Bart Van Assche <bvanassche@acm.org> says:
Hi Martin,
This patch series optimizes the hot path of the UFS driver by making
struct scsi_cmnd and struct ufshcd_lrb adjacent. Making these two data
structures adjacent is realized as follows:
@@ -9040,6 +9046,7 @@ static const struct scsi_host_template ufshcd_driver_template = {
.name = UFSHCD,
.proc_name = UFSHCD,
.map_queues = ufshcd_map_queues,
+ .cmd_size = sizeof(struct ufshcd_lrb),
.init_cmd_priv = ufshcd_init_cmd_priv,
.queuecommand = ufshcd_queuecommand,
.mq_poll = ufshcd_poll,
The following changes had to be made prior to making these two data
structures adjacent:
* Add support for driver-internal and reserved commands in the SCSI core.
* Instead of making the reserved command slot (hba->reserved_slot)
invisible to the SCSI core, let the SCSI core allocate a reserved command.
* Remove all UFS data structure members that are no longer needed
because struct scsi_cmnd and struct ufshcd_lrb are now adjacent
* Call ufshcd_init_lrb() from inside the code for queueing a command instead of
calling this function before I/O starts. This is necessary because
ufshcd_memory_alloc() allocates fewer instances than the block layer
allocates requests. See also the following code in the block layer
core:
if (blk_mq_init_request(set, hctx->fq->flush_rq, hctx_idx,
hctx->numa_node))
Although the UFS driver could be modified such that ufshcd_init_lrb()
is called from ufshcd_init_cmd_priv(), realizing this would require
moving the memory allocations that happen from inside
ufshcd_memory_alloc() into ufshcd_init_cmd_priv(). That would make
this patch series even larger. Although ufshcd_init_lrb() is called for each
command, the benefits of reduced indirection and better cache efficiency
outweigh the small overhead of per-command lrb initialization.
* ufshcd_add_scsi_host() happens now before any device management
commands are submitted. This change is necessary because this patch
makes device management command allocation happen when the SCSI host
is allocated.
* Allocate as many command slots as the host controller supports. Decrease
host->cmds_per_lun if necessary once it is clear whether or not the UFS
device supports less command slots than the host controller.
Please consider this patch series for the next merge window.
Thanks,
Bart.
Link: https://patch.msgid.link/20251031204029.2883185-1-bvanassche@acm.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Instead of storing the tag of the reserved command in hba->reserved_slot,
use scsi_get_internal_cmd() and scsi_put_internal_cmd() to allocate the
tag for the reserved command dynamically. Add
ufshcd_queue_reserved_command() for submitting reserved commands. Add
support in ufshcd_abort() for device management commands. Use
blk_execute_rq() for submitting reserved commands. Remove the code and
data structures that became superfluous. This includes
ufshcd_wait_for_dev_cmd(), hba->reserved_slot and ufs_dev_cmd.complete.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20251031204029.2883185-29-bvanassche@acm.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Set .cmd_size in the SCSI host template such that the SCSI core makes
struct scsi_cmnd and struct ufshcd_lrb adjacent. Convert the cmd->lrbp
and lrbp->cmd memory loads into pointer offset calculations. Remove the
data structure members that became superfluous, namely ufshcd_lrb.cmd
and ufs_hba.lrb. Since ufshcd_lrb.cmd is removed, this pointer cannot be
used anymore to test whether or not a command is a SCSI command.
Introduce a new function for this purpose, namely ufshcd_is_scsi_cmd().
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20251031204029.2883185-24-bvanassche@acm.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Ajay Neeli <ajay.neeli@amd.com> says:
This patch series adds support for the UFS driver on the AMD Versal
Gen 2 SoC. It includes:
- Device tree bindings and driver implementation.
- Secure read support for the secure retrieval of UFS calibration
values.
The UFS host driver is based upon the Synopsis DesignWare (DWC) UFS
architecture, utilizing the existing UFSHCD_DWC and UFSHCD_PLATFORM
drivers.
Link: https://patch.msgid.link/20251021113003.13650-1-ajay.neeli@amd.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Bart Van Assche <bvanassche@acm.org> says:
Hi Martin,
This patch series includes two bug fixes for this development cycle
and six small patches that are intended for the next merge window. If
applying the first two patches only during the current development
cycle would be inconvenient, postponing all patches until the next
merge window is fine with me.
