MHI specification shows a state machine with support for STOP channel command
and the validity of certain state transitions. MHI host currently does not
provide any mechanism to stop a channel and restart it without resetting it.
There are also times when the device moves on to a different execution
environment while client drivers on the host are unaware of it and still
attempt to reset the channels facing unnecessary timeouts.
This series addresses the above areas to provide support for stopping an MHI
channel, resuming it back, improved documentation and improving upon channel
state machine handling in general.
This set of patches was tested on arm64 architecture.
v6:
-Dropped the patch which introduced start/stop transfer APIs for lack of users
-Updated error handling and debug prints on channel handling improvements patch
-Improved commit text to better explain certain patches based on review comments
-Removed references to new APIs from the documentation improvement patch
v5:
-Added reviewed-by tags from Hemant I missed earlier
-Added patch to prevent kernel warnings on clearing channel context twice
v4:
-Updated commit text/descriptions and addressed checkpatch checks
-Added context validity check before starting/stopping channels from new API
-Added patch to clear channel context configuration after reset/unprepare
v3:
-Updated documentation for channel transfer APIs to highlight differences
-Create separate patch for "allowing channel to be disabled from stopped state"
v2:
-Renamed the newly introduced APIs to mhi_start_transfer() / mhi_stop_transfer()
-Added improved documentation to avoid confusion with the new APIs
-Removed the __ prefix from mhi_unprepare_channel() API for consistency.
Bhaumik Bhatt (8):
bus: mhi: core: Allow sending the STOP channel command
bus: mhi: core: Clear context for stopped channels from remove()
bus: mhi: core: Improvements to the channel handling state machine
bus: mhi: core: Clear configuration from channel context during reset
bus: mhi: core: Check channel execution environment before issuing
reset
bus: mhi: core: Remove __ prefix for MHI channel unprepare function
bus: mhi: Improve documentation on channel transfer setup APIs
bus: mhi: core: Do not clear channel context more than once
drivers/bus/mhi/core/init.c | 23 ++++-
drivers/bus/mhi/core/internal.h | 12 +++
drivers/bus/mhi/core/main.c | 192 ++++++++++++++++++++++++----------------
include/linux/mhi.h | 18 +++-
4 files changed, 164 insertions(+), 81 deletions(-)
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
Clearing a channel context can happen twice if the client driver
unprepares and reset the channels from the remove() callback from
a controller requested MHI power down sequence. If there are
multiple attempts at calling the mhi_free_coherent() API, we see
kernel warnings such as "trying to free invalid coherent area".
Example for one such client is the QRTR MHI driver. Avoid these
warnings by skipping mhi_deinit_chan_ctxt() API call and prevent
extra work from MHI as the channels are already disabled.
Signed-off-by: Bhaumik Bhatt <[email protected]>
---
drivers/bus/mhi/core/init.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c
index 30eef19..272f350 100644
--- a/drivers/bus/mhi/core/init.c
+++ b/drivers/bus/mhi/core/init.c
@@ -1314,6 +1314,7 @@ static int mhi_driver_remove(struct device *dev)
if ((ch_state[dir] == MHI_CH_STATE_ENABLED ||
ch_state[dir] == MHI_CH_STATE_STOP) &&
+ mhi_chan->ch_state != MHI_CH_STATE_DISABLED &&
!mhi_chan->offload_ch)
mhi_deinit_chan_ctxt(mhi_cntrl, mhi_chan);
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
When clearing up the channel context after client drivers are
done using channels, the configuration is currently not being
reset entirely. Ensure this is done to appropriately handle
issues where clients unaware of the context state end up calling
functions which expect a context.
