| Message ID | 20231215235725.1247350-4-tanmay.shah@amd.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | add zynqmp TCM bindings | expand |
On Fri, Dec 15, 2023 at 03:57:25PM -0800, Tanmay Shah wrote: > ZynqMP TCM information is fixed in driver. Now ZynqMP TCM information s/"is fixed in driver"/"was fixed in driver" > is available in device-tree. Parse TCM information in driver > as per new bindings. > > Signed-off-by: Tanmay Shah <tanmay.shah@amd.com> > --- > > Changes in v8: > - parse power-domains property from device-tree and use EEMI calls > to power on/off TCM instead of using pm domains framework > - Remove checking of pm_domain_id validation to power on/off tcm > - Remove spurious change > > Changes in v7: > - move checking of pm_domain_id from previous patch > - fix mem_bank_data memory allocation > > drivers/remoteproc/xlnx_r5_remoteproc.c | 154 +++++++++++++++++++++++- > 1 file changed, 148 insertions(+), 6 deletions(-) > > diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c > index 4395edea9a64..36d73dcd93f0 100644 > --- a/drivers/remoteproc/xlnx_r5_remoteproc.c > +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c > @@ -74,8 +74,8 @@ struct mbox_info { > }; > > /* > - * Hardcoded TCM bank values. This will be removed once TCM bindings are > - * accepted for system-dt specifications and upstreamed in linux kernel > + * Hardcoded TCM bank values. This will stay in driver to maintain backward > + * compatibility with device-tree that does not have TCM information. > */ > static const struct mem_bank_data zynqmp_tcm_banks_split[] = { > {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */ > @@ -878,6 +878,139 @@ static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev) > return ERR_PTR(ret); > } > > +static int zynqmp_r5_get_tcm_node_from_dt(struct zynqmp_r5_cluster *cluster) > +{ > + struct of_phandle_args out_args; > + int tcm_reg_per_r5, tcm_pd_idx; > + struct zynqmp_r5_core *r5_core; > + int i, j, tcm_bank_count, ret; > + struct platform_device *cpdev; > + struct mem_bank_data *tcm; > + struct device_node *np; > + struct resource *res; > + u64 abs_addr, size; > + struct device *dev; > + > + for (i = 0; i < cluster->core_count; i++) { > + r5_core = cluster->r5_cores[i]; > + dev = r5_core->dev; > + np = of_node_get(dev_of_node(dev)); > + tcm_pd_idx = 1; > + > + /* we have address cell 2 and size cell as 2 */ > + tcm_reg_per_r5 = of_property_count_elems_of_size(np, "reg", > + 4 * sizeof(u32)); > + if (tcm_reg_per_r5 <= 0) { > + dev_err(dev, "can't get reg property err %d\n", tcm_reg_per_r5); > + return -EINVAL; > + } > + > + /* > + * In lockstep mode, r5 core 0 will use r5 core 1 TCM > + * power domains as well. so allocate twice of per core TCM Twice of what? Please use proper english in your multi line comments, i.e a capital letter for the first word and a dot at the end. > + */ > + if (cluster->mode == LOCKSTEP_MODE) > + tcm_bank_count = tcm_reg_per_r5 * 2; > + else > + tcm_bank_count = tcm_reg_per_r5; > + > + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count, > + sizeof(struct mem_bank_data *), > + GFP_KERNEL); > + if (!r5_core->tcm_banks) > + ret = -ENOMEM; > + > + r5_core->tcm_bank_count = tcm_bank_count; > + for (j = 0; j < tcm_bank_count; j++) { > + tcm = devm_kzalloc(dev, sizeof(struct mem_bank_data), > + GFP_KERNEL); > + if (!tcm) > + return -ENOMEM; > + > + r5_core->tcm_banks[j] = tcm; > + > + /* > + * In lockstep mode, get second core's TCM power domains id > + * after first core TCM parsing is done as There seems to be words missing at the end of the sentence, and there is no dot. > + */ > + if (j == tcm_reg_per_r5) { > + /* dec first core node */ > + of_node_put(np); > + > + /* get second core node */ > + np = of_get_next_child(cluster->dev->of_node, np); > + > + /* > + * reset index of power-domains property list > + * for second core > + */ > + tcm_pd_idx = 1; > + } > + > + /* get power-domains id of tcm */ > + ret = of_parse_phandle_with_args(np, "power-domains", > + "#power-domain-cells", > + tcm_pd_idx, > + &out_args); > + if (ret) { > + dev_err(r5_core->dev, > + "failed to get tcm %d pm domain, ret %d\n", > + j, ret); > + of_node_put(out_args.np); I'm pretty sure this isn't needed in error conditions since @out_args would not have been assigned. > + return ret; > + } > + tcm->pm_domain_id = out_args.args[0]; > + of_node_put(out_args.np); > + tcm_pd_idx++; > + > + /* > + * In lockstep mode, we only need second core's power domain > + * ids. Other information from second core isn't needed so > + * ignore it. This forms table as zynqmp_tcm_banks_lockstep I don't understand