| Message ID | 12bfdd23772c49530b8b0993cc82bc89b3eb4ada.1706296015.git.quic_uchalich@quicinc.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | LLCC: Support for Broadcast_AND region | expand |
On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala <quic_uchalich@quicinc.com> wrote: > > To support CSR programming, a broadcast interface is used to program > all channels in a single command. Until SM8450 there was only one > broadcast region (Broadcast_OR) used to broadcast write and check > for status bit 0. From SM8450 onwards another broadcast region > (Broadcast_AND) has been added which checks for status bit 1. > > Update llcc_drv_data structure with new regmap for Broadcast_AND > region and initialize regmap for Broadcast_AND region when HW block > version is greater than 4.1 for backwards compatibility. > > Switch from broadcast_OR to broadcast_AND region for checking > status bit 1 as Broadcast_OR region checks only for bit 0. This breaks backwards compatibility with the existing DT files, doesn't it? > While at it, also check return value after reading Broadcast_OR > region in llcc_update_act_ctrl(). Separate patch, Fixes tag. > > Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> > --- > drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- > include/linux/soc/qcom/llcc-qcom.h | 4 +++- > 2 files changed, 14 insertions(+), 2 deletions(-) > > diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c > index 4ca88eaebf06..5a2dac2d4772 100644 > --- a/drivers/soc/qcom/llcc-qcom.c > +++ b/drivers/soc/qcom/llcc-qcom.c > @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, > return ret; > > if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > - ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > + ret = regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, > slice_status, (slice_status & ACT_COMPLETE), > 0, LLCC_STATUS_READ_DELAY); > if (ret) > @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, > ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > slice_status, !(slice_status & status), > 0, LLCC_STATUS_READ_DELAY); > + if (ret) > + return ret; > > if (drv_data->version >= LLCC_VERSION_4_1_0_0) > ret = regmap_write(drv_data->bcast_regmap, act_clear_reg, > @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct platform_device *pdev) > > drv_data->version = version; > > + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > + drv_data->bcast_and_regmap = qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); > + if (IS_ERR(drv_data->bcast_and_regmap)) { > + ret = PTR_ERR(drv_data->bcast_and_regmap); > + goto err; > + } > + } > + > llcc_cfg = cfg->sct_data; > sz = cfg->size; > > diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h > index 1a886666bbb6..9e9f528b1370 100644 > --- a/include/linux/soc/qcom/llcc-qcom.h > +++ b/include/linux/soc/qcom/llcc-qcom.h > @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { > /** > * struct llcc_drv_data - Data associated with the llcc driver > * @regmaps: regmaps associated with the llcc device > - * @bcast_regmap: regmap associated with llcc broadcast offset > + * @bcast_regmap: regmap associated with llcc broadcast OR offset > + * @bcast_and_regmap: regmap associated with llcc broadcast AND offset > * @cfg: pointer to the data structure for slice configuration > * @edac_reg_offset: Offset of the LLCC EDAC registers > * @lock: mutex associated with each slice > @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { > struct llcc_drv_data { > struct regmap **regmaps; > struct regmap *bcast_regmap; > + struct regmap *bcast_and_regmap; > const struct llcc_slice_config *cfg; > const struct llcc_edac_reg_offset *edac_reg_offset; > struct mutex lock; > -- > 2.25.1 > >
On 1/26/2024 12:29 PM, Dmitry Baryshkov wrote: > On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala > <quic_uchalich@quicinc.com> wrote: >> >> To support CSR programming, a broadcast interface is used to program >> all channels in a single command. Until SM8450 there was only one >> broadcast region (Broadcast_OR) used to broadcast write and check >> for status bit 0. From SM8450 onwards another broadcast region >> (Broadcast_AND) has been added which checks for status bit 1. >> >> Update llcc_drv_data structure with new regmap for Broadcast_AND >> region and initialize regmap for Broadcast_AND region when HW block >> version is greater than 4.1 for backwards compatibility. >> >> Switch from broadcast_OR to