Re: [PATCH V10 2/2] scsi: ufs: Add configfs support for UFS provisioning

From: Sayali Lokhande
Date: Tue Aug 28 2018 - 08:38:18 EST

On 8/28/2018 5:36 PM, Adrian Hunter wrote:
On 28/08/18 14:40, Sayali Lokhande wrote:
On 8/28/2018 4:56 PM, Adrian Hunter wrote:
On 28/08/18 13:23, Sayali Lokhande wrote:
Hi Adrian,

On 8/24/2018 2:15 PM, Adrian Hunter wrote:
On 24/08/18 08:33, Sayali Lokhande wrote:
This patch adds configfs support to provision UFS device at
runtime. This feature can be primarily useful in factory or
assembly line as some devices may be required to be configured
multiple times during initial system development phase.
Configuration Descriptors can be written multiple times until
bConfigDescrLock attribute is zero.

Configuration descriptor buffer consists of Device and Unit
descriptor configurable parameters which are parsed from vendor
specific provisioning file and then passed via configfs node at
runtime to provision ufs device.
CONFIG_CONFIGFS_FS needs to be enabled for using this feature.

1) To read current configuration descriptor :
ÂÂÂÂ cat /config/XXXX.ufshc/ufs_provision
2) To provision ufs device:
ÂÂÂÂ echo <config_desc_buf> > /config/XXXX.ufshc/ufs_provision

Signed-off-by: Sayali Lokhande <sayalil@xxxxxxxxxxxxxx>
ÂÂ Documentation/ABI/testing/configfs-driver-ufs |Â 18 +++
ÂÂ drivers/scsi/ufs/KconfigÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ |Â 10 ++
ÂÂ drivers/scsi/ufs/MakefileÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ |ÂÂ 1 +
ÂÂ drivers/scsi/ufs/ufs-configfs.cÂÂÂÂÂÂÂÂÂÂÂÂÂÂ | 155
ÂÂ drivers/scsi/ufs/ufshcd.cÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ |ÂÂ 3 +-
ÂÂ drivers/scsi/ufs/ufshcd.hÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ |Â 17 +++
ÂÂ 6 files changed, 203 insertions(+), 1 deletion(-)
ÂÂ create mode 100644 Documentation/ABI/testing/configfs-driver-ufs
ÂÂ create mode 100644 drivers/scsi/ufs/ufs-configfs.c

diff --git a/Documentation/ABI/testing/configfs-driver-ufs
new file mode 100644
index 0000000..eeee499c
--- /dev/null
+++ b/Documentation/ABI/testing/configfs-driver-ufs
@@ -0,0 +1,18 @@
+What:ÂÂÂÂÂÂÂ /config/ufshcd/ufs_provision
You are using the device name as the config subsystem name, so this is

ÂÂÂÂÂWhat:ÂÂÂÂÂÂÂ /config/*/ufs_provision
Agree. Will update.
+Date:ÂÂÂÂÂÂÂ Jun 2018
+KernelVersion:ÂÂÂ 4.14
4.19 maybe
+ÂÂÂÂÂÂÂ This file shows the current ufs configuration descriptor set in
It is only the first config descriptor though isn't it? There is no
for the other 3. Shouldn't we support config descriptors fully?
Yes, we should support config descriptors fully. My current implementation
only supports config desc (index = 00h).
I m thinking I can add separate "index" field along with "ufs_provision"
field and can use this index (instead of hard coding to 0)
to write repective config_desc. In this way, we should first set the "index"
and then read/write to "ufs_provision".
Let me know if you have any concern with this approach ?
One file for each config descriptor would seem more normal.
I can make separate file for each config descriptor index(say
ufs_provision_0/1/2/3), but still show() and store() api's can remain common.
Will try it out and update.
+ÂÂÂÂÂÂÂ This can be used to provision ufs device if bConfigDescrLock
is 0.
+ÂÂÂÂÂÂÂ For more details, refer Configuration Descriptor and
+ÂÂÂÂÂÂÂ Table 14-12 - Unit Descriptor configurable parameters from specs
+ÂÂÂÂÂÂÂ description of each configuration descriptor parameter.
+ÂÂÂÂÂÂÂ Configuration descriptor buffer needs to be passed in space
+ÂÂÂÂÂÂÂ format specificied as below:
specificied -> specified
+ÂÂÂÂÂÂÂ echo <bLength> <bDescriptorIDN> <bConfDescContinue> <bBootEnable>
+ÂÂÂÂÂÂÂ <bDescrAccessEn> <bInitPowerMode> <bHighPriorityLUN>
+ÂÂÂÂÂÂÂ <bSecureRemovalType> <bInitActiveICCLevel> <wPeriodicRTCUpdate>
+ÂÂÂÂÂÂÂ <0Bh:0Fh_ReservedAs_0> <bLUEnable> <bBootLunID> <bLUWriteProtect>
+ÂÂÂÂÂÂÂ <bMemoryType> <dNumAllocUnits> <bDataReliability>
+ÂÂÂÂÂÂÂ <bProvisioningType> <wContextCapabilities> <0Dh:0Fh_ReservedAs_0>
This format is not exactly correct anymore because each byte has to be
represented separately. Also theoretically it can be change, so maybe
better just to refer the user to the UFS specification.
Agree. Will update to just refer to UFS spec (for Config_desc format).

