Variables real & fake port do not convey their purpose, thus they are
replaced with a pointer to the root hub port 'struct xhci_port *rhub_port'.
'rhub_port' contains real & fake ports in zero-based format, which happens
to be more widely used inside the xHCI driver:
- 'real_port' is ('rhub_port->hw_portnum' + 1)
- 'fake_port' is ('rhub_port->hcd_portnum' + 1)
One reason for real port being one-based, is to signal other functions in
case struct 'xhci_virt_device' initialization failed, in this case the
value will remain 0. This is no longer needed, instead we check whether
or not 'rhub_port' is 'NULL'.
Signed-off-by: Niklas Neronin <niklas.neronin@intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Link: https://lore.kernel.org/r/20240229141438.619372-3-mathias.nyman@linux.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Calculate the IN/OUT budget separately to improve the bandwidth schedule,
meanwhile should avoid Start-Split token overlap between IN and OUT
endpoints, and take into account the FS/LS bandwidth boundary in each
microframe and also in each FS frame.
Calculate the budget for SS of OUT eps and CS of IN eps, but not include
extra-cs, and always add at most extra-cs allowed.
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/20230830122820.18859-2-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This happens when @udev->reset_resume is set to true, when usb resume,
the flow as below:
- hub_resume
- usb_disable_interface
- usb_disable_endpoint
- usb_hcd_disable_endpoint
- xhci_endpoint_disable // it set @ep->hcpriv to NULL
Then when reset usb device, it will drop allocated endpoints,
the flow as below:
- usb_reset_and_verify_device
- usb_hcd_alloc_bandwidth
- xhci_mtk_drop_ep
but @ep->hcpriv is already set to NULL, the bandwidth will be not
released anymore.
Due to the added endponts are stored in hash table, we can drop the check
of @ep->hcpriv.
Fixes: 4ce186665e ("usb: xhci-mtk: Do not use xhci's virt_dev in drop_endpoint")
Cc: stable <stable@kernel.org>
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/20220819080556.32215-2-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently uses the worst case byte budgets on FS/LS bus bandwidth,
for example, for an isochronos IN endpoint with 192 bytes budget, it
will consume the whole 5 uframes(188 * 5) while the actual FS bus
budget should be just 192 bytes. It cause that many usb audio headsets
with 3 interfaces (audio input, audio output, and HID) cannot be
configured.
To improve it, changes to use "approximate" best case budget for FS/LS
bandwidth management. For the same endpoint from the above example,
the approximate best case budget is now reduced to (188 * 2) bytes.
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Cc: stable <stable@kernel.org>
Link: https://lore.kernel.org/r/20220819080556.32215-1-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The bandwidth budget table is introduced to trace ideal bandwidth used
by each INT/ISOC endpoint, but in fact the endpoint may consume more
bandwidth and cause data transfer error, so it's better to leave some
margin. Obviously it's difficult to find the best margin for all cases,
instead take use of the worst-case scenario.
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/20220512064931.31670-2-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci-mtk has 64 slots for periodic bandwidth calculations and each
slot represents byte budgets on a microframe. When an endpoint's
allocation sits on the boundary of the table, byte budgets' slot
can be rolled over but the current implementation doesn't.
This patch allows the microframe index rollover and prevent
out-of-bounds array access.
Signed-off-by: Ikjoon Jang <ikjn@chromium.org>
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/20210827033105.26595-1-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci-mtk depends on xhci's internal virt_dev when it retrieves its
internal data from usb_host_endpoint both in add_endpoint and
drop_endpoint callbacks. But when setup packet was retired by
transaction errors in xhci_setup_device() path, a virt_dev for the slot
is newly created with real_port 0. This leads to xhci-mtks's NULL pointer
dereference from drop_endpoint callback as xhci-mtk assumes that virt_dev's
real_port is always started from one. The similar problems were addressed
by [1] but that can't cover the failure cases from setup_device.
This patch drops the usages of xhci's virt_dev in xhci-mtk's drop_endpoint
callback by adopting hashtable for searching mtk's schedule entity
from a given usb_host_endpoint pointer instead of searching a linked list.
So mtk's drop_endpoint callback doesn't have to rely on virt_dev at all.
[1] f351f4b63d ("usb: xhci-mtk: fix oops when unbind driver")
Signed-off-by: Ikjoon Jang <ikjn@chromium.org>
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/20210826025144.51992-5-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
I find the patch introduce some issues, e.g.
1. oops happens when xhci_gen_setup() failed, and hash is not init
but try to destroy it;
2. memory leakage happens when fail to insert ep, need free sch_ep,
or insert ep after insert int list;
3. memory leakage happens when fail to allocate sch_array, need destroy
rhashtable;
4. it's better to check ep->hcpriv when drop ep;
so prefer to revert this patch, and resend it after the issues are fixed.
This reverts commit b873120995.