the last sentence of this comment and it is missing a dot at the end. Comments should be enlightening, the ones I found in this patch are sowing confusion. > + */ > + if (j >= tcm_reg_per_r5) > + contiue; > + This condition and the one above (j == tcm_reg_per_r5) is brittle and almost guaranteed to cause maintenance problems in the future. I understand your will to reuse as much code as possible but this is one of the rare cases where duplicating code is probably better. Please introduce two new functions, i.e zynqmp_r5_get_tcm_node_from_dt_split() and zynqmp_r5_get_tcm_node_from_dt_lockstep(), and do all the necessary processing based on the use case. Thanks, Mathieu > + /* get tcm address without translation */ > + ret = of_property_read_reg(np, j, &abs_addr, &size); > + if (ret) { > + of_node_put(np); > + dev_err(dev, "failed to get reg property\n"); > + return ret; > + } > + > + /* > + * remote processor can address only 32 bits > + * so convert 64-bits into 32-bits. This will discard > + * any unwanted upper 32-bits. > + */ > + tcm->da = (u32)abs_addr; > + tcm->size = (u32)size; > + > + cpdev = to_platform_device(dev); > + res = platform_get_resource(cpdev, IORESOURCE_MEM, j); > + if (!res) { > + of_node_put(np); > + dev_err(dev, "failed to get tcm resource\n"); > + return -EINVAL; > + } > + > + tcm->addr = (u32)res->start; > + tcm->bank_name = (char *)res->name; > + res = devm_request_mem_region(dev, tcm->addr, tcm->size, > + tcm->bank_name); > + if (!res) { > + dev_err(dev, "failed to request tcm resource\n"); > + of_node_put(np); > + return -EINVAL; > + } > + } > + } > + > + of_node_put(np); > + return 0; > +} > + > /** > * zynqmp_r5_get_tcm_node() > * Ideally this function should parse tcm node and store information > @@ -956,10 +1089,19 @@ static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster, > struct zynqmp_r5_core *r5_core; > int ret, i; > > - ret = zynqmp_r5_get_tcm_node(cluster); > - if (ret < 0) { > - dev_err(dev, "can't get tcm node, err %d\n", ret); > - return ret; > + r5_core = cluster->r5_cores[0]; > + if (of_find_property(r5_core->np, "reg", NULL)) { > + ret = zynqmp_r5_get_tcm_node_from_dt(cluster); > + if (ret) { > + dev_err(dev, "can't get tcm node from dt, err %d\n", ret); > + return ret; > + } > + } else { > + ret = zynqmp_r5_get_tcm_node(cluster); > + if (ret < 0) { > + dev_err(dev, "can't get tcm node, err %d\n", ret); > + return ret; > + } > } > > for (i = 0; i < cluster->core_count; i++) { > -- > 2.25.1 >
On 1/3/24 12:17 PM, Mathieu Poirier wrote: > On Fri, Dec 15, 2023 at 03:57:25PM -0800, Tanmay Shah wrote: > > ZynqMP TCM information is fixed in driver. Now ZynqMP TCM information > > s/"is fixed in driver"/"was fixed in driver" > > > is available in device-tree. Parse TCM information in driver > > as per new bindings. > > > > Signed-off-by: Tanmay Shah <tanmay.shah@amd.com> > > --- > > > > Changes in v8: > > - parse power-domains property from device-tree and use EEMI calls > > to power on/off TCM instead of using pm domains framework > > - Remove checking of pm_domain_id validation to power on/off tcm > > - Remove spurious change > > > > Changes in v7: > > - move checking of pm_domain_id from previous patch > > - fix mem_bank_data memory allocation > > > > drivers/remoteproc/xlnx_r5_remoteproc.c | 154 +++++++++++++++++++++++- > > 1 file changed, 148 insertions(+), 6 deletions(-) > > > > diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c > > index 4395edea9a64..36d73dcd93f0 100644 > > --- a/drivers/remoteproc/xlnx_r5_remoteproc.c > > +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c > > @@ -74,8 +74,8 @@ struct mbox_info { > > }; > > > > /* > > - * Hardcoded TCM bank values. This will be removed once TCM bindings are > > - * accepted for system-dt specifications and upstreamed in linux kernel > > + * Hardcoded TCM bank values. This will stay in driver to maintain backward > > + * compatibility with device-tree that does not have TCM information. > > */ > > static const struct mem_bank_data zynqmp_tcm_banks_split[] = { > > {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */ > > @@ -878,6 +878,139 @@ static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev) > > return ERR_PTR(ret); > > } > > > > +static int zynqmp_r5_get_tcm_node_from_dt(struct zynqmp_r5_cluster *cluster) > > +{ > > + struct of_phandle_args out_args; > > + int tcm_reg_per_r5, tcm_pd_idx; > > + struct zynqmp_r5_core *r5_core; > > + int i, j, tcm_bank_count, ret; > > + struct platform_device *cpdev; > > + struct mem_bank_data *tcm; > > + struct device_node *np; > > + struct resource *res; > > + u64 