+ÂÂÂÂÂÂÂ > /config/XXXX.ufshc/ufs_provision
diff --git a/drivers/scsi/ufs/Kconfig b/drivers/scsi/ufs/Kconfig
index e27b4d4..a9a0a58 100644
--- a/drivers/scsi/ufs/Kconfig
+++ b/drivers/scsi/ufs/Kconfig
@@ -100,3 +100,13 @@ config SCSI_UFS_QCOM
ÂÂ ÂÂÂÂÂÂÂ Select this if you have UFS controller on QCOM chipset.
ÂÂÂÂÂÂÂÂ If unsure, say N.
+ÂÂÂ tristate "Runtime UFS Provisioning support"
+ÂÂÂ help
+ÂÂÂÂÂ This enables runtime UFS provisioning support. This can be used
+ÂÂÂÂÂ primarily during assembly line as some devices may be required to
+ÂÂÂÂÂ be configured multiple times during initial development phase.
+ÂÂÂÂÂ If unsure, say N.
diff --git a/drivers/scsi/ufs/Makefile b/drivers/scsi/ufs/Makefile
index 918f579..00110ea 100644
--- a/drivers/scsi/ufs/Makefile
+++ b/drivers/scsi/ufs/Makefile
@@ -5,5 +5,6 @@ obj-$(CONFIG_SCSI_UFS_DWC_TC_PLATFORM) +=
tc-dwc-g210-pltfrm.o ufshcd-dwc.o tc-d
ÂÂ obj-$(CONFIG_SCSI_UFS_QCOM) += ufs-qcom.o
ÂÂ obj-$(CONFIG_SCSI_UFSHCD) += ufshcd-core.o
ÂÂ ufshcd-core-objs := ufshcd.o ufs-sysfs.o
+obj-$(CONFIG_UFS_PROVISION) += ufs-configfs.o
ÂÂ obj-$(CONFIG_SCSI_UFSHCD_PCI) += ufshcd-pci.o
ÂÂ obj-$(CONFIG_SCSI_UFSHCD_PLATFORM) += ufshcd-pltfrm.o
diff --git a/drivers/scsi/ufs/ufs-configfs.c
new file mode 100644
index 0000000..74563a4
--- /dev/null
+++ b/drivers/scsi/ufs/ufs-configfs.c
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: GPL-2.0+
+// Copyright (c) 2018, Linux Foundation.
+#include <linux/configfs.h>
+#include <linux/err.h>
+#include <linux/string.h>
+#include "ufs.h"
+#include "ufshcd.h"
+static inline struct ufs_hba *config_item_to_hba(struct config_item
+ÂÂÂ struct config_group *group = to_config_group(item);
+ÂÂÂ struct configfs_subsystem *subsys = to_configfs_subsystem(group);
+ÂÂÂ struct ufs_hba *hba = container_of(subsys, struct ufs_hba, subsys);
+ÂÂÂ return subsys ? hba : NULL;
+static ssize_t ufs_provision_show(struct config_item *item, char *buf)
Is it really ok to hard-code the descriptor size?
I will use ufshcd_read_desc_length() api for each index as required. Will
remove this hard-code.
Alternatively, I see there is already hba->desc_size.conf_desc
hba->desc_size.conf_desc is actually size for index 0. I thought what if
desc size various per index.
So better to read for each index. What say ?
Based on the 'B' and 'L' values defined in the spec., all configuration
descriptors have the same length.
Ok. Got it. Then I will just use "hba->desc_size.conf_desc" as desc length.
+ÂÂÂ int i, ret, curr_len = 0;
+ÂÂÂ struct ufs_hba *hba = config_item_to_hba(item);
+ÂÂÂ ret = ufshcd_query_descriptor_retry(hba, UPIU_QUERY_OPCODE_READ_DESC,
+ÂÂÂÂÂÂÂ QUERY_DESC_IDN_CONFIGURATION, 0, 0, desc_buf, &desc_buf_len);
