Re: [RFC PATCH v6 04/12] cxl/memscrub: Add CXL device patrol scrub control feature

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

On Thu, 15 Feb 2024 19:14:46 +0800
<shiju.jose@huawei.com> wrote:

> From: Shiju Jose <shiju.jose@huawei.com>
> 
> CXL spec 3.1 section 8.2.9.9.11.1 describes the device patrol scrub control
> feature. The device patrol scrub proactively locates and makes corrections
> to errors in regular cycle. The patrol scrub control allows the request to
> configure patrol scrub input configurations.
> 
> The patrol scrub control allows the requester to specify the number of
> hours for which the patrol scrub cycles must be completed, provided that
> the requested number is not less than the minimum number of hours for the
> patrol scrub cycle that the device is capable of. In addition, the patrol
> scrub controls allow the host to disable and enable the feature in case
> disabling of the feature is needed for other purposes such as
> performance-aware operations which require the background operations to be
> turned off.
> 
> Signed-off-by: Shiju Jose <shiju.jose@huawei.com>
Hi Shiju

Various comments inline. Sorry I didn't get to this on earlier versions!

Jonathan

> ---
>  drivers/cxl/Kconfig         |  17 +++
>  drivers/cxl/core/Makefile   |   1 +
>  drivers/cxl/core/memscrub.c | 266 ++++++++++++++++++++++++++++++++++++
>  drivers/cxl/cxlmem.h        |   8 ++
>  drivers/cxl/pci.c           |   5 +
>  5 files changed, 297 insertions(+)
>  create mode 100644 drivers/cxl/core/memscrub.c
> 
> diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
> index 67998dbd1d46..873bdda5db32 100644
> --- a/drivers/cxl/Kconfig
> +++ b/drivers/cxl/Kconfig
> @@ -157,4 +157,21 @@ config CXL_PMU
>  	  monitoring units and provide standard perf based interfaces.
>  
>  	  If unsure say 'm'.
> +
> +config CXL_SCRUB
> +	bool "CXL: Memory scrub feature"
> +	depends on CXL_PCI
> +	depends on CXL_MEM
> +	help
> +	  The CXL memory scrub control is an optional feature allows host to
> +	  control the scrub configurations of CXL Type 3 devices, which
> +	  support patrol scrub and/or DDR5 ECS(Error Check Scrub).
> +
> +	  Say 'y/n' to enable/disable the CXL memory scrub driver that will
> +	  attach to CXL.mem devices for memory scrub control feature. See
> +	  sections 8.2.9.9.11.1 and 8.2.9.9.11.2 in the CXL 3.1 specification
> +	  for a detailed description of CXL memory scrub control features.
> +
> +	  If unsure say 'n'.

No need for negative here I think.  It's a reasonable thing to turn on
and hardware should provide minimum guarantees that stop it being dangerous.


> +
>  endif
> diff --git a/drivers/cxl/core/Makefile b/drivers/cxl/core/Makefile
> index 9259bcc6773c..e0fc814c3983 100644
> --- a/drivers/cxl/core/Makefile
> +++ b/drivers/cxl/core/Makefile
> @@ -16,3 +16,4 @@ cxl_core-y += pmu.o
>  cxl_core-y += cdat.o
>  cxl_core-$(CONFIG_TRACING) += trace.o
>  cxl_core-$(CONFIG_CXL_REGION) += region.o
> +cxl_core-$(CONFIG_CXL_SCRUB) += memscrub.o
> diff --git a/drivers/cxl/core/memscrub.c b/drivers/cxl/core/memscrub.c
> new file mode 100644
> index 000000000000..be8d9a9743eb
> --- /dev/null
> +++ b/drivers/cxl/core/memscrub.c
> @@ -0,0 +1,266 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later
> +/*
> + * cxl_memscrub.c - CXL memory scrub driver

No point in a filename (which incidentally is wrong ;)
in a file (because they are often wrong and add nothing)

> + *
> + * Copyright (c) 2023 HiSilicon Limited.

2024 probably appropriate now.

> + *
> + *  - Provides functions to configure patrol scrub
> + *    feature of the CXL memory devices.
Very short line wrap.

> + */
> +
> +#define pr_fmt(fmt)	"CXL_MEM_SCRUB: " fmt
> +
> +#include <cxlmem.h>
> +
> +/* CXL memory scrub feature common definitions */
> +#define CXL_SCRUB_MAX_ATTR_RANGE_LENGTH	128
> +
> +static int cxl_mem_get_supported_feature_entry(struct cxl_memdev *cxlmd, const uuid_t *feat_uuid,
> +					       struct cxl_mbox_supp_feat_entry *feat_entry_out)
> +{
> +	struct cxl_mbox_get_supp_feats_out *feats_out __free(kvfree) = NULL;
> +	struct cxl_mbox_supp_feat_entry *feat_entry;
> +	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> +	struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds);
> +	struct cxl_mbox_get_supp_feats_in pi;
> +	int feat_index, count;
> +	int nentries;
> +	int ret;
> +
> +	feat_index = 0;
> +	pi.count = sizeof(struct cxl_mbox_get_supp_feats_out) +
> +			  sizeof(struct cxl_mbox_supp_feat_entry);
> +	feats_out = kvmalloc(pi.count, GFP_KERNEL);

Not very big. kmalloc should be fine I think.

