| Message ID | 20221030142333.31019-1-quic_shazhuss@quicinc.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v1] clk: qcom: gcc-sc8280xp: add cxo as parent for gcc_ufs_ref_clkref_clk | expand |
Quoting Shazad Hussain (2022-10-30 07:23:33) > Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref So we should have a Fixes tag for this commit? Or really back to the beginning of the driver? > clock for ufs_mem_phy")' we need to explicitly make cxo as parent to > gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. >
On Tue, Nov 01, 2022 at 11:23:59AM -0700, Stephen Boyd wrote: > Quoting Shazad Hussain (2022-10-30 07:23:33) > > Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref > > So we should have a Fixes tag for this commit? Or really back to the > beginning of the driver? > > > clock for ufs_mem_phy")' we need to explicitly make cxo as parent to > > gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. The commit message is slightly misleading as this affects the other UFS PHY as well. If CX is indeed a parent of this clock then the issue has been there since the clock driver was added. (And otherwise, the PHY binding may need to be amended instead.) Johan
On 11/2/2022 12:46 PM, Johan Hovold wrote: > On Tue, Nov 01, 2022 at 11:23:59AM -0700, Stephen Boyd wrote: >> Quoting Shazad Hussain (2022-10-30 07:23:33) >>> Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref >> >> So we should have a Fixes tag for this commit? Or really back to the >> beginning of the driver? >> >>> clock for ufs_mem_phy")' we need to explicitly make cxo as parent to >>> gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. > > The commit message is slightly misleading as this affects the other UFS > PHY as well. > > If CX is indeed a parent of this clock then the issue has been there > since the clock driver was added. (And otherwise, the PHY binding may > need to be amended instead.) > > Johan CX is not the actual parent of this clk. GCC_UFS_REF_CLKREF_CLK is an external clk to the device, which needs to be voted. If we use the GCC_UFS_REF_CLKREF_CLK as ref clk, we don't have explicit vote for CX from ufs_mem_phy. If no client votes for CX,(very unlikely) then it's won't be ON for ufs_mem_phy as well right ! So to maintain the voting to CX, we make this as parent to ref clk. Shazad
On Wed, Nov 02, 2022 at 01:28:48PM +0530, Shazad Hussain wrote: > On 11/2/2022 12:46 PM, Johan Hovold wrote: > > On Tue, Nov 01, 2022 at 11:23:59AM -0700, Stephen Boyd wrote: > >> Quoting Shazad Hussain (2022-10-30 07:23:33) > >>> Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref > >> > >> So we should have a Fixes tag for this commit? Or really back to the > >> beginning of the driver? > >> > >>> clock for ufs_mem_phy")' we need to explicitly make cxo as parent to > >>> gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. > > > > The commit message is slightly misleading as this affects the other UFS > > PHY as well. > > > > If CX is indeed a parent of this clock then the issue has been there > > since the clock driver was added. (And otherwise, the PHY binding may > > need to be amended instead.) > CX is not the actual parent of this clk. GCC_UFS_REF_CLKREF_CLK is an > external clk to the device, which needs to be voted. If we use the > GCC_UFS_REF_CLKREF_CLK as ref clk, we don't have explicit vote for CX > from ufs_mem_phy. > > If no client votes for CX,(very unlikely) then it's won't be ON for > ufs_mem_phy as well right ! So to maintain the voting to CX, we make > this as parent to ref clk. Right, but if the PHYs really requires CX and it is not an ancestor of the refclk then this should be described by the binding (and not be hidden away in the clock driver). Johan
On 11/2/2022 1:43 PM, Johan Hovold wrote: > On Wed, Nov 02, 2022 at 01:28:48PM +0530, Shazad Hussain wrote: >> On 11/2/2022 12:46 PM, Johan Hovold wrote: >>> On Tue, Nov 01, 2022 at 11:23:59AM -0700, Stephen Boyd wrote: >>>> Quoting Shazad Hussain (2022-10-30 07:23:33) >>>>> Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref >>>> >>>> So we should have a Fixes tag for this commit? Or really back to the >>>> beginning of the driver? >>>> >>>>> clock for ufs_mem_phy")' we need to explicitly make cxo as parent to >>>>> gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. >>> >>> The commit message is slightly misleading as this affects the other UFS >>> PHY as well. >>> >>> If CX is indeed a parent of this clock then the