| Message ID | 1586207077-22361-4-git-send-email-jhugo@codeaurora.org (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Misc MHI fixes | expand |
Hi Jeff, On Mon, Apr 06, 2020 at 03:04:37PM -0600, Jeffrey Hugo wrote: > If the MHI core detects invalid data due to a PCI read, it calls into > the controller via link_status() to double check that the link is infact > down. All in all, this is pretty pointless, and racy. There are no good > reasons for this, and only drawbacks. > > Its pointless because chances are, the controller is going to do the same > thing to determine if the link is down - attempt a PCI access and compare > the result. This does not make the link status decision any smarter. > > Its racy because its possible that the link was down at the time of the > MHI core access, but then recovered before the controller access. In this > case, the controller will indicate the link is not down, and the MHI core > will precede to use a bad value as the MHI core does not attempt to retry > the access. > > Retrying the access in the MHI core is a bad idea because again, it is > racy - what if the link is down again? Furthermore, there may be some > higher level state associated with the link status, that is now invalid > because the link went down. > > The only reason why the MHI core could see "invalid" data when doing a PCI > access, that is actually valid, is if the register actually contained the > PCI spec defined sentinel for an invalid access. In this case, it is > arguable that the MHI implementation broken, and should be fixed, not > worked around. > > Therefore, remove the link_status() callback before anyone attempts to > implement it. > > Signed-off-by: Jeffrey Hugo <jhugo@codeaurora.org> LGTM. But as per the IRC discussion I'd like the mhi_reg_read() to be implemented as a callback in mhi_controller struct inorder to truly make MHI a PCI agnostic bus. Since we don't have any controller driver in mainline, I think it is the good time to do this change. For this, Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org> Thanks, Mani > --- > drivers/bus/mhi/core/init.c | 6 ++---- > drivers/bus/mhi/core/main.c | 5 ++--- > include/linux/mhi.h | 2 -- > 3 files changed, 4 insertions(+), 9 deletions(-) > > diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c > index b38359c..2af08d57 100644 > --- a/drivers/bus/mhi/core/init.c > +++ b/drivers/bus/mhi/core/init.c > @@ -812,10 +812,8 @@ int mhi_register_controller(struct mhi_controller *mhi_cntrl, > if (!mhi_cntrl) > return -EINVAL; > > - if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put) > - return -EINVAL; > - > - if (!mhi_cntrl->status_cb || !mhi_cntrl->link_status) > + if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put || > + !mhi_cntrl->status_cb) > return -EINVAL; > > ret = parse_config(mhi_cntrl, config); > diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c > index eb4256b..473278b8 100644 > --- a/drivers/bus/mhi/core/main.c > +++ b/drivers/bus/mhi/core/main.c > @@ -20,9 +20,8 @@ int __must_check mhi_read_reg(struct mhi_controller *mhi_cntrl, > { > u32 tmp = readl(base + offset); > > - /* If there is any unexpected value, query the link status */ > - if (PCI_INVALID_READ(tmp) && > - mhi_cntrl->link_status(mhi_cntrl)) > + /* If the value is invalid, the link is down */ > + if (PCI_INVALID_READ(tmp)) > return -EIO; > > *out = tmp; > diff --git a/include/linux/mhi.h b/include/linux/mhi.h > index ad19960..be704a4 100644 > --- a/include/linux/mhi.h > +++ b/include/linux/mhi.h > @@ -335,7 +335,6 @@ struct mhi_controller_config { > * @syserr_worker: System error worker > * @state_event: State change event > * @status_cb: CB function to notify power states of the device (required) > - * @link_status: CB function to query link status of the device (required) > * @wake_get: CB function to assert device wake (optional) > * @wake_put: CB function to de-assert device wake (optional) > * @wake_toggle: CB function to assert and de-assert device wake (optional) > @@ -417,7 +416,6 @@ struct mhi_controller { > > void (*status_cb)(struct mhi_controller *mhi_cntrl, > enum mhi_callback cb); > - int (*link_status)(struct mhi_controller *mhi_cntrl); > void (*wake_get)(struct mhi_controller *mhi_cntrl, bool override); > void (*wake_put)(struct mhi_controller *mhi_cntrl, bool override); > void (*wake_toggle)(struct mhi_controller *mhi_cntrl); > -- > Qualcomm Technologies, Inc. is a member of the > Code Aurora Forum, a Linux Foundation Collaborative Project.
