Re: [PATCH v3 10/20] cxl: indicate probe deferral

[Alejandro Lucero Palau <alucerop@amd.com>]

On 9/17/24 04:31, Li, Ming4 wrote:
> On 9/16/2024 4:24 PM, Alejandro Lucero Palau wrote:
>> On 9/10/24 07:37, Li, Ming4 wrote:
>>> On 9/7/2024 4:18 PM, alejandro.lucero-palau@amd.com wrote:
>>>> From: Alejandro Lucero <alucerop@amd.com>
>>>>
>>>> The first stop for a CXL accelerator driver that wants to establish new
>>>> CXL.mem regions is to register a 'struct cxl_memdev. That kicks off
>>>> cxl_mem_probe() to enumerate all 'struct cxl_port' instances in the
>>>> topology up to the root.
>>>>
>>>> If the root driver has not attached yet the expectation is that the
>>>> driver waits until that link is established. The common cxl_pci_driver
>>>> has reason to keep the 'struct cxl_memdev' device attached to the bus
>>>> until the root driver attaches. An accelerator may want to instead defer
>>>> probing until CXL resources can be acquired.
>>>>
>>>> Use the @endpoint attribute of a 'struct cxl_memdev' to convey when
>>>> accelerator driver probing should be deferred vs failed. Provide that
>>>> indication via a new cxl_acquire_endpoint() API that can retrieve the
>>>> probe status of the memdev.
>>>>
>>>> Based on https://lore.kernel.org/linux-cxl/168592155270.1948938.11536845108449547920.stgit@dwillia2-xfh.jf.intel.com/
>>>>
>>>> Signed-off-by: Alejandro Lucero <alucerop@amd.com>
>>>> Co-developed-by: Dan Williams <dan.j.williams@intel.com>
>>>> ---
>>>>    drivers/cxl/core/memdev.c | 67 +++++++++++++++++++++++++++++++++++++++
>>>>    drivers/cxl/core/port.c   |  2 +-
>>>>    drivers/cxl/mem.c         |  4 ++-
>>>>    include/linux/cxl/cxl.h   |  2 ++
>>>>    4 files changed, 73 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/cxl/core/memdev.c b/drivers/cxl/core/memdev.c
>>>> index 5f8418620b70..d4406cf3ed32 100644
>>>> --- a/drivers/cxl/core/memdev.c
>>>> +++ b/drivers/cxl/core/memdev.c
>>>> @@ -5,6 +5,7 @@
>>>>    #include <linux/io-64-nonatomic-lo-hi.h>
>>>>    #include <linux/firmware.h>
>>>>    #include <linux/device.h>
>>>> +#include <linux/delay.h>
>>>>    #include <linux/slab.h>
>>>>    #include <linux/idr.h>
>>>>    #include <linux/pci.h>
>>>> @@ -23,6 +24,8 @@ static DECLARE_RWSEM(cxl_memdev_rwsem);
>>>>    static int cxl_mem_major;
>>>>    static DEFINE_IDA(cxl_memdev_ida);
>>>>    +static unsigned short endpoint_ready_timeout = HZ;
>>>> +
>>>>    static void cxl_memdev_release(struct device *dev)
>>>>    {
>>>>        struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
>>>> @@ -1163,6 +1166,70 @@ struct cxl_memdev *devm_cxl_add_memdev(struct device *host,
>>>>    }
>>>>    EXPORT_SYMBOL_NS_GPL(devm_cxl_add_memdev, CXL);
>>>>    +/*
>>>> + * Try to get a locked reference on a memdev's CXL port topology
>>>> + * connection. Be careful to observe when cxl_mem_probe() has deposited
>>>> + * a probe deferral awaiting the arrival of the CXL root driver.
>>>> + */
>>>> +struct cxl_port *cxl_acquire_endpoint(struct cxl_memdev *cxlmd)
>>>> +{
>>>> +    struct cxl_port *endpoint;
>>>> +    unsigned long timeout;
>>>> +    int rc = -ENXIO;
>>>> +
>>>> +    /*
>>>> +     * A memdev creation triggers ports creation through the kernel
>>>> +     * device object model. An endpoint port could not be created yet
>>>> +     * but coming. Wait here for a gentle space of time for ensuring
>>>> +     * and endpoint port not there is due to some error and not because
>>>> +     * the race described.
>>>> +     *
>>>> +     * Note this is a similar case this function is implemented for, but
>>>> +     * instead of the race with the root port, this is against its own
>>>> +     * endpoint port.
>>>> +     */
>>>> +    timeout = jiffies + endpoint_ready_timeout;
>>>> +    do {
>>>> +        device_lock(&cxlmd->dev);
>>>> +        endpoint = cxlmd->endpoint;
>>>> +        if (endpoint)
>>>> +            break;
>>>> +        device_unlock(&cxlmd->dev);
>>>> +        if (msleep_interruptible(100)) {
>>>> +            device_lock(&cxlmd->dev);
>>>> +            break;
>>> Can exit directly. not need to hold the lock of cxlmd->dev then break.
>>
>> Not sure if it is safe to do device_unlock twice, but even if so, it looks better to my eyes to get the lock or if not to add another error path.
>>
> why device_unlock will be called twice? directly return the value of rc like below if the sleep is interrupted.
>
>      if (msleep_interruptible(100))
>
>              return ERR_PTR(rc);
>
>

You are right.