Signed-off-by: Bhaumik Bhatt <[email protected]>
Reviewed-by: Hemant Kumar <[email protected]>
Reviewed-by: Manivannan Sadhasivam <[email protected]>
---
drivers/bus/mhi/core/init.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c
index 482b365..30eef19 100644
--- a/drivers/bus/mhi/core/init.c
+++ b/drivers/bus/mhi/core/init.c
@@ -558,6 +558,7 @@ void mhi_deinit_chan_ctxt(struct mhi_controller *mhi_cntrl,
struct mhi_ring *buf_ring;
struct mhi_ring *tre_ring;
struct mhi_chan_ctxt *chan_ctxt;
+ u32 tmp;
buf_ring = &mhi_chan->buf_ring;
tre_ring = &mhi_chan->tre_ring;
@@ -568,7 +569,19 @@ void mhi_deinit_chan_ctxt(struct mhi_controller *mhi_cntrl,
vfree(buf_ring->base);
buf_ring->base = tre_ring->base = NULL;
+ tre_ring->ctxt_wp = NULL;
chan_ctxt->rbase = 0;
+ chan_ctxt->rlen = 0;
+ chan_ctxt->rp = 0;
+ chan_ctxt->wp = 0;
+
+ tmp = chan_ctxt->chcfg;
+ tmp &= ~CHAN_CTX_CHSTATE_MASK;
+ tmp |= (MHI_CH_STATE_DISABLED << CHAN_CTX_CHSTATE_SHIFT);
+ chan_ctxt->chcfg = tmp;
+
+ /* Update to all cores */
+ smp_wmb();
}
int mhi_init_chan_ctxt(struct mhi_controller *mhi_cntrl,
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
If a channel was explicitly stopped but not reset and a driver
remove is issued, clean up the channel context such that it is
reflected on the device. This move is useful if a client driver
module is unloaded or a device crash occurs with the host having
placed the channel in a stopped state.
Signed-off-by: Bhaumik Bhatt <[email protected]>
Reviewed-by: Hemant Kumar <[email protected]>
Reviewed-by: Manivannan Sadhasivam <[email protected]>
---
drivers/bus/mhi/core/init.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c
index aa575d3..03c5786 100644
--- a/drivers/bus/mhi/core/init.c
+++ b/drivers/bus/mhi/core/init.c
@@ -1293,7 +1293,8 @@ static int mhi_driver_remove(struct device *dev)
mutex_lock(&mhi_chan->mutex);
- if (ch_state[dir] == MHI_CH_STATE_ENABLED &&
+ if ((ch_state[dir] == MHI_CH_STATE_ENABLED ||
+ ch_state[dir] == MHI_CH_STATE_STOP) &&
!mhi_chan->offload_ch)
mhi_deinit_chan_ctxt(mhi_cntrl, mhi_chan);
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
A client can attempt to unprepare certain channels for transfer even
after the execution environment they are supposed to run in has changed.
In the event that happens, the device need not be notified of the reset
and the host can proceed with clean up for the channel context and
memory allocated for it on the host as the device will no longer be able
to respond to such a request.
Signed-off-by: Bhaumik Bhatt <[email protected]>
Reviewed-by: Hemant Kumar <[email protected]>
---
drivers/bus/mhi/core/main.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c
index 2f6fdb2..f511e3a 100644
--- a/drivers/bus/mhi/core/main.c
+++ b/drivers/bus/mhi/core/main.c
@@ -1283,6 +1283,12 @@ static void __mhi_unprepare_channel(struct mhi_controller *mhi_cntrl,
mutex_lock(&mhi_chan->mutex);
+ if (!(BIT(mhi_cntrl->ee) & mhi_chan->ee_mask)) {
+ dev_dbg(dev, "Current EE: %s Required EE Mask: 0x%x\n",
+ TO_MHI_EXEC_STR(mhi_cntrl->ee), mhi_chan->ee_mask);
+ goto exit_unprepare_channel;
+ }
+
/* no more processing events for this channel */
ret = mhi_update_channel_state(mhi_cntrl, mhi_chan,
MHI_CH_STATE_TYPE_RESET);
@@ -1290,6 +1296,11 @@ static void __mhi_unprepare_channel(struct mhi_controller *mhi_cntrl,
dev_err(dev, "%d: Failed to reset channel, still resetting\n",
mhi_chan->chan);
+exit_unprepare_channel:
+ write_lock_irq(&mhi_chan->lock);
+ mhi_chan->ch_state = MHI_CH_STATE_DISABLED;
+ write_unlock_irq(&mhi_chan->lock);
+
if (!mhi_chan->offload_ch) {
mhi_reset_chan(mhi_cntrl, mhi_chan);
mhi_deinit_chan_ctxt(mhi_cntrl, mhi_chan);
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
The __mhi_unprepare_channel() API does not require the __ prefix.