Please consider including these patches in the upstream kernel.
Thanks,
Bart.
[mkp: Applied patches #1 and #2 to 6.18/scsi-fixes]
Link: https://patch.msgid.link/20251014200118.3390839-1-bvanassche@acm.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Peter Wang <peter.wang@mediatek.com says>:
Improves the UFS Mediatek driver by correcting clock scaling with PM
QoS, and adjusting power management flows. It addresses
shutdown/suspend race conditions, and removes redundant
functions. Support for new platforms is added with the MMIO_OTSD_CTRL
register, and MT6991 performance is optimized with MRTT and random
improvements. These changes collectively enhance driver performance,
stability, and compatibility.
Changes since v1:
1. Remove two patches that will be fixed in UFS core.
- ufs: host: mediatek: Fix runtime suspend error deadlock
- ufs: host: mediatek: Enable interrupts for MCQ mode
2. Use hba->shutting_down instead of ufshcd_is_user_access_allowed
v1:
https://patch.msgid.link/20250918104000.208856-1-peter.wang@mediatek.com
Link: https://patch.msgid.link/20250924094527.2992256-1-peter.wang@mediatek.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Bao D. Nguyen <quic_nguyenb@quicinc.com> says:
Multiple ufs device manufacturers request support for the
UFS_DEVICE_QUIRK_DELAY_AFTER_LPM quirk in the Qualcomm's platform
driver. After checking further with the major UFS manufacturers
engineering teams such as Samsung, Kioxia, SK Hynix and Micron, all
the manufacturers require this quirk. Since the quirk is needed by all
the ufs device manufacturers, remove the quirk in the ufs core driver
and implement a universal delay for all the ufs devices.
In addition to verifying with the public device's datasheets, the ufs
device manufacturer's engineering teams confirmed the required vcc
power-off time for the devices is a minimum of 1ms before vcc can be
powered on again. The existing 5ms delay implemented in the ufs core
driver seems too conservative, so replace the hard coded 5ms delay
with a variable default to 2ms setting to improve the system resume
latency. The platform drivers can override this setting as needed.
Link: https://patch.msgid.link/cover.1760383740.git.quic_nguyenb@quicinc.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
After the UFS device VCC is turned off, all the UFS device manufacturers
require a period of power-off time before the VCC can be turned on
again. This requirement has been confirmed with all the UFS device
manufacturer's datasheets.
Remove the UFS_DEVICE_QUIRK_DELAY_AFTER_LPM quirk in the UFS core driver
and implement a universal delay that is required by all the UFS device
manufacturers. In addition, remove the support for this quirk in the
platform drivers.
Signed-off-by: Bao D. Nguyen <quic_nguyenb@quicinc.com>
Reviewed-by: Peter Wang <peter.wang@mediatek.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/25f134d5a42e8b8365be64d512d1bb5fc2bce6ff.1760383740.git.quic_nguyenb@quicinc.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When a UTP error occurs in isolation, UFS is not currently recoverable.
This is because the UTP error is not considered fatal in the error
handling code, leading to either an I/O timeout or an OCS error.
Add the UTP error flag to INT_FATAL_ERRORS so the controller will be
reset in this situation.
sd 0:0:0:0: [sda] tag#38 UNKNOWN(0x2003) Result: hostbyte=0x07
driverbyte=DRIVER_OK cmd_age=0s
sd 0:0:0:0: [sda] tag#38 CDB: opcode=0x28 28 00 00 51 24 e2 00 00 08 00
I/O error, dev sda, sector 42542864 op 0x0:(READ) flags 0x80700 phys_seg
8 prio class 2
OCS error from controller = 9 for tag 39
pa_err[1] = 0x80000010 at 2667224756 us
pa_err: total cnt=2
dl_err[0] = 0x80000002 at 2667148060 us
dl_err[1] = 0x80002000 at 2667282844 us
No record of nl_err
No record of tl_err
No record of dme_err
No record of auto_hibern8_err
fatal_err[0] = 0x804 at 2667282836 us
---------------------------------------------------
REGISTER
---------------------------------------------------
NAME OFFSET VALUE
STD HCI SFR 0xfffffff0 0x0
AHIT 0x18 0x814
INTERRUPT STATUS 0x20 0x1000
INTERRUPT ENABLE 0x24 0x70ef5
[mkp: commit desc]
Signed-off-by: Hoyoung Seo <hy50.seo@samsung.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Message-Id: <20250930061428.617955-1-hy50.seo@samsung.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The cpu_latency_qos_add/remove/update_request interfaces lack internal
synchronization by design, requiring the caller to ensure thread safety.