abs_addr, size; > > + struct device *dev; > > + > > + for (i = 0; i < cluster->core_count; i++) { > > + r5_core = cluster->r5_cores[i]; > > + dev = r5_core->dev; > > + np = of_node_get(dev_of_node(dev)); > > + tcm_pd_idx = 1; > > + > > + /* we have address cell 2 and size cell as 2 */ > > + tcm_reg_per_r5 = of_property_count_elems_of_size(np, "reg", > > + 4 * sizeof(u32)); > > + if (tcm_reg_per_r5 <= 0) { > > + dev_err(dev, "can't get reg property err %d\n", tcm_reg_per_r5); > > + return -EINVAL; > > + } > > + > > + /* > > + * In lockstep mode, r5 core 0 will use r5 core 1 TCM > > + * power domains as well. so allocate twice of per core TCM > > Twice of what? Please use proper english in your multi line comments, i.e a > capital letter for the first word and a dot at the end. > > > + */ > > + if (cluster->mode == LOCKSTEP_MODE) > > + tcm_bank_count = tcm_reg_per_r5 * 2; > > + else > > + tcm_bank_count = tcm_reg_per_r5; > > + > > + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count, > > + sizeof(struct mem_bank_data *), > > + GFP_KERNEL); > > + if (!r5_core->tcm_banks) > > + ret = -ENOMEM; > > + > > + r5_core->tcm_bank_count = tcm_bank_count; > > + for (j = 0; j < tcm_bank_count; j++) { > > + tcm = devm_kzalloc(dev, sizeof(struct mem_bank_data), > > + GFP_KERNEL); > > + if (!tcm) > > + return -ENOMEM; > > + > > + r5_core->tcm_banks[j] = tcm; > > + > > + /* > > + * In lockstep mode, get second core's TCM power domains id > > + * after first core TCM parsing is done as > > There seems to be words missing at the end of the sentence, and there is no dot. > > > + */ > > + if (j == tcm_reg_per_r5) { > > + /* dec first core node */ > > + of_node_put(np); > > + > > + /* get second core node */ > > + np = of_get_next_child(cluster->dev->of_node, np); > > + > > + /* > > + * reset index of power-domains property list > > + * for second core > > + */ > > + tcm_pd_idx = 1; > > + } > > + > > + /* get power-domains id of tcm */ > > + ret = of_parse_phandle_with_args(np, "power-domains", > > + "#power-domain-cells", > > + tcm_pd_idx, > > + &out_args); > > + if (ret) { > > + dev_err(r5_core->dev, > > + "failed to get tcm %d pm domain, ret %d\n", > > + j, ret); > > + of_node_put(out_args.np); > > I'm pretty sure this isn't needed in error conditions since @out_args would not > have been assigned. > > > + return ret; > > + } > > + tcm->pm_domain_id = out_args.args[0]; > > + of_node_put(out_args.np); > > + tcm_pd_idx++; > > + > > + /* > > + * In lockstep mode, we only need second core's power domain > > + * ids. Other information from second core isn't needed so > > + * ignore it. This forms table as zynqmp_tcm_banks_lockstep > > I don't understand the last sentence of this comment and it is missing a dot at > the end. Comments should be enlightening, the ones I found in this patch are > sowing confusion. > > > + */ > > + if (j >= tcm_reg_per_r5) > > + contiue; > > + > > This condition and the one above (j == tcm_reg_per_r5) is brittle and almost > guaranteed to cause maintenance problems in the future. > > I understand your will to reuse as much code as possible but this is one of the > rare cases where duplicating code is probably better. Please introduce two new > functions, i.e zynqmp_r5_get_tcm_node_from_dt_split() and > zynqmp_r5_get_tcm_node_from_dt_lockstep(), and do all the necessary processing > based on the use case. Ack will fix the documentation and tcm parsing as suggested. > Thanks, > Mathieu > > > + /* get tcm address without translation */ > > + ret = of_property_read_reg(np, j, &abs_addr, &size); > > + if (ret) { > > + of_node_put(np); > > + dev_err(dev, "failed to get reg property\n"); > > + return ret; > > + } > > + > > + /* > > + * remote processor can address only 32 bits > > + * so convert 64-bits into 32-bits. This will discard > > + * any unwanted upper 32-bits. > > + */ > > + tcm->da = (u32)abs_addr; > > + tcm->size = (u32)size; > > + > > + cpdev = to_platform_device(dev); > > + res = platform_get_resource(cpdev, IORESOURCE_MEM, j); > > + if (!res) { > > + of_node_put(np); > > + dev_err(dev, "failed to get tcm resource\n"); > > + return -EINVAL; > > + } > > + > > + tcm->addr = (u32)res->start; > > + tcm->bank_name = (char *)res->name; > > + res = devm_request_mem_region(dev, tcm->addr, tcm->size, > > + tcm->bank_name); > > + if (!res) { > > + dev_err(dev, "failed to request tcm resource\n"); > > + of_node_put(np); > > + return -EINVAL; > > + } > > + } > > + } > > + > > + of_node_put(np); > > + return 0; > > +} > > + > > /** > > * zynqmp_r5_get_tcm_node() > > * Ideally this function should parse tcm node and store