| Message ID | 20171006155136.4682-1-srinivas.kandagatla@linaro.org |
|---|---|
| Headers | show |
| Series | Introduce framework for SLIMbus device drivers | expand |
Hi, some trivial comments below. On Fri, Oct 06, 2017 at 05:51:33PM +0200, srinivas.kandagatla@linaro.org wrote: > From: Sagar Dharia <sdharia@codeaurora.org> > > Per slimbus specification, a reconfiguration sequence known as > 'clock pause' needs to be broadcast over the bus while entering low- > power mode. Clock-pause is initiated by the controller driver. > To exit clock-pause, controller typically wakes up the framer device. > Since wakeup precedure is controller-specific, framework calls it via > controller's function pointer to invoke it. > > Signed-off-by: Sagar Dharia <sdharia@codeaurora.org> > Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> > --- [...] > @@ -429,6 +444,14 @@ void slim_return_tx(struct slim_controller *ctrl, int err) > cur.cb(cur.ctx, err); > > up(&ctrl->tx_sem); > + if (!cur.clk_pause && (!cur.need_tid || err)) { > + /** This isn't really a kerneldoc comment. > + * remove runtime-pm vote if this was TX only, or > + * if there was error during this transaction > + */ > + pm_runtime_mark_last_busy(ctrl->dev.parent); > + pm_runtime_put_autosuspend(ctrl->dev.parent); > + } > } > EXPORT_SYMBOL_GPL(slim_return_tx); > [...] > +/** > + * slim_ctrl_clk_pause: Called by slimbus controller to enter/exit 'clock pause' > + * Slimbus specification needs this sequence to turn-off clocks for the bus. > + * The sequence involves sending 3 broadcast messages (reconfiguration > + * sequence) to inform all devices on the bus. > + * To exit clock-pause, controller typically wakes up active framer device. > + * @ctrl: controller requesting bus to be paused or woken up > + * @wakeup: Wakeup this controller from clock pause. > + * @restart: Restart time value per spec used for clock pause. This value > + * isn't used when controller is to be woken up. > + * This API executes clock pause reconfiguration sequence if wakeup is false. > + * If wakeup is true, controller's wakeup is called. > + * For entering clock-pause, -EBUSY is returned if a message txn in pending. > + */ > +int slim_ctrl_clk_pause(struct slim_controller *ctrl, bool wakeup, u8 restart) > +{ > + int i, ret = 0; > + unsigned long flags; > + struct slim_sched *sched = &ctrl->sched; > + struct slim_val_inf msg = {0, 0, NULL, NULL, NULL, NULL}; > + > + DEFINE_SLIM_BCAST_TXN(txn, SLIM_MSG_MC_BEGIN_RECONFIGURATION, > + 3, SLIM_LA_MANAGER, &msg); > + > + if (wakeup == false && restart > SLIM_CLK_UNSPECIFIED) > + return -EINVAL; > + > + mutex_lock(&sched->m_reconf); > + if (wakeup) { > + if (sched->clk_state == SLIM_CLK_ACTIVE) { > + mutex_unlock(&sched->m_reconf); > + return 0; > + } > + > + /** ditto > + * Fine-tune calculation based on clock gear, > + * message-bandwidth after bandwidth management > + */ > + ret = wait_for_completion_timeout(&sched->pause_comp, > + msecs_to_jiffies(100)); > + if (!ret) { > + mutex_unlock(&sched->m_reconf); > + pr_err("Previous clock pause did not finish"); > + return -ETIMEDOUT; > + } > + ret = 0; > + > + /** ditto > + * Slimbus framework will call controller wakeup > + * Controller should make sure that it sets active framer > + * out of clock pause > + */ > + if (sched->clk_state == SLIM_CLK_PAUSED && ctrl->wakeup) > + ret = ctrl->wakeup(ctrl); > + if (!ret) > + sched->clk_state = SLIM_CLK_ACTIVE; > + mutex_unlock(&sched->m_reconf); > + > + return ret; > + } [...] Thanks, Jonathan Neuschäfer
Hi, some more trivial comments below. On Fri, Oct 06, 2017 at 05:51:34PM +0200, srinivas.kandagatla@linaro.org wrote: > From: Sagar Dharia <sdharia@codeaurora.org> > > Slimbus HW mandates that clock-pause sequence has to be executed > before disabling relevant interface and core clocks. > Runtime-PM's autosuspend feature is used here to enter/exit low > power mode for Qualcomm's Slimbus controller. Autosuspend feature > enables driver to avoid changing power-modes too frequently since > entering clock-pause is an expensive sequence > > Signed-off-by: Sagar Dharia <sdharia@codeaurora.org> > Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> > --- [...] > +static int msm_clk_pause_wakeup(struct slim_controller *ctrl) > +{ > + struct msm_slim_ctrl *dev = slim_get_ctrldata(ctrl); > + > + clk_prepare_enable(dev->hclk); > + clk_prepare_enable(dev->rclk); > + enable_irq(dev->irq); > + > + writel_relaxed(1, dev->base + FRM_WAKEUP); > + /* Make sure framer wakeup write goes through before ISR fires */ > + mb(); > + /** This isn't really a kerneldoc comment. > + * HW Workaround: Currently, slave is reporting lost-sync messages > + * after slimbus comes out of clock pause. > + * Transaction with slave fail before slave reports that message > + * Give some time for that report to come > + * Slimbus wakes up in clock gear 10 at 24.576MHz. With each superframe > + * being 250 usecs, we wait for 5-10 superframes here to ensure > + * we get the message > + */ > + usleep_range(1250, 2500); > + return 0; > +} [...] > +#ifdef CONFIG_PM_SLEEP > +static int msm_slim_suspend(struct device *dev) > +{ > + int ret = 0; > + > + if (!pm_runtime_enabled(dev) || > + (!pm_runtime_suspended(dev))) { > + dev_dbg(dev, "system suspend"); > + ret = msm_slim_runtime_suspend(dev); > + } > + if (ret == -EISCONN) { > + /** ditto. Also, it looks misindented. > + * If the clock pause failed due to active channels, there is > + * a possibility that some audio stream is active during suspend. > + * (e.g. modem usecase during suspend) > + * We dont want to return suspend failure in that case so that > + * display and relevant components can still go to suspend. > + * If there is some other error, then it should prevent > + * system level suspend > + */ > + ret = 0; > + } > + return ret; > +} Thanks, Jonathan Neuschäfer
Thanks for the review comments On 07/10/17 09:22, Jonathan Neuschäfer wrote: > Hi, some more trivial comments below. > > On Fri, Oct 06, 2017 at 05:51:34PM +0200, srinivas.kandagatla@linaro.org wrote: >> From: Sagar Dharia <sdharia@codeaurora.org> >> >> Slimbus HW mandates that clock-pause sequence has to be executed >> before disabling relevant interface and core clocks. >> Runtime-PM's autosuspend feature is used here to enter/exit low >> power mode for Qualcomm's Slimbus controller. Autosuspend feature >> enables driver to avoid changing power-modes too frequently since >> entering clock-pause is an expensive sequence >> >> Signed-off-by: Sagar Dharia <sdharia@codeaurora.org> >> Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> >> --- > [...] >> +static int msm_clk_pause_wakeup(struct slim_controller *ctrl) >> +{ >> + struct msm_slim_ctrl *dev = slim_get_ctrldata(ctrl); >> + >> + clk_prepare_enable(dev->hclk); >> + clk_prepare_enable(dev->rclk); >> + enable_irq(dev->irq); >> + >> + writel_relaxed(1, dev->base + FRM_WAKEUP); >> + /* Make sure framer wakeup write goes through before ISR fires */ >> + mb(); >> + /** > > This isn't really a kerneldoc comment. Yep I agree will fix all such instances. > >> + * HW Workaround: Currently, slave is reporting lost-sync messages >> + * after slimbus comes out of clock pause. >> + * Transaction with slave fail before slave reports that message >> + * Give some time for that report to come >> + * Slimbus wakes up in clock gear 10 at 24.576MHz. With each superframe >> + * being 250 usecs, we wait for 5-10 superframes here to ensure >> + * we get the message >> + */ >> + usleep_range(1250, 2500); >> + return 0; >> +} > [...] >> +#ifdef CONFIG_PM_SLEEP >> +static int msm_slim_suspend(struct device *dev) >> +{ >> + int ret = 0; >> + >> + if (!pm_runtime_enabled(dev) || >> + (!pm_runtime_suspended(dev))) { >> + dev_dbg(dev, "system suspend"); >> + ret = msm_slim_runtime_suspend(dev); >> + } >> + if (ret == -EISCONN) { >> + /** > > ditto. > Also, it looks misindented. Yep. will fix this too. > >> + * If the clock pause failed due to active channels, there is >> + * a possibility that some audio stream is active during suspend. >> + * (e.g. modem usecase during suspend) >> + * We dont want to return suspend failure in that case so that >> + * display and relevant components can still go to suspend. >> + * If there is some other error, then it should prevent >> + * system level suspend >> + */ >> + ret = 0; >> + } >> + return ret; >> +} > > > Thanks, > Jonathan Neuschäfer > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Oct 06, 2017 at 05:51:30PM +0200, srinivas.kandagatla@linaro.org wrote: > From: Sagar Dharia <sdharia@codeaurora.org> > > SLIMbus (Serial Low Power Interchip Media Bus) is a specification > developed by MIPI (Mobile Industry Processor Interface) alliance. > SLIMbus is a 2-wire implementation, which is used to communicate with > peripheral components like audio-codec. > SLIMbus uses Time-Division-Multiplexing to accommodate multiple data > channels, and control channel. Control channel has messages to do > device-enumeration, messages to send/receive control-data to/from > slimbus devices, messages for port/channel management, and messages to > do bandwidth allocation. > The framework supports multiple instances of the bus (1 controller per > bus), and multiple slave devices per controller. > > This patch does device enumeration, logical address assignment, > informing device when the device reports present/absent etc. > Reporting present may need the driver to do the needful (e.g. turning > on voltage regulators powering the device). Additionally device is > probed when it reports present if that device doesn't need any such > steps mentioned above. > > Signed-off-by: Sagar Dharia <sdharia@codeaurora.org> > Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> > --- > Documentation/devicetree/bindings/slimbus/bus.txt | 57 ++ > Documentation/slimbus/summary | 109 ++++ > drivers/Kconfig | 2 + > drivers/Makefile | 1 + > drivers/slimbus/Kconfig | 11 + > drivers/slimbus/Makefile | 5 + > drivers/slimbus/slim-core.c | 695 ++++++++++++++++++++++ > include/linux/mod_devicetable.h | 13 + > include/linux/slimbus.h | 299 ++++++++++ > 9 files changed, 1192 insertions(+) > create mode 100644 Documentation/devicetree/bindings/slimbus/bus.txt > create mode 100644 Documentation/slimbus/summary > create mode 100644 drivers/slimbus/Kconfig > create mode 100644 drivers/slimbus/Makefile > create mode 100644 drivers/slimbus/slim-core.c > create mode 100644 include/linux/slimbus.h > > diff --git a/Documentation/devicetree/bindings/slimbus/bus.txt b/Documentation/devicetree/bindings/slimbus/bus.txt > new file mode 100644 > index 0000000..cb658bb > --- /dev/null > +++ b/Documentation/devicetree/bindings/slimbus/bus.txt > @@ -0,0 +1,57 @@ > +SLIM(Serial Low Power Interchip Media Bus) bus > + > +SLIMbus is a 2-wire bus, and is used to communicate