broadcast_AND region for checking >> status bit 1 as Broadcast_OR region checks only for bit 0. > > This breaks backwards compatibility with the existing DT files, doesn't it? > It shouldn't as checking for status bit 1 is happening only when the block version is greater than 4.1, which is when Broadcast_AND region support is added. >> While at it, also check return value after reading Broadcast_OR >> region in llcc_update_act_ctrl(). > > Separate patch, Fixes tag. > Ack. Will remove this from existing patch. Thanks for the review Dmitry! >> >> Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> >> --- >> drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- >> include/linux/soc/qcom/llcc-qcom.h | 4 +++- >> 2 files changed, 14 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c >> index 4ca88eaebf06..5a2dac2d4772 100644 >> --- a/drivers/soc/qcom/llcc-qcom.c >> +++ b/drivers/soc/qcom/llcc-qcom.c >> @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, >> return ret; >> >> if (drv_data->version >= LLCC_VERSION_4_1_0_0) { >> - ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, >> + ret = regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, >> slice_status, (slice_status & ACT_COMPLETE), >> 0, LLCC_STATUS_READ_DELAY); >> if (ret) >> @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, >> ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, >> slice_status, !(slice_status & status), >> 0, LLCC_STATUS_READ_DELAY); >> + if (ret) >> + return ret; >> >> if (drv_data->version >= LLCC_VERSION_4_1_0_0) >> ret = regmap_write(drv_data->bcast_regmap, act_clear_reg, >> @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct platform_device *pdev) >> >> drv_data->version = version; >> >> + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { >> + drv_data->bcast_and_regmap = qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); >> + if (IS_ERR(drv_data->bcast_and_regmap)) { >> + ret = PTR_ERR(drv_data->bcast_and_regmap); >> + goto err; >> + } >> + } >> + >> llcc_cfg = cfg->sct_data; >> sz = cfg->size; >> >> diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h >> index 1a886666bbb6..9e9f528b1370 100644 >> --- a/include/linux/soc/qcom/llcc-qcom.h >> +++ b/include/linux/soc/qcom/llcc-qcom.h >> @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { >> /** >> * struct llcc_drv_data - Data associated with the llcc driver >> * @regmaps: regmaps associated with the llcc device >> - * @bcast_regmap: regmap associated with llcc broadcast offset >> + * @bcast_regmap: regmap associated with llcc broadcast OR offset >> + * @bcast_and_regmap: regmap associated with llcc broadcast AND offset >> * @cfg: pointer to the data structure for slice configuration >> * @edac_reg_offset: Offset of the LLCC EDAC registers >> * @lock: mutex associated with each slice >> @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { >> struct llcc_drv_data { >> struct regmap **regmaps; >> struct regmap *bcast_regmap; >> + struct regmap *bcast_and_regmap; >> const struct llcc_slice_config *cfg; >> const struct llcc_edac_reg_offset *edac_reg_offset; >> struct mutex lock; >> -- >> 2.25.1 >> >> > >
On Mon, 29 Jan 2024 at 20:17, Unnathi Chalicheemala <quic_uchalich@quicinc.com> wrote: > > > > On 1/26/2024 12:29 PM, Dmitry Baryshkov wrote: > > On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala > > <quic_uchalich@quicinc.com> wrote: > >> > >> To support CSR programming, a broadcast interface is used to program > >> all channels in a single command. Until SM8450 there was only one > >> broadcast region (Broadcast_OR) used to broadcast write and check > >> for status bit 0. From SM8450 onwards another broadcast region > >> (Broadcast_AND) has been added which checks for status bit 1. > >> > >> Update llcc_drv_data structure with new regmap for Broadcast_AND > >> region and initialize regmap for Broadcast_AND region when HW block > >> version is greater than 4.1 for backwards compatibility. > >> > >> Switch from broadcast_OR to broadcast_AND region for checking > >> status bit 1 as Broadcast_OR region checks only for bit 0. > > > > This breaks backwards compatibility with the existing DT files, doesn't it? > > > > It shouldn't as checking for status bit 1 is happening only when the block > version is greater than 4.1, which is when Broadcast_AND region support > is added. Let me reiterate, please: with the existing DT files. You are patching DT files in patches 2-4, but this is not