information > > @@ -956,10 +1089,19 @@ static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster, > > struct zynqmp_r5_core *r5_core; > > int ret, i; > > > > - ret = zynqmp_r5_get_tcm_node(cluster); > > - if (ret < 0) { > > - dev_err(dev, "can't get tcm node, err %d\n", ret); > > - return ret; > > + r5_core = cluster->r5_cores[0]; > > + if (of_find_property(r5_core->np, "reg", NULL)) { > > + ret = zynqmp_r5_get_tcm_node_from_dt(cluster); > > + if (ret) { > > + dev_err(dev, "can't get tcm node from dt, err %d\n", ret); > > + return ret; > > + } > > + } else { > > + ret = zynqmp_r5_get_tcm_node(cluster); > > + if (ret < 0) { > > + dev_err(dev, "can't get tcm node, err %d\n", ret); > > + return ret; > > + } > > } > > > > for (i = 0; i < cluster->core_count; i++) { > > -- > > 2.25.1 > >
On 1/3/24 12:17 PM, Mathieu Poirier wrote: > On Fri, Dec 15, 2023 at 03:57:25PM -0800, Tanmay Shah wrote: > > ZynqMP TCM information is fixed in driver. Now ZynqMP TCM information > > s/"is fixed in driver"/"was fixed in driver" > > > is available in device-tree. Parse TCM information in driver > > as per new bindings. > > > > Signed-off-by: Tanmay Shah <tanmay.shah@amd.com> > > --- > > > > Changes in v8: > > - parse power-domains property from device-tree and use EEMI calls > > to power on/off TCM instead of using pm domains framework > > - Remove checking of pm_domain_id validation to power on/off tcm > > - Remove spurious change > > > > Changes in v7: > > - move checking of pm_domain_id from previous patch > > - fix mem_bank_data memory allocation > > > > drivers/remoteproc/xlnx_r5_remoteproc.c | 154 +++++++++++++++++++++++- > > 1 file changed, 148 insertions(+), 6 deletions(-) > > > > diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c > > index 4395edea9a64..36d73dcd93f0 100644 > > --- a/drivers/remoteproc/xlnx_r5_remoteproc.c > > +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c > > @@ -74,8 +74,8 @@ struct mbox_info { > > }; > > > > /* > > - * Hardcoded TCM bank values. This will be removed once TCM bindings are > > - * accepted for system-dt specifications and upstreamed in linux kernel > > + * Hardcoded TCM bank values. This will stay in driver to maintain backward > > + * compatibility with device-tree that does not have TCM information. > > */ > > static const struct mem_bank_data zynqmp_tcm_banks_split[] = { > > {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */ > > @@ -878,6 +878,139 @@ static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev) > > return ERR_PTR(ret); > > } > > > > +static int zynqmp_r5_get_tcm_node_from_dt(struct zynqmp_r5_cluster *cluster) > > +{ > > + struct of_phandle_args out_args; > > + int tcm_reg_per_r5, tcm_pd_idx; > > + struct zynqmp_r5_core *r5_core; > > + int i, j, tcm_bank_count, ret; > > + struct platform_device *cpdev; > > + struct mem_bank_data *tcm; > > + struct device_node *np; > > + struct resource *res; > > + u64 abs_addr, size; > > + struct device *dev; > > + > > + for (i = 0; i < cluster->core_count; i++) { > > + r5_core = cluster->r5_cores[i]; > > + dev = r5_core->dev; > > + np = of_node_get(dev_of_node(dev)); > > + tcm_pd_idx = 1; > > + > > + /* we have address cell 2 and size cell as 2 */ > > + tcm_reg_per_r5 = of_property_count_elems_of_size(np, "reg", > > + 4 * sizeof(u32)); > > + if (tcm_reg_per_r5 <= 0) { > > + dev_err(dev, "can't get reg property err %d\n", tcm_reg_per_r5); > > + return -EINVAL; > > + } > > + > > + /* > > + * In lockstep mode, r5 core 0 will use r5 core 1 TCM > > + * power domains as well. so allocate twice of per core TCM > > Twice of what? Please use proper english in your multi line comments, i.e a > capital letter for the first word and a dot at the end. > > > + */ > > + if (cluster->mode == LOCKSTEP_MODE) > > + tcm_bank_count = tcm_reg_per_r5 * 2; > > + else > > + tcm_bank_count = tcm_reg_per_r5; > > + > > + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count, > > + sizeof(struct mem_bank_data *), > > + GFP_KERNEL); > > + if (!r5_core->tcm_banks) > > + ret = -ENOMEM; > > + > > + r5_core->tcm_bank_count = tcm_bank_count; > > + for (j = 0; j < tcm_bank_count; j++) { > > + tcm = devm_kzalloc(dev, sizeof(struct mem_bank_data), > > + GFP_KERNEL); > > + if (!tcm) > > + return -ENOMEM; > > + > > + r5_core->tcm_banks[j] = tcm; > > + > > + /* > > + * In lockstep mode, get second core's TCM power domains id > > + * after first core TCM parsing is done as > > There seems to be words missing at the end of the sentence, and there is no dot. > > > + */ > > + if (j == tcm_reg_per_r5) { > > + /* dec first core node */ > > + of_node_put(np); > > + > > + /* get