| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CABGxfO73LpQkRt-LWMcejuM4T8pacsD7JWyzaozgJiubr-8upA@mail.gmail.com>
Date: Mon, 11 Oct 2021 08:01:46 +0200
From: Saheed Bolarinwa <refactormyself@...il.com>
To: Bjorn Helgaas <helgaas@...nel.org>
Cc: linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [RFC PATCH v2 1/4] PCI/ASPM: Remove struct pcie_link_state.parent
On Fri, Oct 1, 2021 at 12:40 AM Bjorn Helgaas <helgaas@...nel.org> wrote:
>
> On Wed, Sep 29, 2021 at 02:43:12AM +0200, Saheed O. Bolarinwa wrote:
> > From: "Bolarinwa O. Saheed" <refactormyself@...il.com>
> >
> > Information cached in struct pcie_link_state.parent is accessible
> > via struct pci_dev.
> >
> > This patch:
> > - removes *parent* from the *struct pcie_link_state*
> > - creates pci_get_parent() which returns the parent of a pci_dev
> > - replaces references to pcie_link_state.parent with a call to
> > pci_get_parent()
> > - removes BUG_ON(root->parent), instead uses the parent's root
> >
> > Signed-off-by: Bolarinwa O. Saheed <refactormyself@...il.com>
> > ---
> > drivers/pci/pcie/aspm.c | 36 ++++++++++++++++++++++++++----------
> > 1 file changed, 26 insertions(+), 10 deletions(-)
> >
> > diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> > index 013a47f587ce..414c04ffe962 100644
> > --- a/drivers/pci/pcie/aspm.c
> > +++ b/drivers/pci/pcie/aspm.c
> > @@ -50,7 +50,6 @@ struct pcie_link_state {
> > struct pci_dev *pdev; /* Upstream component of the Link */
> > struct pci_dev *downstream; /* Downstream component, function 0 */
> > struct pcie_link_state *root; /* pointer to the root port link */
> > - struct pcie_link_state *parent; /* pointer to the parent Link state */
> > struct list_head sibling; /* node in link_list */
> >
> > /* ASPM state */
> > @@ -139,6 +138,14 @@ static int policy_to_clkpm_state(struct pcie_link_state *link)
> > return 0;
> > }
> >
> > +static struct pci_dev *pci_get_parent(struct pci_dev *pdev)
> > +{
> > + if (!pdev || !pdev->bus->parent || !pdev->bus->parent->self)
> > + return NULL;
> > +
> > + return pdev->bus->parent->self;
> > +}
>
> I LOVE the idea of getting rid of the pcie_link_state.parent pointer.
> I think it's dumb to maintain a shadow hierarchy when we already HAVE
> a hierarchy in struct pci_dev.
>
> I'm not in love with the pci_get_parent() name because a pci_dev
> doesn't really have a "parent." The closest thing to a parent would
> be the bridge upstream from the device, and that's not what this
> returns.
>
> This actually has to start from a Downstream Port (not an Endpoint)
> because the struct pcie_link_state is always associated with the
> upstream end of the link.
>
> And it actually returns the bridge that is *two* levels up, because
> that's the upstream end of the next link, so it's more like the
> "grandparent" of pdev, not the "parent."
>
> Example from my laptop:
>
> 0a:04.0 Downstream Port (switch A) to [bus 0c-3d]
> 0c:00.0 Upstream Port (switch B) to [bus 0d-3d]
> 0d:01.0 Downstream Port (switch B) to [bus 0e]
> 0e:00.0 Upstream Port (Endpoint) USB controller
>
> Here there are two links:
>
> 0a:04.0 --- 0c:00.0
> 0d:01.0 --- 0e:00.0
>
> and the pcie_link_states are associated with 0a:04.0 and 0d:01.0.
>
> If we start from 0d:01.0, which is the upstream end of the last link:
>
> "pdev" is a pci_dev of a downstream port, e.g., 0d:01.0.
> "pdev->bus" is the pci_bus pdev is on: [bus 0d].
> "pdev->bus->self" is the bridge leading to "bus": 0c:00.0.
> "pdev->bus->parent" is the parent pci_bus of [bus 0d]: [bus 0c].
> "pdev->bus->parent->self" is the bridge leading to [bus 0c]: 0a:04.0.
>
> Sorry for the rambling, just trying to get this all clear in my head.
Thanks, it is very helpful for me.