enough. DT files are considered to be ABI. As such old DT files must continue to work with newer kernels. > > >> While at it, also check return value after reading Broadcast_OR > >> region in llcc_update_act_ctrl(). > > > > Separate patch, Fixes tag. > > > > Ack. Will remove this from existing patch. > Thanks for the review Dmitry! > > >> > >> Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> > >> --- > >> drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- > >> include/linux/soc/qcom/llcc-qcom.h | 4 +++- > >> 2 files changed, 14 insertions(+), 2 deletions(-) > >> > >> diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c > >> index 4ca88eaebf06..5a2dac2d4772 100644 > >> --- a/drivers/soc/qcom/llcc-qcom.c > >> +++ b/drivers/soc/qcom/llcc-qcom.c > >> @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, > >> return ret; > >> > >> if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > >> - ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > >> + ret = regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, > >> slice_status, (slice_status & ACT_COMPLETE), > >> 0, LLCC_STATUS_READ_DELAY); > >> if (ret) > >> @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, > >> ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > >> slice_status, !(slice_status & status), > >> 0, LLCC_STATUS_READ_DELAY); > >> + if (ret) > >> + return ret; > >> > >> if (drv_data->version >= LLCC_VERSION_4_1_0_0) > >> ret = regmap_write(drv_data->bcast_regmap, act_clear_reg, > >> @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct platform_device *pdev) > >> > >> drv_data->version = version; > >> > >> + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > >> + drv_data->bcast_and_regmap = qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); > >> + if (IS_ERR(drv_data->bcast_and_regmap)) { > >> + ret = PTR_ERR(drv_data->bcast_and_regmap); > >> + goto err; > >> + } > >> + } > >> + > >> llcc_cfg = cfg->sct_data; > >> sz = cfg->size; > >> > >> diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h > >> index 1a886666bbb6..9e9f528b1370 100644 > >> --- a/include/linux/soc/qcom/llcc-qcom.h > >> +++ b/include/linux/soc/qcom/llcc-qcom.h > >> @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { > >> /** > >> * struct llcc_drv_data - Data associated with the llcc driver > >> * @regmaps: regmaps associated with the llcc device > >> - * @bcast_regmap: regmap associated with llcc broadcast offset > >> + * @bcast_regmap: regmap associated with llcc broadcast OR offset > >> + * @bcast_and_regmap: regmap associated with llcc broadcast AND offset > >> * @cfg: pointer to the data structure for slice configuration > >> * @edac_reg_offset: Offset of the LLCC EDAC registers > >> * @lock: mutex associated with each slice > >> @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { > >> struct llcc_drv_data { > >> struct regmap **regmaps; > >> struct regmap *bcast_regmap; > >> + struct regmap *bcast_and_regmap; > >> const struct llcc_slice_config *cfg; > >> const struct llcc_edac_reg_offset *edac_reg_offset; > >> struct mutex lock; > >> -- > >> 2.25.1 > >> > >> > > > >
On 1/29/2024 2:03 PM, Dmitry Baryshkov wrote: > On Mon, 29 Jan 2024 at 20:17, Unnathi Chalicheemala > <quic_uchalich@quicinc.com> wrote: >> >> >> >> On 1/26/2024 12:29 PM, Dmitry Baryshkov wrote: >>> On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala >>> <quic_uchalich@quicinc.com> wrote: >>>> >>>> To support CSR programming, a broadcast interface is used to program >>>> all channels in a single command. Until SM8450 there was only one >>>> broadcast region (Broadcast_OR) used to broadcast write and check >>>> for status bit 0. From SM8450 onwards another broadcast region >>>> (Broadcast_AND) has been added which checks for status bit 1. >>>> >>>> Update llcc_drv_data structure with new regmap for Broadcast_AND >>>> region and initialize regmap for Broadcast_AND region when HW block >>>> version is greater than 4.1 for backwards compatibility. >>>> >>>> Switch from broadcast_OR to broadcast_AND region for checking >>>> status bit 1 as Broadcast_OR region checks only for bit 0. >>> >>> This breaks backwards compatibility with the existing DT files, doesn't it? >>> >> >> It shouldn't as checking for status bit 1 is happening only when the block >> version is greater than 4.1, which is when Broadcast_AND region support >> is added. > > Let me reiterate, please: with the existing DT files. You are patching > DT