The current implementation relies on the 'pm_qos_enabled' flag, which is
insufficient to prevent concurrent access and cannot serve as a proper
synchronization mechanism. This has led to data races and list
corruption issues.
A typical race condition call trace is:
[Thread A]
ufshcd_pm_qos_exit()
--> cpu_latency_qos_remove_request()
--> cpu_latency_qos_apply();
--> pm_qos_update_target()
--> plist_del <--(1) delete plist node
--> memset(req, 0, sizeof(*req));
--> hba->pm_qos_enabled = false;
[Thread B]
ufshcd_devfreq_target
--> ufshcd_devfreq_scale
--> ufshcd_scale_clks
--> ufshcd_pm_qos_update <--(2) pm_qos_enabled is true
--> cpu_latency_qos_update_request
--> pm_qos_update_target
--> plist_del <--(3) plist node use-after-free
Introduces a dedicated mutex to serialize PM QoS operations, preventing
data races and ensuring safe access to PM QoS resources, including sysfs
interface reads.
Fixes: 2777e73fc1 ("scsi: ufs: core: Add CPU latency QoS support for UFS driver")
Signed-off-by: Zhongqiu Han <zhongqiu.han@oss.qualcomm.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Huan Tang <tanghuan@vivo.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Move the MCQ interrupt enable process to
ufshcd_mcq_make_queues_operational() to ensure that interrupts are set
correctly when making queues operational, similar to
ufshcd_make_hba_operational(). This change addresses the issue where
ufshcd_mcq_make_queues_operational() was not fully operational due to
missing interrupt enablement.
This change only affects host drivers that call
ufshcd_mcq_make_queues_operational(), i.e. ufs-mediatek.
Signed-off-by: Peter Wang <peter.wang@mediatek.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Nitin Rawat <nitin.rawat@oss.qualcomm.com> says:
The patch series simplifies the UFS MCQ (Multi Circular Queue)
resource mapping in the Qualcomm UFS host controller driver by
replacing the complex multi-resource approach with a streamlined
single-resource implementation.
The current MCQ implementation uses multiple separate resource
mappings (RES_UFS, RES_MCQ, RES_MCQ_SQD, RES_MCQ_VS) with dynamic
resource allocation, which increases code complexity and potential for
resource mapping errors. This approach also doesn't align with the
device tree binding specification that defines a single 'mcq' memory
region.
Replace the multi-resource mapping with a single "mcq" resource that
encompasses the entire MCQ configuration space. The doorbell registers
(SQD, CQD, SQIS, CQIS) are accessed using predefined offsets relative
to the MCQ base address, providing clearer memory layout organization.
Tested on Qualcomm platforms SM8650 and SM8750 with UFS MCQ enabled.
Changes from v3:
1. Addressed Krzysztof comment to separate device tree and driver
patch independently in different patch series. This series caters
driver changes.
2. Addressed Manivannan's change to update commit text and remove
redundant null check in mcq code.
3. Addressed Manivannan's to Update few offsets as fixed definition
instead of enum.
Changes from v2:
1. Removed dt-bindings patch as existing binding supports required
reg-names format.
2. Added patch to refactor MCQ register dump logic for new resource
mapping.
3. Added patch to remove unused ufshcd_res_info structure from UFS core.
4. Changed reg-names from "ufs_mem" to "std" in device tree patches.
5. Reordered patches with driver changes first, then device tree changes.
6. Updated SM8750 MCQ region size from 0x2000 to 0x15000
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Refactor MCQ register dump to align with the new resource mapping. As
part of refactor, below changes are done:
- Update ufs_qcom_dump_regs() function signature to accept direct
base address instead of resource ID enum
- Modify ufs_qcom_dump_mcq_hci_regs() to use hba->mcq_base and
calculated addresses from MCQ operation info
- Replace enum ufshcd_res with direct memory-mapped I/O addresses
Additionally remove the ufshcd_res_info structure and associated enum
ufshcd_res definitions from the UFS host controller header. These were
previously used for MCQ resource mapping but are no longer needed
following recent refactoring to use direct base addresses instead of
multiple separate resource regions.
