| Message ID | 20210611071805.9169-1-yuan.fang@verisilicon.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drivers/usb/dwc3: Set PHY ready after soft reset done | expand |
On Fri, Jun 11, 2021 at 03:18:05PM +0800, Yuan Fang wrote: > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access > it, But i have a wrong PHY clock default setting in PHY > register which means PHY is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite > usb get PHYs with return -ENODEV, and set phys_ready=true > but at this point, we can not say phys is ready or at least > it not safe to do that. > then, go on with dwc3_core_soft_reset,dwc3 reset core and > PHY, as phy->init is NULL in this case, so, usb_phy_init > do nothing and go on with a while loop which cost 1000* > 20 ms, and return with -ETIMEDOUT > check phys_ready is set to TRUE but actually not. > Move phys_ready after soft reset done is more reasonable > and don't see side-effect yet. > > Signed-off-by: Yuan Fang <yuan.fang@verisilicon.com> What commit does this "fix"? Should it also be backported to older kernels? If so, how far back? thanks, greg k-h
Hi Maintainer, Thanks so much for the quick reply. I would like to back ported to older kernels, but I can not figure out to which version is suitable, I can have a check later But merge into the new release version is also acceptable as I think. Thanks again. B.R Yuan Fang -----邮件原件----- 发件人: Greg KH [mailto:gregkh@linuxfoundation.org] 发送时间: 2021年6月11日 15:26 收件人: Yuan Fang 抄送: linux-usb@vger.kernel.org; balbi@kernel.org; Fang, Yuan 主题: Re: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done On Fri, Jun 11, 2021 at 03:18:05PM +0800, Yuan Fang wrote: > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true but at this point, > we can not say phys is ready or at least it not safe to do that. > then, go on with dwc3_core_soft_reset,dwc3 reset core and PHY, as > phy->init is NULL in this case, so, usb_phy_init do nothing and go on > with a while loop which cost 1000* > 20 ms, and return with -ETIMEDOUT > check phys_ready is set to TRUE but actually not. > Move phys_ready after soft reset done is more reasonable and don't see > side-effect yet. > > Signed-off-by: Yuan Fang <yuan.fang@verisilicon.com> What commit does this "fix"? Should it also be backported to older kernels? If so, how far back? thanks, greg k-h
Yuan Fang <fangyuanseu@gmail.com> writes: > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access > it, But i have a wrong PHY clock default setting in PHY > register which means PHY is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite > usb get PHYs with return -ENODEV, and set phys_ready=true your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist.
Hi Balbi, > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true > your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist. Sorry for the late reply, my explain may be not clear, let me describe it once more 1, I don’t coding a PHY driver, I have a PHY which do NOT need SW operation All I should do is set the related PHY ctrl regs correctly. So I do not return anything to generic PHY framework. 2, In my above case, dwc3_core_get_phy will return 0, and set dwc->phys_ready=true, And then, dwc3_core_soft_reset will cost retries*20ms to check if soft-reset is done , and return -ETIMEDOUT if time out, is that correct? My point is, in this case(PHY do not set correctly, which is not rare during development stage I think), PHY initial fail led to soft-reset timeout but dwc->phys_ready is true, So I think is more safe to say "PHY is ready" after soft-reset done, please let me know your opinion, thanks a lot Yuan Fang -----邮件原件----- 发件人: Felipe Balbi [mailto:balbi@kernel.org] 发送时间: 2021年6月11日 21:03 收件人: Yuan Fang; linux-usb@vger.kernel.org 抄送: Fang, Yuan 主题: Re: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done Yuan Fang <fangyuanseu@gmail.com> writes: > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist. -- balbi