files in patches 2-4, but this is not enough. DT files are > considered to be ABI. As such old DT files must continue to work with > newer kernels. > I'm sorry, I think I'm not understanding this right. >> >>>> While at it, also check return value after reading Broadcast_OR >>>> region in llcc_update_act_ctrl(). >>> >>> Separate patch, Fixes tag. >>> >> >> Ack. Will remove this from existing patch. >> Thanks for the review Dmitry! >> >>>> >>>> Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> >>>> --- >>>> drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- >>>> include/linux/soc/qcom/llcc-qcom.h | 4 +++- >>>> 2 files changed, 14 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c >>>> index 4ca88eaebf06..5a2dac2d4772 100644 >>>> --- a/drivers/soc/qcom/llcc-qcom.c >>>> +++ b/drivers/soc/qcom/llcc-qcom.c >>>> @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, >>>> return ret; >>>> >>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) { >>>> - ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, >>>> + ret = regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, >>>> slice_status, (slice_status & ACT_COMPLETE), >>>> 0, LLCC_STATUS_READ_DELAY); Above if condition will be true only for SM8450, 8550 and 8650 - whose DT files have been changed. It would never check for other existing DT files - I guess I'm failing to understand why the code would break with other DeviceTree files. >>>> if (ret) >>>> @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, >>>> ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, >>>> slice_status, !(slice_status & status), >>>> 0, LLCC_STATUS_READ_DELAY); >>>> + if (ret) >>>> + return ret; >>>> >>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) >>>> ret = regmap_write(drv_data->bcast_regmap, act_clear_reg, >>>> @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct platform_device *pdev) >>>> >>>> drv_data->version = version; >>>> >>>> + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { >>>> + drv_data->bcast_and_regmap = qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); >>>> + if (IS_ERR(drv_data->bcast_and_regmap)) { >>>> + ret = PTR_ERR(drv_data->bcast_and_regmap); >>>> + goto err; >>>> + } >>>> + } I have added a similar check in the probe function above; are you saying this too will break with existing DT files? >>>> + >>>> llcc_cfg = cfg->sct_data; >>>> sz = cfg->size; >>>> >>>> diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h >>>> index 1a886666bbb6..9e9f528b1370 100644 >>>> --- a/include/linux/soc/qcom/llcc-qcom.h >>>> +++ b/include/linux/soc/qcom/llcc-qcom.h >>>> @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { >>>> /** >>>> * struct llcc_drv_data - Data associated with the llcc driver >>>> * @regmaps: regmaps associated with the llcc device >>>> - * @bcast_regmap: regmap associated with llcc broadcast offset >>>> + * @bcast_regmap: regmap associated with llcc broadcast OR offset >>>> + * @bcast_and_regmap: regmap associated with llcc broadcast AND offset >>>> * @cfg: pointer to the data structure for slice configuration >>>> * @edac_reg_offset: Offset of the LLCC EDAC registers >>>> * @lock: mutex associated with each slice >>>> @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { >>>> struct llcc_drv_data { >>>> struct regmap **regmaps; >>>> struct regmap *bcast_regmap; >>>> + struct regmap *bcast_and_regmap; >>>> const struct llcc_slice_config *cfg; >>>> const struct llcc_edac_reg_offset *edac_reg_offset; >>>> struct mutex lock; >>>> -- >>>> 2.25.1 >>>> >>>> >>> >>> > > > -- Thanks & Warm Regards, Unnathi
On Tue, 30 Jan 2024 at 19:52, Unnathi Chalicheemala <quic_uchalich@quicinc.com> wrote: > > On 1/29/2024 2:03 PM, Dmitry Baryshkov wrote: > > On Mon, 29 Jan 2024 at 20:17, Unnathi Chalicheemala > > <quic_uchalich@quicinc.com> wrote: > >> > >> > >> > >> On 1/26/2024 12:29 PM, Dmitry Baryshkov wrote: > >>> On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala > >>> <quic_uchalich@quicinc.com> wrote: > >>>> > >>>> To support CSR programming, a broadcast interface is used to program > >>>> all channels in a single command. Until SM8450 there was only one > >>>> broadcast region (Broadcast_OR) used to broadcast write and check > >>>> for status bit 0. From SM8450 onwards another broadcast region > >>>> (Broadcast_AND) has been added which checks for status bit 1. > >>>> > >>>> Update llcc_drv_data structure with new regmap for Broadcast_AND > >>>> region and