On Tue, 14 Jun 2022 20:29:40 +0800
Kai Ye via Linux-accelerators <linux-accelerators@xxxxxxxxxxxxxxxx> wrote:
> Define the device isolation strategy by the device driver. if the
> AER error frequency exceeds the value of setting for a certain
> period of time, The device will not be available in user space. The VF
> device use the PF device isolation strategy. All the hardware errors
> are processed by PF driver.
>
> Signed-off-by: Kai Ye <yekai13@xxxxxxxxxx>
I'll try and avoid duplicating Greg's feedback but might well overlap a bit!
> ---
> drivers/crypto/hisilicon/qm.c | 157 +++++++++++++++++++++++++++++++---
> include/linux/hisi_acc_qm.h | 9 ++
> 2 files changed, 152 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/crypto/hisilicon/qm.c b/drivers/crypto/hisilicon/qm.c
> index ad83c194d664..47c41fa52693 100644
> --- a/drivers/crypto/hisilicon/qm.c
> +++ b/drivers/crypto/hisilicon/qm.c
> @@ -12,7 +12,6 @@
> #include <linux/pm_runtime.h>
> #include <linux/seq_file.h>
> #include <linux/slab.h>
> -#include <linux/uacce.h>
I assume you do this because you are now relying on hisi_acc_qm.h including
uacce.h? Generally it is better to include most headers that we use
directly so it this still uses stuff from uacce.h then keep the include.
> #include <linux/uaccess.h>
> #include <uapi/misc/uacce/hisi_qm.h>
> #include <linux/hisi_acc_qm.h>
> @@ -417,6 +416,16 @@ struct hisi_qm_resource {
> struct list_head list;
> };
>
> +/**
> + * struct qm_hw_err - structure of describes the device err
> + * @list: hardware error list
> + * @tick_stamp: timestamp when the error occurred
tick? Perhaps just call it timestamp if that is what it is...
> + */
> +struct qm_hw_err {
> + struct list_head list;
> + unsigned long long tick_stamp;
> +};
> +
>
> +/**
> + * qm_hw_err_isolate() - Try to isolate the uacce device with its VFs
> + * @qm: The qm which we want to configure.
> + *
> + * according to user's configuration of isolation strategy. Warning: this
Rewrite to make it full sentence.
> + * API should be called while there is no user on the device, or the users
> + * on this device are suspended by slot resetting preparation of PCI AER.
> + */
> +static int qm_hw_err_isolate(struct hisi_qm *qm)
> +{
> + struct qm_hw_err *err, *tmp, *hw_err;
> + struct qm_err_isolate *isolate;
> + u32 count = 0;
> +
> + isolate = &qm->isolate_data;
> +
> +#define SECONDS_PER_HOUR 3600
> +
> + /* All the hw errs are processed by PF driver */
> + if (qm->uacce->is_vf || atomic_read(&isolate->is_isolate) ||
> + !isolate->hw_err_isolate_hz)
> + return 0;
> +
> + hw_err = kzalloc(sizeof(*hw_err), GFP_ATOMIC);
> + if (!hw_err)
> + return -ENOMEM;
blank line here to separate error handling from next bit of code.
> + hw_err->tick_stamp = jiffies;
> + list_for_each_entry_safe(err, tmp, &qm->uacce_hw_errs, list) {
These are ordered (I think). Could take advantage of that by
maintaining count of elements in parallel to the list then walking
list in right direction + stop when you reach last one to need
deleting.
> + if ((hw_err->tick_stamp - err->tick_stamp) / HZ >
> + SECONDS_PER_HOUR) {
> + list_del(&err->list);
> + kfree(err);
> + } else {
> + count++;
> + }
> + }
> + list_add(&hw_err->list, &qm->uacce_hw_errs);
> +
> + if (count >= isolate->hw_err_isolate_hz)
> + atomic_set(&isolate->is_isolate, 1);
> +
> + return 0;
> +}
> +
...