>
> Almost all the calls of pci_get_parent() look like this:
>
> parent = pci_get_parent(link->pdev);
> link = parent ? parent->link_state : NULL;
>
> What if you made something like this:
>
> struct pcie_link_state *pcie_upstream_link(struct pcie_link_state *link)
This is better, especially using pci_upstream_bridge() directly.
By returning struct pci_dev, I was trying to avoid "struct pcie_link_state",
so as to ease the journey towards eliminating it.
Also, will it be helpful to handle NULL values within pci_upstream_bridge()?
> {
> struct pci_dev *bridge;
>
> bridge = pci_upstream_bridge(link->pdev);
> if (!bridge)
> return NULL;
>
> bridge = pci_upstream_bridge(bridge);
> return bridge ? bridge->link_state : NULL;
> }
>
> > static void pcie_set_clkpm_nocheck(struct pcie_link_state *link, int enable)
> > {
> > struct pci_dev *child;
> > @@ -379,6 +386,7 @@ static void encode_l12_threshold(u32 threshold_us, u32 *scale, u32 *value)
> > static void pcie_aspm_check_latency(struct pci_dev *endpoint)
> > {
> > u32 latency, l1_switch_latency = 0;
> > + struct pci_dev *parent;
> > struct aspm_latency *acceptable;
> > struct pcie_link_state *link;
> >
> > @@ -419,7 +427,8 @@ static void pcie_aspm_check_latency(struct pci_dev *endpoint)
> > link->aspm_capable &= ~ASPM_STATE_L1;
> > l1_switch_latency += 1000;
> >
> > - link = link->parent;
> > + parent = pci_get_parent(link->pdev);
> > + link = parent ? parent->link_state : NULL;
> > }
> > }
> >
> > @@ -793,9 +802,11 @@ static void pcie_config_aspm_link(struct pcie_link_state *link, u32 state)
> >
> > static void pcie_config_aspm_path(struct pcie_link_state *link)
> > {
> > + struct pci_dev *parent;
>
> Missing a blank line here.
>
> > while (link) {
> > pcie_config_aspm_link(link, policy_to_aspm_state(link));
> > - link = link->parent;
> > + parent = pci_get_parent(link->pdev);
> > + link = parent ? parent->link_state : NULL;
> > }
> > }
> >
> > @@ -864,16 +875,15 @@ static struct pcie_link_state *alloc_pcie_link_state(struct pci_dev *pdev)
> > !pdev->bus->parent->self) {
> > link->root = link;
> > } else {
> > - struct pcie_link_state *parent;
> > + struct pci_dev *parent;
> >
> > - parent = pdev->bus->parent->self->link_state;
> > - if (!parent) {
> > + parent = pci_get_parent(pdev);
> > + if (!parent->link_state) {
> > kfree(link);
> > return NULL;
> > }
> >
> > - link->parent = parent;
> > - link->root = link->parent->root;
> > + link->root = parent->link_state->root;
> > }
> >
> > list_add(&link->sibling, &link_list);
> > @@ -962,7 +972,11 @@ void pcie_aspm_init_link_state(struct pci_dev *pdev)
> > static void pcie_update_aspm_capable(struct pcie_link_state *root)
> > {
> > struct pcie_link_state *link;
> > - BUG_ON(root->parent);
> > + struct pci_dev *parent = pci_get_parent(root->pdev);
> > +
> > + if (parent && parent->link_state)
> > + root = parent->link_state->root;
> > +
> > list_for_each_entry(link, &link_list, sibling) {
> > if (link->root != root)
> > continue;
> > @@ -985,6 +999,7 @@ static void pcie_update_aspm_capable(struct pcie_link_state *root)
> > /* @pdev: the endpoint device */
> > void pcie_aspm_exit_link_state(struct pci_dev *pdev)
> > {
> > + struct pci_dev *parent_dev;
> > struct pci_dev *parent = pdev->bus->self;
> > struct pcie_link_state *link, *root, *parent_link;
> >
> > @@ -1002,7 +1017,8 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
> >
> > link = parent->link_state;
> > root = link->root;
> > - parent_link = link->parent;
> > + parent_dev = pci_get_parent(link->pdev);
> > + parent_link = parent_dev ? parent_dev->link_state : NULL;
> >
> > /* All functions are removed, so just disable ASPM for the link */
> > pcie_config_aspm_link(link, 0);
> > --
> > 2.20.1
> >
Powered by blists - more mailing lists