Hi Maintainner, Could you please share your thoughts with me? Thanks a lot. -----邮件原件----- 发件人: Fang, Yuan 发送时间: 2021年6月16日 10:56 收件人: 'Felipe Balbi'; Yuan Fang; linux-usb@vger.kernel.org 主题: 答复: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done Hi Balbi, > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true > your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist. Sorry for the late reply, my explain may be not clear, let me describe it once more 1, I don’t coding a PHY driver, I have a PHY which do NOT need SW operation All I should do is set the related PHY ctrl regs correctly. So I do not return anything to generic PHY framework. 2, In my above case, dwc3_core_get_phy will return 0, and set dwc->phys_ready=true, And then, dwc3_core_soft_reset will cost retries*20ms to check if soft-reset is done , and return -ETIMEDOUT if time out, is that correct? My point is, in this case(PHY do not set correctly, which is not rare during development stage I think), PHY initial fail led to soft-reset timeout but dwc->phys_ready is true, So I think is more safe to say "PHY is ready" after soft-reset done, please let me know your opinion, thanks a lot Yuan Fang -----邮件原件----- 发件人: Felipe Balbi [mailto:balbi@kernel.org] 发送时间: 2021年6月11日 21:03 收件人: Yuan Fang; linux-usb@vger.kernel.org 抄送: Fang, Yuan 主题: Re: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done Yuan Fang <fangyuanseu@gmail.com> writes: > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist. -- balbi
Hi Maintainer, May I know your thoughts on this topic? Thanks a lot. -----邮件原件----- 发件人: Fang, Yuan 发送时间: 2021年6月29日 10:42 收件人: Fang, Yuan; 'Felipe Balbi'; 'Yuan Fang'; 'linux-usb@vger.kernel.org' 主题: 答复: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done Hi Maintainner, Could you please share your thoughts with me? Thanks a lot. -----邮件原件----- 发件人: Fang, Yuan 发送时间: 2021年6月16日 10:56 收件人: 'Felipe Balbi'; Yuan Fang; linux-usb@vger.kernel.org 主题: 答复: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done Hi Balbi, > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true > your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist. Sorry for the late reply, my explain may be not clear, let me describe it once more 1, I don’t coding a PHY driver, I have a PHY which do NOT need SW operation All I should do is set the related PHY ctrl regs correctly. So I do not return anything to generic PHY framework. 2, In my above case, dwc3_core_get_phy will return 0, and set dwc->phys_ready=true, And then, dwc3_core_soft_reset will cost retries*20ms to check if soft-reset is done , and return -ETIMEDOUT if time out, is that correct? My point is, in this case(PHY do not set correctly, which is not rare during development stage I think), PHY initial fail led to soft-reset timeout but dwc->phys_ready is true, So I think is more safe to say "PHY is ready" after soft-reset done, please let me know your opinion, thanks a lot Yuan Fang -----邮件原件----- 发件人: Felipe Balbi [mailto:balbi@kernel.org] 发送时间: 2021年6月11日 21:03 收件人: Yuan Fang; linux-usb@vger.kernel.org 抄送: Fang, Yuan 主题: Re: [PATCH] drivers/usb/dwc3: Set PHY ready after soft reset done Yuan Fang <fangyuanseu@gmail.com> writes: > It's more safe to set PHY ready after soft reset done > > Let's consider a test case like this: > I have a usb PHY which don't need SW initial before access it, But i > have a wrong PHY clock default setting in PHY register which means PHY > is not ready in fact. > in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get > PHYs with return -ENODEV, and set phys_ready=true your PHY driver shouldn't return -ENODEV here. Why are you returning -ENODEV? The PHY isn't ready but it _does_ exist. -- balbi
diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c index b6e53d8212cd..04b1bbaf694a 100644 --- a/drivers/usb/dwc3/core.c +++ b/drivers/usb/dwc3/core.c @@ -324,6 +324,7 @@ static int dwc3_core_soft_reset(struct dwc3 *dwc) if (DWC3_VER_IS_WITHIN(DWC31, ANY, 180A)) msleep(50); + dwc->phys_ready = true; return 0; } @@ -979,7 +980,6 @@ static int dwc3_core_init(struct dwc3 *dwc) ret = dwc3_core_get_phy(dwc); if (ret) goto err0a; - dwc->phys_ready = true; } ret = dwc3_core_soft_reset(dwc);
It's more safe to set PHY ready after soft reset done Let's consider a test case like this: I have a usb PHY which don't need SW initial before access it, But i have a wrong PHY clock default setting in PHY register which means PHY is not ready in fact. in dwc3_core_init, dwc3_core_get_phy will return 0 despite usb get PHYs with return -ENODEV, and set phys_ready=true but at this point, we can not say phys is ready or at least it not safe to do that. then, go on with dwc3_core_soft_reset,dwc3 reset core and PHY, as phy->init is NULL in this case, so, usb_phy_init do nothing and go on with a while loop which cost 1000* 20 ms, and return with -ETIMEDOUT check phys_ready is set to TRUE but actually not. Move phys_ready after soft reset done is more reasonable and don't see side-effect yet. Signed-off-by: Yuan Fang <yuan.fang@verisilicon.com> --- Hi, please take a look at issue description above and kindly review this CL if any side-effect, many thanks. B.R Yuan Fang drivers/usb/dwc3/core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)