On 4/6/2020 11:58 PM, Manivannan Sadhasivam wrote: > Hi Jeff, > > On Mon, Apr 06, 2020 at 03:04:37PM -0600, Jeffrey Hugo wrote: >> If the MHI core detects invalid data due to a PCI read, it calls into >> the controller via link_status() to double check that the link is infact >> down. All in all, this is pretty pointless, and racy. There are no good >> reasons for this, and only drawbacks. >> >> Its pointless because chances are, the controller is going to do the same >> thing to determine if the link is down - attempt a PCI access and compare >> the result. This does not make the link status decision any smarter. >> >> Its racy because its possible that the link was down at the time of the >> MHI core access, but then recovered before the controller access. In this >> case, the controller will indicate the link is not down, and the MHI core >> will precede to use a bad value as the MHI core does not attempt to retry >> the access. >> >> Retrying the access in the MHI core is a bad idea because again, it is >> racy - what if the link is down again? Furthermore, there may be some >> higher level state associated with the link status, that is now invalid >> because the link went down. >> >> The only reason why the MHI core could see "invalid" data when doing a PCI >> access, that is actually valid, is if the register actually contained the >> PCI spec defined sentinel for an invalid access. In this case, it is >> arguable that the MHI implementation broken, and should be fixed, not >> worked around. >> >> Therefore, remove the link_status() callback before anyone attempts to >> implement it. >> >> Signed-off-by: Jeffrey Hugo <jhugo@codeaurora.org> > > LGTM. But as per the IRC discussion I'd like the mhi_reg_read() to be > implemented as a callback in mhi_controller struct inorder to truly make MHI > a PCI agnostic bus. > > Since we don't have any controller driver in mainline, I think it is the > good time to do this change. No problem. I thought you might prefer that approach, hence the discussion. :) Do you want that included in this change, or as a follow up? > > For this, > > Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org> > > Thanks, > Mani > >> --- >> drivers/bus/mhi/core/init.c | 6 ++---- >> drivers/bus/mhi/core/main.c | 5 ++--- >> include/linux/mhi.h | 2 -- >> 3 files changed, 4 insertions(+), 9 deletions(-) >> >> diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c >> index b38359c..2af08d57 100644 >> --- a/drivers/bus/mhi/core/init.c >> +++ b/drivers/bus/mhi/core/init.c >> @@ -812,10 +812,8 @@ int mhi_register_controller(struct mhi_controller *mhi_cntrl, >> if (!mhi_cntrl) >> return -EINVAL; >> >> - if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put) >> - return -EINVAL; >> - >> - if (!mhi_cntrl->status_cb || !mhi_cntrl->link_status) >> + if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put || >> + !mhi_cntrl->status_cb) >> return -EINVAL; >> >> ret = parse_config(mhi_cntrl, config); >> diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c >> index eb4256b..473278b8 100644 >> --- a/drivers/bus/mhi/core/main.c >> +++ b/drivers/bus/mhi/core/main.c >> @@ -20,9 +20,8 @@ int __must_check mhi_read_reg(struct mhi_controller *mhi_cntrl, >> { >> u32 tmp = readl(base + offset); >> >> - /* If there is any unexpected value, query the link status */ >> - if (PCI_INVALID_READ(tmp) && >> - mhi_cntrl->link_status(mhi_cntrl)) >> + /* If the value is invalid, the link is down */ >> + if (PCI_INVALID_READ(tmp)) >> return -EIO; >> >> *out = tmp; >> diff --git a/include/linux/mhi.h b/include/linux/mhi.h >> index ad19960..be704a4 100644 >> --- a/include/linux/mhi.h >> +++ b/include/linux/mhi.h >> @@ -335,7 +335,6 @@ struct mhi_controller_config { >> * @syserr_worker: System error worker >> * @state_event: State change event >> * @status_cb: CB function to notify power states of the device (required) >> - * @link_status: CB function to query link status of the device (required) >> * @wake_get: CB function to assert device wake (optional) >> * @wake_put: CB function to de-assert device wake (optional) >> * @wake_toggle: CB function to assert and de-assert device wake (optional) >> @@ -417,7 +416,6 @@ struct mhi_controller { >> >> void (*status_cb)(struct mhi_controller *mhi_cntrl, >> enum mhi_callback cb); >> - int (*link_status)(struct mhi_controller *mhi_cntrl); >> void (*wake_get)(struct mhi_controller *mhi_cntrl, bool override); >> void (*wake_put)(struct mhi_controller *mhi_cntrl, bool override); >> void (*wake_toggle)(struct mhi_controller *mhi_cntrl); >> -- >> Qualcomm Technologies, Inc. is a member of the >> Code Aurora Forum, a Linux Foundation Collaborative Project.
