Hi David, On 7/10/19 10:10 PM, David Lechner wrote: > On 7/7/19 10:52 PM, Suman Anna wrote: >> The PRUSS INTC receives a number of system input interrupt source events >> and supports individual control configuration and hardware >> prioritization. >> These input events can be mapped to some output host interrupts through 2 >> levels of many-to-one mapping i.e. events to channel mapping and channels >> to host interrupts. >> >> This mapping information is provided through the PRU firmware that is >> loaded onto a PRU core/s or through the device tree node of the PRU > Thanks for the thorough review and alternate solutions/suggestions. > What will the device tree bindings for this look like? They would be as in the below patch you already figured. > > Looking back at Rob's comment on the initial series [1], I still think > that increasing the #interrupt-cells sounds like a reasonable solution. > > [1]: https://patchwork.kernel.org/patch/10697705/#22375155 So, there are couple of reasons why I did not use an extended #interrupt-cells: 1. There is only one irq descriptor associated with each event, and the usage of events is typically per application. And the descriptor mapping is done once. We can have two different applications use the same event with different mappings. So we want this programming done at application's usage of PRU (so done when a consumer driver acquires a PRU processor(s) which are treated as an exclusive resource). All the different application properties that you saw in [1] are configured at the time of acquiring a PRU and reset when they release a PRU. 2. The configuration is performed by Linux for all host interrupts and channels, and this was primarily done to save the very limited IRAM space for those needed by the PRUs. From firmware's point of view, this was offloaded to the ARM OS driver/infrastructure, but in general it is a design by contract between a PRU client driver and its firmware. Also, the DT binding semantics using interrupts property and request_irq() typically limits these to interrupts only being requested by MPU, and so will leave out those needed by PRUs. > > > >> application. The mapping is configured by the PRU remoteproc driver, and >> is setup before the PRU core is started and cleaned up after the PRU core >> is stopped. This event mapping configuration logic is optimized to >> program >> the Channel Map Registers (CMRx) and Host-Interrupt Map Registers (HMRx) >> only when a new program is being loaded/started and simply disables the >> same events and interrupt channels without zeroing out the corresponding >> map registers when stopping a PRU. >> >> Add two helper functions: pruss_intc_configure() & >> pruss_intc_unconfigure() >> that the PRU remoteproc driver can use to configure the PRUSS INTC. >> >> Signed-off-by: Suman Anna <s-anna@xxxxxx> >> Signed-off-by: Andrew F. Davis <afd@xxxxxx> >> Signed-off-by: Roger Quadros <rogerq@xxxxxx> >> --- >> drivers/irqchip/irq-pruss-intc.c | 258 ++++++++++++++++++++++++- >> include/linux/irqchip/irq-pruss-intc.h | 33 ++++ >> 2 files changed, 289 insertions(+), 2 deletions(-) >> create mode 100644 include/linux/irqchip/irq-pruss-intc.h >> >> diff --git a/drivers/irqchip/irq-pruss-intc.c >> b/drivers/irqchip/irq-pruss-intc.c >> index 142d01b434e0..8118c2a2ac43 100644 >> --- a/drivers/irqchip/irq-pruss-intc.c >> +++ b/drivers/irqchip/irq-pruss-intc.c >> @@ -9,6 +9,7 @@ >> #include <linux/irq.h> >> #include <linux/irqchip/chained_irq.h> >> +#include <linux/irqchip/irq-pruss-intc.h> >> #include <linux/irqdomain.h> >> #include <linux/module.h> >> #include <linux/of_device.h> >> @@ -24,8 +25,8 @@ >> /* minimum starting host interrupt number for MPU */ >> #define MIN_PRU_HOST_INT 2 >> -/* maximum number of system events */ >> -#define MAX_PRU_SYS_EVENTS 64 >> +/* maximum number of host interrupts */ >> +#define MAX_PRU_HOST_INT 10 >> /* PRU_ICSS_INTC registers */ >> #define PRU_INTC_REVID 0x0000 >> @@ -57,15 +58,29 @@ >> #define PRU_INTC_HINLR(x) (0x1100 + (x) * 4) >> #define PRU_INTC_HIER 0x1500 >> +/* CMR register bit-field macros */ >> +#define CMR_EVT_MAP_MASK 0xf >> +#define CMR_EVT_MAP_BITS 8 >> +#define CMR_EVT_PER_REG 4 >> + >> +/* HMR register bit-field macros */ >> +#define HMR_CH_MAP_MASK 0xf >> +#define HMR_CH_MAP_BITS 8 >> +#define HMR_CH_PER_REG 4 >> + >> /* HIPIR register bit-fields */ >> #define INTC_HIPIR_NONE_HINT 0x80000000 >> +/* use -1 to mark unassigned