initialize regmap for Broadcast_AND region when HW block > >>>> version is greater than 4.1 for backwards compatibility. > >>>> > >>>> Switch from broadcast_OR to broadcast_AND region for checking > >>>> status bit 1 as Broadcast_OR region checks only for bit 0. > >>> > >>> This breaks backwards compatibility with the existing DT files, doesn't it? > >>> > >> > >> It shouldn't as checking for status bit 1 is happening only when the block > >> version is greater than 4.1, which is when Broadcast_AND region support > >> is added. > > > > Let me reiterate, please: with the existing DT files. You are patching > > DT files in patches 2-4, but this is not enough. DT files are > > considered to be ABI. As such old DT files must continue to work with > > newer kernels. > > > > I'm sorry, I think I'm not understanding this right. > > >> > >>>> While at it, also check return value after reading Broadcast_OR > >>>> region in llcc_update_act_ctrl(). > >>> > >>> Separate patch, Fixes tag. > >>> > >> > >> Ack. Will remove this from existing patch. > >> Thanks for the review Dmitry! > >> > >>>> > >>>> Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> > >>>> --- > >>>> drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- > >>>> include/linux/soc/qcom/llcc-qcom.h | 4 +++- > >>>> 2 files changed, 14 insertions(+), 2 deletions(-) > >>>> > >>>> diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c > >>>> index 4ca88eaebf06..5a2dac2d4772 100644 > >>>> --- a/drivers/soc/qcom/llcc-qcom.c > >>>> +++ b/drivers/soc/qcom/llcc-qcom.c > >>>> @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, > >>>> return ret; > >>>> > >>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > >>>> - ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > >>>> + ret = regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, > >>>> slice_status, (slice_status & ACT_COMPLETE), > >>>> 0, LLCC_STATUS_READ_DELAY); > > Above if condition will be true only for SM8450, 8550 and 8650 - whose DT files have been changed. > It would never check for other existing DT files - I guess I'm failing to understand why the code > would break with other DeviceTree files. I'm saying that the driver must continue to work (well, at least not to crash) even if somebody runs the kernel with older DT. > > >>>> if (ret) > >>>> @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, > >>>> ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > >>>> slice_status, !(slice_status & status), > >>>> 0, LLCC_STATUS_READ_DELAY); > >>>> + if (ret) > >>>> + return ret; > >>>> > >>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) > >>>> ret = regmap_write(drv_data->bcast_regmap, act_clear_reg, > >>>> @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct platform_device *pdev) > >>>> > >>>> drv_data->version = version; > >>>> > >>>> + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > >>>> + drv_data->bcast_and_regmap = qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); > >>>> + if (IS_ERR(drv_data->bcast_and_regmap)) { > >>>> + ret = PTR_ERR(drv_data->bcast_and_regmap); > >>>> + goto err; > >>>> + } > >>>> + } > > I have added a similar check in the probe function above; are you saying this too will break with > existing DT files? > > >>>> + > >>>> llcc_cfg = cfg->sct_data; > >>>> sz = cfg->size; > >>>> > >>>> diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h > >>>> index 1a886666bbb6..9e9f528b1370 100644 > >>>> --- a/include/linux/soc/qcom/llcc-qcom.h > >>>> +++ b/include/linux/soc/qcom/llcc-qcom.h > >>>> @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { > >>>> /** > >>>> * struct llcc_drv_data - Data associated with the llcc driver > >>>> * @regmaps: regmaps associated with the llcc device > >>>> - * @bcast_regmap: regmap associated with llcc broadcast offset > >>>> + * @bcast_regmap: regmap associated with llcc broadcast OR offset > >>>> + * @bcast_and_regmap: regmap associated with llcc broadcast AND offset > >>>> * @cfg: pointer to the data structure for slice configuration > >>>> * @edac_reg_offset: Offset of the LLCC EDAC registers > >>>> * @lock: mutex associated with each slice > >>>> @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { > >>>> struct llcc_drv_data { > >>>> struct regmap **regmaps; > >>>> struct regmap *bcast_regmap; > >>>> + struct regmap *bcast_and_regmap; > >>>> const struct llcc_slice_config *cfg; > >>>> const struct llcc_edac_reg_offset *edac_reg_offset; > >>>> struct mutex lock; > >>>> -- > >>>> 2.25.1 > >>>> > >>>> > >>> > >>> > > > > > > > -- > Thanks & Warm Regards, > Unnathi