with peripheral > +components like audio-codec. > + > +Controller is a normal device using binding for whatever bus it is > +on (e.g. platform bus). > +Required property for SLIMbus controller node: > +- compatible - name of SLIMbus controller following generic names > + recommended practice. > +- #address-cells - should be 2 > +- #size-cells - should be 0 > + > +No other properties are required in the SLIMbus controller bus node. > + > +Child nodes: > +Every SLIMbus controller node can contain zero or more child nodes > +representing slave devices on the bus. Every SLIMbus slave device is > +uniquely determined by the enumeration address containing 4 fields: > +Manufacturer ID, Product code, Device index, and Instance value for > +the device. > +If child node is not present and it is instantiated after device > +discovery (slave device reporting itself present). > + > +In some cases it may be necessary to describe non-probeable device > +details such as non-standard ways of powering up a device. In > +such cases, child nodes for those devices will be present as > +slaves of the slimbus-controller, as detailed below. > + > +Required property for SLIMbus child node if it is present: > +- reg - Is Duplex (Device index, Instance ID) from Enumeration > + Address. > + Device Index Uniquely identifies multiple Devices within > + a single Component. > + Instance ID Is for the cases where multiple Devices of the > + same type or Class are attached to the bus. > + > +- compatible -"slimMID,PID". The textual representation of Manufacturer ID, > + Product Code, shall be in lower case hexadecimal with leading > + zeroes suppressed This does sort of make sense but kinda makes the code a bit ugly parsing the MID and PID. Some parts will support SLIMBus and also other control interfaces, which means they would need to add an additional compatible string just for SLIMBus. It also breaks the normal conventions of vendor,part and finally it makes the compatible strings really unreadable which will be a bit annoying when looking at DTs. I think the MID and PID should just be included in the reg field and just leave this as a standard compatible. > +/** This doesn't appear to be a kernel doc comment, so only /*. > + * Report callbacks(device_up, device_down) are implemented by slimbus-devices. > + * The calls are scheduled into a workqueue to avoid holding up controller > + * initialization/tear-down. > + */ > +static void schedule_slim_report(struct slim_controller *ctrl, > + struct slim_device *sb, bool report) > +{ > + struct sb_report_wd *sbw; > + > + dev_dbg(&ctrl->dev, "report:%d for slave:%s\n", report, sb->name); > + > + sbw = kmalloc(sizeof(*sbw), GFP_KERNEL); > + if (!sbw) > + return; > + > + INIT_WORK(&sbw->wd, slim_report); > + sbw->sbdev = sb; > + sbw->report = report; > + if (!queue_work(ctrl->wq, &sbw->wd)) { > + dev_err(&ctrl->dev, "failed to queue report:%d slave:%s\n", > + report, sb->name); > + kfree(sbw); > + } > +} > + > +/** > + * slim_driver_register: Client driver registration with slimbus A - after the function name usually works better, I think kernel-doc doesn't support a colon after the function name. > + * @drv:Client driver to be associated with client-device. > + * @owner: owning module/driver > + * This API will register the client driver with the slimbus > + * It is called from the driver's module-init function. If you don't put a blank line after the arguments kernel doc will treat this as a run on for the description of owner rather than the long description for the function you intended it to be. > + > +/* Helper to get hex Manufacturer ID and Product id from compatible */ > +static unsigned long str2hex(unsigned char *str) > +{ > + int value = 0; > + > + while (*str) { > + char c = *str++; > + > + value = value << 4; > + if (c >= '0' && c <= '9') > + value |= (c - '0'); > + if (c >= 'a' && c <= 'f') > + value |= (c - 'a' + 10); > + > + } > + > + return value; > +} Isn't this just reimplementing kstrtoul? > + > +/* OF helpers for SLIMbus */ > +static void of_register_slim_devices(struct slim_controller *ctrl) > +{ > + struct device *dev = &ctrl->dev; > + struct device_node *node; > + > + if (!ctrl->dev.of_node) > + return; > + > + for_each_child_of_node(ctrl->dev.of_node, node) { > + struct slim_device *slim; > + const char *compat = NULL; > + char *p, *tok; > + int reg[2], ret; > + > + slim = kzalloc(sizeof(*slim), GFP_KERNEL); > + if (!slim) > + continue; > + > + slim->dev.of_node = of_node_get(node); > + > + compat = of_get_property(node, "compatible", NULL); > + if (!compat) > + continue; > + > + p = kasprintf(GFP_KERNEL, "%s", compat + strlen("slim")); > + > + tok = strsep(&p, ","); > + if (!tok) { > + dev_err(dev, "No valid Manufacturer ID found\n"); > + kfree(p); > + continue; > + } > + slim->e_addr.manf_id = str2hex(tok); > + > + tok = strsep(&p, ","); > + if (!tok) { > + dev_err(dev, "No valid Product ID found\n"); > + kfree(p); > + continue; > + } > + slim->e_addr.prod_code = str2hex(tok); > + kfree(p); > + > + ret = of_property_read_u32_array(node, "reg", reg, 2); > + if (ret) { > + dev_err(dev, "Device and Instance id not found:%d\n", > + ret); > + continue; > + } > + slim->e_addr.dev_index = reg[0]; > + slim->e_addr.instance = reg[1]; As I said above, this feels a bit complex compared to just reading the e_addr from reg. > + > + ret = slim_add_device(ctrl, slim); > + if (ret) > + dev_err(dev, "of_slim device register err:%d\n", ret); > + } > +} > + > +static int slim_boot_child(struct device *dev, void *unused) > +{ > + struct slim_driver *sbdrv; > + struct slim_device *sbdev = to_slim_device(dev); > + > + if (sbdev && sbdev->dev.driver) { > + sbdrv = to_slim_driver(sbdev->dev.driver); > + if (sbdrv->boot_device) > + sbdrv->boot_device(sbdev); > + } > + return 0; > +} > + > +static int slim_match_dev(struct device *dev, void *data) > +{ > + struct slim_eaddr *e_addr = data; > + struct slim_device *slim = to_slim_device(dev); > + > + return slim_eaddr_equal(&slim->e_addr, e_addr); > +} Would it make sense to move this down to above slim_query_device, that way all the related code is next to itself? slim_boot_child/slim_framer_booted and slim_match_dev/slim_query_device. > + > +/** > + * slim_framer_booted: This function is called by controller after the active > + * framer has booted (using Bus Reset sequence, or after it has shutdown and has > + * come back up). > + * @ctrl: Controller associated with this framer > + * Components, devices on the bus may be in undefined state, > + * and this function triggers their drivers to do the needful > + * to bring them back in Reset state so that they can acquire sync, report > + * present and be operational again. > + */ > +void slim_framer_booted(struct slim_controller *ctrl) > +{ > + if (!ctrl) > + return; > + > + device_for_each_child(&ctrl->dev, NULL, slim_boot_child); > +} > +EXPORT_SYMBOL_GPL(slim_framer_booted); > + > +/** > + * slim_query_device: Query and get handle to a device. > + * @ctrl: Controller on which this device will be added/queried > + * @e_addr: Enumeration address of the device to be queried > + * Returns pointer to a device if it has already reported. Creates a new > + * device and returns pointer to it if the device has not yet enumerated. > + */ > +struct slim_device *slim_query_device(struct slim_controller *ctrl, > + struct slim_eaddr *e_addr) > +{ > + struct device *dev; > + struct slim_device *slim = NULL; > + > + dev = device_find_child(&ctrl->dev, e_addr, slim_match_dev); > + if (dev) { > + slim = to_slim_device(dev); > + return slim; > + } > + > + slim = kzalloc(sizeof(struct slim_device), GFP_KERNEL); > + if (IS_ERR(slim)) > + return NULL; > + > + slim->e_addr = *e_addr; > + if (slim_add_device(ctrl, slim) != 0) { > + kfree(slim); > + return NULL; > + } > + return slim; > +} > +EXPORT_SYMBOL_GPL(slim_query_device); > + > +static int ctrl_getaddr_entry(struct slim_controller *ctrl, > + struct slim_eaddr *eaddr, u8 *entry) > +{ > + int i; > + > + for (i = 0; i < ctrl->num_dev; i++) { > + if (ctrl->addrt[i].valid && > + slim_eaddr_equal(&ctrl->addrt[i].eaddr, eaddr)) { > + *entry = i; > + return 0; > + } > + } > + return -ENXIO; > +} > + > +/** > + * slim_assign_laddr: Assign logical address to a device enumerated. > + * @ctrl: Controller with which device is enumerated. > + * @e_addr: Enumeration address of the device. > + * @laddr: Return logical address (if valid flag is false) > + * @valid: true if laddr holds a valid address that controller wants to > + * set for this enumeration address. Otherwise framework sets index into > + * address table as logical address. > + * Called by controller in response to REPORT_PRESENT. Framework will assign > + * a logical address to this enumeration address. > + * Function returns -EXFULL to indicate that all logical addresses are already > + * taken. > + */ > +int slim_assign_laddr(struct slim_controller *ctrl, struct slim_eaddr *e_addr, > + u8 *laddr, bool valid) I would be inclined to remove the valid parameter and just use get_laddr in here. Feels weird to have two mechanisms for specify the laddr and presumably if the controller wants to specify the laddr it will need to support get_laddr. Additionally, I would make get_laddr optional which feels more sensible, indeed the code appears otherwise implements with that assumption. > +{ > + int ret; > + u8 i = 0; > + bool exists = false; > + struct slim_device *slim; > + struct slim_addrt *temp; > + > + mutex_lock(&ctrl->m_ctrl); > + /* already assigned */ > + if (ctrl_getaddr_entry(ctrl, e_addr, &i) == 0) { > + *laddr = ctrl->addrt[i].laddr; > + exists = true; > + } else { > + if (ctrl->num_dev >= (SLIM_LA_MANAGER - 1)) { > + ret = -EXFULL; > + goto ret_assigned_laddr; > + } > + for (i = 0; i < ctrl->num_dev; i++) { > + if (ctrl->addrt[i].valid == false) > + break; > + } > + if (i == ctrl->num_dev) { > + temp = krealloc(ctrl->addrt, > + (ctrl->num_dev + 1) * > + sizeof(struct slim_addrt), > + GFP_KERNEL); > + if (!temp) { > + ret = -ENOMEM; > + goto ret_assigned_laddr; > + } > + ctrl->addrt = temp; > + ctrl->num_dev++; > + } > + ctrl->addrt[i].eaddr = *e_addr; > + ctrl->addrt[i].valid = true; > + > + /* Preferred address is index into table */ > + if (!valid) > + *laddr = i; > + } > + > + ret = ctrl->set_laddr(ctrl, &ctrl->addrt[i].eaddr, *laddr); > + if (ret) { > + ctrl->addrt[i].valid = false; > + goto ret_assigned_laddr; > + } > + ctrl->addrt[i].laddr = *laddr; > + > +ret_assigned_laddr: > + mutex_unlock(&ctrl->m_ctrl); > + if (exists || ret) > + return ret; > + > + dev_info(&ctrl->dev, "setting slimbus l-addr:%x, ea:%x,%x,%x,%x\n", > + *laddr, e_addr->manf_id, e_addr->prod_code, > + e_addr->dev_index, e_addr->instance); > + > + /** > + * Add this device to list of devices on this controller if it's > + * not already present > + */ > + slim = slim_query_device(ctrl, e_addr); > + if (!slim) { > + ret = -ENODEV; > + } else { > + struct slim_driver *sbdrv; > + > + slim->laddr = *laddr; > + mutex_lock(&slim->report_lock); > + slim->reported = true; > + if (slim->dev.driver) { > + sbdrv = to_slim_driver(slim->dev.driver); > + if (sbdrv->device_up) > + schedule_slim_report(ctrl, slim, true); > + } > + mutex_unlock(&slim->report_lock); > + } > + return ret; > +} > +EXPORT_SYMBOL_GPL(slim_assign_laddr); > + > +/** > + * slim_get_logical_addr: Return the logical address of a slimbus device. > + * @sb: client handle requesting the address. > + * @e_addr: Enumeration address of the device. > + * @laddr: output buffer to store the address > + * context: can sleep > + * -EINVAL is returned in case of invalid parameters, and -ENXIO is returned if > + * the device with this enumeration address is not found. > + */ > +int slim_get_logical_addr(struct slim_device *sb, struct slim_eaddr *e_addr, > + u8 *laddr) > +{ > + int ret; > + u8 entry; > + struct slim_controller *ctrl = sb->ctrl; > + > + if (!ctrl || !laddr || !e_addr) > + return -EINVAL; > + > + mutex_lock(&ctrl->m_ctrl); > + ret = ctrl_getaddr_entry(ctrl, e_addr, &entry); > + if (!ret) > + *laddr = ctrl->addrt[entry].laddr; > + mutex_unlock(&ctrl->m_ctrl); > + > + if (ret == -ENXIO && ctrl->get_laddr) { > + ret = ctrl->get_laddr(ctrl, e_addr, laddr); > + if (!ret) > + ret = slim_assign_laddr(ctrl, e_addr, laddr, true); > + } > + return ret; > +} > +EXPORT_SYMBOL_GPL(slim_get_logical_addr); I find the interface across these two functions (assign_laddr/get_logical_addr) a little odd. Since get_logical_addr calls assign_laddr if it couldn't find the laddr and then assign_laddr actually does another search for the laddr. Would it perhaps make sense to combine these into one function? Thanks, Charles -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Thanks for the review comments, On 10/10/17 13:19, Charles Keepax wrote: > On Fri, Oct 06, 2017 at 05:51:31PM +0200, srinivas.kandagatla@linaro.org wrote: >> From: Sagar Dharia <sdharia@codeaurora.org> >> >> Slimbus devices use value-element, and information elements to >> control device parameters (e.g. value element is used to represent >> gain for codec, information element is used to represent interrupt >> status for codec when codec interrupt fires). >> Messaging APIs are used to set/get these value and information >> elements. Slimbus specification uses 8-bit "transaction IDs" for >> messages where a read-value is anticipated. Framework uses a table >> of pointers to store those TIDs and responds back to the caller in >> O(1). >> Caller can opt to do synchronous, or asynchronous reads/writes. For >> asynchronous operations, the callback will be called from atomic >> context. >> TX and RX circular rings are used to allow queuing of multiple >> transfers per controller. Controller can choose size of these rings >> based of controller HW implementation. The buffers are coerently >> mapped so that controller can utilize DMA operations for the >> transactions without remapping every transaction buffer. >> Statically allocated rings help to improve performance by avoiding >> overhead of dynamically allocating transactions on need basis. >> >> Signed-off-by: Sagar Dharia <sdharia@codeaurora.org> >> Tested-by: Naveen Kaje <nkaje@codeaurora.org> >> Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> >> --- >> drivers/slimbus/Makefile | 2 +- >> drivers/slimbus/slim-core.c | 15 ++ >> drivers/slimbus/slim-messaging.c | 471 +++++++++++++++++++++++++++++++++++++++ >> include/linux/slimbus.h | 161 +++++++++++++ >> 4 files changed, 648 insertions(+), 1 deletion(-) >> create mode 100644 drivers/slimbus/slim-messaging.c >> >> +/** >> + * slim_processtxn: Process a slimbus-messaging transaction >> + * @ctrl: Controller handle >> + * @txn: Transaction to be sent over SLIMbus >> + * Called by controller to transmit messaging transactions not dealing with >> + * Interface/Value elements. (e.g. transmittting a message to assign logical >> + * address to a slave device >> + * Returns: >> + * -ETIMEDOUT: If transmission of this message timed out (e.g. due to bus lines >> + * not being clocked or driven by controller) >> + * -ENOTCONN: If the transmitted message was not ACKed by destination device. >> + */ >> +int slim_processtxn(struct slim_controller *ctrl, >> + struct slim_msg_txn *txn) > > Can all go on one line. Thats true, Will fix it in next version. > >> +{ >> + int ret, i = 0; >> + unsigned long flags; >> + u8 *buf; >> + bool async = false; >> + struct slim_cb_data cbd; >> + DECLARE_COMPLETION_ONSTACK(done); >> + bool need_tid = slim_tid_txn(txn->mt, txn->mc); >> + >> + if (!txn->msg->comp_cb) { >> + txn->msg->comp_cb = slim_sync_default_cb; >> + cbd.comp = &done; >> + txn->msg->ctx = &cbd; >> + } else { >> + async = true; >> + } >> + >> + buf = slim_get_tx(ctrl, txn, need_tid); >> + if (!buf) >> + return -ENOMEM; >> + >> + if (need_tid) { >> + spin_lock_irqsave(&ctrl->txn_lock, flags); >> + for (i = 0; i < ctrl->last_tid; i++) { >> + if (ctrl->tid_tbl[i] == NULL) >> + break; >> + } >> + if (i >= ctrl->last_tid) { >> + if (ctrl->last_tid == (SLIM_MAX_TIDS - 1)) { >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + slim_return_tx(ctrl, -ENOMEM); >> + return -ENOMEM; >> + } >> + ctrl->last_tid++; >> + } >> + ctrl->tid_tbl[i] = txn->msg; >> + txn->tid = i; >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + } >> + >> + ret = ctrl->xfer_msg(ctrl, txn, buf); >> + >> + if (!ret && !async) { /* sync transaction */ >> + /* Fine-tune calculation after bandwidth management */ >> + unsigned long ms = txn->rl + 100; >> + >> + ret = wait_for_completion_timeout(&done, >> + msecs_to_jiffies(ms)); >> + if (!ret) >> + slim_return_tx(ctrl, -ETIMEDOUT); >> + >> + ret = cbd.ret; >> + } >> + >> + if (ret && need_tid) { >> + spin_lock_irqsave(&ctrl->txn_lock, flags); >> + /* Invalidate the transaction */ >> + ctrl->tid_tbl[txn->tid] = NULL; >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + } >> + if (ret) >> + dev_err(&ctrl->dev, "Tx:MT:0x%x, MC:0x%x, LA:0x%x failed:%d\n", >> + txn->mt, txn->mc, txn->la, ret); >> + if (!async) { >> + txn->msg->comp_cb = NULL; >> + txn->msg->ctx = NULL; >> + } > > What is the intent of this if statement here? We set async > locally so this code only runs if we executed the else on the if > statement at the top. If its just clearing anything the user > might have put in these fields why not do it up there. Its clearing the temporary callback and context field when user wants to read/write on simbus synchronously. If async is false user should not put anything in comp_cb or ctx. comp_cb and ctx are only used when drivers are doing asynchronous read/writes on slimbus. Completion of those are indicated by comp_cb() with context. > >> + return ret; >> +} >> +EXPORT_SYMBOL_GPL(slim_processtxn); >> + >> +int slim_xfer_msg(struct slim_controller *ctrl, >> + struct slim_device *sbdev, struct slim_val_inf *msg, >> + u8 mc) >> +{ >> + DEFINE_SLIM_LDEST_TXN(txn_stack, mc, 6, sbdev->laddr, msg); >> + struct slim_msg_txn *txn = &txn_stack; >> + int ret; >> + u16 sl, cur; >> + >> + ret = slim_val_inf_sanity(ctrl, msg, mc); >> + if (ret) >> + return ret; >> + >> + sl = slim_slicesize(msg->num_bytes); >> + >> + dev_dbg(&ctrl->dev, "SB xfer msg:os:%x, len:%d, MC:%x, sl:%x\n", >> + msg->start_offset, msg->num_bytes, mc, sl); >> + >> + cur = slim_slicecodefromsize(sl); > > cur should probably be removed until it is needed. Yep. > >> + txn->ec = ((sl | (1 << 3)) | ((msg->start_offset & 0xFFF) << 4)); >> + >> + switch (mc) { >> + case SLIM_MSG_MC_REQUEST_CHANGE_VALUE: >> + case SLIM_MSG_MC_CHANGE_VALUE: >> + case SLIM_MSG_MC_REQUEST_CLEAR_INFORMATION: >> + case SLIM_MSG_MC_CLEAR_INFORMATION: >> + txn->rl += msg->num_bytes; >> + default: >> + break; >> + } >> + >> + if (slim_tid_txn(txn->mt, txn->mc)) >> + txn->rl++; >> + >> + return slim_processtxn(ctrl, txn); >> +} >> +EXPORT_SYMBOL_GPL(slim_xfer_msg); >> + >> +/* Message APIs Unicast message APIs used by slimbus slave drivers */ >> + >> +/* >> + * slim_request_val_element: request value element >> + * @sb: client handle requesting elemental message reads, writes. >> + * @msg: Input structure for start-offset, number of bytes to read. >> + * context: can sleep >> + * Returns: >> + * -EINVAL: Invalid parameters >> + * -ETIMEDOUT: If transmission of this message timed out (e.g. due to bus lines >> + * not being clocked or driven by controller) >> + * -ENOTCONN: If the transmitted message was not ACKed by destination device. >> + */ >> +int slim_request_val_element(struct slim_device *sb, >> + struct slim_val_inf *msg) >> +{ >> + struct slim_controller *ctrl = sb->ctrl; >> + >> + if (!ctrl) >> + return -EINVAL; > > You could put this check into slim_xfer_msg and save duplicating > it. Would also then cover cases that call that function directly, > also would let you make these functions either inlines or macros. I will give that a try in next version. > >> + >> + return slim_xfer_msg(ctrl, sb, msg, SLIM_MSG_MC_REQUEST_VALUE); >> +} >> +EXPORT_SYMBOL_GPL(slim_request_val_element); >> + >> +/* Functions to get/return TX, RX buffers for messaging. */ >> + >> +/** >> + * slim_get_rx: To get RX buffers for messaging. >> + * @ctrl: Controller handle >> + * These functions are called by controller to process the RX buffers. >> + * RX buffer is requested by controller when data is received from HW, but is >> + * not processed (e.g. 