+ÂÂÂ if (ret)
+ÂÂÂÂÂÂÂ return ret;
+ÂÂÂ for (i = 0; i < desc_buf_len; i++)
+ÂÂÂÂÂÂÂ curr_len += snprintf((buf + curr_len), (PAGE_SIZE - curr_len),
+ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ "0x%x ", desc_buf[i]);
+ÂÂÂ return curr_len;
+ssize_t ufshcd_desc_configfs_store(struct ufs_hba *hba,
+ÂÂÂÂÂÂÂ const char *buf, size_t count)
+ÂÂÂ char *strbuf;
+ÂÂÂ char *strbuf_copy;
+ÂÂÂ char *token;
+ÂÂÂ int i, ret, value;
+ÂÂÂ u32 config_desc_lock = 0;
+ÂÂÂ /* reserve one byte for null termination */
+ÂÂÂ strbuf = kmalloc(count + 1, GFP_KERNEL);
+ÂÂÂ if (!strbuf)
+ÂÂÂ strbuf_copy = strbuf;
+ÂÂÂ strlcpy(strbuf, buf, count + 1);
+ÂÂÂ /* Just return if bConfigDescrLock is already set */
+ÂÂÂ ret = ufshcd_query_attr(hba, UPIU_QUERY_OPCODE_READ_ATTR,
+ÂÂÂÂÂÂÂ QUERY_ATTR_IDN_CONF_DESC_LOCK, 0, 0, &config_desc_lock);
+ÂÂÂ if (ret)
+ÂÂÂÂÂÂÂ goto out;
+ÂÂÂ if (config_desc_lock) {
+ÂÂÂÂÂÂÂ dev_err(hba->dev, "%s: bConfigDescrLock already set to %u,
cannot re-provision device!\n",
+ÂÂÂÂÂÂÂ __func__, config_desc_lock);
+ÂÂÂÂÂÂÂ goto out;
+ÂÂÂ }
+ÂÂÂÂÂÂÂ token = strsep(&strbuf, " ");
+ÂÂÂÂÂÂÂ if (!token && i)
+ÂÂÂÂÂÂÂ ret = kstrtoint(token, 0, &value);
+ÂÂÂÂÂÂÂ if (ret) {
+ÂÂÂÂÂÂÂÂÂÂÂ dev_err(hba->dev, "%s: kstrtoint failed %d %s\n",
+ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ __func__, ret, token);
+ÂÂÂÂÂÂÂÂÂÂÂ goto out;
+ÂÂÂÂÂÂÂ desc_buf[i] = (u8)value;
+ÂÂÂ }
+ÂÂÂ /* Write configuration descriptor to provision ufs */
+ÂÂÂ ret = ufshcd_query_descriptor_retry(hba,
+ÂÂÂÂÂÂÂ QUERY_DESC_IDN_CONFIGURATION, 0, 0, desc_buf, &desc_buf_len);
+ÂÂÂ if (!ret)
+ÂÂÂÂÂÂÂ dev_info(hba->dev, "%s: UFS Provisioning done, reboot now!\n",
That is not necessarily true - bConfDescContinue might have been used or
bConfigDescrLock might need to be set. Better just to say that
configuration descriptor 0 was written.
Agree. Will update.
+ÂÂÂÂÂÂÂ __func__);
+ÂÂÂ kfree(strbuf_copy);
+ÂÂÂ if (ret)
+ÂÂÂÂÂÂÂ return ret;
+ÂÂÂ return count;
+static ssize_t ufs_provision_store(struct config_item *item,
+ÂÂÂÂÂÂÂ const char *buf, size_t count)
+ÂÂÂ struct ufs_hba *hba = config_item_to_hba(item);
+ÂÂÂ return ufshcd_desc_configfs_store(hba, buf, count);
+static struct configfs_attribute ufshcd_attr_provision = {
+ÂÂÂ .ca_nameÂÂÂ = "ufs_provision",
+ÂÂÂ .ca_modeÂÂÂ = 0644,
+ÂÂÂ .ca_ownerÂÂÂ = THIS_MODULE,
+ÂÂÂ .showÂÂÂÂÂÂÂ = ufs_provision_show,
+ÂÂÂ .storeÂÂÂÂÂÂÂ = ufs_provision_store,
+static struct configfs_attribute *ufshcd_attrs[] = {
+ÂÂÂ &ufshcd_attr_provision,
+static struct config_item_type ufscfg_type = {
+ÂÂÂ .ct_attrsÂÂÂ = ufshcd_attrs,
+ÂÂÂ .ct_ownerÂÂÂ = THIS_MODULE,
+static struct configfs_subsystem ufscfg_subsys = {
+ÂÂÂ .su_group = {
+ÂÂÂÂÂÂÂ .cg_item = {
