On Tue, Feb 1, 2022 at 9:37 AM Ben Widawsky <ben.widawsky@xxxxxxxxx> wrote:
>
> On 22-01-23 16:31:29, Dan Williams wrote:
> > So far the platorm level CXL resources have been enumerated by the
> > cxl_acpi driver, and cxl_pci has gathered all the pre-requisite
> > information it needs to fire up a cxl_mem driver. However, the first
> > thing the cxl_mem driver will be tasked to do is validate that all the
> > PCIe Switches in its ancestry also have CXL capabilities and an CXL.mem
> > link established.
> >
> > Provide a common mechanism for a CXL.mem endpoint driver to enumerate
> > all the ancestor CXL ports in the topology and validate CXL.mem
> > connectivity.
> >
> > Multiple endpoints may end up racing to establish a shared port in the
> > topology. This race is resolved via taking the device-lock on a parent
> > CXL Port before establishing a new child. The winner of the race
> > establishes the port, the loser simply registers its interest in the
> > port via 'struct cxl_ep' place-holder reference.
> >
> > At endpoint teardown the same parent port lock is taken as 'struct
> > cxl_ep' references are deleted. Last endpoint to drop its reference
> > unregisters the port.
> >
> > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > ---
> > drivers/cxl/acpi.c | 17 --
> > drivers/cxl/core/port.c | 379 +++++++++++++++++++++++++++++++++++++++++++++++
> > drivers/cxl/cxl.h | 20 ++
> > 3 files changed, 400 insertions(+), 16 deletions(-)
> >
> > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > index 683f2ca32c97..7bd53dc691ec 100644
> > --- a/drivers/cxl/acpi.c
> > +++ b/drivers/cxl/acpi.c
> > @@ -130,21 +130,6 @@ static int cxl_parse_cfmws(union acpi_subtable_headers *header, void *arg,
> > return 0;
> > }
> >
> > -static struct cxl_dport *find_dport_by_dev(struct cxl_port *port, struct device *dev)
> > -{
> > - struct cxl_dport *dport;
> > -
> > - cxl_device_lock(&port->dev);
> > - list_for_each_entry(dport, &port->dports, list)
> > - if (dport->dport == dev) {
> > - cxl_device_unlock(&port->dev);
> > - return dport;
> > - }
> > -
> > - cxl_device_unlock(&port->dev);
> > - return NULL;
> > -}
> > -
> > __mock struct acpi_device *to_cxl_host_bridge(struct device *host,
> > struct device *dev)
> > {
> > @@ -175,7 +160,7 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> > if (!bridge)
> > return 0;
> >
> > - dport = find_dport_by_dev(root_port, match);
> > + dport = cxl_find_dport_by_dev(root_port, match);
> > if (!dport) {
> > dev_dbg(host, "host bridge expected and not found\n");
> > return 0;
> > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > index 26c3eb9180cd..cd95d9f8c624 100644
> > --- a/drivers/cxl/core/port.c
> > +++ b/drivers/cxl/core/port.c
> > @@ -7,6 +7,7 @@
> > #include <linux/slab.h>
> > #include <linux/idr.h>
> > #include <cxlmem.h>
> > +#include <cxlpci.h>
> > #include <cxl.h>
> > #include "core.h"
> >
> > @@ -267,10 +268,24 @@ struct cxl_decoder *to_cxl_decoder(struct device *dev)
> > }
> > EXPORT_SYMBOL_NS_GPL(to_cxl_decoder, CXL);
> >
> > +static void cxl_ep_release(struct cxl_ep *ep)
> > +{
> > + if (!ep)
> > + return;
> > + list_del(&ep->list);
> > + put_device(ep->ep);
> > + kfree(ep);
> > +}
> > +
> > static void cxl_port_release(struct device *dev)
> > {
> > struct cxl_port *port = to_cxl_port(dev);
> > + struct cxl_ep *ep, *_e;
> >
> > + cxl_device_lock(dev);
> > + list_for_each_entry_safe(ep, _e, &port->endpoints, list)
> > + cxl_ep_release(ep);
> > + cxl_device_unlock(dev);
> > ida_free(&cxl_port_ida, port->id);
> > kfree(port);
> > }
> > @@ -361,6 +376,7 @@ static struct cxl_port *cxl_port_alloc(struct device *uport,
> > port->component_reg_phys = component_reg_phys;
> > ida_init(&port->decoder_ida);
> > INIT_LIST_HEAD(&port->dports);
> > + INIT_LIST_HEAD(&port->endpoints);
> >
> > device_initialize(dev);
> > device_set_pm_not_required(dev);
> > @@ -639,6 +655,369 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
> > }
> > EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
> >
> > +static struct cxl_ep *find_ep(struct cxl_port *port, struct device *ep_dev)
> > +{
> > + struct cxl_ep *ep;
> > +
> > + device_lock_assert(&port->dev);
> > + list_for_each_entry(ep, &port->endpoints, list)
> > + if (ep->ep == ep_dev)
> > + return ep;
> > + return NULL;
> > +}
> > +
> > +static int add_ep(struct cxl_port *port, struct cxl_ep *new)
> > +{
> > + struct cxl_ep *dup;
> > +
> > + cxl_device_lock(&port->dev);
> > + if (port->dead) {
> > + cxl_device_unlock(&port->dev);
> > + return -ENXIO;
> > + }
> > + dup = find_ep(port, new->ep);
> > + if (!dup)
> > + list_add_tail(&new->list, &port->endpoints);
> > + cxl_device_unlock(&port->dev);
> > +
> > + return dup ? -EEXIST : 0;
> > +}
> > +
> > +/**
> > + * cxl_add_ep - register an endpoint's interest in a port
> > + * @port: a port in the endpoint's topology ancestry
> > + * @ep_dev: device representing the endpoint
> > + *
> > + * Intermediate CXL ports are scanned based on the arrival of endpoints.