On 1/30/2024 10:57 AM, Dmitry Baryshkov wrote: > On Tue, 30 Jan 2024 at 19:52, Unnathi Chalicheemala > <quic_uchalich@quicinc.com> wrote: >> On 1/29/2024 2:03 PM, Dmitry Baryshkov wrote: >>> On Mon, 29 Jan 2024 at 20:17, Unnathi Chalicheemala >>> <quic_uchalich@quicinc.com> wrote: >>>> >>>> >>>> On 1/26/2024 12:29 PM, Dmitry Baryshkov wrote: >>>>> On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala >>>>> <quic_uchalich@quicinc.com> wrote: >>>>>> To support CSR programming, a broadcast interface is used to program >>>>>> all channels in a single command. Until SM8450 there was only one >>>>>> broadcast region (Broadcast_OR) used to broadcast write and check >>>>>> for status bit 0. From SM8450 onwards another broadcast region >>>>>> (Broadcast_AND) has been added which checks for status bit 1. >>>>>> >>>>>> Update llcc_drv_data structure with new regmap for Broadcast_AND >>>>>> region and initialize regmap for Broadcast_AND region when HW block >>>>>> version is greater than 4.1 for backwards compatibility. >>>>>> >>>>>> Switch from broadcast_OR to broadcast_AND region for checking >>>>>> status bit 1 as Broadcast_OR region checks only for bit 0. >>>>> This breaks backwards compatibility with the existing DT files, >>>>> doesn't it? >>>>> >>>> It shouldn't as checking for status bit 1 is happening only when the >>>> block >>>> version is greater than 4.1, which is when Broadcast_AND region support >>>> is added. >>> Let me reiterate, please: with the existing DT files. You are patching >>> DT files in patches 2-4, but this is not enough. DT files are >>> considered to be ABI. As such old DT files must continue to work with >>> newer kernels. >>> >> I'm sorry, I think I'm not understanding this right. >> >>>>>> While at it, also check return value after reading Broadcast_OR >>>>>> region in llcc_update_act_ctrl(). >>>>> Separate patch, Fixes tag. >>>>> >>>> Ack. Will remove this from existing patch. >>>> Thanks for the review Dmitry! >>>> >>>>>> Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> >>>>>> --- >>>>>> drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- >>>>>> include/linux/soc/qcom/llcc-qcom.h | 4 +++- >>>>>> 2 files changed, 14 insertions(+), 2 deletions(-) >>>>>> >>>>>> diff --git a/drivers/soc/qcom/llcc-qcom.c >>>>>> b/drivers/soc/qcom/llcc-qcom.c >>>>>> index 4ca88eaebf06..5a2dac2d4772 100644 >>>>>> --- a/drivers/soc/qcom/llcc-qcom.c >>>>>> +++ b/drivers/soc/qcom/llcc-qcom.c >>>>>> @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, >>>>>> return ret; >>>>>> >>>>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) { >>>>>> - ret = >>>>>> regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, >>>>>> + ret = >>>>>> regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, >>>>>> slice_status, (slice_status & >>>>>> ACT_COMPLETE), >>>>>> 0, LLCC_STATUS_READ_DELAY); >> Above if condition will be true only for SM8450, 8550 and 8650 - whose DT >> files have been changed. >> It would never check for other existing DT files - I guess I'm failing to >> understand why the code >> would break with other DeviceTree files. > I'm saying that the driver must continue to work (well, at least not > to crash) even if somebody runs the kernel with older DT. Thanks Dmitry. While I get the ask, wondering why someone would use old DT while DT is also being updated in this series along with the driver change? Unnathi, you add check and make sure to update only when bcast_and region is specified in DT, otherwise, stick to what is being done today, that way backwards compatibility is maintained. >>>>>> if (ret) >>>>>> @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, >>>>>> ret = regmap_read_poll_timeout(drv_data->bcast_regmap, >>>>>> status_reg, >>>>>> slice_status, !