:helgaas@xxxxxxxxxx] > > > > > Sent: Wednesday, March 21, 2018 12:48 PM > > > > > To: Tal Gilboa <talgi@xxxxxxxxxxxx> > > > > > Cc: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>; Linux PCI <linux- > > > > > pci@xxxxxxxxxxxxxxx>; Tariq Toukan <tariqt@xxxxxxxxxxxx>; Keller, Jacob E > > > > > <jacob.e.keller@xxxxxxxxx> > > > > > Subject: Re: [PATCH next V3 3/5] PCI: Print PCI device link status in kernel log > > > > > > > > > > On Wed, Mar 21, 2018 at 09:43:48AM +0200, Tal Gilboa wrote: > > > > > > On 3/20/2018 4:05 PM, Bjorn Helgaas wrote: > > > > > > > [+cc Jacob] > > > > > > > > > > > > > > On Mon, Mar 12, 2018 at 02:06:08PM +0200, Tal Gilboa wrote: > > > > > > > > Add pcie_print_link_status() function for querying and verifying > > > > > > > > a PCI device link status. The PCI speed and width are reported > > > > > > > > in kernel log. > > > > > > > > This provides a unified method for all PCI devices to > > > > > > > > report status and issues, instead of each device reporting in a > > > > > > > > different way, using different code. > > > > > > > > > So let's see if we agree on the steps: > > > > > > 1. my_speed_cap, my_width_cap <- calculate device PCIe caps > > > > > > 2. avail_bw, limiting_dev <- calculate PCIe chain bandwidth > > > > > > 3. my_bw <- my_speed_cap * my_width_cap > > > > > > 4. If avail_bw == my_bw print available bandwidth + PCIe caps > > > > > > 5. Else print available bandwidth + limited by + capable bandwidth + PCIe > > > > > > caps > > > > > > > > > > > > What do you think? > > > > > > > > > > Steps 2 and 3 might need to be smart enough to apply the effect of > > > > > encoding differences between generations. > > > > > > > > > > In step 2, we don't have any current user of the "limiting_dev" > > > > > information, so I'd omit it until we have somebody who wants it. > > > > > > > > > > In step 5, we don't know the "limited by" part (unless you want to add > > > > > that). > > > > > > > > It might be useful to have the limited by information printed, even > > > > if no driver yet bothered to do it today. > > > > > > I wouldn't object to printing that information (although it increases > > > the challenge of making the message pithy), and it's basically free to > > > collect it. > > > > > > > Coding done, currently under internal review. Will submit right after. > > I'm having some trouble printing the limiting device bus. Any > > recommendations on which format to use? dev->bus->name gives me the 6 first > > digits (e.g. "0000:07"). How do I get the last 3 (e.g. "00.0")? > > dev->bus->primary and dev->bus->number seem like good candidates but the > > actual values I get seem off. > > device_name(pdev) should be what you want. You should see two devices > with the lowest bandwidth, i.e., the upstream and downstream ends of > one link. I think it would make the most sense to print the upstream > end. Sorry, it would help if I actually looked at the code first. pci_name(pdev), which calls dev_name(&pdev->dev), is what you want.