+ÂÂÂÂÂÂÂÂÂÂÂ .ci_type = &ufscfg_type,
+ÂÂÂ },
+void ufshcd_configfs_init(struct ufs_hba *hba, const char *name)
+ÂÂÂ int ret;
+ÂÂÂ struct configfs_subsystem *subsys = &hba->subsys;
+ÂÂÂ strncpy(ufscfg_subsys.su_group.cg_item.ci_namebuf, name,
+ÂÂÂ subsys->su_group = ufscfg_subsys.su_group;
configfs_subsystem cannot share su_group with other configfs_subsystem i.e
you are not allowing for there being more than 1 hba
Agree. I think I will have to move above
declarations(ufscfg_subsys/ufscfg_type/ufshcd_attrs) in driver .c/.h file
(to make it per ufs hba/probe).
Will check and update.
+ÂÂÂ config_group_init(&subsys->su_group);
+ÂÂÂ mutex_init(&subsys->su_mutex);
+ÂÂÂ ret = configfs_register_subsystem(subsys);
+ÂÂÂ if (ret)
+ÂÂÂÂÂÂÂ pr_err("Error %d while registering subsystem %s\n",
+ÂÂÂÂÂÂÂÂÂÂÂÂÂÂ subsys->su_group.cg_item.ci_namebuf);
+void ufshcd_configfs_exit(void)
+ÂÂÂ configfs_unregister_subsystem(&ufscfg_subsys);
diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index e01cdc0..32a67b0 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -7711,6 +7711,7 @@ int ufshcd_shutdown(struct ufs_hba *hba)
ÂÂ void ufshcd_remove(struct ufs_hba *hba)
ÂÂ {
ÂÂÂÂÂÂ ufs_sysfs_remove_nodes(hba->dev);
+ÂÂÂ ufshcd_configfs_exit();
ÂÂÂÂÂÂ scsi_remove_host(hba->host);
ÂÂÂÂÂÂ /* disable interrupts */
ÂÂÂÂÂÂ ufshcd_disable_intr(hba, hba->intr_mask);
@@ -7965,7 +7966,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem
*mmio_base, unsigned int irq)
ÂÂ ÂÂÂÂÂ async_schedule(ufshcd_async_scan, hba);
ÂÂÂÂÂÂ ufs_sysfs_add_nodes(hba->dev);
+ÂÂÂ ufshcd_configfs_init(hba, dev_name(hba->dev));
ÂÂÂÂÂÂ return 0;
ÂÂ Â out_remove_scsi_host:
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index 45013b6..2275032 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -37,6 +37,7 @@
ÂÂ #ifndef _UFSHCD_H
ÂÂ #define _UFSHCD_H
ÂÂ +#include <linux/configfs.h>
ÂÂ #include <linux/module.h>
ÂÂ #include <linux/kernel.h>
ÂÂ #include <linux/init.h>
@@ -515,6 +516,9 @@ struct ufs_hba {
ÂÂ ÂÂÂÂÂ struct Scsi_Host *host;
ÂÂÂÂÂÂ struct device *dev;
+ÂÂÂ struct configfs_subsystem subsys;
ÂÂÂÂÂÂÂ * This field is to keep a reference to "scsi_device"
corresponding to
ÂÂÂÂÂÂÂ * "UFS device" W-LU.
@@ -868,6 +872,19 @@ int ufshcd_read_string_desc(struct ufs_hba *hba, int
ÂÂ int ufshcd_hold(struct ufs_hba *hba, bool async);
ÂÂ void ufshcd_release(struct ufs_hba *hba);
ÂÂ +/* Expose UFS configfs API's */
+static inline void ufshcd_configfs_init(struct ufs_hba *hba, const char
+static inline void ufshcd_configfs_exit(void)
+void ufshcd_configfs_init(struct ufs_hba *hba, const char *name);
+void ufshcd_configfs_exit(void);
ÂÂ int ufshcd_map_desc_id_to_length(struct ufs_hba *hba, enum desc_idn
ÂÂÂÂÂÂ int *desc_length);