events and channels */ >> +#define FREE -1 > > It could be helpful to have this macro in the public header. Yes, I can rename it and move it, and I can reuse it in the parsing logic within the PRU remoteproc driver as well. > >> + >> /** >> * struct pruss_intc - PRUSS interrupt controller structure >> * @irqs: kernel irq numbers corresponding to PRUSS host interrupts >> * @base: base virtual address of INTC register space >> * @irqchip: irq chip for this interrupt controller >> * @domain: irq domain for this interrupt controller >> + * @config_map: stored INTC configuration mapping data >> * @lock: mutex to serialize access to INTC >> * @host_mask: indicate which HOST IRQs are enabled >> * @shared_intr: bit-map denoting if the MPU host interrupt is shared >> @@ -76,6 +91,7 @@ struct pruss_intc { >> void __iomem *base; >> struct irq_chip *irqchip; >> struct irq_domain *domain; >> + struct pruss_intc_config config_map; >> struct mutex lock; /* PRUSS INTC lock */ >> u32 host_mask; >> u16 shared_intr; >> @@ -107,6 +123,238 @@ static int pruss_intc_check_write(struct >> pruss_intc *intc, unsigned int reg, >> return 0; >> } >> +static struct pruss_intc *to_pruss_intc(struct device *pru_dev) >> +{ >> + struct device_node *np; >> + struct platform_device *pdev; >> + struct device *pruss_dev = pru_dev->parent; >> + struct pruss_intc *intc = ERR_PTR(-ENODEV); >> + >> + np = of_get_child_by_name(pruss_dev->of_node, >> "interrupt-controller"); >> + if (!np) { >> + dev_err(pruss_dev, "pruss does not have an >> interrupt-controller node\n"); >> + return intc; >> + } >> + >> + pdev = of_find_device_by_node(np); >> + if (!pdev) { >> + dev_err(pruss_dev, "no associated platform device\n"); >> + goto out; >> + } >> + >> + intc = platform_get_drvdata(pdev); >> + if (!intc) { >> + dev_err(pruss_dev, "pruss intc device probe failed?\n"); >> + intc = ERR_PTR(-EINVAL); >> + } >> + >> +out: >> + of_node_put(np); >> + return intc; >> +} >> + >> +/** >> + * pruss_intc_configure() - configure the PRUSS INTC >> + * @dev: pru device pointer >> + * @intc_config: PRU core-specific INTC configuration >> + * >> + * Configures the PRUSS INTC with the provided configuration from >> + * a PRU core. Any existing event to channel mappings or channel to >> + * host interrupt mappings are checked to make sure there are no >> + * conflicting configuration between both the PRU cores. The function >> + * is intended to be used only by the PRU remoteproc driver. >> + * >> + * Returns 0 on success, or a suitable error code otherwise >> + */ >> +int pruss_intc_configure(struct device *dev, > > It seems like this would be easier to use if it took an IRQ number > or struct irq_data * as a parameter instead of struct device *. My > line of thinking is that callers of this function will already be > calling some variant of request_irq() so they will already have > this info. It would cut out the pointer acrobatics in to_pruss_intc. These API are actually not seen by PRU client drivers, but is only limited to the PRU remoteproc driver. The INTC configuration is managed per PRU core and in sync with the life-cycle of the PRU load/start and stop. As I mentioned above, we need to manage the configuration for events generating interrupts to non Linux ARM host as well. > > >> + struct pruss_intc_config *intc_config) >> +{ >> + struct pruss_intc *intc; >> + int i, idx, ret; >> + s8 ch, host; >> + u64 sysevt_mask = 0; >> + u32 ch_mask = 0; >> + u32 host_mask = 0; >> + u32 val; >> + >> + intc = to_pruss_intc(dev); >> + if (IS_ERR(intc)) >> + return PTR_ERR(intc); >> + >> + mutex_lock(&intc->lock); >> + >> + /* >> + * configure channel map registers - each register holds map info >> + * for 4 events, with each event occupying the lower nibble in >> + * a register byte address in little-endian fashion >> + */ >> + for (i = 0; i < ARRAY_SIZE(intc_config->sysev_to_ch); i++) { >> + ch = intc_config->sysev_to_ch[i]; >> + if (ch < 0) >> + continue; >> + >> + /* check if sysevent already assigned */ >> + if (intc->config_map.sysev_to_ch[i] != FREE) { >> + dev_err(dev, "event %d (req. channel %d) already assigned >> to channel %d\n", >> + i, ch, intc->config_map.sysev_to_ch[i]); >> + ret = -EEXIST; >> + goto unlock; > > If we fail here, shouldn't we unwind any previous mappings made? > Otherwise, if we try to map the same event again, it will show as > in use, even though it is not in use. Yeah, I will fix up the unwind