> > + * When those endpoints depart the port can be destroyed once all
> > + * endpoints that care about that port have been removed.
> > + */
> > +static int cxl_add_ep(struct cxl_port *port, struct device *ep_dev)
> > +{
> > + struct cxl_ep *ep;
> > + int rc;
> > +
> > + ep = kzalloc(sizeof(*ep), GFP_KERNEL);
> > + if (!ep)
> > + return -ENOMEM;
> > +
> > + INIT_LIST_HEAD(&ep->list);
> > + ep->ep = get_device(ep_dev);
> > +
> > + rc = add_ep(port, ep);
> > + if (rc)
> > + cxl_ep_release(ep);
> > + return rc;
> > +}
> > +
> > +struct cxl_find_port_ctx {
> > + const struct device *dport_dev;
> > + const struct cxl_port *parent_port;
> > +};
> > +
> > +static int match_port_by_dport(struct device *dev, const void *data)
> > +{
> > + const struct cxl_find_port_ctx *ctx = data;
> > + struct cxl_port *port;
> > +
> > + if (!is_cxl_port(dev))
> > + return 0;
> > + if (ctx->parent_port && dev->parent != &ctx->parent_port->dev)
> > + return 0;
> > +
> > + port = to_cxl_port(dev);
> > + return cxl_find_dport_by_dev(port, ctx->dport_dev) != NULL;
> > +}
> > +
> > +static struct cxl_port *__find_cxl_port(struct cxl_find_port_ctx *ctx)
> > +{
> > + struct device *dev;
> > +
> > + if (!ctx->dport_dev)
> > + return NULL;
> > +
> > + dev = bus_find_device(&cxl_bus_type, NULL, ctx, match_port_by_dport);
> > + if (dev)
> > + return to_cxl_port(dev);
> > + return NULL;
> > +}
> > +
> > +static struct cxl_port *find_cxl_port(struct device *dport_dev)
> > +{
> > + struct cxl_find_port_ctx ctx = {
> > + .dport_dev = dport_dev,
> > + };
> > +
> > + return __find_cxl_port(&ctx);
> > +}
> > +
> > +static struct cxl_port *find_cxl_port_at(struct cxl_port *parent_port,
> > + struct device *dport_dev)
> > +{
> > + struct cxl_find_port_ctx ctx = {
> > + .dport_dev = dport_dev,
> > + .parent_port = parent_port,
> > + };
> > +
> > + return __find_cxl_port(&ctx);
> > +}
> > +
> > +static struct device *grandparent(struct device *dev)
> > +{
> > + if (dev && dev->parent)
> > + return dev->parent->parent;
> > + return NULL;
> > +}
> > +
> > +static void delete_switch_port(struct cxl_memdev *cxlmd, struct cxl_port *port,
> > + struct list_head *dports)
> > +{
> > + struct cxl_dport *dport, *_d;
> > +
> > + dev_dbg(&cxlmd->dev, "delete %s\n", dev_name(&port->dev));
> > + list_for_each_entry_safe(dport, _d, dports, list) {
> > + devm_release_action(&port->dev, cxl_dport_unlink, dport);
> > + devm_release_action(&port->dev, cxl_dport_remove, dport);
> > + devm_kfree(&port->dev, dport);
> > + }
> > + devm_release_action(port->dev.parent, cxl_unlink_uport, port);
> > + devm_release_action(port->dev.parent, unregister_port, port);
> > +}
>
> I'd drop the cxlmd argument here. Let the caller print the dev_dbg if it wants.