issue has been there >>> since the clock driver was added. (And otherwise, the PHY binding may >>> need to be amended instead.) > >> CX is not the actual parent of this clk. GCC_UFS_REF_CLKREF_CLK is an >> external clk to the device, which needs to be voted. If we use the >> GCC_UFS_REF_CLKREF_CLK as ref clk, we don't have explicit vote for CX >> from ufs_mem_phy. >> >> If no client votes for CX,(very unlikely) then it's won't be ON for >> ufs_mem_phy as well right ! So to maintain the voting to CX, we make >> this as parent to ref clk. > > Right, but if the PHYs really requires CX and it is not an ancestor of > the refclk then this should be described by the binding (and not be > hidden away in the clock driver). > > Johan This makes sense, will be posting v2 post for the same. I assume this should use the Fixes tag then ! Shazad
On Wed, Nov 02, 2022 at 02:15:26PM +0530, Shazad Hussain wrote: > On 11/2/2022 1:43 PM, Johan Hovold wrote: > > Right, but if the PHYs really requires CX and it is not an ancestor of > > the refclk then this should be described by the binding (and not be > > hidden away in the clock driver). > This makes sense, will be posting v2 post for the same. > I assume this should use the Fixes tag then ! Yeah, I guess to you can add a fixes tag for the commits adding support for sc8280xp to the UFS PHY binding and driver. But please do check with the hardware documentation first so we get this right this time. I've already asked Bjorn to see what he can dig out as it is still not clear how the two "card" refclocks (GCC_UFS_CARD_CLKREF_CLK and GCC_UFS_1_CARD_CLKREF_CLK) are supposed to be used. Johan
On Wed, Nov 02, 2022 at 10:26:13AM +0100, Johan Hovold wrote: > On Wed, Nov 02, 2022 at 02:15:26PM +0530, Shazad Hussain wrote: > > On 11/2/2022 1:43 PM, Johan Hovold wrote: > > > > Right, but if the PHYs really requires CX and it is not an ancestor of > > > the refclk then this should be described by the binding (and not be > > > hidden away in the clock driver). > > > This makes sense, will be posting v2 post for the same. > > I assume this should use the Fixes tag then ! > > Yeah, I guess to you can add a fixes tag for the commits adding support > for sc8280xp to the UFS PHY binding and driver. > > But please do check with the hardware documentation first so we get this > right this time. > > I've already asked Bjorn to see what he can dig out as it is still not > clear how the two "card" refclocks (GCC_UFS_CARD_CLKREF_CLK and > GCC_UFS_1_CARD_CLKREF_CLK) are supposed to be used. > We've come full circle and Shazad's patch came from that discussion :) In line with the downstream dts, we have GCC_UFS{,_1}_CARD_CLKREF_CLK providing a reference clock to the two phys. Then GCC_UFS_REF_CLKREF_CLK feeds the UFS refclock pads (both of them), which connect to the memory device(s). In other words, GCC_UFS{,_1}_CARD_CLKREF_CLK should be "ref" in respective phy. GCC_UFS_REF_CLKREF_CLK is the clock to the devices, but as we don't represent the memory device explicitly it seems suitable to use as "ref_clk" in the ufshc nodes - which would then match the special handling of the "link clock" in the UFS driver. All three clocks are sourced off the CXO pad, so I would like this patch to cover at least all of these. And Fixes: d65d005f9a6c ("clk: qcom: add sc8280xp GCC driver") seems to be in order for such patch. @Johan, would you mind writing a dts patch flipping the clocks around and Shazad can update this patch? Regards, Bjorn
On Sun, Oct 30, 2022 at 07:53:33PM +0530, Shazad Hussain wrote: > Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref > clock for ufs_mem_phy")' we need to explicitly make cxo as parent to > gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. > Prior to that change we relied on both cxo and gcc_ufs_ref_clkref_clk being voted for. So I think the reasoning for this patch should simply be to express the fact that the clkref is fed from CXO. Regards, Bjorn > Signed-off-by: Shazad Hussain <quic_shazhuss@quicinc.com> > --- > drivers/clk/qcom/gcc-sc8280xp.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/drivers/clk/qcom/gcc-sc8280xp.c b/drivers/clk/qcom/gcc-sc8280xp.c > index a18ed88f3b82..72b545121c57 100644 > --- a/drivers/clk/qcom/gcc-sc8280xp.c > +++ b/drivers/clk/qcom/gcc-sc8280xp.c > @@ -5848,6 +5848,8 @@ static struct clk_branch gcc_ufs_ref_clkref_clk = { > .enable_mask = BIT(0), > .hw.init = &(const struct clk_init_data) { > .name = "gcc_ufs_ref_clkref_clk", > + .parent_data = &gcc_parent_data_tcxo, > + .num_parents = 1, > .ops = &clk_branch2_ops, > }, > }, > -- > 2.38.0 >