logic. I intended for the callers to invoke the unconfigure upon failures, but even that has some unneeded operations, so it is better to unwind the operations here for a cleaner style. > >> + } >> + >> + intc->config_map.sysev_to_ch[i] = ch; >> + >> + idx = i / CMR_EVT_PER_REG; >> + val = pruss_intc_read_reg(intc, PRU_INTC_CMR(idx)); >> + val &= ~(CMR_EVT_MAP_MASK << >> + ((i % CMR_EVT_PER_REG) * CMR_EVT_MAP_BITS)); >> + val |= ch << ((i % CMR_EVT_PER_REG) * CMR_EVT_MAP_BITS); >> + pruss_intc_write_reg(intc, PRU_INTC_CMR(idx), val); >> + sysevt_mask |= BIT_ULL(i); >> + ch_mask |= BIT(ch); >> + >> + dev_dbg(dev, "SYSEV%d -> CH%d (CMR%d 0x%08x)\n", i, ch, idx, >> + pruss_intc_read_reg(intc, PRU_INTC_CMR(idx))); >> + } >> + >> + /* >> + * set host map registers - each register holds map info for >> + * 4 channels, with each channel occupying the lower nibble in >> + * a register byte address in little-endian fashion >> + */ >> + for (i = 0; i < ARRAY_SIZE(intc_config->ch_to_host); i++) { >> + host = intc_config->ch_to_host[i]; >> + if (host < 0) >> + continue; >> + >> + /* check if channel already assigned */ >> + if (intc->config_map.ch_to_host[i] != FREE) { >> + dev_err(dev, "channel %d (req. intr_no %d) already >> assigned to intr_no %d\n", >> + i, host, intc->config_map.ch_to_host[i]); >> + ret = -EEXIST; >> + goto unlock; > > Same comment about unwinding here and below. Yep, will fix this up as well in the next version. > >> + } >> + >> + /* check if host intr is already in use by other PRU */ > > It seems like there would be use cases where someone might want to map > multiple PRU system events, and therefore multiple channels, to a single > host interrupt. Yes, that is in general supported but for a given PRU. The idea here was to partition the host events separately between two PRUs and this is done to simplify the life-cycle per host event and their mappings between two different PRUs potentially running two different unrelated co-operative applications. > >> + if (intc->host_mask & (1U << host)) { >> + dev_err(dev, "%s: host intr %d already in use\n", >> + __func__, host); >> + ret = -EEXIST; >> + goto unlock; >> + } >> + > > --snip-- > >> diff --git a/include/linux/irqchip/irq-pruss-intc.h >> b/include/linux/irqchip/irq-pruss-intc.h >> new file mode 100644 >> index 000000000000..f1f1bb150100 >> --- /dev/null >> +++ b/include/linux/irqchip/irq-pruss-intc.h >> @@ -0,0 +1,33 @@ >> +/* SPDX-License-Identifier: GPL-2.0 */ >> +/* >> + * PRU-ICSS sub-system private interfaces >> + * >> + * Copyright (C) 2019 Texas Instruments Incorporated - >> http://www.ti.com/ >> + * Suman Anna <s-anna@xxxxxx> >> + */ >> + >> +#ifndef __LINUX_IRQ_PRUSS_INTC_H >> +#define __LINUX_IRQ_PRUSS_INTC_H >> + >> +/* maximum number of system events */ >> +#define MAX_PRU_SYS_EVENTS 64 >> + >> +/* maximum number of interrupt channels */ >> +#define MAX_PRU_CHANNELS 10 >> + >> +/** >> + * struct pruss_intc_config - INTC configuration info >> + * @sysev_to_ch: system events to channel mapping information >> + * @ch_to_host: interrupt channel to host interrupt information >> + */ >> +struct pruss_intc_config { >> + s8 sysev_to_ch[MAX_PRU_SYS_EVENTS]; >> + s8 ch_to_host[MAX_PRU_CHANNELS]; >> +}; >> + >> +int pruss_intc_configure(struct device *dev, >> + struct pruss_intc_config *intc_config); >> +int pruss_intc_unconfigure(struct device *dev, >> + struct pruss_intc_config *intc_config); >> + >> +#endif /* __LINUX_IRQ_PRUSS_INTC_H */ >> > > FYI, on AM18xx, events 0 to 31 can be muxed via CFGCHIP3[3].PRUSSEVTSEL > so an additional bit of information will be needed in this struct for > the mux selection. I don't see a probably with adding that later though. Yeah, there are different input pinmux'ing options controlling different number of input events on different SoCs. On AM18xx it is a SoC-level CHIPCFG register, and on other SoCs, it is a PRUSS CFG register (Standard mode vs MII mode) both of which are registers outside of the INTC module. I see these again as an application-level configuration, and this is what the last bullet item in the feature list in my cover-letter is about. I did think about adding a separate property to INTC node to configure a default value at INTC probe time, and then allow it to be overwritten as per a PRU application need. The latter is going to be needed anyway, so I dropped the idea of a default configuration, and leave it at POR values. regards Suman