> +	if (!feats_out)
> +		return -ENOMEM;
> +
> +	do {
> +		pi.start_index = feat_index;
> +		memset(feats_out, 0, pi.count);
> +		ret = cxl_get_supported_features(mds, &pi, feats_out);
> +		if (ret)
> +			return ret;
> +
> +		nentries = feats_out->entries;
> +		if (!nentries)
> +			break;

I'd return here.

> +
> +		/* Check CXL memdev supports the feature */
> +		feat_entry = (void *)feats_out->feat_entries;

Cast is odd.  I think type is correct already.  


> +		for (count = 0; count < nentries; count++, feat_entry++) {
> +			if (uuid_equal(&feat_entry->uuid, feat_uuid)) {
> +				memcpy(feat_entry_out, feat_entry, sizeof(*feat_entry_out));
Long line.  add a break after feat_entry

> +				return 0;
> +			}
> +		}
> +		feat_index += nentries;
> +	} while (nentries);

Given exit on !entries don't think you can exit via normal while condition path. So
make this while (true)

> +
> +	return -ENOTSUPP;
> +}
> +
> +/* CXL memory patrol scrub control definitions */
> +#define CXL_MEMDEV_PS_GET_FEAT_VERSION	0x01
> +#define CXL_MEMDEV_PS_SET_FEAT_VERSION	0x01
> +
> +static const uuid_t cxl_patrol_scrub_uuid =
> +	UUID_INIT(0x96dad7d6, 0xfde8, 0x482b, 0xa7, 0x33, 0x75, 0x77, 0x4e,     \
> +		  0x06, 0xdb, 0x8a);
> +
> +/* CXL memory patrol scrub control functions */
> +struct cxl_patrol_scrub_context {
> +	struct device *dev;
> +	u16 get_feat_size;
> +	u16 set_feat_size;
> +	bool scrub_cycle_changeable;
> +};
> +
> +/**
> + * struct cxl_memdev_ps_params - CXL memory patrol scrub parameter data structure.
> + * @enable:     [IN] enable(1)/disable(0) patrol scrub.

In and out I think.

> + * @scrub_cycle_changeable: [OUT] scrub cycle attribute of patrol scrub is changeable.
> + * @rate:       [IN] Requested patrol scrub cycle in hours.
> + *              [OUT] Current patrol scrub cycle in hours.
> + * @min_rate:[OUT] minimum patrol scrub cycle, in hours, supported.
> + * @rate_avail:[OUT] Supported patrol scrub cycle in hours.
> + */
> +struct cxl_memdev_ps_params {
> +	bool enable;
> +	bool scrub_cycle_changeable;
> +	u16 rate;
> +	u16 min_rate;
> +	char rate_avail[CXL_SCRUB_MAX_ATTR_RANGE_LENGTH];
> +};
> +
> +enum {
> +	CXL_MEMDEV_PS_PARAM_ENABLE,
> +	CXL_MEMDEV_PS_PARAM_RATE,
> +};
> +
> +#define	CXL_MEMDEV_PS_SCRUB_CYCLE_CHANGE_CAP_MASK	BIT(0)
> +#define	CXL_MEMDEV_PS_SCRUB_CYCLE_REALTIME_REPORT_CAP_MASK	BIT(1)
> +#define	CXL_MEMDEV_PS_CUR_SCRUB_CYCLE_MASK	GENMASK(7, 0)
> +#define	CXL_MEMDEV_PS_MIN_SCRUB_CYCLE_MASK	GENMASK(15, 8)
> +#define	CXL_MEMDEV_PS_FLAG_ENABLED_MASK	BIT(0)
> +
> +struct cxl_memdev_ps_feat_read_attrs {
> +	u8 scrub_cycle_cap;
> +	__le16 scrub_cycle;
> +	u8 scrub_flags;
> +}  __packed;
> +
> +struct cxl_memdev_ps_set_feat_pi {
> +	struct cxl_mbox_set_feat_in pi;

Maybe rename this in earlier patch to make it clear it is a header.
Not sure why it is called pi vs attrs term used for read.