On Wed, Nov 02, 2022 at 09:49:49PM -0500, Bjorn Andersson wrote: > On Wed, Nov 02, 2022 at 10:26:13AM +0100, Johan Hovold wrote: > > On Wed, Nov 02, 2022 at 02:15:26PM +0530, Shazad Hussain wrote: > > > On 11/2/2022 1:43 PM, Johan Hovold wrote: > > > > > > Right, but if the PHYs really requires CX and it is not an ancestor of > > > > the refclk then this should be described by the binding (and not be > > > > hidden away in the clock driver). > > > > > This makes sense, will be posting v2 post for the same. > > > I assume this should use the Fixes tag then ! > > > > Yeah, I guess to you can add a fixes tag for the commits adding support > > for sc8280xp to the UFS PHY binding and driver. > > > > But please do check with the hardware documentation first so we get this > > right this time. > > > > I've already asked Bjorn to see what he can dig out as it is still not > > clear how the two "card" refclocks (GCC_UFS_CARD_CLKREF_CLK and > > GCC_UFS_1_CARD_CLKREF_CLK) are supposed to be used. > > > > We've come full circle and Shazad's patch came from that discussion :) Ah, good. :) > In line with the downstream dts, we have GCC_UFS{,_1}_CARD_CLKREF_CLK > providing a reference clock to the two phys. Then GCC_UFS_REF_CLKREF_CLK > feeds the UFS refclock pads (both of them), which connect to the memory > device(s). > > In other words, GCC_UFS{,_1}_CARD_CLKREF_CLK should be "ref" in > respective phy. > > GCC_UFS_REF_CLKREF_CLK is the clock to the devices, but as we don't > represent the memory device explicitly it seems suitable to use as > "ref_clk" in the ufshc nodes - which would then match the special > handling of the "link clock" in the UFS driver. Thanks for clearing that up. Using GCC_UFS_REF_CLKREF_CLK as ref_clk for the controller sounds reasonable. I guess the only missing piece is which "card" ref clock is used by which PHY. The ADP dts uses: phy ref clock phy@1d87000 (UFS_PHY) GCC_UFS_CARD_CLKREF_CLK phy@1da7000 (UFS_CARD) GCC_UFS_1_CARD_CLKREF_CLK but that is not what the firmware on ADP and CRD seem to enable. Both the ADP and CRD fw leaves GCC_UFS_1_CARD_CLKREF_CLK enabled, while GCC_UFS_CARD_CLKREF_CLK is only enabled on ADP (which unlike the CRD also uses the UFS_CARD controller). Does the ADP dts have these clocks switched or is the firmware confused? (Also note that experiments suggest that neither refclock appears to has to be explicitly enabled for the controllers to function.) > All three clocks are sourced off the CXO pad, so I would like this patch > to cover at least all of these. And > > Fixes: d65d005f9a6c ("clk: qcom: add sc8280xp GCC driver") > > seems to be in order for such patch. > > > @Johan, would you mind writing a dts patch flipping the clocks around > and Shazad can update this patch? I'll do so, but I'll wait with posting until you can confirm which clkref is which. Johan
On Thu, Nov 03, 2022 at 10:06:20AM +0100, Johan Hovold wrote: > On Wed, Nov 02, 2022 at 09:49:49PM -0500, Bjorn Andersson wrote: > > On Wed, Nov 02, 2022 at 10:26:13AM +0100, Johan Hovold wrote: > > > On Wed, Nov 02, 2022 at 02:15:26PM +0530, Shazad Hussain wrote: > > > > On 11/2/2022 1:43 PM, Johan Hovold wrote: > > > > > > > > Right, but if the PHYs really requires CX and it is not an ancestor of > > > > > the refclk then this should be described by the binding (and not be > > > > > hidden away in the clock driver). > > > > > > > This makes sense, will be posting v2 post for the same. > > > > I assume this should use the Fixes tag then ! > > > > > > Yeah, I guess to you can add a fixes tag for the commits adding support > > > for sc8280xp to the UFS PHY binding and driver. > > > > > > But please do check with the hardware documentation first so we get this > > > right this time. > > > > > > I've already asked Bjorn to see what he can dig out as it is still not > > > clear how the two "card" refclocks (GCC_UFS_CARD_CLKREF_CLK and > > > GCC_UFS_1_CARD_CLKREF_CLK) are supposed to be used. > > > > > > > We've come full circle and Shazad's patch came from that discussion :) > > Ah, good. :) > > > In line with the downstream dts, we have GCC_UFS{,_1}_CARD_CLKREF_CLK > > providing a reference clock to the two phys. Then GCC_UFS_REF_CLKREF_CLK > > feeds the UFS refclock pads (both of them), which connect to the memory > > device(s). > > > > In