second core node */ > > + np = of_get_next_child(cluster->dev->of_node, np); > > + > > + /* > > + * reset index of power-domains property list > > + * for second core > > + */ > > + tcm_pd_idx = 1; > > + } > > + > > + /* get power-domains id of tcm */ > > + ret = of_parse_phandle_with_args(np, "power-domains", > > + "#power-domain-cells", > > + tcm_pd_idx, > > + &out_args); > > + if (ret) { > > + dev_err(r5_core->dev, > > + "failed to get tcm %d pm domain, ret %d\n", > > + j, ret); > > + of_node_put(out_args.np); > > I'm pretty sure this isn't needed in error conditions since @out_args would not > have been assigned. > > > + return ret; > > + } > > + tcm->pm_domain_id = out_args.args[0]; > > + of_node_put(out_args.np); > > + tcm_pd_idx++; > > + > > + /* > > + * In lockstep mode, we only need second core's power domain > > + * ids. Other information from second core isn't needed so > > + * ignore it. This forms table as zynqmp_tcm_banks_lockstep > > I don't understand the last sentence of this comment and it is missing a dot at > the end. Comments should be enlightening, the ones I found in this patch are > sowing confusion. > > > + */ > > + if (j >= tcm_reg_per_r5) > > + contiue; > > + > > This condition and the one above (j == tcm_reg_per_r5) is brittle and almost > guaranteed to cause maintenance problems in the future. > > I understand your will to reuse as much code as possible but this is one of the > rare cases where duplicating code is probably better. Please introduce two new > functions, i.e zynqmp_r5_get_tcm_node_from_dt_split() and > zynqmp_r5_get_tcm_node_from_dt_lockstep(), and do all the necessary processing > based on the use case. Hi Mathieu, I tried to implement this and it still looks hacky, as in lockstep mode unnecessary TCM is being allocated just to store power-domains. Instead, I am taking another cleaner approach where, TCM is parsed in uniform way in both modes from device-tree during zynqmp_r5_core_init. However, during "add_tcm_carveout_lockstep_mode" call, I will simply parse second core's TCM power-domains from device-tree and turn it on. I will implement this and send v9 after successful testing. I wanted to give you heads up on this approach. I hope it is fine. Thanks, Tanmay > > Thanks, > Mathieu > > > + /* get tcm address without translation */ > > + ret = of_property_read_reg(np, j, &abs_addr, &size); > > + if (ret) { > > + of_node_put(np); > > + dev_err(dev, "failed to get reg property\n"); > > + return ret; > > + } > > + > > + /* > > + * remote processor can address only 32 bits > > + * so convert 64-bits into 32-bits. This will discard > > + * any unwanted upper 32-bits. > > + */ > > + tcm->da = (u32)abs_addr; > > + tcm->size = (u32)size; > > + > > + cpdev = to_platform_device(dev); > > + res = platform_get_resource(cpdev, IORESOURCE_MEM, j); > > + if (!res) { > > + of_node_put(np); > > + dev_err(dev, "failed to get tcm resource\n"); > > + return -EINVAL; > > + } > > + > > + tcm->addr = (u32)res->start; > > + tcm->bank_name = (char *)res->name; > > + res = devm_request_mem_region(dev, tcm->addr, tcm->size, > > + tcm->bank_name); > > + if (!res) { > > + dev_err(dev, "failed to request tcm resource\n"); > > + of_node_put(np); > > + return -EINVAL; > > + } > > + } > > + } > > + > > + of_node_put(np); > > + return 0; > > +} > > + > > /** > > * zynqmp_r5_get_tcm_node() > > * Ideally this function should parse tcm node and store information > > @@ -956,10 +1089,19 @@ static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster, > > struct zynqmp_r5_core *r5_core; > > int ret, i; > > > > - ret = zynqmp_r5_get_tcm_node(cluster); > > - if (ret < 0) { > > - dev_err(dev, "can't get tcm node, err %d\n", ret); > > - return ret; > > + r5_core = cluster->r5_cores[0]; > > + if (of_find_property(r5_core->np, "reg", NULL)) { > > + ret = zynqmp_r5_get_tcm_node_from_dt(cluster); > > + if (ret) { > > + dev_err(dev, "can't get tcm node from dt, err %d\n", ret); > > + return ret; > > + } > > + } else { > > + ret = zynqmp_r5_get_tcm_node(cluster); > > + if (ret < 0) { > > + dev_err(dev, "can't get tcm node, err %d\n", ret); > > + return ret; > > + } > > } > > > > for (i = 0; i < cluster->core_count; i++) { > > -- > > 2.25.1 > >