Cc: Ikjoon Jang <ikjn@chromium.org>
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/20210820065913.64490-2-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently xhci-mtk needs software-managed bandwidth allocation for
periodic endpoints, it allocates the microframe index for the first
start-split packet for each endpoint. As this index allocation logic
should avoid the conflicts with other full/low-speed periodic endpoints,
it uses the worst case byte budgets on high-speed bus bandwidth
For example, for an isochronos IN endpoint with 192 bytes budget,
it will consume the whole 4 u-frames(188 * 4) while the actual
full-speed bus budget should be just 192bytes.
This patch changes the low/full-speed bandwidth allocation logic
to use "approximate" best case budget for lower speed bandwidth
management. For the same endpoint from the above example, the
approximate best case budget is now reduced to (188 * 2) bytes.
Without this patch, many usb audio headsets with 3 interfaces
(audio input, audio output, and HID) cannot be configured
on xhci-mtk.
Signed-off-by: Ikjoon Jang <ikjn@chromium.org>
Link: https://lore.kernel.org/r/20210805133937.1.Ia8174b875bc926c12ce427a5a1415dea31cc35ae@changeid
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci-mtk depends on xhci's internal virt_dev when it retrieves its
internal data from usb_host_endpoint both in add_endpoint and
drop_endpoint callbacks. But when setup packet was retired by
transaction errors in xhci_setup_device() path, a virt_dev for the slot
is newly created with real_port 0. This leads to xhci-mtks's NULL pointer
dereference from drop_endpoint callback as xhci-mtk assumes that virt_dev's
real_port is always started from one. The similar problems were addressed
by [1] but that can't cover the failure cases from setup_device.
This patch drops the usages of xhci's virt_dev in xhci-mtk's drop_endpoint
callback by adopting rhashtable for searching mtk's schedule entity
from a given usb_host_endpoint pointer instead of searching a linked list.
So mtk's drop_endpoint callback doesn't have to rely on virt_dev at all.
[1] https://lore.kernel.org/r/1617179142-2681-2-git-send-email-chunfeng.yun@mediatek.com
Signed-off-by: Ikjoon Jang <ikjn@chromium.org>
Link: https://lore.kernel.org/r/20210805133731.1.Icc0f080e75b1312692d4c7c7d25e7df9fe1a05c2@changeid
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The oops happens when unbind driver through sysfs as following,
because xhci_mtk_drop_ep() try to drop the endpoint of root hub
which is not added by xhci_add_endpoint() and the virtual device
is not allocated, in fact also needn't drop it, so should skip it.
Call trace:
xhci_mtk_drop_ep+0x1b8/0x298
usb_hcd_alloc_bandwidth+0x1d8/0x380
usb_disable_device_endpoints+0x8c/0xe0
usb_disable_device+0x128/0x168
usb_disconnect+0xbc/0x2c8
usb_remove_hcd+0xd8/0x210
xhci_mtk_remove+0x98/0x108
platform_remove+0x28/0x60
device_release_driver_internal+0x110/0x1e8
device_driver_detach+0x18/0x28
unbind_store+0xd4/0x108
drv_attr_store+0x24/0x38
Fixes: 14295a1500 ("usb: xhci-mtk: support to build xhci-mtk-hcd.ko")
Reported-by: Eddie Hung <eddie.hung@mediatek.com>
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/1617179142-2681-2-git-send-email-chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In USB2 Spec:
"11.18.5 TT Response Generation
In general, there will be two (or more) complete-split
transactions scheduled for a periodic endpoint.
However, for interrupt endpoints, the maximum size of
the full-/low-speed transaction guarantees that it can
never require more than two complete-split transactions.
Two complete-split transactions are only required
when the transaction spans a microframe boundary."
Due to the maxp is 64, and less then 188 (at most in one
microframe), seems never span boundary, so use only one CS
for FS/LS interrupt transfer, this will save some bandwidth.
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Link: https://lore.kernel.org/r/5b9ff09f53d23cf9e5c5437db4ffc18b798bf60c.1615170625.git.chunfeng.yun@mediatek.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci-mtk needs XHCI_MTK_HOST quirk functions in add_endpoint() and
drop_endpoint() to handle its own sw bandwidth management.
It stores bandwidth data into an internal table every time
add_endpoint() is called, and drops those in drop_endpoint().
But when bandwidth allocation fails at one endpoint, all earlier
allocation from the same interface could still remain at the table.
This patch moves bandwidth management codes to check_bandwidth() and
reset_bandwidth() path. To do so, this patch also adds those functions
to xhci_driver_overrides and lets mtk-xhci to release all failed
endpoints in reset_bandwidth() path.
Fixes: 08e469de87 ("usb: xhci-mtk: supports bandwidth scheduling with multi-TT")
Signed-off-by: Ikjoon Jang <ikjn@chromium.org>
Link: https://lore.kernel.org/r/20210113180444.v6.1.Id0d31b5f3ddf5e734d2ab11161ac5821921b1e1e@changeid
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Supports SSP scheduling only for SSP device directly connected
to root hub but not through external USB3 gen2 hub which need
use a new scheduling way.
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Mainly improve SuperSpeed ISOC bandwidth in last microframe,
and LowSpeed/FullSpeed IN INT/ISOC bandwidth in split and
idle microframes by introduing a bandwidth budget table;
Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