Signed-off-by: Nitin Rawat <quic_nitirawa@quicinc.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Manivannan Sadhasivam <mani@kernel.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Peter Wang <peter.wang@mediatek.com> says:
These patches collectively enhance the UFS host driver's reliability,
power management efficiency, and error recovery mechanisms on MediaTek
platforms. They address critical issues and introduce optimizations
that improve system stability and performance.
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Improve the recovery process for hibernation exit failures. Trigger the
error handler and break the suspend operation to ensure effective
recovery from hibernation errors. Activate the error handling mechanism
by ufshcd_force_error_recovery and scheduling the error handler work.
Signed-off-by: Peter Wang <peter.wang@mediatek.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The <ufs/ufs.h> header file defines constants and data structures
related to the UFS standard. Move the enumeration types related to
tracing into a new header file because these are not defined in the UFS
standard. An intended side effect of this patch is that the tracing
enumeration types are no longer visible to UFS host drivers.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20250829153841.2201700-1-bvanassche@acm.org
Reviewed-by: Avri Altman <avri.altman@sandisk.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Pull in fixes from 6.15 and resolve a few conflicts so we can have a
clean base for UFS patches.
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Follow JESD220G, support a WB buffer resize function through sysfs. The
host can obtain resize hint and resize status, and enable the resize
operation. Add three sysfs nodes:
1. wb_resize_enable
2. wb_resize_hint
3. wb_resize_status
The detailed definition of the three nodes can be found in the sysfs
documentation.
Signed-off-by: Huan Tang <tanghuan@vivo.com>
Signed-off-by: Lu Hongfei <luhongfei@vivo.com>
Link: https://lore.kernel.org/r/20250411092924.1116-1-tanghuan@vivo.com
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The ufs device JEDEC specification version 4.1 adds support for the
device level exception events. To support this new device level
exception feature, expose two new sysfs nodes below to provide the user
space access to the device level exception information.
/sys/bus/platform/drivers/ufshcd/*/device_lvl_exception_count
/sys/bus/platform/drivers/ufshcd/*/device_lvl_exception_id
The device_lvl_exception_count sysfs node reports the number of device
level exceptions that have occurred since the last time this variable is
reset. Writing a value of 0 will reset it. The device_lvl_exception_id
reports the exception ID which is the qDeviceLevelExceptionID attribute
of the device JEDEC specifications version 4.1 and later. The user space
application can query these sysfs nodes to get more information about
the device level exception.
Signed-off-by: Bao D. Nguyen <quic_nguyenb@quicinc.com>
Link: https://lore.kernel.org/r/6278d7c125b2f0cf5056f4a647a4b9c1fdd24fc7.1743198325.git.quic_nguyenb@quicinc.com
Reviewed-by: Peter Wang <peter.wang@mediatek.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Arthur Simchaev <arthur.simchaev@sandisk.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
There is a TOCTOU race in ufshcd_compl_one_cqe(): hba->dev_cmd.complete may
be cleared from another thread after it has been checked and before it is
used. Fix this race by moving the device command completion from the stack
of the device command submitter into struct ufs_hba. This patch fixes the
following kernel crash:
Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
Call trace:
_raw_spin_lock_irqsave+0x34/0x80
complete+0x24/0xb8
ufshcd_compl_one_cqe+0x13c/0x4f0
ufshcd_mcq_poll_cqe_lock+0xb4/0x108
ufshcd_intr+0x2f4/0x444
__handle_irq_event_percpu+0xbc/0x250
handle_irq_event+0x48/0xb0
Fixes: 5a0b0cb9be ("[SCSI] ufs: Add support for sending NOP OUT UPIU")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20250314225206.1487838-1-bvanassche@acm.org
Reviewed-by: Peter Wang <peter.wang@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Ziqi Chen <quic_ziqichen@quicinc.com> says:
With OPP V2 enabled, devfreq can scale clocks amongst multiple frequency
plans. However, the gear speed is only toggled between min and max during
clock scaling. Enable multi-level gear scaling by mapping clock frequencies
to gear speeds, so that when devfreq scales clock frequencies we can put
the UFS link at the appropraite gear speeds accordingly.