On Thu, 4 Jan 2024 at 09:14, Tanmay Shah <tanmay.shah@amd.com> wrote: > > > On 1/3/24 12:17 PM, Mathieu Poirier wrote: > > On Fri, Dec 15, 2023 at 03:57:25PM -0800, Tanmay Shah wrote: > > > ZynqMP TCM information is fixed in driver. Now ZynqMP TCM information > > > > s/"is fixed in driver"/"was fixed in driver" > > > > > is available in device-tree. Parse TCM information in driver > > > as per new bindings. > > > > > > Signed-off-by: Tanmay Shah <tanmay.shah@amd.com> > > > --- > > > > > > Changes in v8: > > > - parse power-domains property from device-tree and use EEMI calls > > > to power on/off TCM instead of using pm domains framework > > > - Remove checking of pm_domain_id validation to power on/off tcm > > > - Remove spurious change > > > > > > Changes in v7: > > > - move checking of pm_domain_id from previous patch > > > - fix mem_bank_data memory allocation > > > > > > drivers/remoteproc/xlnx_r5_remoteproc.c | 154 +++++++++++++++++++++++- > > > 1 file changed, 148 insertions(+), 6 deletions(-) > > > > > > diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c > > > index 4395edea9a64..36d73dcd93f0 100644 > > > --- a/drivers/remoteproc/xlnx_r5_remoteproc.c > > > +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c > > > @@ -74,8 +74,8 @@ struct mbox_info { > > > }; > > > > > > /* > > > - * Hardcoded TCM bank values. This will be removed once TCM bindings are > > > - * accepted for system-dt specifications and upstreamed in linux kernel > > > + * Hardcoded TCM bank values. This will stay in driver to maintain backward > > > + * compatibility with device-tree that does not have TCM information. > > > */ > > > static const struct mem_bank_data zynqmp_tcm_banks_split[] = { > > > {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */ > > > @@ -878,6 +878,139 @@ static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev) > > > return ERR_PTR(ret); > > > } > > > > > > +static int zynqmp_r5_get_tcm_node_from_dt(struct zynqmp_r5_cluster *cluster) > > > +{ > > > + struct of_phandle_args out_args; > > > + int tcm_reg_per_r5, tcm_pd_idx; > > > + struct zynqmp_r5_core *r5_core; > > > + int i, j, tcm_bank_count, ret; > > > + struct platform_device *cpdev; > > > + struct mem_bank_data *tcm; > > > + struct device_node *np; > > > + struct resource *res; > > > + u64 abs_addr, size; > > > + struct device *dev; > > > + > > > + for (i = 0; i < cluster->core_count; i++) { > > > + r5_core = cluster->r5_cores[i]; > > > + dev = r5_core->dev; > > > + np = of_node_get(dev_of_node(dev)); > > > + tcm_pd_idx = 1; > > > + > > > + /* we have address cell 2 and size cell as 2 */ > > > + tcm_reg_per_r5 = of_property_count_elems_of_size(np, "reg", > > > + 4 * sizeof(u32)); > > > + if (tcm_reg_per_r5 <= 0) { > > > + dev_err(dev, "can't get reg property err %d\n", tcm_reg_per_r5); > > > + return -EINVAL; > > > + } > > > + > > > + /* > > > + * In lockstep mode, r5 core 0 will use r5 core 1 TCM > > > + * power domains as well. so allocate twice of per core TCM > > > > Twice of what? Please use proper english in your multi line comments, i.e a > > capital letter for the first word and a dot at the end. > > > > > + */ > > > + if (cluster->mode == LOCKSTEP_MODE) > > > + tcm_bank_count = tcm_reg_per_r5 * 2; > > > + else > > > + tcm_bank_count = tcm_reg_per_r5; > > > + > > > + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count, > > > + sizeof(struct mem_bank_data *), > > > + GFP_KERNEL); > > > + if (!r5_core->tcm_banks) > > > + ret = -ENOMEM; > > > + > > > + r5_core->tcm_bank_count = tcm_bank_count; > > > + for (j = 0; j < tcm_bank_count; j++) { > > > + tcm = devm_kzalloc(dev, sizeof(struct mem_bank_data), > > > + GFP_KERNEL); > > > + if (!tcm) > > > + return -ENOMEM; > > > + > > > + r5_core->tcm_banks[j] = tcm; > > > + > > > + /* > > > + * In lockstep mode, get second core's TCM power domains id > > > + * after first core TCM parsing is done as > > > > There seems to be words missing at the end of the sentence, and there is no dot. > > > > > + */ > > > + if (j == tcm_reg_per_r5) { > > > + /* dec first core node */ > > > + of_node_put(np); > > > + > > > + /* get second core node */ > > > + np = of_get_next_child(cluster->dev->of_node, np); > > > + > > > + /* > > > + * reset index of power-domains property list > > > + * for second core > > > + */ > > > + tcm_pd_idx = 1; > > > + } > > > + > > > + /* get power-domains id of tcm */ > > > + ret = of_parse_phandle_with_args(np, "power-domains", > > > + "#power-domain-cells", > > > + tcm_pd_idx, > > > + &out_args); > > > + if (ret) { > > > + dev_err(r5_core->dev, > > > + "failed to get tcm %d pm domain, ret %d\n", > > > + j, ret); > > > + of_node_put(out_args.np); > > > > I'm pretty sure this isn't needed in error conditions since @out_args would not > > have been assigned. > > > > > + return ret; > > > + } > > > + tcm->pm_domain_id = out_args.args[0]; > > > + of_node_put(out_args.np); > > > + tcm_pd_idx++; > > > + > > > + /* > > > + * In lockstep mode, we only need second core's power