On 5/7/24 6:48 AM, Lk Sii wrote:
On 2024/5/4 05:51, quic_zijuhu wrote:
On 5/4/2024 5:25 AM, Luiz Augusto von Dentz wrote:
Hi,
On Fri, May 3, 2024 at 4:18 PM quic_zijuhu <quic_zijuhu@xxxxxxxxxxx> wrote:
On 5/4/2024 3:22 AM, Luiz Augusto von Dentz wrote:
Hi Zijun,
On Thu, May 2, 2024 at 10:06 AM Zijun Hu <quic_zijuhu@xxxxxxxxxxx> wrote:
Commit 272970be3dab ("Bluetooth: hci_qca: Fix driver shutdown on closed
serdev") will cause below regression issue:
BT can't be enabled after below steps:
cold boot -> enable BT -> disable BT -> warm reboot -> BT enable failure
if property enable-gpios is not configured within DT|ACPI for QCA6390.
The commit is to fix a use-after-free issue within qca_serdev_shutdown()
during reboot, but also introduces this regression issue regarding above
steps since the VSC is not sent to reset controller during warm reboot.
Fixed by sending the VSC to reset controller within qca_serdev_shutdown()
once BT was ever enabled, and the use-after-free issue is also be fixed
by this change since serdev is still opened when send to serdev.
Fixes: 272970be3dab ("Bluetooth: hci_qca: Fix driver shutdown on closed serdev")
Cc: stable@xxxxxxxxxxxxxxx
Reported-by: Wren Turkal <wt@xxxxxxxxxxxxxxxx>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=218726
Signed-off-by: Zijun Hu <quic_zijuhu@xxxxxxxxxxx>
Tested-by: Wren Turkal <wt@xxxxxxxxxxxxxxxx>
---
drivers/bluetooth/hci_qca.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/drivers/bluetooth/hci_qca.c b/drivers/bluetooth/hci_qca.c
index 0c9c9ee56592..8e35c9091486 100644
--- a/drivers/bluetooth/hci_qca.c
+++ b/drivers/bluetooth/hci_qca.c
@@ -2450,13 +2450,12 @@ static void qca_serdev_shutdown(struct device *dev)
struct qca_serdev *qcadev = serdev_device_get_drvdata(serdev);
struct hci_uart *hu = &qcadev->serdev_hu;
struct hci_dev *hdev = hu->hdev;
- struct qca_data *qca = hu->priv;
const u8 ibs_wake_cmd[] = { 0xFD };
const u8 edl_reset_soc_cmd[] = { 0x01, 0x00, 0xFC, 0x01, 0x05 };
if (qcadev->btsoc_type == QCA_QCA6390) {
- if (test_bit(QCA_BT_OFF, &qca->flags) ||
- !test_bit(HCI_RUNNING, &hdev->flags))
This probably deserves a comment on why you end up with
HCI_QUIRK_NON_PERSISTENT_SETUP and HCI_SETUP flags here, also why you
are removing the flags above since that was introduce to prevent
use-after-free this sort of revert it so I do wonder how serdev can
still be open if you haven't tested for QCA_BT_OFF for example?
okay, let me give comments at next version.
this design logic is shown below. you maybe review it.
if HCI_QUIRK_NON_PERSISTENT_SETUP is set, it means that hdev->setup()
is able to be invoked by every open() to initializate SoC without any
help. so we don't need to send the VSC to reset SoC into initial and
clean state for the next hdev->setup() call success.
otherwise, namely, HCI_QUIRK_NON_PERSISTENT_SETUP is not set.
if HCI_SETUP is set, it means hdev->setup() was never be invoked, so the
SOC is already in the initial and clean state, so we also don't need to
send the VSC to reset SOC.
otherwise, we need to send the VSC to reset Soc into a initial and clean
state for hdev->setup() call success after "warm reboot -> enable BT"
for the case commit message cares about, the only factor which decide to
send the VSC is that SoC is a initial and clean state or not after warm
reboot, any other factors are irrelevant to this decision.
why the serdev is still open after go through
(test_bit(HCI_QUIRK_NON_PERSISTENT_SETUP, &hdev->quirks)
|| hci_dev_test_flag(hdev, HCI_SETUP) checking is that
serdev is not closed by hci_uart_close().
Sounds like a logical jump to me, in fact hci_uart_close doesn't
really change any of those flags, beside these flags are not really
meant to tell the driver if serdev_device_close has been called or not
which seems to be the intention with HCI_UART_PROTO_READY so how about
we use that instead?
sorry for that i maybe not give good explanation, let me explain again.
hci_uart_close() is the only point which maybe close serdev before
qca_serdev_shutdown() is called, but for our case that
HCI_QUIRK_NON_PERSISTENT_SETUP is NOT set, hci_uart_close() will not
close serdev for our case, so serdev must be open state before sending
the VSC. so should not need other checking.
hello, i have paid attention to your discussion for a long time, i would
like to join this discussion now.