> +	u8 scrub_cycle_hr;
> +	u8 scrub_flags;
> +}  __packed;
> +
> +static int cxl_mem_ps_get_attrs(struct device *dev,
> +				struct cxl_memdev_ps_params *params)
> +{
> +	struct cxl_memdev_ps_feat_read_attrs *rd_attrs __free(kvfree) = NULL;
> +	struct cxl_mbox_get_feat_in pi = {
> +		.uuid = cxl_patrol_scrub_uuid,
> +		.offset = 0,
> +		.count = sizeof(struct cxl_memdev_ps_feat_read_attrs),
> +		.selection = CXL_GET_FEAT_SEL_CURRENT_VALUE,
> +	};
> +	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
> +	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> +	struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds);
> +	int ret;
> +
> +	if (!mds)
> +		return -EFAULT;
> +
> +	rd_attrs = kvmalloc(pi.count, GFP_KERNEL);

Small so I don't see need for kvmalloc.
In general that might not be true for a feature, but in this case
we know it is.


> +	if (!rd_attrs)
> +		return -ENOMEM;
> +
> +	ret = cxl_get_feature(mds, &pi, rd_attrs);
> +	if (ret) {
> +		params->enable = 0;
> +		params->rate = 0;

The cxl_get_feature() should not have side effects on failure to read. 
As such, these parameters should be left in original state if there is
a problem. Initialize them to these states and we should be fine unless
a read succeeds in updating them.

> +		snprintf(params->rate_avail, CXL_SCRUB_MAX_ATTR_RANGE_LENGTH,
> +			"Unavailable");
> +		return ret;
> +	}
> +	params->scrub_cycle_changeable = FIELD_GET(CXL_MEMDEV_PS_SCRUB_CYCLE_CHANGE_CAP_MASK,
> +						   rd_attrs->scrub_cycle_cap);
> +	params->enable = FIELD_GET(CXL_MEMDEV_PS_FLAG_ENABLED_MASK,
> +				   rd_attrs->scrub_flags);
> +	params->rate = FIELD_GET(CXL_MEMDEV_PS_CUR_SCRUB_CYCLE_MASK,
> +				 rd_attrs->scrub_cycle);
> +	params->min_rate  = FIELD_GET(CXL_MEMDEV_PS_MIN_SCRUB_CYCLE_MASK,
> +				      rd_attrs->scrub_cycle);
> +	snprintf(params->rate_avail, CXL_SCRUB_MAX_ATTR_RANGE_LENGTH,
> +		 "Minimum scrub cycle = %d hour", params->min_rate);
> +
> +	return 0;
> +}
> +
> +static int __maybe_unused
> +cxl_mem_ps_set_attrs(struct device *dev, struct cxl_memdev_ps_params *params,
> +		     u8 param_type)
> +{
> +	struct cxl_memdev_ps_set_feat_pi set_pi = {
> +		.pi.uuid = cxl_patrol_scrub_uuid,
> +		.pi.flags = CXL_SET_FEAT_FLAG_MOD_VALUE_SAVED_ACROSS_RESET |
> +			    CXL_SET_FEAT_FLAG_FULL_DATA_TRANSFER,
> +		.pi.offset = 0,
> +		.pi.version = CXL_MEMDEV_PS_SET_FEAT_VERSION,
> +	};
> +	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
> +	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> +	struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds);
> +	struct cxl_memdev_ps_params rd_params;
> +	int ret;
> +
> +	if (!mds)
> +		return -EFAULT;
> +
> +	ret = cxl_mem_ps_get_attrs(dev, &rd_params);
> +	if (ret) {
> +		dev_err(dev, "Get cxlmemdev patrol scrub params fail ret=%d\n",
> +			ret);
> +		return ret;
> +	}
> +
> +	switch (param_type) {
> +	case CXL_MEMDEV_PS_PARAM_ENABLE:
> +		set_pi.scrub_flags = FIELD_PREP(CXL_MEMDEV_PS_FLAG_ENABLED_MASK,
> +						   params->enable);
> +		set_pi.scrub_cycle_hr = FIELD_PREP(CXL_MEMDEV_PS_CUR_SCRUB_CYCLE_MASK,
> +						      rd_params.rate);
> +		break;
> +	case CXL_MEMDEV_PS_PARAM_RATE:
> +		if (params->rate < rd_params.min_rate) {
> +			dev_err(dev, "Invalid CXL patrol scrub cycle(%d) to set\n",
> +				params->rate);
> +			dev_err(dev, "Minimum supported CXL patrol scrub cycle in hour %d\n",
> +			       params->min_rate);
> +			return -EINVAL;
> +		}
> +		set_pi.scrub_cycle_hr = FIELD_PREP(CXL_MEMDEV_PS_CUR_SCRUB_CYCLE_MASK,
> +						      params->rate);
> +		set_pi.scrub_flags = FIELD_PREP(CXL_MEMDEV_PS_FLAG_ENABLED_MASK,
> +						   rd_params.enable);
> +		break;
> +	default:
> +		dev_err(dev, "Invalid CXL patrol scrub parameter to set\n");
> +		return -EINVAL;
> +	}
> +
> +	ret = cxl_set_feature(mds, &set_pi, sizeof(set_pi));
> +	if (ret) {
> +		dev_err(dev, "CXL patrol scrub set feature fail ret=%d\n",
> +			ret);
> +		return ret;
> +	}
> +
> +	/* Verify attribute set successfully */
Why? Is there a specification defined reason it might not give an error return
but still fail to set the attribute? (rounding or similar perhaps?)
If so add a comment here. If not drop this check.