domain > > > + * ids. Other information from second core isn't needed so > > > + * ignore it. This forms table as zynqmp_tcm_banks_lockstep > > > > I don't understand the last sentence of this comment and it is missing a dot at > > the end. Comments should be enlightening, the ones I found in this patch are > > sowing confusion. > > > > > + */ > > > + if (j >= tcm_reg_per_r5) > > > + contiue; > > > + > > > > This condition and the one above (j == tcm_reg_per_r5) is brittle and almost > > guaranteed to cause maintenance problems in the future. > > > > I understand your will to reuse as much code as possible but this is one of the > > rare cases where duplicating code is probably better. Please introduce two new > > functions, i.e zynqmp_r5_get_tcm_node_from_dt_split() and > > zynqmp_r5_get_tcm_node_from_dt_lockstep(), and do all the necessary processing > > based on the use case. > > Hi Mathieu, > > I tried to implement this and it still looks hacky, as in lockstep mode unnecessary TCM is being allocated just to store power-domains. > > Instead, I am taking another cleaner approach where, TCM is parsed in uniform way in both modes from device-tree during > > zynqmp_r5_core_init. However, during "add_tcm_carveout_lockstep_mode" call, I will simply parse second core's TCM power-domains > > from device-tree and turn it on. > > I will implement this and send v9 after successful testing. I wanted to give you heads up on this approach. I hope it is fine. > Thanks for letting me know - now I don't have to pull out what's left of hair on my head to figure out what is going on. > Thanks, > Tanmay > > > > > > Thanks, > > Mathieu > > > > > + /* get tcm address without translation */ > > > + ret = of_property_read_reg(np, j, &abs_addr, &size); > > > + if (ret) { > > > + of_node_put(np); > > > + dev_err(dev, "failed to get reg property\n"); > > > + return ret; > > > + } > > > + > > > + /* > > > + * remote processor can address only 32 bits > > > + * so convert 64-bits into 32-bits. This will discard > > > + * any unwanted upper 32-bits. > > > + */ > > > + tcm->da = (u32)abs_addr; > > > + tcm->size = (u32)size; > > > + > > > + cpdev = to_platform_device(dev); > > > + res = platform_get_resource(cpdev, IORESOURCE_MEM, j); > > > + if (!res) { > > > + of_node_put(np); > > > + dev_err(dev, "failed to get tcm resource\n"); > > > + return -EINVAL; > > > + } > > > + > > > + tcm->addr = (u32)res->start; > > > + tcm->bank_name = (char *)res->name; > > > + res = devm_request_mem_region(dev, tcm->addr, tcm->size, > > > + tcm->bank_name); > > > + if (!res) { > > > + dev_err(dev, "failed to request tcm resource\n"); > > > + of_node_put(np); > > > + return -EINVAL; > > > + } > > > + } > > > + } > > > + > > > + of_node_put(np); > > > + return 0; > > > +} > > > + > > > /** > > > * zynqmp_r5_get_tcm_node() > > > * Ideally this function should parse tcm node and store information > > > @@ -956,10 +1089,19 @@ static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster, > > > struct zynqmp_r5_core *r5_core; > > > int ret, i; > > > > > > - ret = zynqmp_r5_get_tcm_node(cluster); > > > - if (ret < 0) { > > > - dev_err(dev, "can't get tcm node, err %d\n", ret); > > > - return ret; > > > + r5_core = cluster->r5_cores[0]; > > > + if (of_find_property(r5_core->np, "reg", NULL)) { > > > + ret = zynqmp_r5_get_tcm_node_from_dt(cluster); > > > + if (ret) { > > > + dev_err(dev, "can't get tcm node from dt, err %d\n", ret); > > > + return ret; > > > + } > > > + } else { > > > + ret = zynqmp_r5_get_tcm_node(cluster); > > > + if (ret < 0) { > > > + dev_err(dev, "can't get tcm node, err %d\n", ret); > > > + return ret; > > > + } > > > } > > > > > > for (i = 0; i < cluster->core_count; i++) { > > > -- > > > 2.25.1 > > >
diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c index 4395edea9a64..36d73dcd93f0 100644 --- a/drivers/remoteproc/xlnx_r5_remoteproc.c +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c @@ -74,8 +74,8 @@ struct mbox_info { }; /* - * Hardcoded TCM bank values. This will be removed once TCM bindings are - * accepted for system-dt specifications and upstreamed in linux kernel + * Hardcoded TCM bank values. This will stay in driver to maintain backward + * compatibility with device-tree that does not have TCM information. */ static const struct mem_bank_data zynqmp_tcm_banks_split[] = { {0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */ @@ -878,6 +878,139 @@ static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev) return ERR_PTR(ret); } +static int zynqmp_r5_get_tcm_node_from_dt(struct zynqmp_r5_cluster *cluster) +{ + struct of_phandle_args