The serdev is still open before sending the VSC for this patch.
are you agree with above Zijun's point?
I will say that this part of the discussion seems to be addressing KK's
concerns.
@KK, is this accurate?
@Zijun, this description along with the info about the baud rate issues
should probably be part of the commit message. In these last two
messages, you have been much clearer about why this logic is needed and
correct. I wish you'd provided this description from the beginning when
KK asked for more information about the logic change.
Another thing that is troubling me is that having traffic on shutdown
is not common, specially if you are going to reboot, etc, and even if
it doesn't get power cycle why don't you reset on probe rather than
shutdown? That way we don't have to depend on what has been done in a
previous boot, which can really become a problem in case of multi-OS
where you have another system that may not be doing what you expect.
as you know, BT UART are working at 3M baudrate for normal usage.
we can't distinguish if SoC expects 3M or default 11.52K baudarate
during probe() after reboot. so we send the VSC within shutdown to make
sure SoC enter a initial state with 11.52 baudrate.
for cold boot, SOC expects default 11.52K baudrate for probe().
for Enable BT -> warm boot, SOC expects 3M baudrate for probe().
we can't tell these two case within probe(). so need to send the VSC
within shutdown().
it seems the traffic within qca_serdev_shutdown() actually does software
reset for BT SOC.
From Zijun's points. the reasons why to do software reset within
shutdown() instead of probe() maybe be shown below
1) it is impossible to do software reset within probe().
2) it seems it is easier to do it within shutdown() than probe.
Zijun's simple fix only change the condition and does NOT change the
location to send the VSC, i think it maybe be other topic about location
where(probe() or shutdown()) to do software reset.
are you agree with this point?
From a practical standpoint, this change does seem to fix the warm boot
issue on my laptop. I do not think that it would fix the issue of
booting from an OS that puts the hardware into an unknown state.
For concern about multi-OS, i would like to show my points.
the patch is for linux kernel, we maybe only need to care about linux
OS. it maybe out-of-scope to make assumptions about other OSs vendor
announced supported such as Windows OS we don't known much about.
are you agree with this point?
I would not fully agree with this point. However, I would agree
completely with your proposal for moving forward (i.e. landing the change).
I would say that it is a problem if the kernel is not doing something to
setup the hardware correctly in any case where it is technically
possible. That is clearly a bug in the driver and Qualcomm should take
responsibility for fixing this poor design. Luiz is totally right here.
Having said that, I don't see that bug as a blocker for a logic fix that
creates an obvious (in my mind) UX improvement.
Here's my view of the situation. Right now, I experience a bug on every
single warm boot or module reload).
After Zijun's improvement commit, I might experience this problem iif
the right set of rare circumstances occurs (i.e. whenever I warm boot
from an OS that puts the hardware in an unknown state, like the current
mainline kernel to a kernel with the improvement).
In the world where the design problem of the init/shutdown sequences are
fixed AND this logic change is applied, I might never see this problem
again.
These seem like two different problems in my head. They don't seem
directly logically related, and the blast radii of the problems are very
different.
Here's the facts as I see them:
1. Logic improvement problem and init/shutdown sequence problem are 2
orthogonal problems
2. Logic improvement greatly users' chances of running into the bad UX
of the current code.
3. Init/shutdown isn't a trivial extension from the logic improvement.
4. Logic improvement has a pretty low cost to apply.
5. Init/shutdown sequence fix seems to be more fundamental and more
intrusive.
All of these together indicate to me that the logic improvement should
be landed. The init/shutdown issue should also be fixed, but a fix for
that issue should not block the logic improvement.
My basic reasoning for this is that a visible UX fix should not e
blocked for a change like the init/shutdown logic fix when the logic fix
will help users.
However, Qualcomm does need to feel some pressure to fix their driver
code. I would like to think that user's will not be held hostage for
putting pressure on a vendor in this case as the balance seems very
intrusive for users. It certainly feels intrusive as a user to have to
be very careful about how I reboot my laptop.
see hci_uart_close() within drivers/bluetooth/hci_serdev.c
static int hci_uart_close(struct hci_dev *hdev)
{
......
/* When QUIRK HCI_QUIRK_NON_PERSISTENT_SETUP is set by driver,
* BT SOC is completely powered OFF during BT OFF, holding port
* open may drain the battery.
*/
if (test_bit(HCI_QUIRK_NON_PERSISTENT_SETUP, &hdev->quirks)) {
clear_bit(HCI_UART_PROTO_READY, &hu->flags);
serdev_device_close(hu->serdev);
}
return 0;
}
+ if (test_bit(HCI_QUIRK_NON_PERSISTENT_SETUP, &hdev->quirks) ||
+ hci_dev_test_flag(hdev, HCI_SETUP))
return;
serdev_device_write_flush(serdev);
--
2.7.4
--
You're more amazing than you think!