On Tue, Apr 07, 2020 at 09:03:28AM -0600, Jeffrey Hugo wrote: > On 4/6/2020 11:58 PM, Manivannan Sadhasivam wrote: > > Hi Jeff, > > > > On Mon, Apr 06, 2020 at 03:04:37PM -0600, Jeffrey Hugo wrote: > > > If the MHI core detects invalid data due to a PCI read, it calls into > > > the controller via link_status() to double check that the link is infact > > > down. All in all, this is pretty pointless, and racy. There are no good > > > reasons for this, and only drawbacks. > > > > > > Its pointless because chances are, the controller is going to do the same > > > thing to determine if the link is down - attempt a PCI access and compare > > > the result. This does not make the link status decision any smarter. > > > > > > Its racy because its possible that the link was down at the time of the > > > MHI core access, but then recovered before the controller access. In this > > > case, the controller will indicate the link is not down, and the MHI core > > > will precede to use a bad value as the MHI core does not attempt to retry > > > the access. > > > > > > Retrying the access in the MHI core is a bad idea because again, it is > > > racy - what if the link is down again? Furthermore, there may be some > > > higher level state associated with the link status, that is now invalid > > > because the link went down. > > > > > > The only reason why the MHI core could see "invalid" data when doing a PCI > > > access, that is actually valid, is if the register actually contained the > > > PCI spec defined sentinel for an invalid access. In this case, it is > > > arguable that the MHI implementation broken, and should be fixed, not > > > worked around. > > > > > > Therefore, remove the link_status() callback before anyone attempts to > > > implement it. > > > > > > Signed-off-by: Jeffrey Hugo <jhugo@codeaurora.org> > > > > LGTM. But as per the IRC discussion I'd like the mhi_reg_read() to be > > implemented as a callback in mhi_controller struct inorder to truly make MHI > > a PCI agnostic bus. > > > > Since we don't have any controller driver in mainline, I think it is the > > good time to do this change. > > No problem. I thought you might prefer that approach, hence the discussion. > :) > > Do you want that included in this change, or as a follow up? > You can add a patch in this series itself. Thanks, Mani > > > > For this, > > > > Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org> > > > > Thanks, > > Mani > > > > > --- > > > drivers/bus/mhi/core/init.c | 6 ++---- > > > drivers/bus/mhi/core/main.c | 5 ++--- > > > include/linux/mhi.h | 2 -- > > > 3 files changed, 4 insertions(+), 9 deletions(-) > > > > > > diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c > > > index b38359c..2af08d57 100644 > > > --- a/drivers/bus/mhi/core/init.c > > > +++ b/drivers/bus/mhi/core/init.c > > > @@ -812,10 +812,8 @@ int mhi_register_controller(struct mhi_controller *mhi_cntrl, > > > if (!mhi_cntrl) > > > return -EINVAL; > > > - if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put) > > > - return -EINVAL; > > > - > > > - if (!mhi_cntrl->status_cb || !mhi_cntrl->link_status) > > > + if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put || > > > + !mhi_cntrl->status_cb) > > > return -EINVAL; > > > ret = parse_config(mhi_cntrl, config); > > > diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c > > > index eb4256b..473278b8 100644 > > > --- a/drivers/bus/mhi/core/main.c > > > +++ b/drivers/bus/mhi/core/main.c > > > @@ -20,9 +20,8 @@ int __must_check mhi_read_reg(struct mhi_controller *mhi_cntrl, > > > { > > > u32 tmp = readl(base + offset); > > > - /* If there is any unexpected value, query the link status */ > > > - if (PCI_INVALID_READ(tmp) && > > > - mhi_cntrl->link_status(mhi_cntrl)) > > > + /* If the value is invalid, the link is down */ > > > + if (PCI_INVALID_READ(tmp)) > > > return -EIO; > > > *out = tmp; > > > diff --git a/include/linux/mhi.h b/include/linux/mhi.h > > > index ad19960..be704a4 100644 > > > --- a/include/linux/mhi.h > > > +++ b/include/linux/mhi.h > > > @@ -335,7 +335,6 @@ struct mhi_controller_config { > > > * @syserr_worker: System error worker > > > * @state_event: State change event > > > * @status_cb: CB function to notify power states of the device (required) > > > - * @link_status: CB function to query link status of the device (required) > > > * @wake_get: CB function to assert device wake (optional) > > > * @wake_put: CB function to de-assert device wake (optional) > > > * @wake_toggle: CB function to assert and de-assert device wake (optional) > > > @@ -417,7 +416,6 @@ struct mhi_controller { > > > void (*status_cb)(struct mhi_controller *mhi_cntrl, > > > enum mhi_callback cb); > > > - int (*link_status)(struct mhi_controller *mhi_cntrl); > > > void (*wake_get)(struct mhi_controller *mhi_cntrl, bool override); > > > void (*wake_put)(struct mhi_controller *mhi_cntrl, bool override); > > > void (*wake_toggle)(struct mhi_controller *mhi_cntrl); > > > -- > > > Qualcomm Technologies, Inc. is a member of the > > > Code Aurora Forum, a Linux Foundation Collaborative Project. > > > -- > Jeffrey Hugo > Qualcomm Technologies, Inc. is a member of the > Code Aurora Forum, a Linux Foundation Collaborative Project.
diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c index b38359c..2af08d57 100644 --- a/drivers/bus/mhi/core/init.c +++ b/drivers/bus/mhi/core/init.c @@ -812,10 +812,8 @@ int mhi_register_controller(struct mhi_controller *mhi_cntrl, if (!mhi_cntrl) return -EINVAL; - if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put) - return -EINVAL; - - if (!mhi_cntrl->status_cb || !mhi_cntrl->link_status) + if (!mhi_cntrl->runtime_get || !mhi_cntrl->runtime_put || + !mhi_cntrl->status_cb) return -EINVAL; ret = parse_config(mhi_cntrl, config); diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c index eb4256b..473278b8 100644 --- a/drivers/bus/mhi/core/main.c +++ b/drivers/bus/mhi/core/main.c @@ -20,9 +20,8 @@ int __must_check mhi_read_reg(struct mhi_controller *mhi_cntrl, { u32 tmp = readl(base + offset); - /* If there is any unexpected value, query the link status */ - if (PCI_INVALID_READ(tmp) && - mhi_cntrl->link_status(mhi_cntrl)) + /* If the value is invalid, the link is down */ + if (PCI_INVALID_READ(tmp)) return -EIO; *out = tmp; diff --git a/include/linux/mhi.h b/include/linux/mhi.h index ad19960..be704a4 100644 --- a/include/linux/mhi.h +++ b/include/linux/mhi.h @@ -335,7 +335,6 @@ struct mhi_controller_config { * @syserr_worker: System error worker * @state_event: State change event * @status_cb: CB function to notify power states of the device (required) - * @link_status: CB function to query link status of the device (required) * @wake_get: CB function to assert device wake (optional) * @wake_put: CB function to de-assert device wake (optional) * @wake_toggle: CB function to assert and de-assert device wake (optional) @@ -417,7 +416,6 @@ struct mhi_controller { void (*status_cb)(struct mhi_controller *mhi_cntrl, enum mhi_callback cb); - int (*link_status)(struct mhi_controller *mhi_cntrl); void (*wake_get)(struct mhi_controller *mhi_cntrl, bool override); void (*wake_put)(struct mhi_controller *mhi_cntrl, bool override); void (*wake_toggle)(struct mhi_controller *mhi_cntrl);
If the MHI core detects invalid data due to a PCI read, it calls into the controller via link_status() to double check that the link is infact down. All in all, this is pretty pointless, and racy. There are no good reasons for this, and only drawbacks. Its pointless because chances are, the controller is going to do the same thing to determine if the link is down - attempt a PCI access and compare the result. This does not make the link status decision any smarter. Its racy because its possible that the link was down at the time of the MHI core access, but then recovered before the controller access. In this case, the controller will indicate the link is not down, and the MHI core will precede to use a bad value as the MHI core does not attempt to retry the access. Retrying the access in the MHI core is a bad idea because again, it is racy - what if the link is down again? Furthermore, there may be some higher level state associated with the link status, that is now invalid because the link went down. The only reason why the MHI core could see "invalid" data when doing a PCI access, that is actually valid, is if the register actually contained the PCI spec defined sentinel for an invalid access. In this case, it is arguable that the MHI implementation broken, and should be fixed, not worked around. Therefore, remove the link_status() callback before anyone attempts to implement it. Signed-off-by: Jeffrey Hugo <jhugo@codeaurora.org> --- drivers/bus/mhi/core/init.c | 6 ++---- drivers/bus/mhi/core/main.c | 5 ++--- include/linux/mhi.h | 2 -- 3 files changed, 4 insertions(+), 9 deletions(-)