> +static int hisi_qm_isolate_strategy_write(struct uacce_device *uacce,
> + const char *buf)
> +{
> + struct hisi_qm *qm = uacce->priv;
> + unsigned long val = 0;
> +
> +#define MAX_ISOLATE_STRATEGY 65535
> +
> + if (atomic_read(&qm->uacce_ref))
> + return -EBUSY;
> +
> + /* must be set by PF */
> + if (atomic_read(&qm->isolate_data.is_isolate) || uacce->is_vf)
Why is the file visible on the vf? Hide it or don't register it for vfs.
> + return -EINVAL;
> +
> + if (kstrtoul(buf, 0, &val) < 0)
> + return -EINVAL;
> +
> + if (val > MAX_ISOLATE_STRATEGY)
> + return -EINVAL;
> +
> + qm->isolate_data.hw_err_isolate_hz = val;
> + dev_info(&qm->pdev->dev,
> + "the value of isolate_strategy is set to %lu.\n", val);
This is just noise in the log. If someone wants to check they should read
the sysfs file back and it reflect the new state.
> +
> + return 0;
> +}
> +
...
> static int qm_alloc_uacce(struct hisi_qm *qm)
> {
> struct pci_dev *pdev = qm->pdev;
> @@ -3433,6 +3554,7 @@ static int qm_alloc_uacce(struct hisi_qm *qm)
> };
> int ret;
>
> + INIT_LIST_HEAD(&qm->uacce_hw_errs);
> ret = strscpy(interface.name, dev_driver_string(&pdev->dev),
> sizeof(interface.name));
> if (ret < 0)
> @@ -3446,8 +3568,7 @@ static int qm_alloc_uacce(struct hisi_qm *qm)
> qm->use_sva = true;
> } else {
> /* only consider sva case */
> - uacce_remove(uacce);
> - qm->uacce = NULL;
> + qm_remove_uacce(qm);
> return -EINVAL;
> }
>
> @@ -5109,6 +5230,12 @@ static int qm_controller_reset_prepare(struct hisi_qm *qm)
> return ret;
> }
>
> + if (qm->use_sva) {
> + ret = qm_hw_err_isolate(qm);
> + if (ret)
> + pci_err(pdev, "failed to isolate hw err!\n");
> + }
> +
> ret = qm_wait_vf_prepare_finish(qm);
> if (ret)
> pci_err(pdev, "failed to stop by vfs in soft reset!\n");
> @@ -5436,19 +5563,24 @@ static int qm_controller_reset(struct hisi_qm *qm)
> ret = qm_soft_reset(qm);
> if (ret) {
> pci_err(pdev, "Controller reset failed (%d)\n", ret);
This is printed below as well - probably best to drop this one and then you
can remove the brackets as well.
> - qm_reset_bit_clear(qm);
> - return ret;
> + goto err_reset;
> }
>
> ret = qm_controller_reset_done(qm);
> - if (ret) {
> - qm_reset_bit_clear(qm);
> - return ret;
> - }
> + if (ret)
> + goto err_reset;
>
> pci_info(pdev, "Controller reset complete\n");
> -
Avoid noise via white space changes like this. The white space was
good and generally don't change white space in a patch doing anything else.
> return 0;
> +
> +err_reset:
> + pci_err(pdev, "Controller reset failed (%d)\n", ret);
> + qm_reset_bit_clear(qm);
> +
> + /* if resetting fails, isolate the device */
> + if (qm->use_sva && !qm->uacce->is_vf)
> + atomic_set(&qm->isolate_data.is_isolate, 1);
> + return ret;
> }
>
> /**
> @@ -6246,10 +6378,7 @@ int hisi_qm_init(struct hisi_qm *qm)
> err_free_qm_memory:
> hisi_qm_memory_uninit(qm);
> err_alloc_uacce:
> - if (qm->use_sva) {
> - uacce_remove(qm->uacce);
> - qm->uacce = NULL;
> - }
> + qm_remove_uacce(qm);
> err_irq_register:
> qm_irq_unregister(qm);
> err_pci_init:
