| Message ID | 20230411110002.19824-9-sumitg@nvidia.com (mailing list archive) |
|---|---|
| State | Handled Elsewhere, archived |
| Headers | show |
| Series | Tegra234 Memory interconnect support | expand |
On Tue, Apr 11, 2023 at 04:30:01PM +0530, Sumit Gupta wrote: > Add support to request DRAM bandwidth with Memory Interconnect > in Tegra234 SoC. The DRAM BW required for different modes depends > on speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). > Currently, no latency is observed in data transfer with PCI as the > DRAM Freq is always set to max. But that results in high power > consumption. Now for Tegra234, we are enabling the dynamic scaling > of the DRAM Freq based on requests from Clients instead of running > at the max Freq always. This change does that for PCI MC client. I am sorry but this is still unclear to me. The sentence above makes me think that you are *adding* latency to the data transfer trading it with lower power consumption; probably that's a wrong parsing of what you are saying - so please explain what you want to say with "no latency is observed" and whether this patch changes that (which is not allowed because that would count as a regression). Thanks, Lorenzo > > Suggested-by: Manikanta Maddireddy <mmaddireddy@nvidia.com> > Signed-off-by: Sumit Gupta <sumitg@nvidia.com> > --- > drivers/pci/controller/dwc/pcie-tegra194.c | 51 +++++++++++++++------- > 1 file changed, 35 insertions(+), 16 deletions(-) > > diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c > index e6eec85480ca..4fdadc7b045f 100644 > --- a/drivers/pci/controller/dwc/pcie-tegra194.c > +++ b/drivers/pci/controller/dwc/pcie-tegra194.c > @@ -14,6 +14,7 @@ > #include <linux/delay.h> > #include <linux/gpio.h> > #include <linux/gpio/consumer.h> > +#include <linux/interconnect.h> > #include <linux/interrupt.h> > #include <linux/iopoll.h> > #include <linux/kernel.h> > @@ -288,6 +289,7 @@ struct tegra_pcie_dw { > unsigned int pex_rst_irq; > int ep_state; > long link_status; > + struct icc_path *icc_path; > }; > > static inline struct tegra_pcie_dw *to_tegra_pcie(struct dw_pcie *pci) > @@ -310,6 +312,27 @@ struct tegra_pcie_soc { > enum dw_pcie_device_mode mode; > }; > > +static void tegra_pcie_icc_set(struct tegra_pcie_dw *pcie) > +{ > + struct dw_pcie *pci = &pcie->pci; > + u32 val, speed, width; > + > + val = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA); > + > + speed = FIELD_GET(PCI_EXP_LNKSTA_CLS, val); > + width = FIELD_GET(PCI_EXP_LNKSTA_NLW, val); > + > + val = width * (PCIE_SPEED2MBS_ENC(pcie_link_speed[speed]) / BITS_PER_BYTE); > + > + if (icc_set_bw(pcie->icc_path, MBps_to_icc(val), 0)) > + dev_err(pcie->dev, "can't set bw[%u]\n", val); > + > + if (speed >= ARRAY_SIZE(pcie_gen_freq)) > + speed = 0; > + > + clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); > +} > + > static void apply_bad_link_workaround(struct dw_pcie_rp *pp) > { > struct dw_pcie *pci = to_dw_pcie_from_pp(pp); > @@ -453,18 +476,12 @@ static irqreturn_t tegra_pcie_ep_irq_thread(int irq, void *arg) > struct tegra_pcie_dw *pcie = arg; > struct dw_pcie_ep *ep = &pcie->pci.ep; > struct dw_pcie *pci = &pcie->pci; > - u32 val, speed; > + u32 val; > > if (test_and_clear_bit(0, &pcie->link_status)) > dw_pcie_ep_linkup(ep); > > - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & > - PCI_EXP_LNKSTA_CLS; > - > - if (speed >= ARRAY_SIZE(pcie_gen_freq)) > - speed = 0; > - > - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); > + tegra_pcie_icc_set(pcie); > > if (pcie->of_data->has_ltr_req_fix) > return IRQ_HANDLED; > @@ -950,9 +967,9 @@ static int tegra_pcie_dw_host_init(struct dw_pcie_rp *pp) > > static int tegra_pcie_dw_start_link(struct dw_pcie *pci) > { > - u32 val, offset, speed, tmp; > struct tegra_pcie_dw *pcie = to_tegra_pcie(pci); > struct dw_pcie_rp *pp = &pci->pp; > + u32 val, offset, tmp; > bool retry = true; > > if (pcie->of_data->mode == DW_PCIE_EP_TYPE) { > @@ -1023,13 +1040,7 @@ static int tegra_pcie_dw_start_link(struct dw_pcie *pci) > goto retry_link; > } > > - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & > - PCI_EXP_LNKSTA_CLS; > - > - if (speed >= ARRAY_SIZE(pcie_gen_freq)) > - speed = 0; > - > - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); > + tegra_pcie_icc_set(pcie); > > tegra_pcie_enable_interrupts(pp); > > @@ -2233,6 +2244,14 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev) > > platform_set_drvdata(pdev, pcie); > > + pcie->icc_path = devm_of_icc_get(&pdev->dev, "write"); > + ret = PTR_ERR_OR_ZERO(pcie->icc_path); > + if (ret) { > + tegra_bpmp_put(pcie->bpmp); > + dev_err_probe(&pdev->dev, ret, "failed to get write interconnect\n"); > + return ret; > + } > + > switch (pcie->of_data->mode) { > case DW_PCIE_RC_TYPE: > ret = devm_request_irq(dev, pp->irq, tegra_pcie_rp_irq_handler, > -- > 2.17.1 >
On 13/04/23 20:22, Lorenzo Pieralisi wrote: > External email: Use caution opening links or attachments > > > On Tue, Apr 11, 2023 at 04:30:01PM +0530, Sumit Gupta wrote: >> Add support to request DRAM bandwidth with Memory Interconnect >> in Tegra234 SoC. The DRAM BW required for different modes depends >> on speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). >> Currently, no latency is observed in data transfer with PCI as the >> DRAM Freq is always set to max. But that results in high power >> consumption. Now for Tegra234, we are enabling the dynamic scaling >> of the DRAM Freq based on requests from Clients instead of running >> at the max Freq always. This change does that for PCI MC client. > > I am sorry but this is still unclear to me. The sentence above makes > me think that you are *adding* latency to the data transfer trading > it with lower power consumption; probably that's a wrong parsing of > what you are saying - so please explain what you want to say > with "no latency is observed" and whether this patch changes that > (which is not allowed because that would count as a regression). > > Thanks, > Lorenzo > Rephrased as below. Please suggest if it is clear now. Add support to request DRAM bandwidth with Memory Interconnect in Tegra234 SoC. The DRAM BW required for different modes depends on speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). Currently, the DRAM freq is always set to max but that results in higher power consumption. Memory Interconnect feature adds capability for MC clients to request BW and scale the DRAM freq dynamically to provide the requested BW. Thank you, Sumit Gupta >> >> Suggested-by: Manikanta Maddireddy <mmaddireddy@nvidia.com> >> Signed-off-by: Sumit Gupta <sumitg@nvidia.com> >> --- >> drivers/pci/controller/dwc/pcie-tegra194.c | 51 +++++++++++++++------- >> 1 file changed, 35 insertions(+), 16 deletions(-) >> >> diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c >> index e6eec85480ca..4fdadc7b045f 100644 >> --- a/drivers/pci/controller/dwc/pcie-tegra194.c >> +++ b/drivers/pci/controller/dwc/pcie-tegra194.c >> @@ -14,6 +14,7 @@ >> #include <linux/delay.h> >> #include <linux/gpio.h> >> #include <linux/gpio/consumer.h> >> +#include <linux/interconnect.h> >> #include <linux/interrupt.h> >> #include <linux/iopoll.h> >> #include <linux/kernel.h> >> @@ -288,6 +289,7 @@ struct tegra_pcie_dw { >> unsigned int pex_rst_irq; >> int ep_state; >> long link_status; >> + struct icc_path *icc_path; >> }; >> >> static inline struct tegra_pcie_dw *to_tegra_pcie(struct dw_pcie *pci) >> @@ -310,6 +312,27 @@ struct tegra_pcie_soc { >> enum dw_pcie_device_mode mode; >> }; >> >> +static void tegra_pcie_icc_set(struct tegra_pcie_dw *pcie) >> +{ >> + struct dw_pcie *pci = &pcie->pci; >> + u32 val, speed, width; >> + >> + val = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA); >> + >> + speed = FIELD_GET(PCI_EXP_LNKSTA_CLS, val); >> + width = FIELD_GET(PCI_EXP_LNKSTA_NLW, val); >> + >> + val = width * (PCIE_SPEED2MBS_ENC(pcie_link_speed[speed]) / BITS_PER_BYTE); >> + >> + if (icc_set_bw(pcie->icc_path, MBps_to_icc(val), 0)) >> + dev_err(pcie->dev, "can't set bw[%u]\n", val); >> + >> + if (speed >= ARRAY_SIZE(pcie_gen_freq)) >> + speed = 0; >> + >> + clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); >> +} >> + >> static void apply_bad_link_workaround(struct dw_pcie_rp *pp) >> { >> struct dw_pcie *pci = to_dw_pcie_from_pp(pp); >> @@ -453,18 +476,12 @@ static irqreturn_t tegra_pcie_ep_irq_thread(int irq, void *arg) >> struct tegra_pcie_dw *pcie = arg; >> struct dw_pcie_ep *ep = &pcie->pci.ep; >> struct dw_pcie *pci = &pcie->pci; >> - u32 val, speed; >> + u32 val; >> >> if (test_and_clear_bit(0, &pcie->link_status)) >> dw_pcie_ep_linkup(ep); >> >> - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & >> - PCI_EXP_LNKSTA_CLS; >> - >> - if (speed >= ARRAY_SIZE(pcie_gen_freq)) >> - speed = 0; >> - >> - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); >> + tegra_pcie_icc_set(pcie); >> >> if (pcie->of_data->has_ltr_req_fix) >> return IRQ_HANDLED; >> @@ -950,9 +967,9 @@ static int tegra_pcie_dw_host_init(struct dw_pcie_rp *pp) >> >> static int tegra_pcie_dw_start_link(struct dw_pcie *pci) >> { >> - u32 val, offset, speed, tmp; >> struct tegra_pcie_dw *pcie = to_tegra_pcie(pci); >> struct dw_pcie_rp *pp = &pci->pp; >> + u32 val, offset, tmp; >> bool retry = true; >> >> if (pcie->of_data->mode == DW_PCIE_EP_TYPE) { >> @@ -1023,13 +1040,7 @@ static int tegra_pcie_dw_start_link(struct dw_pcie *pci) >> goto retry_link; >> } >> >> - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & >> - PCI_EXP_LNKSTA_CLS; >> - >> - if (speed >= ARRAY_SIZE(pcie_gen_freq)) >> - speed = 0; >> - >> - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); >> + tegra_pcie_icc_set(pcie); >> >> tegra_pcie_enable_interrupts(pp); >> >> @@ -2233,6 +2244,14 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev) >> >> platform_set_drvdata(pdev, pcie); >> >> + pcie->icc_path = devm_of_icc_get(&pdev->dev, "write"); >> + ret = PTR_ERR_OR_ZERO(pcie->icc_path); >> + if (ret) { >> + tegra_bpmp_put(pcie->bpmp); >> + dev_err_probe(&pdev->dev, ret, "failed to get write interconnect\n"); >> + return ret; >> + } >> + >> switch (pcie->of_data->mode) { >> case DW_PCIE_RC_TYPE: >> ret = devm_request_irq(dev, pp->irq, tegra_pcie_rp_irq_handler, >> -- >> 2.17.1 >>
On Fri, Apr 14, 2023 at 04:24:02PM +0530, Sumit Gupta wrote: > > > On 13/04/23 20:22, Lorenzo Pieralisi wrote: > > External email: Use caution opening links or attachments > > > > > > On Tue, Apr 11, 2023 at 04:30:01PM +0530, Sumit Gupta wrote: > > > Add support to request DRAM bandwidth with Memory Interconnect > > > in Tegra234 SoC. The DRAM BW required for different modes depends > > > on speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). > > > Currently, no latency is observed in data transfer with PCI as the > > > DRAM Freq is always set to max. But that results in high power > > > consumption. Now for Tegra234, we are enabling the dynamic scaling > > > of the DRAM Freq based on requests from Clients instead of running > > > at the max Freq always. This change does that for PCI MC client. > > > > I am sorry but this is still unclear to me. The sentence above makes > > me think that you are *adding* latency to the data transfer trading > > it with lower power consumption; probably that's a wrong parsing of > > what you are saying - so please explain what you want to say > > with "no latency is observed" and whether this patch changes that > > (which is not allowed because that would count as a regression). > > > > Thanks, > > Lorenzo > > > > Rephrased as below. Please suggest if it is clear now. > > Add support to request DRAM bandwidth with Memory Interconnect in > Tegra234 SoC. The DRAM BW required for different modes depends on > speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). > Currently, the DRAM freq is always set to max but that results in > higher power consumption. Memory Interconnect feature adds capability > for MC clients to request BW and scale the DRAM freq dynamically to > provide the requested BW. What does "Memory Interconnect" stand for ? Is that HW technology or a Linux framework (or both) ? What does MC stand for ? Update the commit log below accordingly. "Add support to request DRAM bandwidth with Memory Interconnect in Tegra234 SoC. The DRAM BW required for different modes depends on the link speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). Currently, the DRAM frequency is always set to the maximum available but that results in the highest power consumption. The Memory Interconnect feature adds the capability for MC clients to request bandwidth and therefore scale DRAM frequency dynamically depending on the required link speed so that the DRAM energy consumption can be optimized". With that: Acked-by: Lorenzo Pieralisi <lpieralisi@kernel.org> > > Thank you, > Sumit Gupta > > > > > > > Suggested-by: Manikanta Maddireddy <mmaddireddy@nvidia.com> > > > Signed-off-by: Sumit Gupta <sumitg@nvidia.com> > > > --- > > > drivers/pci/controller/dwc/pcie-tegra194.c | 51 +++++++++++++++------- > > > 1 file changed, 35 insertions(+), 16 deletions(-) > > > > > > diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c > > > index e6eec85480ca..4fdadc7b045f 100644 > > > --- a/drivers/pci/controller/dwc/pcie-tegra194.c > > > +++ b/drivers/pci/controller/dwc/pcie-tegra194.c > > > @@ -14,6 +14,7 @@ > > > #include <linux/delay.h> > > > #include <linux/gpio.h> > > > #include <linux/gpio/consumer.h> > > > +#include <linux/interconnect.h> > > > #include <linux/interrupt.h> > > > #include <linux/iopoll.h> > > > #include <linux/kernel.h> > > > @@ -288,6 +289,7 @@ struct tegra_pcie_dw { > > > unsigned int pex_rst_irq; > > > int ep_state; > > > long link_status; > > > + struct icc_path *icc_path; > > > }; > > > > > > static inline struct tegra_pcie_dw *to_tegra_pcie(struct dw_pcie *pci) > > > @@ -310,6 +312,27 @@ struct tegra_pcie_soc { > > > enum dw_pcie_device_mode mode; > > > }; > > > > > > +static void tegra_pcie_icc_set(struct tegra_pcie_dw *pcie) > > > +{ > > > + struct dw_pcie *pci = &pcie->pci; > > > + u32 val, speed, width; > > > + > > > + val = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA); > > > + > > > + speed = FIELD_GET(PCI_EXP_LNKSTA_CLS, val); > > > + width = FIELD_GET(PCI_EXP_LNKSTA_NLW, val); > > > + > > > + val = width * (PCIE_SPEED2MBS_ENC(pcie_link_speed[speed]) / BITS_PER_BYTE); > > > + > > > + if (icc_set_bw(pcie->icc_path, MBps_to_icc(val), 0)) > > > + dev_err(pcie->dev, "can't set bw[%u]\n", val); > > > + > > > + if (speed >= ARRAY_SIZE(pcie_gen_freq)) > > > + speed = 0; > > > + > > > + clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); > > > +} > > > + > > > static void apply_bad_link_workaround(struct dw_pcie_rp *pp) > > > { > > > struct dw_pcie *pci = to_dw_pcie_from_pp(pp); > > > @@ -453,18 +476,12 @@ static irqreturn_t tegra_pcie_ep_irq_thread(int irq, void *arg) > > > struct tegra_pcie_dw *pcie = arg; > > > struct dw_pcie_ep *ep = &pcie->pci.ep; > > > struct dw_pcie *pci = &pcie->pci; > > > - u32 val, speed; > > > + u32 val; > > > > > > if (test_and_clear_bit(0, &pcie->link_status)) > > > dw_pcie_ep_linkup(ep); > > > > > > - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & > > > - PCI_EXP_LNKSTA_CLS; > > > - > > > - if (speed >= ARRAY_SIZE(pcie_gen_freq)) > > > - speed = 0; > > > - > > > - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); > > > + tegra_pcie_icc_set(pcie); > > > > > > if (pcie->of_data->has_ltr_req_fix) > > > return IRQ_HANDLED; > > > @@ -950,9 +967,9 @@ static int tegra_pcie_dw_host_init(struct dw_pcie_rp *pp) > > > > > > static int tegra_pcie_dw_start_link(struct dw_pcie *pci) > > > { > > > - u32 val, offset, speed, tmp; > > > struct tegra_pcie_dw *pcie = to_tegra_pcie(pci); > > > struct dw_pcie_rp *pp = &pci->pp; > > > + u32 val, offset, tmp; > > > bool retry = true; > > > > > > if (pcie->of_data->mode == DW_PCIE_EP_TYPE) { > > > @@ -1023,13 +1040,7 @@ static int tegra_pcie_dw_start_link(struct dw_pcie *pci) > > > goto retry_link; > > > } > > > > > > - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & > > > - PCI_EXP_LNKSTA_CLS; > > > - > > > - if (speed >= ARRAY_SIZE(pcie_gen_freq)) > > > - speed = 0; > > > - > > > - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); > > > + tegra_pcie_icc_set(pcie); > > > > > > tegra_pcie_enable_interrupts(pp); > > > > > > @@ -2233,6 +2244,14 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev) > > > > > > platform_set_drvdata(pdev, pcie); > > > > > > + pcie->icc_path = devm_of_icc_get(&pdev->dev, "write"); > > > + ret = PTR_ERR_OR_ZERO(pcie->icc_path); > > > + if (ret) { > > > + tegra_bpmp_put(pcie->bpmp); > > > + dev_err_probe(&pdev->dev, ret, "failed to get write interconnect\n"); > > > + return ret; > > > + } > > > + > > > switch (pcie->of_data->mode) { > > > case DW_PCIE_RC_TYPE: > > > ret = devm_request_irq(dev, pp->irq, tegra_pcie_rp_irq_handler, > > > -- > > > 2.17.1 > > >
On 21/04/23 18:41, Lorenzo Pieralisi wrote: > External email: Use caution opening links or attachments > > > On Fri, Apr 14, 2023 at 04:24:02PM +0530, Sumit Gupta wrote: >> >> >> On 13/04/23 20:22, Lorenzo Pieralisi wrote: >>> External email: Use caution opening links or attachments >>> >>> >>> On Tue, Apr 11, 2023 at 04:30:01PM +0530, Sumit Gupta wrote: >>>> Add support to request DRAM bandwidth with Memory Interconnect >>>> in Tegra234 SoC. The DRAM BW required for different modes depends >>>> on speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). >>>> Currently, no latency is observed in data transfer with PCI as the >>>> DRAM Freq is always set to max. But that results in high power >>>> consumption. Now for Tegra234, we are enabling the dynamic scaling >>>> of the DRAM Freq based on requests from Clients instead of running >>>> at the max Freq always. This change does that for PCI MC client. >>> >>> I am sorry but this is still unclear to me. The sentence above makes >>> me think that you are *adding* latency to the data transfer trading >>> it with lower power consumption; probably that's a wrong parsing of >>> what you are saying - so please explain what you want to say >>> with "no latency is observed" and whether this patch changes that >>> (which is not allowed because that would count as a regression). >>> >>> Thanks, >>> Lorenzo >>> >> >> Rephrased as below. Please suggest if it is clear now. >> >> Add support to request DRAM bandwidth with Memory Interconnect in >> Tegra234 SoC. The DRAM BW required for different modes depends on >> speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). >> Currently, the DRAM freq is always set to max but that results in >> higher power consumption. Memory Interconnect feature adds capability >> for MC clients to request BW and scale the DRAM freq dynamically to >> provide the requested BW. > > What does "Memory Interconnect" stand for ? > > Is that HW technology or a Linux framework (or both) ? > > What does MC stand for ? > > Update the commit log below accordingly. > > "Add support to request DRAM bandwidth with Memory Interconnect > in Tegra234 SoC. > > The DRAM BW required for different modes depends on the link > speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). > > Currently, the DRAM frequency is always set to the maximum available > but that results in the highest power consumption. > > The Memory Interconnect feature adds the capability for MC clients to > request bandwidth and therefore scale DRAM frequency dynamically > depending on the required link speed so that the DRAM energy consumption > can be optimized". > > With that: > > Acked-by: Lorenzo Pieralisi <lpieralisi@kernel.org> > Thank you for the inputs. Have added changes in v7. >> >> Thank you, >> Sumit Gupta >> >>>> >>>> Suggested-by: Manikanta Maddireddy <mmaddireddy@nvidia.com> >>>> Signed-off-by: Sumit Gupta <sumitg@nvidia.com> >>>> --- >>>> drivers/pci/controller/dwc/pcie-tegra194.c | 51 +++++++++++++++------- >>>> 1 file changed, 35 insertions(+), 16 deletions(-) >>>> >>>> diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c >>>> index e6eec85480ca..4fdadc7b045f 100644 >>>> --- a/drivers/pci/controller/dwc/pcie-tegra194.c >>>> +++ b/drivers/pci/controller/dwc/pcie-tegra194.c >>>> @@ -14,6 +14,7 @@ >>>> #include <linux/delay.h> >>>> #include <linux/gpio.h> >>>> #include <linux/gpio/consumer.h> >>>> +#include <linux/interconnect.h> >>>> #include <linux/interrupt.h> >>>> #include <linux/iopoll.h> >>>> #include <linux/kernel.h> >>>> @@ -288,6 +289,7 @@ struct tegra_pcie_dw { >>>> unsigned int pex_rst_irq; >>>> int ep_state; >>>> long link_status; >>>> + struct icc_path *icc_path; >>>> }; >>>> >>>> static inline struct tegra_pcie_dw *to_tegra_pcie(struct