>>
>>>> +        }
>>>> +    } while (!time_after(jiffies, timeout));
> Another issue I noticed is that above loop will not hold the device lock if timeout happened(without msleep interrupted), but below "goto err" will call device_unlock() for the device.
>
> I think below 'if (!endpoint)' can also return the value of rc. Combine above changes, I think the code should be:
>
>      do {
>
>          ......
>
>          if (msleep_interruptible(100))
>
>                  break;
>
>      } while (!time_after(jiffies, timeout));
>
>      if (!endpoint)
>
>                  return ERR_PTR(rc);
>
>
> Does it make more sense?


Right again.

I can see it now.

Thank you!


>
>>>> +
>>>> +    if (!endpoint)
>>>> +        goto err;
>>>> +
>>>> +    if (IS_ERR(endpoint)) {
>>>> +        rc = PTR_ERR(endpoint);
>>>> +        goto err;
>>>> +    }
>>>> +
>>>> +    device_lock(&endpoint->dev);
>>>> +    if (!endpoint->dev.driver)
>>>> +        goto err_endpoint;
>>>> +
>>>> +    return endpoint;
>>>> +
>>>> +err_endpoint:
>>>> +    device_unlock(&endpoint->dev);
>>>> +err:
>>>> +    device_unlock(&cxlmd->dev);
>>>> +    return ERR_PTR(rc);
>>>> +}
>>>> +EXPORT_SYMBOL_NS(cxl_acquire_endpoint, CXL);
>>>> +
>>>> +void cxl_release_endpoint(struct cxl_memdev *cxlmd, struct cxl_port *endpoint)
>>>> +{
>>>> +    device_unlock(&endpoint->dev);
>>>> +    device_unlock(&cxlmd->dev);
>>>> +}
>>>> +EXPORT_SYMBOL_NS(cxl_release_endpoint, CXL);
>>>> +
>>>>    static void sanitize_teardown_notifier(void *data)
>>>>    {
>>>>        struct cxl_memdev_state *mds = data;
>>>> diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
>>>> index 39b20ddd0296..ca2c993faa9c 100644
>>>> --- a/drivers/cxl/core/port.c
>>>> +++ b/drivers/cxl/core/port.c
>>>> @@ -1554,7 +1554,7 @@ static int add_port_attach_ep(struct cxl_memdev *cxlmd,
>>>>             */
>>>>            dev_dbg(&cxlmd->dev, "%s is a root dport\n",
>>>>                dev_name(dport_dev));
>>>> -        return -ENXIO;
>>>> +        return -EPROBE_DEFER;
>>>>        }
>>>>          parent_port = find_cxl_port(dparent, &parent_dport);
>>>> diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
>>>> index 5c7ad230bccb..56fd7a100c2f 100644
>>>> --- a/drivers/cxl/mem.c
>>>> +++ b/drivers/cxl/mem.c
>>>> @@ -145,8 +145,10 @@ static int cxl_mem_probe(struct device *dev)
>>>>            return rc;
>>>>          rc = devm_cxl_enumerate_ports(cxlmd);
>>>> -    if (rc)
>>>> +    if (rc) {
>>>> +        cxlmd->endpoint = ERR_PTR(rc);
>>>>            return rc;
>>>> +    }
>>>>          parent_port = cxl_mem_find_port(cxlmd, &dport);
>>>>        if (!parent_port) {
>>>> diff --git a/include/linux/cxl/cxl.h b/include/linux/cxl/cxl.h
>>>> index fc0859f841dc..7e4580fb8659 100644
>>>> --- a/include/linux/cxl/cxl.h
>>>> +++ b/include/linux/cxl/cxl.h
>>>> @@ -57,4 +57,6 @@ int cxl_release_resource(struct cxl_dev_state *cxlds, enum cxl_resource type);
>>>>    void cxl_set_media_ready(struct cxl_dev_state *cxlds);
>>>>    struct cxl_memdev *devm_cxl_add_memdev(struct device *host,
>>>>                           struct cxl_dev_state *cxlds);
>>>> +struct cxl_port *cxl_acquire_endpoint(struct cxl_memdev *cxlmd);
>>>> +void cxl_release_endpoint(struct cxl_memdev *cxlmd, struct cxl_port *endpoint);
>>>>    #endif