Get rid of it and make the internal function consistent with the
other function names.
Signed-off-by: Bhaumik Bhatt <[email protected]>
Reviewed-by: Manivannan Sadhasivam <[email protected]>
Reviewed-by: Hemant Kumar <[email protected]>
---
drivers/bus/mhi/core/main.c | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c
index f511e3a..0ac4512 100644
--- a/drivers/bus/mhi/core/main.c
+++ b/drivers/bus/mhi/core/main.c
@@ -1275,8 +1275,8 @@ static int mhi_update_channel_state(struct mhi_controller *mhi_cntrl,
return ret;
}
-static void __mhi_unprepare_channel(struct mhi_controller *mhi_cntrl,
- struct mhi_chan *mhi_chan)
+static void mhi_unprepare_channel(struct mhi_controller *mhi_cntrl,
+ struct mhi_chan *mhi_chan)
{
int ret;
struct device *dev = &mhi_chan->mhi_dev->dev;
@@ -1386,7 +1386,7 @@ int mhi_prepare_channel(struct mhi_controller *mhi_cntrl,
error_pre_alloc:
mutex_unlock(&mhi_chan->mutex);
- __mhi_unprepare_channel(mhi_cntrl, mhi_chan);
+ mhi_unprepare_channel(mhi_cntrl, mhi_chan);
return ret;
}
@@ -1503,7 +1503,7 @@ int mhi_prepare_for_transfer(struct mhi_device *mhi_dev)
if (!mhi_chan)
continue;
- __mhi_unprepare_channel(mhi_cntrl, mhi_chan);
+ mhi_unprepare_channel(mhi_cntrl, mhi_chan);
}
return ret;
@@ -1521,7 +1521,7 @@ void mhi_unprepare_from_transfer(struct mhi_device *mhi_dev)
if (!mhi_chan)
continue;
- __mhi_unprepare_channel(mhi_cntrl, mhi_chan);
+ mhi_unprepare_channel(mhi_cntrl, mhi_chan);
}
}
EXPORT_SYMBOL_GPL(mhi_unprepare_from_transfer);
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
The mhi_prepare_for_transfer() and mhi_unprepare_from_transfer()
APIs could use better explanation. Add details on what MHI does
when these APIs are used.
Signed-off-by: Bhaumik Bhatt <[email protected]>
Reviewed-by: Hemant Kumar <[email protected]>
---
include/linux/mhi.h | 18 ++++++++++++++++--
1 file changed, 16 insertions(+), 2 deletions(-)
diff --git a/include/linux/mhi.h b/include/linux/mhi.h
index d26acc8..56c4c52 100644
--- a/include/linux/mhi.h
+++ b/include/linux/mhi.h
@@ -712,13 +712,27 @@ int mhi_device_get_sync(struct mhi_device *mhi_dev);
void mhi_device_put(struct mhi_device *mhi_dev);
/**
- * mhi_prepare_for_transfer - Setup channel for data transfer
+ * mhi_prepare_for_transfer - Setup UL and DL channels for data transfer.
+ * Allocate and initialize the channel context and
+ * also issue the START channel command to both
+ * channels. Channels can be started only if both
+ * host and device execution environments match and
+ * channels are in a DISABLED state.
* @mhi_dev: Device associated with the channels
*/
int mhi_prepare_for_transfer(struct mhi_device *mhi_dev);
/**
- * mhi_unprepare_from_transfer - Unprepare the channels
+ * mhi_unprepare_from_transfer - Reset UL and DL channels for data transfer.
+ * Issue the RESET channel command and let the
+ * device clean-up the context so no incoming
+ * transfers are seen on the host. Free memory
+ * associated with the context on host. If device
+ * is unresponsive, only perform a host side
+ * clean-up. Channels can be reset only if both
+ * host and device execution environments match
+ * and channels are in an ENABLED, STOPPED or
+ * SUSPENDED state.