'report-present message was sent by HW in ISR and SW >> + * needs more time to process the buffer to assign Logical Address) >> + * RX buffer is returned back to the pool when associated RX action >> + * is taken (e.g. Received message is decoded and client's >> + * response buffer is filled in.) >> + */ >> +void *slim_get_rx(struct slim_controller *ctrl) >> +{ >> + unsigned long flags; >> + int idx; >> + >> + spin_lock_irqsave(&ctrl->rx.lock, flags); >> + if ((ctrl->rx.tail + 1) % ctrl->rx.n == ctrl->rx.head) { >> + spin_unlock_irqrestore(&ctrl->rx.lock, flags); >> + dev_err(&ctrl->dev, "RX QUEUE full!"); >> + return NULL; >> + } >> + idx = ctrl->rx.tail; >> + ctrl->rx.tail = (ctrl->rx.tail + 1) % ctrl->rx.n; >> + spin_unlock_irqrestore(&ctrl->rx.lock, flags); >> + >> + return ctrl->rx.base + (idx * ctrl->rx.sl_sz); >> +} >> +EXPORT_SYMBOL_GPL(slim_get_rx); >> + >> +int slim_return_rx(struct slim_controller *ctrl, void *buf) >> +{ >> + unsigned long flags; >> + >> + spin_lock_irqsave(&ctrl->rx.lock, flags); >> + if (ctrl->rx.tail == ctrl->rx.head) { >> + spin_unlock_irqrestore(&ctrl->rx.lock, flags); >> + return -ENODATA; >> + } >> + memcpy(buf, ctrl->rx.base + (ctrl->rx.head * ctrl->rx.sl_sz), >> + ctrl->rx.sl_sz); >> + ctrl->rx.head = (ctrl->rx.head + 1) % ctrl->rx.n; >> + spin_unlock_irqrestore(&ctrl->rx.lock, flags); >> + >> + return 0; >> +} >> +EXPORT_SYMBOL_GPL(slim_return_rx); > > I find the combination of get/return a bit odd, would get/put > maybe more idiomatic? Also the return could use some kernel doc. If that makes it more readable I can rename the functions as suggested, and I will also add kernel doc in next version. > >> + >> +void slim_return_tx(struct slim_controller *ctrl, int err) >> +{ >> + unsigned long flags; >> + int idx; >> + struct slim_pending cur; >> + >> + spin_lock_irqsave(&ctrl->tx.lock, flags); >> + idx = ctrl->tx.head; >> + ctrl->tx.head = (ctrl->tx.head + 1) % ctrl->tx.n; >> + cur = ctrl->pending_wr[idx]; >> + spin_unlock_irqrestore(&ctrl->tx.lock, flags); >> + >> + if (!cur.cb) >> + dev_err(&ctrl->dev, "NULL Transaction or completion"); >> + else >> + cur.cb(cur.ctx, err); >> + >> + up(&ctrl->tx_sem); >> +} >> +EXPORT_SYMBOL_GPL(slim_return_tx); >> + >> +/** >> + * slim_get_tx: To get TX buffers for messaging. >> + * @ctrl: Controller handle >> + * These functions are called by controller to process the TX buffers. >> + * TX buffer is requested by controller when it's filled-in and sent to the >> + * HW. When HW has finished processing this buffer, controller should return it >> + * back to the pool. >> + */ >> +void *slim_get_tx(struct slim_controller *ctrl, struct slim_msg_txn *txn, >> + bool need_tid) >> +{ >> + unsigned long flags; >> + int ret, idx; >> + >> + ret = down_interruptible(&ctrl->tx_sem); >> + if (ret < 0) { >> + dev_err(&ctrl->dev, "TX semaphore down returned:%d", ret); >> + return NULL; >> + } >> + spin_lock_irqsave(&ctrl->tx.lock, flags); >> + >> + if (((ctrl->tx.head + 1) % ctrl->tx.n) == ctrl->tx.tail) { >> + spin_unlock_irqrestore(&ctrl->tx.lock, flags); >> + dev_err(&ctrl->dev, "controller TX buf unavailable"); >> + up(&ctrl->tx_sem); >> + return NULL; >> + } >> + idx = ctrl->tx.tail; >> + ctrl->tx.tail = (ctrl->tx.tail + 1) % ctrl->tx.n; >> + ctrl->pending_wr[idx].cb = txn->msg->comp_cb; >> + ctrl->pending_wr[idx].ctx = txn->msg->ctx; >> + ctrl->pending_wr[idx].need_tid = need_tid; >> + spin_unlock_irqrestore(&ctrl->tx.lock, flags); >> + >> + return ctrl->tx.base + (idx * ctrl->tx.sl_sz); >> +} >> +EXPORT_SYMBOL_GPL(slim_get_tx); > > The rx calls seem ok that is basically the controller's job to > call those, but with these two calls it seems sometimes the > framework calls them sometimes the controller driver has to. Is > there anyway we can simplify that a bit? Or at least include some > documentation as to when the controller should call them. I will try to do add some details in the document in next version. > >> diff --git a/include/linux/slimbus.h b/include/linux/slimbus.h >> index b5064b6..080d86a 100644 >> --- a/include/linux/slimbus.h >> +++ b/include/linux/slimbus.h >> @@ -15,6 +15,7 @@ >> #include <linux/module.h> >> #include <linux/device.h> >> #include <linux/mutex.h> >> +#include <linux/semaphore.h> >> #include <linux/mod_devicetable.h> >> >> /** >> @@ -34,6 +35,9 @@ extern struct bus_type slimbus_type; >> #define SLIM_FRM_SLOTS_PER_SUPERFRAME 16 >> #define SLIM_GDE_SLOTS_PER_SUPERFRAME 2 >> >> +/* MAX in-flight transactions neededing transaction ID (8-bit, per spec) */ > > s/neededing/needing/ > Will fix this in next version. >> + >> +/* Frequently used message transaction structures */ >> +#define DEFINE_SLIM_LDEST_TXN(name, mc, rl, la, msg) \ >> + struct slim_msg_txn name = { rl, 0, mc, SLIM_MSG_DEST_LOGICALADDR, 0,\ >> + 0, la, msg, } >> + >> +#define DEFINE_SLIM_BCAST_TXN(name, mc, rl, la, msg) \ >> + struct slim_msg_txn name = { rl, 0, mc, SLIM_MSG_DEST_BROADCAST, 0,\ >> + 0, la, msg, } > > If the LA isn't used in broadcast messages wouldn't it be simpler > to set it to a fixed value in this macro? > Yep, if the destination type is broadcast we should not set la or ea in the header. may be set 0 here. >> + >> +#define DEFINE_SLIM_EDEST_TXN(name, mc, rl, la, msg) \ >> + struct slim_msg_txn name = { rl, 0, mc, SLIM_MSG_DEST_ENUMADDR, 0,\ >> + 0, la, msg, } >> + > > Also one final overall comment this commit contains a lot of two > and three letter abreviations that are not always clear. I would > probably suggest expanding a few of the less standard ones out to > make the code a little easier to follow. Will do that!! thanks srini > > Thanks, > Charles > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Oct 10, 2017 at 06:21:34PM +0100, Srinivas Kandagatla wrote: > On 10/10/17 17:49, Vinod Koul wrote: > >>>>+static int slim_device_probe(struct device *dev) > >>>>+{ > >>>>+ struct slim_device *sbdev; > >>>>+ struct slim_driver *sbdrv; > >>>>+ int status = 0; > >>>>+ > >>>>+ sbdev = to_slim_device(dev); > >>>>+ sbdrv = to_slim_driver(dev->driver); > >>>>+ > >>>>+ sbdev->driver = sbdrv; > >>>>+ > >>>>+ if (sbdrv->probe) > >>>>+ status = sbdrv->probe(sbdev); > >>>>+ > >>>>+ if (status) > >>>>+ sbdev->driver = NULL; > >>>>+ else if (sbdrv->device_up) > >>>>+ schedule_slim_report(sbdev->ctrl, sbdev, true); > >>> > >>>can you please explain what this is trying to do? > >> > >>It is scheduling a device_up() callback in workqueue for reporting > >>discovered device. > > > >any reason for that? Would the device not announce itself on the bus and > >then you can synchronously update the device. > You are correct, Device should announce itself in all cases. core should > only call this callback only when device is announced, it does not make > sense for this call in slim_device_probe(). Will remove it from here in next > version. Okay great. Btw do you need a notify being scheduled in those cases? I guess your controller would get an interrupt and you will handle that in bottom half and then cll this, so why not call in the bottom half? > >>>>+/** > >>>>+ * slim_register_controller: Controller bring-up and registration. > ... > >>>>+ > >>>>+ mutex_init(&ctrl->m_ctrl); > >>>>+ ret = device_register(&ctrl->dev); > >>> > >>>one more device_register?? Can you explain why > >>> > >> > >>This is a device for each controller. > > > >wont the controller have its own platform_device? > > Reason could be that slim_register controller can be called from any code > not just platform devices.. ah which cases would those be. I was expecting that you would have a platform_device as a slimbus controller which would call slim_register? -- ~Vinod -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 11/10/17 05:07, Vinod Koul wrote: > On Tue, Oct 10, 2017 at 06:21:34PM +0100, Srinivas Kandagatla wrote: >> On 10/10/17 17:49, Vinod Koul wrote: > >>>>>> +static int slim_device_probe(struct device *dev) >>>>>> +{ >>>>>> + struct slim_device *sbdev; >>>>>> + struct slim_driver *sbdrv; >>>>>> + int status = 0; >>>>>> + >>>>>> + sbdev = to_slim_device(dev); >>>>>> + sbdrv = to_slim_driver(dev->driver); >>>>>> + >>>>>> + sbdev->driver = sbdrv; >>>>>> + >>>>>> + if (sbdrv->probe) >>>>>> + status = sbdrv->probe(sbdev); >>>>>> + >>>>>> + if (status) >>>>>> + sbdev->driver = NULL; >>>>>> + else if (sbdrv->device_up) >>>>>> + schedule_slim_report(sbdev->ctrl, sbdev, true); >>>>> >>>>> can you please explain what this is trying to do? >>>> >>>> It is scheduling a device_up() callback in workqueue for reporting >>>> discovered device. >>> >>> any reason for that? Would the device not announce itself on the bus and >>> then you can synchronously update the device. >> You are correct, Device should announce itself in all cases. core should >> only call this callback only when device is announced, it does not make >> sense for this call in slim_device_probe(). Will remove it from here in next >> version. > > Okay great. Btw do you need a notify being scheduled in those cases? I guess > your controller would get an interrupt and you will handle that in bottom > half and then cll this, so why not call in the bottom half? > That makes sense, I will optimize this path, It looks like there are 2 workqueues in this path. We should be able to get rid of one work-queue. >>>>>> +/** >>>>>> + * slim_register_controller: Controller bring-up and registration. >> ... >>>>>> + >>>>>> + mutex_init(&ctrl->m_ctrl); >>>>>> + ret = device_register(&ctrl->dev); >>>>> >>>>> one more device_register?? Can you explain why >>>>> >>>> >>>> This is a device for each controller. >>> >>> wont the controller have its own platform_device? >> >> Reason could be that slim_register controller can be called from any code >> not just platform devices.. > > ah which cases would those be. I was expecting that you would have a > platform_device as a slimbus controller which would call slim_register? As of now there is only one controller which uses platform driver, but in future there might be more, but this is something which makes the slimbus core more flexible. > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 11/10/17 11:21, Vinod Koul wrote: >>> ah which cases would those be. I was expecting that you would have a >>> platform_device as a slimbus controller which would call slim_register? >> As of now there is only one controller which uses platform driver, but in >> future there might be more, but this is something which makes the slimbus >> core more flexible. > even if you have more controllers wouldn't we have similar number of platform > devices. Each instance of the link/controller would have its device node. > Yep, I will give at try and see in next version! thanks, srini > Thanks > -- ~Vinod -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Oct 06, 2017 at 05:51:30PM +0200, srinivas.kandagatla@linaro.org wrote: > From: Sagar Dharia <sdharia@codeaurora.org> > > SLIMbus (Serial Low Power Interchip Media Bus) is a specification > developed by MIPI (Mobile Industry Processor Interface) alliance. > SLIMbus is a 2-wire implementation, which is used to communicate with > peripheral components like audio-codec. > SLIMbus uses Time-Division-Multiplexing to accommodate multiple data > channels, and control channel. Control channel has messages to do > device-enumeration, messages to send/receive control-data to/from > slimbus devices, messages for port/channel management, and messages to > do bandwidth allocation. > The framework supports multiple instances of the bus (1 controller per > bus), and multiple slave devices per controller. > > This patch does device enumeration, logical address assignment, > informing device when the device reports present/absent etc. > Reporting present may need the driver to do the needful (e.g. turning > on voltage regulators powering the device). Additionally device is > probed when it reports present if that device doesn't need any such > steps mentioned above. > > Signed-off-by: Sagar Dharia <sdharia@codeaurora.org> > Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> > --- > Documentation/devicetree/bindings/slimbus/bus.txt | 57 ++ > Documentation/slimbus/summary | 109 ++++ > drivers/Kconfig | 2 + > drivers/Makefile | 1 + > drivers/slimbus/Kconfig | 11 + > drivers/slimbus/Makefile | 5 + > drivers/slimbus/slim-core.c | 695 ++++++++++++++++++++++ > include/linux/mod_devicetable.h | 13 + > include/linux/slimbus.h | 299 ++++++++++ > 9 files changed, 1192 insertions(+) > create mode 100644 Documentation/devicetree/bindings/slimbus/bus.txt > create mode 100644 Documentation/slimbus/summary > create mode 100644 drivers/slimbus/Kconfig > create mode 100644 drivers/slimbus/Makefile > create mode 100644 drivers/slimbus/slim-core.c > create mode 100644 include/linux/slimbus.h > > diff --git a/Documentation/devicetree/bindings/slimbus/bus.txt b/Documentation/devicetree/bindings/slimbus/bus.txt > new file mode 100644 > index 0000000..cb658bb > --- /dev/null > +++ b/Documentation/devicetree/bindings/slimbus/bus.txt > @@ -0,0 +1,57 @@ > +SLIM(Serial Low Power Interchip Media Bus) bus > + > +SLIMbus is a 2-wire bus, and is used to communicate with peripheral > +components like audio-codec. > + > +Controller is a normal device using binding for whatever bus it is > +on (e.g. platform bus). > +Required property for SLIMbus controller node: > +- compatible - name of SLIMbus controller following generic names > + recommended practice. > +- #address-cells - should be 2 > +- #size-cells - should be 0 > + > +No other properties are required in the SLIMbus controller bus node. > + > +Child nodes: > +Every SLIMbus controller node can contain zero or more child nodes > +representing slave devices on the bus. Every SLIMbus slave device is > +uniquely determined by the enumeration address containing 4 fields: > +Manufacturer ID, Product code, Device index, and Instance value for > +the device. > +If child node is not present and it is instantiated after device > +discovery (slave device reporting itself present). > + > +In some cases it may be necessary to describe non-probeable device > +details such as non-standard ways of powering up a device. In > +such cases, child nodes for those devices will be present as > +slaves of the slimbus-controller, as detailed below. > + > +Required property for SLIMbus child node if it is present: > +- reg - Is Duplex (Device index, Instance ID) from Enumeration > + Address. > + Device Index Uniquely identifies multiple Devices within > + a single Component. > + Instance ID Is for the cases where multiple Devices of the > + same type or Class are attached to the bus. > + > +- compatible -"slimMID,PID". The textual representation of Manufacturer ID, > + Product Code, shall be in lower case hexadecimal with leading > + zeroes suppressed > + > +SLIMbus example for Qualcomm's slimbus manager component: > + > + slim@28080000 { > + compatible = "qcom,slim-msm"; > + reg = <0x28080000 0x2000>, > + interrupts = <0 33 0>; > + clocks = <&lcc SLIMBUS_SRC>, <&lcc AUDIO_SLIMBUS_CLK>; > + clock-names = "iface_clk", "core_clk"; > + #address-cells = <2>; > + #size-cells = <0>; > + > + codec: wcd9310@1{ > + compatible = "slim217,60""; > + reg = <1 0>; > + }; > + }; > diff --git a/Documentation/slimbus/summary b/Documentation/slimbus/summary > new file mode 100644 > index 0000000..e7f90bb > --- /dev/null > +++ b/Documentation/slimbus/summary > @@ -0,0 +1,109 @@ > +Overview of Linux kernel SLIMbus support > +======================================== > + > +What is SLIMbus? > +---------------- > +SLIMbus (Serial Low Power Interchip Media Bus) is a specification developed by > +MIPI (Mobile Industry Processor Interface) alliance. The bus uses master/slave > +configuration, and is a 2-wire multi-drop implementation (clock, and data). > + > +Currently, SLIMbus is used to interface between application processors of SoCs > +(System-on-Chip) and peripheral components (typically codec).SLIMbus uses > +Time-Division-Multiplexing to accommodate multiple data channels, and > +a control channel. > + > +The control channel is used for various control functions such as bus > +management, configuration and status updates.These messages can be unicast (e.g. > +reading/writing device specific values), or multicast (e.g. data channel > +reconfiguration sequence is a broadcast message announced to all devices) > + > +A data channel is used for data-transfer between 2 Slimbus devices. Data > +channel uses dedicated ports on the device. > + > +Hardware description: > +--------------------- > +Slimbus specification has different types of device classifications based on > +their capabilities. > +A manager device is responsible for enumeration, configuration, and dynamic > +channel allocation. Every bus has 1 active manager. > + > +A generic device is a device providing application functionality (e.g. codec). > + > +Framer device is responsible for clocking the bus, and transmitting frame-sync > +and framing information on the bus. > + > +Each SLIMbus component has an interface device for monitoring physical layer. > + > +Typically each SoC contains SLIMbus component having 1 manager, 1 framer device, > +1 generic device (for data channel support), and 1 interface device. > +External peripheral SLIMbus component usually has 1 generic device (for > +functionality/data channel support), and an associated interface device. > +The generic device's registers are mapped as 'value elements' so that they can > +be written/read using Slimbus control channel exchanging control/status type of > +information. > +In case there are multiple framer devices on the same bus, manager device is > +responsible to select the active-framer for clocking the bus. > + > +Per specification, Slimbus uses "clock gears" to do power management based on > +current frequency and bandwidth requirements. There are 10 clock gears and each > +gear changes the Slimbus frequency to be twice its previous gear. Does the spec mandate 10 clock gears or its controller property? > + > +Each device has a 6-byte enumeration-address and the manager assigns every > +device with a 1-byte logical address after the devices report presence on the > +bus. > + > +Software description: > +--------------------- > +There are 2 types of SLIMbus drivers: > + > +slim_controller represents a 'controller' for SLIMbus. This driver should > +implement duties needed by the SoC (manager device, associated > +interface device for monitoring the layers and reporting errors, default > +framer device). > + > +slim_device represents the 'generic device/component' for SLIMbus, and a > +slim_driver should implement driver for that slim_device. > + > +Device notifications to the driver: > +----------------------------------- > +Since SLIMbus devices have mechanisms for reporting their presence, the > +framework allows drivers to bind when corresponding devices report their > +presence on the bus. > +However, it is possible that the driver needs to be probed > +first so that it can enable corresponding SLIMbus devie (e.g. power it up and/or > +take it out of reset). To support that behavior, the framework allows drivers > +to probe first as well (e.g. using standard DeviceTree compatbility field). > +This creates the necessity for the driver to know when the device is functional > +(i.e. reported present). device_up callback is used for that reason when the > +device reports present and is assigned a logical address by the controller. > + > +Similarly, SLIMbus devices 'report absent' when they go down. A 'device_down' > +callback notifies the driver when the device reports absent and its logical > +address assignment is invalidated by the controller. Is the same logical address assign when it reports present again? > + > +Another notification "boot_device" is used to notify the slim_driver when > +controller resets the bus. This notification allows the driver to take necessary > +steps to boot the device so that it's functional after the bus has been reset. > + > +Clock-pause: > +------------ > +SLIMbus mandates that a reconfiguration sequence (known as clock-pause) be > +broadcast to all active devices on the bus before the bus can enter low-power > +mode. Controller uses this sequence when it decides to enter low-power mode so > +that corresponding clocks and/or power-rails can be turned off to save power. > +Clock-pause is exited by waking up framer device (if controller driver initiates > +exiting low power mode), or by toggling the data line (if a slave device wants > +to initiate it). > + > +Messaging APIs: > +--------------- > +The framework supports APIs to exchange control-information with a SLIMbus > +device. APIs can be synchronous or asynchronous. > +From controller's perspective, multiple buffers can be queued to/from > +hardware for sending/receiving data using slim_ctrl_buf circular buffer. > +The header file <linux/slimbus.h> has more documentation about messaging APIs. > + > +----------------------------------------------------------------- > +<Sections will be added to this document when port/channel bandwidth management > +support, multi-xfer APIs are added to the framework> > +------------------------------------------------------------------ > diff --git a/drivers/Kconfig b/drivers/Kconfig > index 505c676..8010c67 100644 > --- a/drivers/Kconfig > +++ b/drivers/Kconfig > @@ -208,4 +208,6 @@ source "drivers/tee/Kconfig" > > source "drivers/mux/Kconfig" > > +source "drivers/slimbus/Kconfig" > + > endmenu > diff --git a/drivers/Makefile b/drivers/Makefile > index d90fdc4..0449c7c 100644 > --- a/drivers/Makefile > +++ b/drivers/Makefile > @@ -86,6 +86,7 @@ obj-$(CONFIG_MTD) += mtd/ > obj-$(CONFIG_SPI) += spi/ > obj-$(CONFIG_SPMI) += spmi/ > obj-$(CONFIG_HSI) += hsi/ > +obj-$(CONFIG_SLIMBUS) += slimbus/ > obj-y += net/ > obj-$(CONFIG_ATM) += atm/ > obj-$(CONFIG_FUSION) += message/ > diff --git a/drivers/slimbus/Kconfig b/drivers/slimbus/Kconfig > new file mode 100644 > index 0000000..f0b118a > --- /dev/null > +++ b/drivers/slimbus/Kconfig > @@ -0,0 +1,11 @@ > +# > +# SLIMBUS driver configuration > +# > +menuconfig SLIMBUS > + tristate "Slimbus support" > + help > + Slimbus is standard interface between System-on-Chip and audio codec, > + and other peripheral components in typical embedded systems. > + > + If unsure, choose N. > + > diff --git a/drivers/slimbus/Makefile b/drivers/slimbus/Makefile > new file mode 100644 > index 0000000..f580704 > --- /dev/null > +++ b/drivers/slimbus/Makefile > @@ -0,0 +1,5 @@ > +# > +# Makefile for kernel slimbus framework. > +# > +obj-$(CONFIG_SLIMBUS) += slimbus.o > +slimbus-y := slim-core.o > diff --git a/drivers/slimbus/slim-core.c b/drivers/slimbus/slim-core.c > new file mode 100644 > index 0000000..de3ef79 > --- /dev/null > +++ b/drivers/slimbus/slim-core.c > @@ -0,0 +1,695 @@ > +/* Copyright (c) 2011-2017, The Linux Foundation. All rights reserved. > + * > + * This