(slice_status & >>>>>> status), >>>>>> 0, LLCC_STATUS_READ_DELAY); >>>>>> + if (ret) >>>>>> + return ret; >>>>>> >>>>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) >>>>>> ret = regmap_write(drv_data->bcast_regmap, >>>>>> act_clear_reg, >>>>>> @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct >>>>>> platform_device *pdev) >>>>>> >>>>>> drv_data->version = version; >>>>>> >>>>>> + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { >>>>>> + drv_data->bcast_and_regmap = >>>>>> qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); >>>>>> + if (IS_ERR(drv_data->bcast_and_regmap)) { >>>>>> + ret = PTR_ERR(drv_data->bcast_and_regmap); >>>>>> + goto err; >>>>>> + } >>>>>> + } >> I have added a similar check in the probe function above; are you saying >> this too will break with >> existing DT files? >> >>>>>> + >>>>>> llcc_cfg = cfg->sct_data; >>>>>> sz = cfg->size; >>>>>> >>>>>> diff --git a/include/linux/soc/qcom/llcc-qcom.h >>>>>> b/include/linux/soc/qcom/llcc-qcom.h >>>>>> index 1a886666bbb6..9e9f528b1370 100644 >>>>>> --- a/include/linux/soc/qcom/llcc-qcom.h >>>>>> +++ b/include/linux/soc/qcom/llcc-qcom.h >>>>>> @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { >>>>>> /** >>>>>> * struct llcc_drv_data - Data associated with the llcc driver >>>>>> * @regmaps: regmaps associated with the llcc device >>>>>> - * @bcast_regmap: regmap associated with llcc broadcast offset >>>>>> + * @bcast_regmap: regmap associated with llcc broadcast OR offset >>>>>> + * @bcast_and_regmap: regmap associated with llcc broadcast AND >>>>>> offset >>>>>> * @cfg: pointer to the data structure for slice configuration >>>>>> * @edac_reg_offset: Offset of the LLCC EDAC registers >>>>>> * @lock: mutex associated with each slice >>>>>> @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { >>>>>> struct llcc_drv_data { >>>>>> struct regmap **regmaps; >>>>>> struct regmap *bcast_regmap; >>>>>> + struct regmap *bcast_and_regmap; >>>>>> const struct llcc_slice_config *cfg; >>>>>> const struct llcc_edac_reg_offset *edac_reg_offset; >>>>>> struct mutex lock; >>>>>> -- >>>>>> 2.25.1 >>>>>> >>>>>> >>>>> >>> >>> >> -- >> Thanks & Warm Regards, >> Unnathi > >
On Mon, 5 Feb 2024 at 02:01, Satya Durga Srinivasu Prabhala <quic_satyap@quicinc.com> wrote: > > > On 1/30/2024 10:57 AM, Dmitry Baryshkov wrote: > > On Tue, 30 Jan 2024 at 19:52, Unnathi Chalicheemala > > <quic_uchalich@quicinc.com> wrote: > >> On 1/29/2024 2:03 PM, Dmitry Baryshkov wrote: > >>> On Mon, 29 Jan 2024 at 20:17, Unnathi Chalicheemala > >>> <quic_uchalich@quicinc.com> wrote: > >>>> > >>>> > >>>> On 1/26/2024 12:29 PM, Dmitry Baryshkov wrote: > >>>>> On Fri, 26 Jan 2024 at 21:48, Unnathi Chalicheemala > >>>>> <quic_uchalich@quicinc.com> wrote: > >>>>>> To support CSR programming, a broadcast interface is used to program > >>>>>> all channels in a single command. Until SM8450 there was only one > >>>>>> broadcast region (Broadcast_OR) used to broadcast write and check > >>>>>> for status bit 0. From SM8450 onwards another broadcast region > >>>>>> (Broadcast_AND) has been added which checks for status bit 1. > >>>>>> > >>>>>> Update llcc_drv_data structure with new regmap for Broadcast_AND > >>>>>> region and initialize regmap for Broadcast_AND region when HW block > >>>>>> version is greater than 4.1 for backwards compatibility. > >>>>>> > >>>>>> Switch from broadcast_OR to broadcast_AND region for checking > >>>>>> status bit 1 as Broadcast_OR region checks only for bit 0. > >>>>> This breaks backwards compatibility with the existing DT files, > >>>>> doesn't it? > >>>>> > >>>> It shouldn't as checking for status bit 1 is happening only when the > >>>> block > >>>> version is greater than 4.1, which is when Broadcast_AND region support > >>>> is added. > >>> Let me reiterate, please: with the existing DT files. You are patching > >>> DT files in patches 2-4, but this is not enough. DT files are > >>> considered to be ABI. As such old DT files must continue to work with > >>> newer kernels. > >>> > >> I'm sorry, I think I'm not understanding this right. > >> > >>>>>> While at it, also check return value after reading Broadcast_OR > >>>>>> region in llcc_update_act_ctrl(). > >>>>> Separate patch, Fixes tag. > >>>>> > >>>> Ack. Will remove this from existing patch. > >>>> Thanks for the review Dmitry! > >>>> > >>>>>> Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> > >>>>>> --- > >>>>>> drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- > >>>>>> include/linux/soc/qcom/llcc-qcom.h | 4 +++- > >>>>>> 2 files changed, 14 insertions(+), 2 deletions(-) > >>>>>> > >>>>>> diff --git a/drivers/soc/qcom/llcc-qcom.c > >>>>>> b/drivers/soc/qcom/llcc-qcom.c > >>>>>> index 4ca88eaebf06..5a2dac2d4772 100644 > >>>>>> --- a/drivers/soc/qcom/llcc-qcom.c > >>>>>> +++ b/drivers/soc/qcom/llcc-qcom.c > >>>>>> @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, > >>>>>> return ret; > >>>>>> > >>>>>> if (drv_data->version >= LLCC_VERSION_4_1_0_0) { > >>>>>> - ret = > >>>>>> regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, > >>>>>> + ret = > >>>>>> regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, > >>>>>> slice_status, (slice_status & > >>>>>> ACT_COMPLETE), > >>>>>> 0, LLCC_STATUS_READ_DELAY); > >> Above if condition will be true only for SM8450, 8550 and 8650 - whose DT > >> files have been changed. > >> It would never check for other existing DT files - I guess I'm failing to > >> understand why the code > >> would break with other DeviceTree files. > > I'm saying that the driver must continue to work (well, at least not > > to crash) even if somebody runs the kernel with older DT. > Thanks Dmitry. While I get the ask, wondering why someone would use old > DT while DT > is also being updated in this series along with the driver change? This is quite simple. The DT might be coming from the sources other than just the kernel. It might be a part of the firmware. It might be stored separately in the /boot partition, etc. The rule of thumb is that you must never depend on the DT being updated together with the kernel. This is one of the reasons why DT maintainers spend so much time and effort to get DT bindings right in the first place. Once they are merged, we have to support old DT nearly forever. -- With best wishes Dmitry
diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c index 4ca88eaebf06..5a2dac2d4772 100644 --- a/drivers/soc/qcom/llcc-qcom.c +++ b/drivers/soc/qcom/llcc-qcom.c @@ -849,7 +849,7 @@ static int llcc_update_act_ctrl(u32 sid, return ret; if (drv_data->version >= LLCC_VERSION_4_1_0_0) { - ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, + ret = regmap_read_poll_timeout(drv_data->bcast_and_regmap, status_reg, slice_status, (slice_status & ACT_COMPLETE), 0, LLCC_STATUS_READ_DELAY); if (ret) @@ -859,6 +859,8 @@ static int llcc_update_act_ctrl(u32 sid, ret = regmap_read_poll_timeout(drv_data->bcast_regmap, status_reg, slice_status, !(slice_status & status), 0, LLCC_STATUS_READ_DELAY); + if (ret) + return ret; if (drv_data->version >= LLCC_VERSION_4_1_0_0) ret = regmap_write(drv_data->bcast_regmap, act_clear_reg, @@ -1282,6 +1284,14 @@ static int qcom_llcc_probe(struct platform_device *pdev) drv_data->version = version; + if (drv_data->version >= LLCC_VERSION_4_1_0_0) { + drv_data->bcast_and_regmap = qcom_llcc_init_mmio(pdev, i + 1, "llcc_broadcast_and_base"); + if (IS_ERR(drv_data->bcast_and_regmap)) { + ret = PTR_ERR(drv_data->bcast_and_regmap); + goto err; + } + } + llcc_cfg = cfg->sct_data; sz = cfg->size; diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h index 1a886666bbb6..9e9f528b1370 100644 --- a/include/linux/soc/qcom/llcc-qcom.h +++ b/include/linux/soc/qcom/llcc-qcom.h @@ -115,7 +115,8 @@ struct llcc_edac_reg_offset { /** * struct llcc_drv_data - Data associated with the llcc driver * @regmaps: regmaps associated with the llcc device - * @bcast_regmap: regmap associated with llcc broadcast offset + * @bcast_regmap: regmap associated with llcc broadcast OR offset + * @bcast_and_regmap: regmap associated with llcc broadcast AND offset * @cfg: pointer to the data structure for slice configuration * @edac_reg_offset: Offset of the LLCC EDAC registers * @lock: mutex associated with each slice @@ -129,6 +130,7 @@ struct llcc_edac_reg_offset { struct llcc_drv_data { struct regmap **regmaps; struct regmap *bcast_regmap; + struct regmap *bcast_and_regmap; const struct llcc_slice_config *cfg; const struct llcc_edac_reg_offset *edac_reg_offset; struct mutex lock;
To support CSR programming, a broadcast interface is used to program all channels in a single command. Until SM8450 there was only one broadcast region (Broadcast_OR) used to broadcast write and check for status bit 0. From SM8450 onwards another broadcast region (Broadcast_AND) has been added which checks for status bit 1. Update llcc_drv_data structure with new regmap for Broadcast_AND region and initialize regmap for Broadcast_AND region when HW block version is greater than 4.1 for backwards compatibility. Switch from broadcast_OR to broadcast_AND region for checking status bit 1 as Broadcast_OR region checks only for bit 0. While at it, also check return value after reading Broadcast_OR region in llcc_update_act_ctrl(). Signed-off-by: Unnathi Chalicheemala <quic_uchalich@quicinc.com> --- drivers/soc/qcom/llcc-qcom.c | 12 +++++++++++- include/linux/soc/qcom/llcc-qcom.h | 4 +++- 2 files changed, 14 insertions(+), 2 deletions(-)