* @mhi_dev: Device associated with the channels
*/
void mhi_unprepare_from_transfer(struct mhi_device *mhi_dev);
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
On 2/4/21 12:28 PM, Bhaumik Bhatt wrote:
> Clearing a channel context can happen twice if the client driver
> unprepares and reset the channels from the remove() callback from
> a controller requested MHI power down sequence. If there are
> multiple attempts at calling the mhi_free_coherent() API, we see
> kernel warnings such as "trying to free invalid coherent area".
> Example for one such client is the QRTR MHI driver. Avoid these
> warnings by skipping mhi_deinit_chan_ctxt() API call and prevent
> extra work from MHI as the channels are already disabled.
>
> Signed-off-by: Bhaumik Bhatt <[email protected]>
Reviewed-by: Hemant Kumar <[email protected]>
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
On Thu, Feb 04, 2021 at 12:28:06PM -0800, Bhaumik Bhatt wrote:
> Clearing a channel context can happen twice if the client driver
> unprepares and reset the channels from the remove() callback from
> a controller requested MHI power down sequence. If there are
> multiple attempts at calling the mhi_free_coherent() API, we see
> kernel warnings such as "trying to free invalid coherent area".
> Example for one such client is the QRTR MHI driver. Avoid these
> warnings by skipping mhi_deinit_chan_ctxt() API call and prevent
> extra work from MHI as the channels are already disabled.
>
> Signed-off-by: Bhaumik Bhatt <[email protected]>
Is this patch still valid? We merged a similar fix from Loic for v5.11.
Thanks,
Mani
> ---
> drivers/bus/mhi/core/init.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c
> index 30eef19..272f350 100644
> --- a/drivers/bus/mhi/core/init.c
> +++ b/drivers/bus/mhi/core/init.c
> @@ -1314,6 +1314,7 @@ static int mhi_driver_remove(struct device *dev)
>
> if ((ch_state[dir] == MHI_CH_STATE_ENABLED ||
> ch_state[dir] == MHI_CH_STATE_STOP) &&
> + mhi_chan->ch_state != MHI_CH_STATE_DISABLED &&
> !mhi_chan->offload_ch)
> mhi_deinit_chan_ctxt(mhi_cntrl, mhi_chan);
>
> --
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
> a Linux Foundation Collaborative Project
>
On 2021-02-24 02:10 AM, Manivannan Sadhasivam wrote:
> On Thu, Feb 04, 2021 at 12:28:06PM -0800, Bhaumik Bhatt wrote:
>> Clearing a channel context can happen twice if the client driver
>> unprepares and reset the channels from the remove() callback from
>> a controller requested MHI power down sequence. If there are
>> multiple attempts at calling the mhi_free_coherent() API, we see
>> kernel warnings such as "trying to free invalid coherent area".
>> Example for one such client is the QRTR MHI driver. Avoid these
>> warnings by skipping mhi_deinit_chan_ctxt() API call and prevent
>> extra work from MHI as the channels are already disabled.
>>
>> Signed-off-by: Bhaumik Bhatt <[email protected]>
>
> Is this patch still valid? We merged a similar fix from Loic for v5.11.
>
Yes, both the patches work to solve the same problem. This one just does
it
sooner and relies on internal state variable rather than context being
NULL.
> Thanks,
> Mani
>
>> ---
>> drivers/bus/mhi/core/init.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c
>> index 30eef19..272f350 100644
>> --- a/drivers/bus/mhi/core/init.c
>> +++ b/drivers/bus/mhi/core/init.c
>> @@ -1314,6 +1314,7 @@ static int mhi_driver_remove(struct device *dev)
>>
>> if ((ch_state[dir] == MHI_CH_STATE_ENABLED ||
>> ch_state[dir] == MHI_CH_STATE_STOP) &&
>> + mhi_chan->ch_state != MHI_CH_STATE_DISABLED &&
>> !mhi_chan->offload_ch)
>> mhi_deinit_chan_ctxt(mhi_cntrl, mhi_chan);
>>
>> --
>> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
>> Forum,
>> a Linux Foundation Collaborative Project
>>
Thanks,
Bhaumik
---
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
Forum,
a Linux Foundation Collaborative Project