other words, GCC_UFS{,_1}_CARD_CLKREF_CLK should be "ref" in > > respective phy. > > > > GCC_UFS_REF_CLKREF_CLK is the clock to the devices, but as we don't > > represent the memory device explicitly it seems suitable to use as > > "ref_clk" in the ufshc nodes - which would then match the special > > handling of the "link clock" in the UFS driver. > > Thanks for clearing that up. Using GCC_UFS_REF_CLKREF_CLK as ref_clk for > the controller sounds reasonable. > > I guess the only missing piece is which "card" ref clock is used by > which PHY. > > The ADP dts uses: > > phy ref clock > > phy@1d87000 (UFS_PHY) GCC_UFS_CARD_CLKREF_CLK > phy@1da7000 (UFS_CARD) GCC_UFS_1_CARD_CLKREF_CLK > This matches the documentation. Regards, Bjorn > but that is not what the firmware on ADP and CRD seem to enable. > > Both the ADP and CRD fw leaves GCC_UFS_1_CARD_CLKREF_CLK enabled, while > GCC_UFS_CARD_CLKREF_CLK is only enabled on ADP (which unlike the CRD > also uses the UFS_CARD controller). > > Does the ADP dts have these clocks switched or is the firmware confused? > > (Also note that experiments suggest that neither refclock appears to > has to be explicitly enabled for the controllers to function.) > > > All three clocks are sourced off the CXO pad, so I would like this patch > > to cover at least all of these. And > > > > Fixes: d65d005f9a6c ("clk: qcom: add sc8280xp GCC driver") > > > > seems to be in order for such patch. > > > > > > @Johan, would you mind writing a dts patch flipping the clocks around > > and Shazad can update this patch? > > I'll do so, but I'll wait with posting until you can confirm which > clkref is which. > > Johan
On Thu, Nov 03, 2022 at 10:23:55AM -0500, Bjorn Andersson wrote: > On Thu, Nov 03, 2022 at 10:06:20AM +0100, Johan Hovold wrote: > > On Wed, Nov 02, 2022 at 09:49:49PM -0500, Bjorn Andersson wrote: > > > In line with the downstream dts, we have GCC_UFS{,_1}_CARD_CLKREF_CLK > > > providing a reference clock to the two phys. Then GCC_UFS_REF_CLKREF_CLK > > > feeds the UFS refclock pads (both of them), which connect to the memory > > > device(s). > > > > > > In other words, GCC_UFS{,_1}_CARD_CLKREF_CLK should be "ref" in > > > respective phy. > > > > > > GCC_UFS_REF_CLKREF_CLK is the clock to the devices, but as we don't > > > represent the memory device explicitly it seems suitable to use as > > > "ref_clk" in the ufshc nodes - which would then match the special > > > handling of the "link clock" in the UFS driver. > > > > Thanks for clearing that up. Using GCC_UFS_REF_CLKREF_CLK as ref_clk for > > the controller sounds reasonable. > > > > I guess the only missing piece is which "card" ref clock is used by > > which PHY. > > > > The ADP dts uses: > > > > phy ref clock > > > > phy@1d87000 (UFS_PHY) GCC_UFS_CARD_CLKREF_CLK > > phy@1da7000 (UFS_CARD) GCC_UFS_1_CARD_CLKREF_CLK > > > > This matches the documentation. Thanks for checking. > > > All three clocks are sourced off the CXO pad, so I would like this patch > > > to cover at least all of these. And > > > > > > Fixes: d65d005f9a6c ("clk: qcom: add sc8280xp GCC driver") > > > > > > seems to be in order for such patch. > > > > > > > > > @Johan, would you mind writing a dts patch flipping the clocks around > > > and Shazad can update this patch? > > > > I'll do so, but I'll wait with posting until you can confirm which > > clkref is which. I've know posted a patch fixing the devicetree here: https://lore.kernel.org/lkml/20221104092045.17410-1-johan+linaro@kernel.org/ Note that we need to get Shazad's clock driver fix in first as the UFS controller driver expects a valid frequency for the device ref clock. Johan
diff --git a/drivers/clk/qcom/gcc-sc8280xp.c b/drivers/clk/qcom/gcc-sc8280xp.c index a18ed88f3b82..72b545121c57 100644 --- a/drivers/clk/qcom/gcc-sc8280xp.c +++ b/drivers/clk/qcom/gcc-sc8280xp.c @@ -5848,6 +5848,8 @@ static struct clk_branch gcc_ufs_ref_clkref_clk = { .enable_mask = BIT(0), .hw.init = &(const struct clk_init_data) { .name = "gcc_ufs_ref_clkref_clk", + .parent_data = &gcc_parent_data_tcxo, + .num_parents = 1, .ops = &clk_branch2_ops, }, },
Since 'commit f3aa975e230e ("arm64: dts: qcom: sc8280xp: correct ref clock for ufs_mem_phy")' we need to explicitly make cxo as parent to gcc_ufs_ref_clkref_clk to have an independent vote from ufs_mem_phy. Signed-off-by: Shazad Hussain <quic_shazhuss@quicinc.com> --- drivers/clk/qcom/gcc-sc8280xp.c | 2 ++ 1 file changed, 2 insertions(+)