Ok.
>
> > +
> > +static void cxl_remove_ep(void *data)
> > +{
> > + struct cxl_memdev *cxlmd = data;
> > + struct device *iter;
> > +
> > + for (iter = &cxlmd->dev; iter; iter = grandparent(iter)) {
> > + struct device *dport_dev = grandparent(iter);
> > + struct cxl_port *port, *parent_port;
> > + LIST_HEAD(reap_dports);
> > + struct cxl_ep *ep;
> > +
> > + if (!dport_dev)
> > + break;
> > +
> > + port = find_cxl_port(dport_dev);
> > + if (!port || is_cxl_root(port)) {
> > + put_device(&port->dev);
> > + continue;
>
> Is there ever a case that continue != break for this case? It seems to be this
> deserves a WARN_ON(grandparent(iter)) or some such, but I'd be curious to see if
> my understanding is off.
Recall that while outside the lock we have top-down removal racing
potentially multiple bottom-up removals. So the "!port" case can just
be a race loss, but needs to continue up the chain. The root check
does assume that grandparent goes NULL. WARN_ON() would be a nop here
because if grandparent() for a root dport resolves to !NULL the port
registration would not have worked in the first instance.
I added a comment to delete_switch_port() about the races that can be
happening in this function.
>
> > + }
> > +
> > + parent_port = to_cxl_port(port->dev.parent);
> > + cxl_device_lock(&parent_port->dev);
> > + if (!parent_port->dev.driver) {
> > + cxl_device_unlock(&parent_port->dev);
> > + put_device(&port->dev);
> > + continue;
>
> Similar to above, the parent_port must still exist at this point and have a
> driver bound, correct?
Not necessarily.
>
> > + }
> > +
> > + cxl_device_lock(&port->dev);
> > + ep = find_ep(port, &cxlmd->dev);
> > + dev_dbg(&cxlmd->dev, "disconnect %s from %s\n",
> > + ep ? dev_name(ep->ep) : "", dev_name(&port->dev));
> > + cxl_ep_release(ep);
> > + if (ep && !port->dead && list_empty(&port->endpoints) &&
> > + !is_cxl_root(parent_port)) {
> > + /*
> > + * This was the last ep attached to a dynamically
> > + * enumerated port. Block new cxl_add_ep() and garbage
> > + * collect the port.
> > + */
> > + port->dead = true;
> > + list_splice_init(&port->dports, &reap_dports);
> > + }
> > + cxl_device_unlock(&port->dev);
> > +
> > + if (!list_empty(&reap_dports))
> > + delete_switch_port(cxlmd, port, &reap_dports);
>
> I admit I tried to make all this simpler and couldn't figure anything better
> out.
I am working on a unit test to try to stress the above races.
>
> > + put_device(&port->dev);
> > + cxl_device_unlock(&parent_port->dev);
> > + }
> > +}
> > +
> > +static resource_size_t find_component_registers(struct device *dev)
> > +{
> > + struct cxl_register_map map;
> > + struct pci_dev *pdev;
> > +
> > + /*
> > + * Theoretically, CXL component registers can be hosted on a
>
> I believe this is factually incorrect. The spec requires it be on a PCIe device.
> Does it make sense to introduce a dev_is_cxltest()?
I also consider ACPI devices non-PCIe CXL devices. The ACPI CFMWS
already shows how to bridge PCI enumerable resources into PCIe
enumerable CXL domain, so it's not just cxl_test in that
"Theoretical".
>
> > + * non-PCI device, in practice, only cxl_test hits this case.
> > + */
> > + if (!dev_is_pci(dev))
> > + return CXL_RESOURCE_NONE;
> > +
> > + pdev = to_pci_dev(dev);
> > +
> > + cxl_find_regblock(pdev, CXL_REGLOC_RBI_COMPONENT, &map);
> > + return cxl_regmap_to_base(pdev, &map);
> > +}
> > +
> > +static int add_port_register_ep(struct cxl_memdev *cxlmd,
> > + struct device *uport_dev,
> > + struct device *dport_dev)
> > +{
> > + struct cxl_port *port, *parent_port;
> > + resource_size_t component_reg_phys;
> > + int rc;
> > +
> > + parent_port = find_cxl_port(grandparent(dport_dev));
>
> This is confusing (thought correct). dport's parent is a port, and port's parent
> is a port. Logically though, I'd expect a grandparent of a device to be of the
> same type. I wonder if there is some way to straighten that out, or if I'm the
> only one that finds it confusing.