program is free software; you can redistribute it and/or modify > + * it under the terms of the GNU General Public License version 2 and > + * only version 2 as published by the Free Software Foundation. > + * > + * This program is distributed in the hope that it will be useful, > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + * GNU General Public License for more details. > + */ > + > +#include <linux/kernel.h> > +#include <linux/errno.h> > +#include <linux/slab.h> > +#include <linux/init.h> > +#include <linux/completion.h> > +#include <linux/idr.h> > +#include <linux/pm_runtime.h> > +#include <linux/slimbus.h> > +#include <linux/of.h> > +#include <linux/of_device.h> > + > +static DEFINE_MUTEX(slim_lock); > +static DEFINE_IDR(ctrl_idr); > + > +static bool slim_eaddr_equal(struct slim_eaddr *a, struct slim_eaddr *b) > +{ > + > + return (a->manf_id == b->manf_id && > + a->prod_code == b->prod_code && > + a->dev_index == b->dev_index && > + a->instance == b->instance); > +} > + > +static const struct slim_device_id *slim_match(const struct slim_device_id *id, > + const struct slim_device *sbdev) > +{ > + while (id->manf_id != 0 || id->prod_code != 0) { > + if (id->manf_id == sbdev->e_addr.manf_id && > + id->prod_code == sbdev->e_addr.prod_code && > + id->dev_index == sbdev->e_addr.dev_index) > + return id; > + id++; > + } > + return NULL; > +} > + > +static int slim_device_match(struct device *dev, struct device_driver *drv) > +{ > + struct slim_device *sbdev = to_slim_device(dev); > + struct slim_driver *sbdrv = to_slim_driver(drv); > + > + /* Attempt an OF style match first */ > + if (of_driver_match_device(dev, drv)) > + return 1; > + > + /* Then try to match against the id table */ > + if (sbdrv->id_table) > + return slim_match(sbdrv->id_table, sbdev) != NULL; > + > + return 0; > +} > + > +struct sb_report_wd { > + struct work_struct wd; > + struct slim_device *sbdev; > + bool report; > +}; > + > +static void slim_report(struct work_struct *work) > +{ > + struct slim_driver *sbdrv; > + struct sb_report_wd *sbw = container_of(work, struct sb_report_wd, wd); > + struct slim_device *sbdev = sbw->sbdev; > + > + mutex_lock(&sbdev->report_lock); > + if (!sbdev->dev.driver) > + goto report_exit; > + > + /* check if device-up or down needs to be called */ > + if ((!sbdev->reported && !sbdev->notified) || > + (sbdev->reported && sbdev->notified)) > + goto report_exit; > + > + sbdrv = to_slim_driver(sbdev->dev.driver); > + > + /** > + * address no longer valid, means device reported absent, whereas > + * address valid, means device reported present > + */ > + if (sbdev->notified && !sbdev->reported) { > + sbdev->notified = false; > + if (sbdrv->device_down) > + sbdrv->device_down(sbdev); > + } else if (!sbdev->notified && sbdev->reported) { > + sbdev->notified = true; > + if (sbdrv->device_up) > + sbdrv->device_up(sbdev); > + } > +report_exit: > + mutex_unlock(&sbdev->report_lock); > + kfree(sbw); > +} > + > +/** > + * Report callbacks(device_up, device_down) are implemented by slimbus-devices. > + * The calls are scheduled into a workqueue to avoid holding up controller > + * initialization/tear-down. > + */ > +static void schedule_slim_report(struct slim_controller *ctrl, > + struct slim_device *sb, bool report) > +{ > + struct sb_report_wd *sbw; > + > + dev_dbg(&ctrl->dev, "report:%d for slave:%s\n", report, sb->name); > + > + sbw = kmalloc(sizeof(*sbw), GFP_KERNEL); > + if (!sbw) > + return; > + > + INIT_WORK(&sbw->wd, slim_report); > + sbw->sbdev = sb; > + sbw->report = report; > + if (!queue_work(ctrl->wq, &sbw->wd)) { > + dev_err(&ctrl->dev, "failed to queue report:%d slave:%s\n", > + report, sb->name); > + kfree(sbw); > + } > +} > + > +static int slim_device_probe(struct device *dev) > +{ > + struct slim_device *sbdev; > + struct slim_driver *sbdrv; > + int status = 0; > + > + sbdev = to_slim_device(dev); > + sbdrv = to_slim_driver(dev->driver); > + > + sbdev->driver = sbdrv; > + > + if (sbdrv->probe) > + status = sbdrv->probe(sbdev); > + > + if (status) > + sbdev->driver = NULL; > + else if (sbdrv->device_up) > + schedule_slim_report(sbdev->ctrl, sbdev, true); > + > + return status; > +} > + > +static int slim_device_remove(struct device *dev) > +{ > + struct slim_device *sbdev; > + struct slim_driver *sbdrv; > + int status = 0; > + > + sbdev = to_slim_device(dev); > + if (!dev->driver) > + return 0; > + > + sbdrv = to_slim_driver(dev->driver); > + if (sbdrv->remove) > + status = sbdrv->remove(sbdev); > + > + mutex_lock(&sbdev->report_lock); > + sbdev->notified = false; > + if (status == 0) > + sbdev->driver = NULL; > + mutex_unlock(&sbdev->report_lock); > + return status; > +} > + > +struct bus_type slimbus_type = { > + .name = "slimbus", > + .match = slim_device_match, > + .probe = slim_device_probe, > + .remove = slim_device_remove, > +}; > +EXPORT_SYMBOL_GPL(slimbus_type); > + > +/** > + * slim_driver_register: Client driver registration with slimbus > + * @drv:Client driver to be associated with client-device. > + * @owner: owning module/driver > + * This API will register the client driver with the slimbus > + * It is called from the driver's module-init function. > + */ > +int __slim_driver_register(struct slim_driver *drv, struct module *owner) > +{ > + drv->driver.bus = &slimbus_type; > + drv->driver.owner = owner; > + return driver_register(&drv->driver); > +} > +EXPORT_SYMBOL_GPL(__slim_driver_register); > + > +/** > + * slim_driver_unregister: Undo effect of slim_driver_register > + * @drv: Client driver to be unregistered > + */ > +void slim_driver_unregister(struct slim_driver *drv) > +{ > + if (drv) > + driver_unregister(&drv->driver); > +} > +EXPORT_SYMBOL_GPL(slim_driver_unregister); > + > +static struct slim_controller *slim_ctrl_get(struct slim_controller *ctrl) > +{ > + if (!ctrl || !get_device(&ctrl->dev)) > + return NULL; > + > + return ctrl; > +} > + > +static void slim_ctrl_put(struct slim_controller *ctrl) > +{ > + if (ctrl) > + put_device(&ctrl->dev); > +} > + > +static void slim_dev_release(struct device *dev) > +{ > + struct slim_device *sbdev = to_slim_device(dev); > + > + slim_ctrl_put(sbdev->ctrl); > + kfree(sbdev->name); > + kfree(sbdev); > +} > + > +static int slim_add_device(struct slim_controller *ctrl, > + struct slim_device *sbdev) > +{ > + sbdev->dev.bus = &slimbus_type; > + sbdev->dev.parent = &ctrl->dev; > + sbdev->dev.release = slim_dev_release; > + sbdev->dev.driver = NULL; > + sbdev->ctrl = ctrl; > + > + slim_ctrl_get(ctrl); > + sbdev->name = kasprintf(GFP_KERNEL, "%x:%x:%x:%x", > + sbdev->e_addr.manf_id, > + sbdev->e_addr.prod_code, > + sbdev->e_addr.dev_index, > + sbdev->e_addr.instance); > + if (!sbdev->name) > + return -ENOMEM; > + > + dev_set_name(&sbdev->dev, "%s", sbdev->name); > + mutex_init(&sbdev->report_lock); > + > + /* probe slave on this controller */ > + return device_register(&sbdev->dev); > +} > + > +/* Helper to get hex Manufacturer ID and Product id from compatible */ > +static unsigned long str2hex(unsigned char *str) > +{ > + int value = 0; > + > + while (*str) { > + char c = *str++; > + > + value = value << 4; > + if (c >= '0' && c <= '9') > + value |= (c - '0'); > + if (c >= 'a' && c <= 'f') > + value |= (c - 'a' + 10); > + > + } > + > + return value; > +} > + > +/* OF helpers for SLIMbus */ > +static void of_register_slim_devices(struct slim_controller *ctrl) > +{ > + struct device *dev = &ctrl->dev; > + struct device_node *node; > + > + if (!ctrl->dev.of_node) > + return; > + > + for_each_child_of_node(ctrl->dev.of_node, node) { > + struct slim_device *slim; > + const char *compat = NULL; > + char *p, *tok; > + int reg[2], ret; > + > + slim = kzalloc(sizeof(*slim), GFP_KERNEL); > + if (!slim) > + continue; > + > + slim->dev.of_node = of_node_get(node); > + > + compat = of_get_property(node, "compatible", NULL); > + if (!compat) > + continue; > + > + p = kasprintf(GFP_KERNEL, "%s", compat + strlen("slim")); > + > + tok = strsep(&p, ","); > + if (!tok) { > + dev_err(dev, "No valid Manufacturer ID found\n"); > + kfree(p); > + continue; > + } > + slim->e_addr.manf_id = str2hex(tok); > + > + tok = strsep(&p, ","); > + if (!tok) { > + dev_err(dev, "No valid Product ID found\n"); > + kfree(p); > + continue; > + } > + slim->e_addr.prod_code = str2hex(tok); > + kfree(p); > + > + ret = of_property_read_u32_array(node, "reg", reg, 2); > + if (ret) { > + dev_err(dev, "Device and Instance id not found:%d\n", > + ret); > + continue; > + } > + slim->e_addr.dev_index = reg[0]; > + slim->e_addr.instance = reg[1]; > + > + ret = slim_add_device(ctrl, slim); > + if (ret) > + dev_err(dev, "of_slim device register err:%d\n", ret); > + } > +} > + > +/** > + * slim_register_controller: Controller bring-up and registration. > + * @ctrl: Controller to be registered. > + * A controller is registered with the framework using this API. > + * If devices on a controller were registered before controller, > + * this will make sure that they get probed when controller is up > + */ > +int slim_register_controller(struct slim_controller *ctrl) > +{ > + int id, ret = 0; > + > + mutex_lock(&slim_lock); > + id = idr_alloc(&ctrl_idr, ctrl, ctrl->nr, -1, GFP_KERNEL); > + mutex_unlock(&slim_lock); > + > + if (id < 0) > + return id; > + > + ctrl->nr = id; > + > + dev_set_name(&ctrl->dev, "sb-%d", ctrl->nr); > + ctrl->num_dev = 0; > + > + if (!ctrl->min_cg) > + ctrl->min_cg = SLIM_MIN_CLK_GEAR; > + if (!ctrl->max_cg) > + ctrl->max_cg = SLIM_MAX_CLK_GEAR; > + > + mutex_init(&ctrl->m_ctrl); > + ret = device_register(&ctrl->dev); > + if (ret) > + goto dev_reg_failed; > + > + dev_dbg(&ctrl->dev, "Bus [%s] registered:dev:%p\n", > + ctrl->name, &ctrl->dev); > + > + ctrl->wq = create_singlethread_workqueue(dev_name(&ctrl->dev)); > + if (!ctrl->wq) > + goto err_workq_failed; > + > + of_register_slim_devices(ctrl); > + > + return 0; > + > +err_workq_failed: > + device_unregister(&ctrl->dev); > +dev_reg_failed: > + mutex_lock(&slim_lock); > + idr_remove(&ctrl_idr, ctrl->nr); > + mutex_unlock(&slim_lock); > + return ret; > +} > +EXPORT_SYMBOL_GPL(slim_register_controller); > + > +/* slim_remove_device: Remove the effect of slim_add_device() */ > +static void slim_remove_device(struct slim_device *sbdev) > +{ > + device_unregister(&sbdev->dev); > +} > + > +static int slim_ctrl_remove_device(struct device *dev, void *null) > +{ > + slim_remove_device(to_slim_device(dev)); > + return 0; > +} > + > +/** > + * slim_del_controller: Controller tear-down. > + * @ctrl: Controller to tear-down. > + */ > +int slim_del_controller(struct slim_controller *ctrl) > +{ > + struct slim_controller *found; > + > + /* First make sure that this bus was added */ > + mutex_lock(&slim_lock); > + found = idr_find(&ctrl_idr, ctrl->nr); > + mutex_unlock(&slim_lock); > + if (found != ctrl) > + return -EINVAL; > + > + /* Remove all clients */ > + device_for_each_child(&ctrl->dev, NULL, slim_ctrl_remove_device); > + > + > + destroy_workqueue(ctrl->wq); > + > + /* free bus id */ > + mutex_lock(&slim_lock); > + idr_remove(&ctrl_idr, ctrl->nr); > + mutex_unlock(&slim_lock); > + > + device_unregister(&ctrl->dev); > + return 0; > +} > +EXPORT_SYMBOL_GPL(slim_del_controller); > + > +/** > + * slim_report_absent: Controller calls this function when a device > + * reports absent, OR when the device cannot be communicated with > + * @sbdev: Device that cannot be reached, or sent report absent > + */ > +void slim_report_absent(struct slim_device *sbdev) > +{ > + struct slim_controller *ctrl; > + int i; > + > + if (!sbdev) > + return; > + ctrl = sbdev->ctrl; > + if (!ctrl) > + return; > + > + /* invalidate logical addresses */ > + mutex_lock(&ctrl->m_ctrl); > + for (i = 0; i < ctrl->num_dev; i++) { > + if (sbdev->laddr == ctrl->addrt[i].laddr) > + ctrl->addrt[i].valid = false; > + } > + mutex_unlock(&ctrl->m_ctrl); > + > + mutex_lock(&sbdev->report_lock); > + sbdev->reported = false; > + schedule_slim_report(ctrl, sbdev, false); > + mutex_unlock(&sbdev->report_lock); > +} > +EXPORT_SYMBOL_GPL(slim_report_absent); > + > +static int slim_boot_child(struct device *dev, void *unused) > +{ > + struct slim_driver *sbdrv; > + struct slim_device *sbdev = to_slim_device(dev); > + > + if (sbdev && sbdev->dev.driver) { > + sbdrv = to_slim_driver(sbdev->dev.driver); > + if (sbdrv->boot_device) > + sbdrv->boot_device(sbdev); > + } > + return 0; > +} > + > +static int slim_match_dev(struct device *dev, void *data) > +{ > + struct slim_eaddr *e_addr = data; > + struct slim_device *slim = to_slim_device(dev); > + > + return slim_eaddr_equal(&slim->e_addr, e_addr); > +} > + > +/** > + * slim_framer_booted: This function is called by controller after the active > + * framer has booted (using Bus Reset sequence, or after it has shutdown and has > + * come back up). > + * @ctrl: Controller associated with this framer > + * Components, devices on the bus may be in undefined state, > + * and this function triggers their drivers to do the needful > + * to bring them back in Reset state so that they can acquire sync, report > + * present and be operational again. > + */ > +void slim_framer_booted(struct slim_controller *ctrl) > +{ > + if (!ctrl) > + return; > + > + device_for_each_child(&ctrl->dev, NULL, slim_boot_child); > +} > +EXPORT_SYMBOL_GPL(slim_framer_booted); > + > +/** > + * slim_query_device: Query and get handle to a device. > + * @ctrl: Controller on which this device will be added/queried > + * @e_addr: Enumeration address of the device to be queried > + * Returns pointer to a device if it has already reported. Creates a new > + * device and returns pointer to it if the device has not yet enumerated. > + */ > +struct slim_device *slim_query_device(struct slim_controller *ctrl, > + struct slim_eaddr *e_addr) > +{ > + struct device *dev; > + struct slim_device *slim = NULL; > + > + dev = device_find_child(&ctrl->dev, e_addr, slim_match_dev); > + if (dev) { > + slim = to_slim_device(dev); > + return slim; > + } > + > + slim = kzalloc(sizeof(struct slim_device), GFP_KERNEL); > + if (IS_ERR(slim)) > + return NULL; > + > + slim->e_addr = *e_addr; > + if (slim_add_device(ctrl, slim) != 0) { > + kfree(slim); > + return NULL; > + } > + return slim; > +} > +EXPORT_SYMBOL_GPL(slim_query_device); > + > +static int ctrl_getaddr_entry(struct slim_controller *ctrl, > + struct slim_eaddr *eaddr, u8 *entry) > +{ > + int i; > + > + for (i = 0; i < ctrl->num_dev; i++) { > + if (ctrl->addrt[i].valid && > + slim_eaddr_equal(&ctrl->addrt[i].eaddr, eaddr)) { > + *entry = i; > + return 0; > + } > + } > + return -ENXIO; > +} > + > +/** > + * slim_assign_laddr: Assign logical address to a device enumerated. > + * @ctrl: Controller with which device is enumerated. > + * @e_addr: Enumeration address of the device. > + * @laddr: Return logical address (if valid flag is false) > + * @valid: true if laddr holds a valid address that controller wants to > + * set for this enumeration address. Otherwise framework sets index into > + * address table as logical address. > + * Called by controller in response to REPORT_PRESENT. Framework will assign > + * a logical address to this enumeration address. > + * Function returns -EXFULL to indicate that all logical addresses are already > + * taken. > + */ > +int slim_assign_laddr(struct slim_controller *ctrl, struct slim_eaddr *e_addr, > + u8 *laddr, bool valid) > +{ > + int ret; > + u8 i = 0; > + bool exists = false; > + struct slim_device *slim; > + struct slim_addrt *temp; > + > + mutex_lock(&ctrl->m_ctrl); > + /* already assigned */ > + if (ctrl_getaddr_entry(ctrl, e_addr, &i) == 0) { > + *laddr = ctrl->addrt[i].laddr; > + exists = true; > + } else { > + if (ctrl->num_dev >= (SLIM_LA_MANAGER - 1)) { > + ret = -EXFULL; > + goto ret_assigned_laddr; > + } > + for (i = 0; i < ctrl->num_dev; i++) { > + if (ctrl->addrt[i].valid == false) > + break; > + } > + if (i == ctrl->num_dev) { > + temp = krealloc(ctrl->addrt, > + (ctrl->num_dev + 1) * > + sizeof(struct slim_addrt), > + GFP_KERNEL); > + if (!temp) { > + ret = -ENOMEM; > + goto ret_assigned_laddr; > + } > + ctrl->addrt = temp; > + ctrl->num_dev++; > + } > + ctrl->addrt[i].eaddr = *e_addr; > + ctrl->addrt[i].valid = true; > + > + /* Preferred address is index into table */ > + if (!valid) > + *laddr = i; > + } > + > + ret = ctrl->set_laddr(ctrl, &ctrl->addrt[i].eaddr, *laddr); > + if (ret) { > + ctrl->addrt[i].valid = false; > + goto ret_assigned_laddr; > + } > + ctrl->addrt[i].laddr = *laddr; > + > +ret_assigned_laddr: > + mutex_unlock(&ctrl->m_ctrl); > + if (exists || ret) > + return ret; > + > + dev_info(&ctrl->dev, "setting slimbus l-addr:%x, ea:%x,%x,%x,%x\n", > + *laddr, e_addr->manf_id, e_addr->prod_code, > + e_addr->dev_index, e_addr->instance); > + > + /** > + * Add this device to list of devices on this controller if it's > + * not already present > + */ > + slim = slim_query_device(ctrl, e_addr); > + if (!slim) { > + ret = -ENODEV; > + } else { > + struct slim_driver *sbdrv; > + > + slim->laddr = *laddr; > + mutex_lock(&slim->report_lock); > + slim->reported = true; > + if (slim->dev.driver) { > + sbdrv = to_slim_driver(slim->dev.driver); > + if (sbdrv->device_up) > + schedule_slim_report(ctrl, slim, true); > + } > + mutex_unlock(&slim->report_lock); > + } > + return ret; > +} > +EXPORT_SYMBOL_GPL(slim_assign_laddr); > + > +/** > + * slim_get_logical_addr: Return the logical address of a slimbus device. > + * @sb: client handle requesting the address. > + * @e_addr: Enumeration address of the device. > + * @laddr: output buffer to store the address > + * context: can sleep > + * -EINVAL is returned in case of invalid parameters, and -ENXIO is returned if > + * the device with this enumeration address is not found. > + */ > +int slim_get_logical_addr(struct slim_device *sb, struct slim_eaddr *e_addr, > + u8 *laddr) > +{ > + int ret; > + u8 entry; > + struct slim_controller *ctrl = sb->ctrl; > + > + if (!ctrl || !laddr || !e_addr) > + return -EINVAL; > + > + mutex_lock(&ctrl->m_ctrl); > + ret = ctrl_getaddr_entry(ctrl, e_addr, &entry); > + if (!ret) > + *laddr = ctrl->addrt[entry].laddr; > + mutex_unlock(&ctrl->m_ctrl); > + > + if (ret == -ENXIO && ctrl->get_laddr) { > + ret = ctrl->get_laddr(ctrl, e_addr, laddr); > + if (!ret) > + ret = slim_assign_laddr(ctrl, e_addr, laddr, true); > + } > + return ret; > +} > +EXPORT_SYMBOL_GPL(slim_get_logical_addr); > + > +static void __exit slimbus_exit(void) > +{ > + bus_unregister(&slimbus_type); > +} > +module_exit(slimbus_exit); > + > +static int __init slimbus_init(void) > +{ > + return bus_register(&slimbus_type); > +} > +postcore_initcall(slimbus_init); > + > +MODULE_LICENSE("GPL v2"); > +MODULE_VERSION("0.1"); > +MODULE_DESCRIPTION("Slimbus module"); > diff --git a/include/linux/mod_devicetable.h b/include/linux/mod_devicetable.h > index 694cebb..015e5f6 100644 > --- a/include/linux/mod_devicetable.h > +++ b/include/linux/mod_devicetable.h > @@ -448,6 +448,19 @@ struct spi_device_id { > kernel_ulong_t driver_data; /* Data private to the driver */ > }; > > +/* SLIMbus */ > + > +#define SLIMBUS_NAME_SIZE 32 > +#define SLIMBUS_MODULE_PREFIX "slim:" > + > +struct slim_device_id { > + __u16 manf_id, prod_code; > + __u8 dev_index, instance; > + > + /* Data private to the driver */ > + kernel_ulong_t driver_data; > +}; > + > #define SPMI_NAME_SIZE 32 > #define SPMI_MODULE_PREFIX "spmi:" > > diff --git a/include/linux/slimbus.h b/include/linux/slimbus.h > new file mode 100644 > index 0000000..b5064b6 > --- /dev/null > +++ b/include/linux/slimbus.h > @@ -0,0 +1,299 @@ > +/* Copyright (c) 2011-2017, The Linux Foundation. All rights reserved. > + * > + * This program is free software; you can redistribute it and/or modify > + * it under the terms of the GNU General Public License version 2 and > + * only version 2 as published by the Free Software Foundation. > + * > + * This program is distributed in the hope that it will be useful, > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + * GNU General Public License for more details. > + */ > + > +#ifndef _LINUX_SLIMBUS_H > +#define _LINUX_SLIMBUS_H > +#include <linux/module.h> > +#include <linux/device.h> > +#include <linux/mutex.h> > +#include <linux/mod_devicetable.h> > + > +/** > + * Interfaces between SLIMbus manager drivers, SLIMbus client drivers, and > + * SLIMbus infrastructure. > + */ > + > +extern struct bus_type slimbus_type; > + > +/* Standard values per SLIMbus spec needed by controllers and devices */ > +#define SLIM_CL_PER_SUPERFRAME 6144 > +#define SLIM_CL_PER_SUPERFRAME_DIV8 (SLIM_CL_PER_SUPERFRAME >> 3) > +#define SLIM_MAX_CLK_GEAR 10 > +#define SLIM_MIN_CLK_GEAR 1 > +#define SLIM_CL_PER_SL 4 > +#define SLIM_SL_PER_SUPERFRAME (SLIM_CL_PER_SUPERFRAME >> 2) > +#define SLIM_FRM_SLOTS_PER_SUPERFRAME 16 > +#define SLIM_GDE_SLOTS_PER_SUPERFRAME 2 > + > +struct slim_controller; > +struct slim_device; > + > +/** > + * struct slim_eaddr: Enumeration address for a slimbus device > + * @manf_id: Manufacturer Id for the device > + * @prod_code: Product code > + * @dev_index: Device index > + * @instance: Instance value > + */ > +struct slim_eaddr { > + u16 manf_id; > + u16 prod_code; > + u8 dev_index; > + u8 instance; > +}; > + > +/** > + * struct slim_framer - Represents Slimbus framer. > + * Every controller may have multiple framers. There is 1 active framer device > + * responsible for clocking the bus. > + * Manager is responsible for framer hand-over. > + * @dev: Driver model representation of the device. > + * @e_addr: Enumeration address of the framer. > + * @rootfreq: Root Frequency at which the framer can run. This is maximum > + * frequency ('clock gear 