From: Pawel Laszczak pawell@cadence.com
commit b69ec50b3e55c4b2a85c8bc46763eaf33060584 upstream
For DEV_VER_V3 version there exist race condition between clearing ep_sts.EP_STS_TRBERR and setting ep_cmd.EP_CMD_DRDY bit. Setting EP_CMD_DRDY will be ignored by controller when EP_STS_TRBERR is set. So, between these two instructions we have a small time gap in which the EP_STS_TRBERR can be set. In such case the transfer will not start after setting doorbell.
Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver") cc: stable@vger.kernel.org # 5.4.x Tested-by: Aswath Govindraju a-govindraju@ti.com Reviewed-by: Aswath Govindraju a-govindraju@ti.com Signed-off-by: Pawel Laszczak pawell@cadence.com --- drivers/usb/cdns3/gadget.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
diff --git a/drivers/usb/cdns3/gadget.c b/drivers/usb/cdns3/gadget.c index f624cc87cbab..5edde6a0e77a 100644 --- a/drivers/usb/cdns3/gadget.c +++ b/drivers/usb/cdns3/gadget.c @@ -807,6 +807,19 @@ static void cdns3_wa1_tray_restore_cycle_bit(struct cdns3_device *priv_dev, cdns3_wa1_restore_cycle_bit(priv_ep); }
+static void cdns3_rearm_drdy_if_needed(struct cdns3_endpoint *priv_ep) +{ + struct cdns3_device *priv_dev = priv_ep->cdns3_dev; + + if (priv_dev->dev_ver < DEV_VER_V3) + return; + + if (readl(&priv_dev->regs->ep_sts) & EP_STS_TRBERR) { + writel(EP_STS_TRBERR, &priv_dev->regs->ep_sts); + writel(EP_CMD_DRDY, &priv_dev->regs->ep_cmd); + } +} + /** * cdns3_ep_run_transfer - start transfer on no-default endpoint hardware * @priv_ep: endpoint object @@ -1003,6 +1016,7 @@ int cdns3_ep_run_transfer(struct cdns3_endpoint *priv_ep, /*clearing TRBERR and EP_STS_DESCMIS before seting DRDY*/ writel(EP_STS_TRBERR | EP_STS_DESCMIS, &priv_dev->regs->ep_sts); writel(EP_CMD_DRDY, &priv_dev->regs->ep_cmd); + cdns3_rearm_drdy_if_needed(priv_ep); trace_cdns3_doorbell_epx(priv_ep->name, readl(&priv_dev->regs->ep_traddr)); }
On Thu, Sep 30, 2021 at 11:42:17AM +0200, Pawel Laszczak wrote:
From: Pawel Laszczak pawell@cadence.com
commit b69ec50b3e55c4b2a85c8bc46763eaf33060584 upstream
For DEV_VER_V3 version there exist race condition between clearing ep_sts.EP_STS_TRBERR and setting ep_cmd.EP_CMD_DRDY bit. Setting EP_CMD_DRDY will be ignored by controller when EP_STS_TRBERR is set. So, between these two instructions we have a small time gap in which the EP_STS_TRBERR can be set. In such case the transfer will not start after setting doorbell.
Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver") cc: stable@vger.kernel.org # 5.4.x Tested-by: Aswath Govindraju a-govindraju@ti.com Reviewed-by: Aswath Govindraju a-govindraju@ti.com Signed-off-by: Pawel Laszczak pawell@cadence.com
drivers/usb/cdns3/gadget.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
What kernel(s) are you wanting this applied to?
thanks,
greg k-h
On Thu, Sep 30, 2021 at 11:42:17AM +0200, Pawel Laszczak wrote:
From: Pawel Laszczak pawell@cadence.com
commit b69ec50b3e55c4b2a85c8bc46763eaf33060584 upstream
For DEV_VER_V3 version there exist race condition between clearing ep_sts.EP_STS_TRBERR and setting ep_cmd.EP_CMD_DRDY bit. Setting EP_CMD_DRDY will be ignored by controller when EP_STS_TRBERR is set. So, between these two instructions we have a small time gap in which the EP_STS_TRBERR can be set. In such case the transfer will not start after setting doorbell.
Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver") cc: stable@vger.kernel.org # 5.4.x Tested-by: Aswath Govindraju a-govindraju@ti.com Reviewed-by: Aswath Govindraju a-govindraju@ti.com Signed-off-by: Pawel Laszczak pawell@cadence.com
drivers/usb/cdns3/gadget.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
What kernel(s) are you wanting this applied to?
To 5.4. I added information in cc: stable@vger.kernel.org tag (# 5.4.x) . Is it sufficient or not? I ask because I need to post this fix also to v5.10.
Thanks, Pawel
thanks,
greg k-h
On Thu, Sep 30, 2021 at 10:22:42AM +0000, Pawel Laszczak wrote:
On Thu, Sep 30, 2021 at 11:42:17AM +0200, Pawel Laszczak wrote:
From: Pawel Laszczak pawell@cadence.com
commit b69ec50b3e55c4b2a85c8bc46763eaf33060584 upstream
For DEV_VER_V3 version there exist race condition between clearing ep_sts.EP_STS_TRBERR and setting ep_cmd.EP_CMD_DRDY bit. Setting EP_CMD_DRDY will be ignored by controller when EP_STS_TRBERR is set. So, between these two instructions we have a small time gap in which the EP_STS_TRBERR can be set. In such case the transfer will not start after setting doorbell.
Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver") cc: stable@vger.kernel.org # 5.4.x Tested-by: Aswath Govindraju a-govindraju@ti.com Reviewed-by: Aswath Govindraju a-govindraju@ti.com Signed-off-by: Pawel Laszczak pawell@cadence.com
drivers/usb/cdns3/gadget.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
What kernel(s) are you wanting this applied to?
To 5.4. I added information in cc: stable@vger.kernel.org tag (# 5.4.x) . Is it sufficient or not? I ask because I need to post this fix also to v5.10.
I queued this up for both 5.10 and 5.4, thanks.
The issue seems to be that in the upstream patch you explicitly stated to go only to 5.12:
cc: stable@vger.kernel.org # 5.12.x
Was that your intent?
On Thu, Sep 30, 2021 at 10:22:42AM +0000, Pawel Laszczak wrote:
On Thu, Sep 30, 2021 at 11:42:17AM +0200, Pawel Laszczak wrote:
From: Pawel Laszczak pawell@cadence.com
commit b69ec50b3e55c4b2a85c8bc46763eaf33060584 upstream
For DEV_VER_V3 version there exist race condition between clearing ep_sts.EP_STS_TRBERR and setting ep_cmd.EP_CMD_DRDY bit. Setting EP_CMD_DRDY will be ignored by controller when EP_STS_TRBERR is set. So, between these two instructions we have a small time gap in which the EP_STS_TRBERR can be set. In such case the transfer will not start after setting doorbell.
Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver") cc: stable@vger.kernel.org # 5.4.x Tested-by: Aswath Govindraju a-govindraju@ti.com Reviewed-by: Aswath Govindraju a-govindraju@ti.com Signed-off-by: Pawel Laszczak pawell@cadence.com
drivers/usb/cdns3/gadget.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
What kernel(s) are you wanting this applied to?
To 5.4. I added information in cc: stable@vger.kernel.org tag (# 5.4.x) . Is it sufficient or not? I ask because I need to post this fix also to v5.10.
I queued this up for both 5.10 and 5.4, thanks.
The issue seems to be that in the upstream patch you explicitly stated to go only to 5.12:
cc: stable@vger.kernel.org # 5.12.x
Was that your intent?
Yes. Patch could not be applied to older kernel version because files name were changed.
Thanks Pawel Laszczak
-- Thanks, Sasha
linux-stable-mirror@lists.linaro.org