You jumped from PCIe to CXL hierarchy. A dport does not exist on the
cxl_bus_type, it is non-device metadata / a link over to a downstream
switch port device in the PCIe domain. The parent of a downstream
switch port is an upstream switch port. the parent of an upstream
switch port is another downstream switch port.
>
> > + if (!parent_port) {
> > + /*
> > + * The root CXL port is added by the CXL platform driver, fail
> > + * for now to be re-probed after platform driver attaches.
> > + */
> > + if (!grandparent(dport_dev)) {
> > + dev_dbg(&cxlmd->dev, "%s is a root dport\n",
> > + dev_name(dport_dev));
> > + return -ENXIO;
> > + }
> > + /* ...otherwise, iterate to create this parent_port */
> > + return -EAGAIN;
> > + }
> > +
> > + cxl_device_lock(&parent_port->dev);
> > + if (!parent_port->dev.driver) {
> > + dev_warn(&cxlmd->dev,
> > + "port %s:%s disabled, failed to enumerate CXL.mem\n",
> > + dev_name(&parent_port->dev), dev_name(uport_dev));
> > + rc = -ENXIO;
> > + goto out;
> > + }
> > +
> > + port = find_cxl_port_at(parent_port, dport_dev);
>
> Again this is a bit tough on the terminology. A dport's parent is a port, but
> not parent_port. I'd definitely suggest a comment for clarifying that.
A dport_dev's parent is not a cxl_port. CXL devices can't parent PCI
devices. I'll add a comment to levelset what some of these variables
are representing and how "grandparent()" works in that context.
>
> > + if (!port) {
> > + component_reg_phys = find_component_registers(uport_dev);
> > + port = devm_cxl_add_port(&parent_port->dev, uport_dev,
> > + component_reg_phys, parent_port);
> > + if (!IS_ERR(port))
> > + get_device(&port->dev);
> > + }
> > +out:
> > + cxl_device_unlock(&parent_port->dev);
> > +
> > + if (IS_ERR(port))
> > + rc = PTR_ERR(port);
> > + else {
> > + dev_dbg(&cxlmd->dev, "add to new port %s:%s\n",
> > + dev_name(&port->dev), dev_name(port->uport));
> > + rc = cxl_add_ep(port, &cxlmd->dev);
> > + if (rc == -EEXIST) {
> > + /*
> > + * "can't" happen, but this error code means
> > + * something to the caller, so translate it.
> > + */
> > + rc = -ENXIO;
>
> "can't" should translate to a WARN IMHO.
WARNs are effectively BUG_ONs in some sites. I wouldn't crash the
kernel for this, a driver load error is sufficient.
>
> > + }
> > + put_device(&port->dev);
> > + }
> > +
> > + put_device(&parent_port->dev);
> > + return rc;
> > +}
> > +
> > +int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > +{
> > + struct device *dev = &cxlmd->dev;
> > + struct device *iter;
> > + int rc;
> > +
> > + rc = devm_add_action_or_reset(&cxlmd->dev, cxl_remove_ep, cxlmd);
> > + if (rc)
> > + return rc;
> > +
> > + /*
> > + * Scan for and add all cxl_ports in this device's ancestry.
> > + * Repeat until no more ports are added. Abort if a port add
> > + * attempt fails.
> > + */
> > +retry:
> > + for (iter = dev; iter; iter = grandparent(iter)) {
> > + struct device *dport_dev = grandparent(iter);
> > + struct device *uport_dev;
> > + struct cxl_port *port;
> > +
> > + if (!dport_dev)
> > + break;
> > + uport_dev = dport_dev->parent;
> > + dev_dbg(dev, "scan: iter: %s dport_dev: %s parent: %s\n",
> > + dev_name(iter), dev_name(dport_dev),
> > + uport_dev ? dev_name(uport_dev) : "'none'");
> > + if (!uport_dev) {
> > + dev_warn(dev, "unexpected topology, no parent for %s\n",
> > + dev_name(dport_dev));
> > + rc = -ENXIO;
> > + break;
> > + }
> > +
> > + port = find_cxl_port(dport_dev);
> > + if (port) {
> > + dev_dbg(&cxlmd->dev,
> > + "found already registered port %s:%s\n",
> > + dev_name(&port->dev), dev_name(port->uport));
> > + rc = cxl_add_ep(port, &cxlmd->dev);
> > +
> > + /*
> > + * If the endpoint already exists in the port's list,
> > + * that's ok, it was added on a previous pass.
> > + * Otherwise, retry in add_port_register_ep() after
> > + * taking the parent_port lock as the current port may
> > + * be being reaped.