out_args; + int tcm_reg_per_r5, tcm_pd_idx; + struct zynqmp_r5_core *r5_core; + int i, j, tcm_bank_count, ret; + struct platform_device *cpdev; + struct mem_bank_data *tcm; + struct device_node *np; + struct resource *res; + u64 abs_addr, size; + struct device *dev; + + for (i = 0; i < cluster->core_count; i++) { + r5_core = cluster->r5_cores[i]; + dev = r5_core->dev; + np = of_node_get(dev_of_node(dev)); + tcm_pd_idx = 1; + + /* we have address cell 2 and size cell as 2 */ + tcm_reg_per_r5 = of_property_count_elems_of_size(np, "reg", + 4 * sizeof(u32)); + if (tcm_reg_per_r5 <= 0) { + dev_err(dev, "can't get reg property err %d\n", tcm_reg_per_r5); + return -EINVAL; + } + + /* + * In lockstep mode, r5 core 0 will use r5 core 1 TCM + * power domains as well. so allocate twice of per core TCM + */ + if (cluster->mode == LOCKSTEP_MODE) + tcm_bank_count = tcm_reg_per_r5 * 2; + else + tcm_bank_count = tcm_reg_per_r5; + + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count, + sizeof(struct mem_bank_data *), + GFP_KERNEL); + if (!r5_core->tcm_banks) + ret = -ENOMEM; + + r5_core->tcm_bank_count = tcm_bank_count; + for (j = 0; j < tcm_bank_count; j++) { + tcm = devm_kzalloc(dev, sizeof(struct mem_bank_data), + GFP_KERNEL); + if (!tcm) + return -ENOMEM; + + r5_core->tcm_banks[j] = tcm; + + /* + * In lockstep mode, get second core's TCM power domains id + * after first core TCM parsing is done as + */ + if (j == tcm_reg_per_r5) { + /* dec first core node */ + of_node_put(np); + + /* get second core node */ + np = of_get_next_child(cluster->dev->of_node, np); + + /* + * reset index of power-domains property list + * for second core + */ + tcm_pd_idx = 1; + } + + /* get power-domains id of tcm */ + ret = of_parse_phandle_with_args(np, "power-domains", + "#power-domain-cells", + tcm_pd_idx, + &out_args); + if (ret) { + dev_err(r5_core->dev, + "failed to get tcm %d pm domain, ret %d\n", + j, ret); + of_node_put(out_args.np); + return ret; + } + tcm->pm_domain_id = out_args.args[0]; + of_node_put(out_args.np); + tcm_pd_idx++; + + /* + * In lockstep mode, we only need second core's power domain + * ids. Other information from second core isn't needed so + * ignore it. This forms table as zynqmp_tcm_banks_lockstep + */ + if (j >= tcm_reg_per_r5) + continue; + + /* get tcm address without translation */ + ret = of_property_read_reg(np, j, &abs_addr, &size); + if (ret) { + of_node_put(np); + dev_err(dev, "failed to get reg property\n"); + return ret; + } + + /* + * remote processor can address only 32 bits + * so convert 64-bits into 32-bits. This will discard + * any unwanted upper 32-bits. + */ + tcm->da = (u32)abs_addr; + tcm->size = (u32)size; + + cpdev = to_platform_device(dev); + res = platform_get_resource(cpdev, IORESOURCE_MEM, j); + if (!res) { + of_node_put(np); + dev_err(dev, "failed to get tcm resource\n"); + return -EINVAL; + } + + tcm->addr = (u32)res->start; + tcm->bank_name = (char *)res->name; + res = devm_request_mem_region(dev, tcm->addr, tcm->size, + tcm->bank_name); + if (!res) { + dev_err(dev, "failed to request tcm resource\n"); + of_node_put(np); + return -EINVAL; + } + } + } + + of_node_put(np); + return 0; +} + /** * zynqmp_r5_get_tcm_node() * Ideally this function should parse tcm node and store information @@ -956,10 +1089,19 @@ static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster, struct zynqmp_r5_core *r5_core; int ret, i; - ret = zynqmp_r5_get_tcm_node(cluster); - if (ret < 0) { - dev_err(dev, "can't get tcm node, err %d\n", ret); - return ret; + r5_core = cluster->r5_cores[0]; + if (of_find_property(r5_core->np, "reg", NULL)) { + ret = zynqmp_r5_get_tcm_node_from_dt(cluster); + if (ret) { + dev_err(dev, "can't get tcm node from dt, err %d\n", ret); + return ret; + } + } else { + ret = zynqmp_r5_get_tcm_node(cluster); + if (ret < 0) { + dev_err(dev, "can't get tcm node, err %d\n", ret); + return ret; + } } for (i = 0; i < cluster->core_count; i++) {
ZynqMP TCM information is fixed in driver. Now ZynqMP TCM information is available in device-tree. Parse TCM information in driver as per new bindings. Signed-off-by: Tanmay Shah <tanmay.shah@amd.com> --- Changes in v8: - parse power-domains property from device-tree and use EEMI calls to power on/off TCM instead of using pm domains framework - Remove checking of pm_domain_id validation to power on/off tcm - Remove spurious change Changes in v7: - move checking of pm_domain_id from previous patch - fix mem_bank_data memory allocation drivers/remoteproc/xlnx_r5_remoteproc.c | 154 +++++++++++++++++++++++- 1 file changed, 148 insertions(+), 6 deletions(-)