dw_pcie *pci) >>>> @@ -310,6 +312,27 @@ struct tegra_pcie_soc { >>>> enum dw_pcie_device_mode mode; >>>> }; >>>> >>>> +static void tegra_pcie_icc_set(struct tegra_pcie_dw *pcie) >>>> +{ >>>> + struct dw_pcie *pci = &pcie->pci; >>>> + u32 val, speed, width; >>>> + >>>> + val = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA); >>>> + >>>> + speed = FIELD_GET(PCI_EXP_LNKSTA_CLS, val); >>>> + width = FIELD_GET(PCI_EXP_LNKSTA_NLW, val); >>>> + >>>> + val = width * (PCIE_SPEED2MBS_ENC(pcie_link_speed[speed]) / BITS_PER_BYTE); >>>> + >>>> + if (icc_set_bw(pcie->icc_path, MBps_to_icc(val), 0)) >>>> + dev_err(pcie->dev, "can't set bw[%u]\n", val); >>>> + >>>> + if (speed >= ARRAY_SIZE(pcie_gen_freq)) >>>> + speed = 0; >>>> + >>>> + clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); >>>> +} >>>> + >>>> static void apply_bad_link_workaround(struct dw_pcie_rp *pp) >>>> { >>>> struct dw_pcie *pci = to_dw_pcie_from_pp(pp); >>>> @@ -453,18 +476,12 @@ static irqreturn_t tegra_pcie_ep_irq_thread(int irq, void *arg) >>>> struct tegra_pcie_dw *pcie = arg; >>>> struct dw_pcie_ep *ep = &pcie->pci.ep; >>>> struct dw_pcie *pci = &pcie->pci; >>>> - u32 val, speed; >>>> + u32 val; >>>> >>>> if (test_and_clear_bit(0, &pcie->link_status)) >>>> dw_pcie_ep_linkup(ep); >>>> >>>> - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & >>>> - PCI_EXP_LNKSTA_CLS; >>>> - >>>> - if (speed >= ARRAY_SIZE(pcie_gen_freq)) >>>> - speed = 0; >>>> - >>>> - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); >>>> + tegra_pcie_icc_set(pcie); >>>> >>>> if (pcie->of_data->has_ltr_req_fix) >>>> return IRQ_HANDLED; >>>> @@ -950,9 +967,9 @@ static int tegra_pcie_dw_host_init(struct dw_pcie_rp *pp) >>>> >>>> static int tegra_pcie_dw_start_link(struct dw_pcie *pci) >>>> { >>>> - u32 val, offset, speed, tmp; >>>> struct tegra_pcie_dw *pcie = to_tegra_pcie(pci); >>>> struct dw_pcie_rp *pp = &pci->pp; >>>> + u32 val, offset, tmp; >>>> bool retry = true; >>>> >>>> if (pcie->of_data->mode == DW_PCIE_EP_TYPE) { >>>> @@ -1023,13 +1040,7 @@ static int tegra_pcie_dw_start_link(struct dw_pcie *pci) >>>> goto retry_link; >>>> } >>>> >>>> - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & >>>> - PCI_EXP_LNKSTA_CLS; >>>> - >>>> - if (speed >= ARRAY_SIZE(pcie_gen_freq)) >>>> - speed = 0; >>>> - >>>> - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); >>>> + tegra_pcie_icc_set(pcie); >>>> >>>> tegra_pcie_enable_interrupts(pp); >>>> >>>> @@ -2233,6 +2244,14 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev) >>>> >>>> platform_set_drvdata(pdev, pcie); >>>> >>>> + pcie->icc_path = devm_of_icc_get(&pdev->dev, "write"); >>>> + ret = PTR_ERR_OR_ZERO(pcie->icc_path); >>>> + if (ret) { >>>> + tegra_bpmp_put(pcie->bpmp); >>>> + dev_err_probe(&pdev->dev, ret, "failed to get write interconnect\n"); >>>> + return ret; >>>> + } >>>> + >>>> switch (pcie->of_data->mode) { >>>> case DW_PCIE_RC_TYPE: >>>> ret = devm_request_irq(dev, pp->irq, tegra_pcie_rp_irq_handler, >>>> -- >>>> 2.17.1 >>>>
diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c index e6eec85480ca..4fdadc7b045f 100644 --- a/drivers/pci/controller/dwc/pcie-tegra194.c +++ b/drivers/pci/controller/dwc/pcie-tegra194.c @@ -14,6 +14,7 @@ #include <linux/delay.h> #include <linux/gpio.h> #include <linux/gpio/consumer.h> +#include <linux/interconnect.h> #include <linux/interrupt.h> #include <linux/iopoll.h> #include <linux/kernel.h> @@ -288,6 +289,7 @@ struct tegra_pcie_dw { unsigned int pex_rst_irq; int ep_state; long link_status; + struct icc_path *icc_path; }; static inline struct tegra_pcie_dw *to_tegra_pcie(struct dw_pcie *pci) @@ -310,6 +312,27 @@ struct tegra_pcie_soc { enum dw_pcie_device_mode mode; }; +static void tegra_pcie_icc_set(struct tegra_pcie_dw *pcie) +{ + struct dw_pcie *pci = &pcie->pci; + u32 val, speed, width; + + val = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA); + + speed = FIELD_GET(PCI_EXP_LNKSTA_CLS, val); + width = FIELD_GET(PCI_EXP_LNKSTA_NLW, val); + + val = width * (PCIE_SPEED2MBS_ENC(pcie_link_speed[speed]) / BITS_PER_BYTE); + + if (icc_set_bw(pcie->icc_path, MBps_to_icc(val), 0)) + dev_err(pcie->dev, "can't set bw[%u]\n", val); + + if (speed >= ARRAY_SIZE(pcie_gen_freq)) + speed = 0; + + clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); +} + static void apply_bad_link_workaround(struct dw_pcie_rp *pp) { struct dw_pcie *pci = to_dw_pcie_from_pp(pp); @@ -453,18 +476,12 @@ static irqreturn_t tegra_pcie_ep_irq_thread(int irq, void *arg) struct tegra_pcie_dw *pcie = arg; struct dw_pcie_ep *ep = &pcie->pci.ep; struct dw_pcie *pci = &pcie->pci; - u32 val, speed; + u32 val; if (test_and_clear_bit(0, &pcie->link_status)) dw_pcie_ep_linkup(ep); - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & - PCI_EXP_LNKSTA_CLS; - - if (speed >= ARRAY_SIZE(pcie_gen_freq)) - speed = 0; - - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); + tegra_pcie_icc_set(pcie); if (pcie->of_data->has_ltr_req_fix) return IRQ_HANDLED; @@ -950,9 +967,9 @@ static int tegra_pcie_dw_host_init(struct dw_pcie_rp *pp) static int tegra_pcie_dw_start_link(struct dw_pcie *pci) { - u32 val, offset, speed, tmp; struct tegra_pcie_dw *pcie = to_tegra_pcie(pci); struct dw_pcie_rp *pp = &pci->pp; + u32 val, offset, tmp; bool retry = true; if (pcie->of_data->mode == DW_PCIE_EP_TYPE) { @@ -1023,13 +1040,7 @@ static int tegra_pcie_dw_start_link(struct dw_pcie *pci) goto retry_link; } - speed = dw_pcie_readw_dbi(pci, pcie->pcie_cap_base + PCI_EXP_LNKSTA) & - PCI_EXP_LNKSTA_CLS; - - if (speed >= ARRAY_SIZE(pcie_gen_freq)) - speed = 0; - - clk_set_rate(pcie->core_clk, pcie_gen_freq[speed]); + tegra_pcie_icc_set(pcie); tegra_pcie_enable_interrupts(pp); @@ -2233,6 +2244,14 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev) platform_set_drvdata(pdev, pcie); + pcie->icc_path = devm_of_icc_get(&pdev->dev, "write"); + ret = PTR_ERR_OR_ZERO(pcie->icc_path); + if (ret) { + tegra_bpmp_put(pcie->bpmp); + dev_err_probe(&pdev->dev, ret, "failed to get write interconnect\n"); + return ret; + } + switch (pcie->of_data->mode) { case DW_PCIE_RC_TYPE: ret = devm_request_irq(dev, pp->irq, tegra_pcie_rp_irq_handler,
Add support to request DRAM bandwidth with Memory Interconnect in Tegra234 SoC. The DRAM BW required for different modes depends on speed (Gen-1/2/3/4) and width/lanes (x1/x2/x4/x8). Currently, no latency is observed in data transfer with PCI as the DRAM Freq is always set to max. But that results in high power consumption. Now for Tegra234, we are enabling the dynamic scaling of the DRAM Freq based on requests from Clients instead of running at the max Freq always. This change does that for PCI MC client. Suggested-by: Manikanta Maddireddy <mmaddireddy@nvidia.com> Signed-off-by: Sumit Gupta <sumitg@nvidia.com> --- drivers/pci/controller/dwc/pcie-tegra194.c | 51 +++++++++++++++------- 1 file changed, 35 insertions(+), 16 deletions(-)