> > + */
> > + if (rc && rc != -EEXIST) {
> > + put_device(&port->dev);
> > + return rc;
> > + }
>
> I could use an explanation on how an endpoint could have been added on a
> previous pass. It seems like !list_empty(&ep->list) here would be a bug.
There's a goto retry after every successful port addition, so
subsequent iterations might find some already completed work.
>
> > +
> > + if (is_cxl_port(port->dev.parent) &&
> > + !is_cxl_root(to_cxl_port(port->dev.parent))) {
> > + put_device(&port->dev);
> > + continue;
> > + }
> > +
> > + put_device(&port->dev);
> > + break;
> > + }
> > +
> > + rc = add_port_register_ep(cxlmd, uport_dev, dport_dev);
> > + /* port missing, try to add parent */
> > + if (rc == -EAGAIN)
> > + continue;
> > + /* failed to add ep or port */
> > + if (rc)
> > + return rc;
> > + /* port added, new descendants possible, start over */
> > + goto retry;
> > + }
> > +
> > + return 0;
> > +}
> > +EXPORT_SYMBOL_NS_GPL(devm_cxl_enumerate_ports, CXL);
> > +
> > +struct cxl_port *cxl_mem_find_port(struct cxl_memdev *cxlmd)
> > +{
> > + return find_cxl_port(grandparent(&cxlmd->dev));
> > +}
> > +EXPORT_SYMBOL_NS_GPL(cxl_mem_find_port, CXL);
> > +
> > +struct cxl_dport *cxl_find_dport_by_dev(struct cxl_port *port,
> > + const struct device *dev)
> > +{
> > + struct cxl_dport *dport;
> > +
> > + cxl_device_lock(&port->dev);
> > + list_for_each_entry(dport, &port->dports, list)
> > + if (dport->dport == dev) {
> > + cxl_device_unlock(&port->dev);
> > + return dport;
> > + }
> > +
> > + cxl_device_unlock(&port->dev);
> > + return NULL;
> > +}
> > +EXPORT_SYMBOL_NS_GPL(cxl_find_dport_by_dev, CXL);
> > +
> > static int decoder_populate_targets(struct cxl_decoder *cxld,
> > struct cxl_port *port, int *target_map)
> > {
> > diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> > index 7c714e559e95..b71d40b68ccd 100644
> > --- a/drivers/cxl/cxl.h
> > +++ b/drivers/cxl/cxl.h
> > @@ -262,8 +262,10 @@ struct cxl_nvdimm {
> > * @uport: PCI or platform device implementing the upstream port capability
> > * @id: id for port device-name
> > * @dports: cxl_dport instances referenced by decoders
> > + * @endpoints: cxl_ep instances, endpoints that are a descendant of this port
> > * @decoder_ida: allocator for decoder ids
> > * @component_reg_phys: component register capability base address (optional)
> > + * @dead: last ep has been removed, force port re-creation
> > * @depth: How deep this port is relative to the root. depth 0 is the root.
> > */
> > struct cxl_port {
> > @@ -271,8 +273,10 @@ struct cxl_port {
> > struct device *uport;
> > int id;
> > struct list_head dports;
> > + struct list_head endpoints;
> > struct ida decoder_ida;
> > resource_size_t component_reg_phys;
> > + bool dead;
> > unsigned int depth;
> > };
> >
> > @@ -292,6 +296,16 @@ struct cxl_dport {
> > struct list_head list;
> > };
> >
> > +/**
> > + * struct cxl_ep - track an endpoint's interest in a port
> > + * @ep: device that hosts a generic CXL endpoint (expander or accelerator)
> > + * @list: node on port->endpoints list
> > + */
> > +struct cxl_ep {
> > + struct device *ep;
> > + struct list_head list;
> > +};
> > +
> > static inline bool is_cxl_root(struct cxl_port *port)
> > {
> > return port->uport == port->dev.parent;
> > @@ -307,9 +321,15 @@ struct cxl_port *devm_cxl_add_port(struct device *host, struct device *uport,
> > resource_size_t component_reg_phys,
> > struct cxl_port *parent_port);
> > struct cxl_port *find_cxl_root(struct device *dev);
> > +int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd);
> > +int cxl_bus_rescan(void);
> > +
> > struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
> > struct device *dport, int port_id,
> > resource_size_t component_reg_phys);
> > +struct cxl_dport *cxl_find_dport_by_dev(struct cxl_port *port,
> > + const struct device *dev);
> > +
> > struct cxl_decoder *to_cxl_decoder(struct device *dev);
> > bool is_root_decoder(struct device *dev);
> > bool is_cxl_decoder(struct device *dev);
> >