This series has been tested on below platforms -
sm8550 mtp + UFS3.1
SM8650 MTP + UFS3.1
SM8750 MTP + UFS4.0
Tested-by: Neil Armstrong <neil.armstrong@linaro.org> # on SM8550-QRD
Tested-by: Neil Armstrong <neil.armstrong@linaro.org> # on SM8550-HDK
Tested-by: Neil Armstrong <neil.armstrong@linaro.org> # on SM8650-HDK
Link: https://lore.kernel.org/r/20250213080008.2984807-1-quic_ziqichen@quicinc.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Martin hi,
The UFS4.1 standard, released on January 8, 2025, added a new exception
event: HEALTH_CRITICAL, which notifies the host of a device's critical
health condition. This notification implies that the device is approaching
the end of its lifetime based on the amount of performed program/erase
cycles.
Once an EOL (End-of-Life) exception event is received, we increment a
designated member, which is exposed via a sysfs entry. This new entry, will
report the number of times a critical health event has been reported by a
UFS device.
To handle this new sysfs entry, userspace applications can use select(),
poll(), or epoll() to monitor changes in the critical_health attribute. The
kernel will call sysfs_notify() to signal changes, allowing the userspace
application to detect and respond to these changes efficiently.
The host can gain further insight into the specific issue by reading one of
the following attributes: bPreEOLInfo, bDeviceLifeTimeEstA,
bDeviceLifeTimeEstB, bWriteBoosterBufferLifeTimeEst, and
bRPMBLifeTimeEst. All those are available for reading via the driver's
sysfs entries or through an applicable utility. It is up to userspace to
read these attributes if needed.
Signed-off-by: Avri Altman <avri.altman@wdc.com>
Link: https://lore.kernel.org/r/20250211065813.58091-1-avri.altman@wdc.com
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
devm_blk_crypto_profile_init() registers a cleanup handler to run when
the associated (platform-) device is being released. For UFS, the
crypto private data and pointers are stored as part of the ufs_hba's
data structure 'struct ufs_hba::crypto_profile'. This structure is
allocated as part of the underlying ufshcd and therefore Scsi_host
allocation.
During driver release or during error handling in ufshcd_pltfrm_init(),
this structure is released as part of ufshcd_dealloc_host() before the
(platform-) device associated with the crypto call above is released.
Once this device is released, the crypto cleanup code will run, using
the just-released 'struct ufs_hba::crypto_profile'. This causes a
use-after-free situation:
Call trace:
kfree+0x60/0x2d8 (P)
kvfree+0x44/0x60
blk_crypto_profile_destroy_callback+0x28/0x70
devm_action_release+0x1c/0x30
release_nodes+0x6c/0x108
devres_release_all+0x98/0x100
device_unbind_cleanup+0x20/0x70
really_probe+0x218/0x2d0
In other words, the initialisation code flow is:
platform-device probe
ufshcd_pltfrm_init()
ufshcd_alloc_host()
scsi_host_alloc()
allocation of struct ufs_hba
creation of scsi-host devices
devm_blk_crypto_profile_init()
devm registration of cleanup handler using platform-device
and during error handling of ufshcd_pltfrm_init() or during driver
removal:
ufshcd_dealloc_host()
scsi_host_put()
put_device(scsi-host)
release of struct ufs_hba
put_device(platform-device)
crypto cleanup handler
To fix this use-after free, change ufshcd_alloc_host() to register a
devres action to automatically cleanup the underlying SCSI device on
ufshcd destruction, without requiring explicit calls to
ufshcd_dealloc_host(). This way:
* the crypto profile and all other ufs_hba-owned resources are
destroyed before SCSI (as they've been registered after)
* a memleak is plugged in tc-dwc-g210-pci.c remove() as a
side-effect
* EXPORT_SYMBOL_GPL(ufshcd_dealloc_host) can be removed fully as
it's not needed anymore
* no future drivers using ufshcd_alloc_host() could ever forget
adding the cleanup
Fixes: cb77cb5abe ("blk-crypto: rename blk_keyslot_manager to blk_crypto_profile")
Fixes: d76d9d7d10 ("scsi: ufs: use devm_blk_ksm_init()")
Cc: stable@vger.kernel.org
Signed-off-by: André Draszik <andre.draszik@linaro.org>
Link: https://lore.kernel.org/r/20250124-ufshcd-fix-v4-1-c5d0144aae59@linaro.org
Reviewed-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Acked-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