> +	if (param_type == CXL_MEMDEV_PS_PARAM_RATE) {
> +		ret = cxl_mem_ps_get_attrs(dev, &rd_params);
> +		if (ret) {
> +			dev_err(dev, "Get cxlmemdev patrol scrub params fail ret=%d\n", ret);
> +			return ret;
> +		}
> +		if (rd_params.rate != params->rate)
> +			return -EFAULT;
> +	}
> +
> +	return 0;
> +}
> +
> +int cxl_mem_patrol_scrub_init(struct cxl_memdev *cxlmd)
> +{
> +	struct cxl_patrol_scrub_context *cxl_ps_ctx;
> +	struct cxl_mbox_supp_feat_entry feat_entry;
> +	struct cxl_memdev_ps_params params;
> +	int ret;
> +
> +	ret = cxl_mem_get_supported_feature_entry(cxlmd, &cxl_patrol_scrub_uuid,
> +						  &feat_entry);
> +	if (ret < 0)
> +		return ret;
> +
> +	if (!(feat_entry.attr_flags & CXL_FEAT_ENTRY_FLAG_CHANGABLE))
> +		return -ENOTSUPP;
> +
> +	cxl_ps_ctx = devm_kzalloc(&cxlmd->dev, sizeof(*cxl_ps_ctx), GFP_KERNEL);
> +	if (!cxl_ps_ctx)
> +		return -ENOMEM;
> +
> +	cxl_ps_ctx->get_feat_size = feat_entry.get_feat_size;
> +	cxl_ps_ctx->set_feat_size = feat_entry.set_feat_size;
> +	ret = cxl_mem_ps_get_attrs(&cxlmd->dev, &params);
> +	if (ret) {
> +		dev_err(&cxlmd->dev, "Get CXL patrol scrub params fail ret=%d\n",
> +			ret);
> +		return ret;

Called from probe so
		return dev_err_probe(&cxlmd->dev, ret,
				     "Get CXL patrol scrub params failed\n");
If you do hit this path, convention is cleanup and devm resources so we don't
waste memory that will never be used.  Rare case where devm_kfree() makes sense.

Or reorganize so you've gotten all the data before doing that allocation.

> +	}
> +	cxl_ps_ctx->scrub_cycle_changeable =  params.scrub_cycle_changeable;
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL_NS_GPL(cxl_mem_patrol_scrub_init, CXL);
> diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> index 2223ef3d3140..7025c4fd66f3 100644
> --- a/drivers/cxl/cxlmem.h
> +++ b/drivers/cxl/cxlmem.h
> @@ -948,6 +948,14 @@ int cxl_trigger_poison_list(struct cxl_memdev *cxlmd);
>  int cxl_inject_poison(struct cxl_memdev *cxlmd, u64 dpa);
>  int cxl_clear_poison(struct cxl_memdev *cxlmd, u64 dpa);
>  
> +/* cxl memory scrub functions */
> +#ifdef CONFIG_CXL_SCRUB
> +int cxl_mem_patrol_scrub_init(struct cxl_memdev *cxlmd);
> +#else
> +static inline int cxl_mem_patrol_scrub_init(struct cxl_memdev *cxlmd)
> +{ return -ENOTSUPP; }

That's a really obscure and little used return code + arch specific.
Probably EONOTSUPP

> +#endif
> +
>  #ifdef CONFIG_CXL_SUSPEND
>  void cxl_mem_active_inc(void);
>  void cxl_mem_active_dec(void);
> diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
> index 233e7c42c161..d2d734d22461 100644
> --- a/drivers/cxl/pci.c
> +++ b/drivers/cxl/pci.c
> @@ -886,6 +886,11 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>  	if (rc)
>  		return rc;
>  
> +	/*
> +	 * Initialize optional CXL scrub features
> +	 */

Single line comment fine, but given naming is obvious, no comment needed. However do
log a dev_dbg() if it fails (probably not for -ENOTSUPP

> +	cxl_mem_patrol_scrub_init(cxlmd);
> +
>  	rc = devm_cxl_sanitize_setup_notifier(&pdev->dev, cxlmd);
>  	if (rc)
>  		return rc;