10') at which the bus can operate. > + * @superfreq: Superframes per root frequency. Every frame is 6144 bits. > + */ > +struct slim_framer { > + struct device dev; > + struct slim_eaddr e_addr; > + int rootfreq; > + int superfreq; > +}; > + > +#define to_slim_framer(d) container_of(d, struct slim_framer, dev) > + > +/** > + * struct slim_addrt: slimbus address used internally by the slimbus framework. > + * @valid: If the device is present. Valid is set to false when device reports > + * absent. > + * @eaddr: Enumeration address > + * @laddr: It is possible that controller will set a predefined logical address > + * rather than the one assigned by framework. (i.e. logical address may > + * not be same as index into this table). This entry will store the > + * logical address value for this enumeration address. > + */ > +struct slim_addrt { > + bool valid; > + struct slim_eaddr eaddr; > + u8 laddr; > +}; > + > +/* SLIMbus message types. Related to interpretation of message code. */ > +#define SLIM_MSG_MT_CORE 0x0 > +#define SLIM_MSG_MT_DEST_REFERRED_CLASS 0x1 > +#define SLIM_MSG_MT_DEST_REFERRED_USER 0x2 > +#define SLIM_MSG_MT_SRC_REFERRED_CLASS 0x5 > +#define SLIM_MSG_MT_SRC_REFERRED_USER 0x6 > + > +/* SLIMbus core type Message Codes. */ > +/* Device management messages used by this framework */ > +#define SLIM_MSG_MC_REPORT_PRESENT 0x1 > +#define SLIM_MSG_MC_ASSIGN_LOGICAL_ADDRESS 0x2 > +#define SLIM_MSG_MC_REPORT_ABSENT 0xF > + > +/* Destination type Values */ > +#define SLIM_MSG_DEST_LOGICALADDR 0 > +#define SLIM_MSG_DEST_ENUMADDR 1 > +#define SLIM_MSG_DEST_BROADCAST 3 > + > +/** > + * struct slim_controller: Controls every instance of SLIMbus > + * (similar to 'master' on SPI) > + * 'Manager device' is responsible for device management, bandwidth > + * allocation, channel setup, and port associations per channel. > + * Device management means Logical address assignment/removal based on > + * enumeration (report-present, report-absent) if a device. > + * Bandwidth allocation is done dynamically by the manager based on active > + * channels on the bus, message-bandwidth requests made by slimbus devices. > + * Based on current bandwidth usage, manager chooses a frequency to run > + * the bus at (in steps of 'clock-gear', 1 through 10, each clock gear > + * representing twice the frequency than the previous gear). > + * Manager is also responsible for entering (and exiting) low-power-mode > + * (known as 'clock pause'). > + * Manager can do handover of framer if there are multiple framers on the > + * bus and a certain usecase warrants using certain framer to avoid keeping > + * previous framer being powered-on. > + * > + * Controller here performs duties of the manager device, and 'interface > + * device'. Interface device is responsible for monitoring the bus and > + * reporting information such as loss-of-synchronization, data > + * slot-collision. > + * @dev: Device interface to this driver > + * @nr: Board-specific number identifier for this controller/bus > + * @list: Link with other slimbus controllers > + * @name: Name for this controller > + * @min_cg: Minimum clock gear supported by this controller (default value: 1) > + * @max_cg: Maximum clock gear supported by this controller (default value: 10) > + * @clkgear: Current clock gear in which this bus is running > + * @a_framer: Active framer which is clocking the bus managed by this controller > + * @m_ctrl: Mutex protecting controller data structures > + * @addrt: Logical address table > + * @num_dev: Number of active slimbus slaves on this bus > + * @wq: Workqueue per controller used to notify devices when they report present > + * @xfer_msg: Transfer a message on this controller (this can be a broadcast > + * control/status message like data channel setup, or a unicast message > + * like value element read/write. xfer_msg element is not present in structure. > + * @set_laddr: Setup logical address at laddr for the slave with elemental > + * address e_addr. Drivers implementing controller will be expected to > + * send unicast message to this device with its logical address. > + * @get_laddr: It is possible that controller needs to set fixed logical > + * address table and get_laddr can be used in that case so that controller > + * can do this assignment. > + */ > +struct slim_controller { > + struct device dev; > + unsigned int nr; > + char name[SLIMBUS_NAME_SIZE]; > + int min_cg; > + int max_cg; > + int clkgear; > + struct slim_framer *a_framer; > + struct mutex m_ctrl; > + struct slim_addrt *addrt; > + u8 num_dev; > + struct workqueue_struct *wq; > + int (*set_laddr)(struct slim_controller *ctrl, > + struct slim_eaddr *ea, u8 laddr); > + int (*get_laddr)(struct slim_controller *ctrl, > + struct slim_eaddr *ea, u8 *laddr); > +}; > + > +#define to_slim_controller(d) container_of(d, struct slim_controller, dev) > + > +/** > + * struct slim_driver: Slimbus 'generic device' (slave) device driver > + * (similar to 'spi_device' on SPI) > + * @probe: Binds this driver to a slimbus device. > + * @remove: Unbinds this driver from the slimbus device. > + * @shutdown: Standard shutdown callback used during powerdown/halt. > + * @suspend: Standard suspend callback used during system suspend > + * @resume: Standard resume callback used during system resume > + * @device_up: This callback is called when the device reports present and > + * gets a logical address assigned to it > + * @device_down: This callback is called when device reports absent, or the > + * bus goes down. Device will report present when bus is up and > + * device_up callback will be called again when that happens > + * @boot_device: This callback is called after framer is booted. > + * Driver should do the needful to boot the device, > + * so that device acquires sync and be operational. > + * @driver: Slimbus device drivers should initialize name and owner field of > + * this structure > + * @id_table: List of slimbus devices supported by this driver > + */ > +struct slim_driver { > + int (*probe)(struct slim_device *sl); > + int (*remove)(struct slim_device *sl); > + void (*shutdown)(struct slim_device *sl); > + int (*suspend)(struct slim_device *sl, > + pm_message_t pmesg); > + int (*resume)(struct slim_device *sl); > + int (*device_up)(struct slim_device *sl); > + int (*device_down)(struct slim_device *sl); > + int (*boot_device)(struct slim_device *sl); > + > + struct device_driver driver; > + const struct slim_device_id *id_table; > +}; > + > +#define to_slim_driver(d) container_of(d, struct slim_driver, driver) > + > +/** > + * Client/device handle (struct slim_device): > + * ------------------------------------------ > + * This is the client/device handle returned when a slimbus > + * device is registered with a controller. > + * Pointer to this structure is used by client-driver as a handle. > + * @dev: Driver model representation of the device. > + * @name: Name of driver to use with this device. > + * @e_addr: Enumeration address of this device. > + * @driver: Device's driver. Pointer to access routines. > + * @ctrl: Slimbus controller managing the bus hosting this device. > + * @laddr: 1-byte Logical address of this device. > + * @reported: Flag to indicate whether this device reported present. The flag > + * is set when device reports present, and is reset when it reports > + * absent. This flag alongwith notified flag below is used to call > + * device_up, or device_down callbacks for driver of this device. > + * @notified: Flag to indicate whether this device has been notified. The > + * device may report present multiple times, but should be notified only > + * first time it has reported present. > + * @report_lock: Lock to protect reporting and notification for this device > + */ > +struct slim_device { > + struct device dev; > + char *name; > + struct slim_eaddr e_addr; > + struct slim_driver *driver; > + struct slim_controller *ctrl; > + u8 laddr; > + bool reported; > + bool notified; > + struct mutex report_lock; > +}; > + > +#define to_slim_device(d) container_of(d, struct slim_device, dev) > + > +/* Manager's logical address is set to 0xFF per spec */ > +#define SLIM_LA_MANAGER 0xFF > + > +int slim_get_logical_addr(struct slim_device *sb, > + struct slim_eaddr *e_addr, u8 *laddr); > + > +/* > + * use a macro to avoid include chaining to get THIS_MODULE > + */ > +#define slim_driver_register(drv) \ > + __slim_driver_register(drv, THIS_MODULE) > + > +int __slim_driver_register(struct slim_driver *drv, struct module *owner); > + > +void slim_driver_unregister(struct slim_driver *drv); > + > +/** > + * module_slim_driver() - Helper macro for registering a slimbus driver > + * @__slimbus_driver: slimbus_driver struct > + * > + * Helper macro for slimbus drivers which do not do anything special in module > + * init/exit. This eliminates a lot of boilerplate. Each module may only > + * use this macro once, and calling it replaces module_init() and module_exit() > + */ > +#define module_slim_driver(__slim_driver) \ > + module_driver(__slim_driver, slim_driver_register, \ > + slim_driver_unregister) > + > +int slim_del_controller(struct slim_controller *ctrl); > +int slim_assign_laddr(struct slim_controller *ctrl, > + struct slim_eaddr *e_addr, u8 *laddr, bool valid); > +void slim_report_absent(struct slim_device *sbdev); > +void slim_framer_booted(struct slim_controller *ctrl); > +int slim_register_controller(struct slim_controller *ctrl); > + > +static inline void *slim_get_ctrldata(const struct slim_controller *dev) > +{ > + return dev_get_drvdata(&dev->dev); > +} > + > +static inline void slim_set_ctrldata(struct slim_controller *dev, void *data) > +{ > + dev_set_drvdata(&dev->dev, data); > +} > + > +static inline void *slim_get_devicedata(const struct slim_device *dev) > +{ > + return dev_get_drvdata(&dev->dev); > +} > + > +static inline void slim_set_devicedata(struct slim_device *dev, void *data) > +{ > + dev_set_drvdata(&dev->dev, data); > +} > + > +#endif /* _LINUX_SLIMBUS_H */ > -- > 2.9.3 > > _______________________________________________ > Alsa-devel mailing list > Alsa-devel@alsa-project.org > http://mailman.alsa-project.org/mailman/listinfo/alsa-devel -- -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Thanks for the Review Bjorn, On 17/10/17 07:23, Bjorn Andersson wrote: > On Fri 06 Oct 08:51 PDT 2017, srinivas.kandagatla@linaro.org wrote: > [..] >> diff --git a/drivers/slimbus/slim-core.c b/drivers/slimbus/slim-core.c > [..] >> +/** >> + * Report callbacks(device_up, device_down) are implemented by slimbus-devices. >> + * The calls are scheduled into a workqueue to avoid holding up controller >> + * initialization/tear-down. >> + */ >> +static void schedule_slim_report(struct slim_controller *ctrl, >> + struct slim_device *sb, bool report) >> +{ >> + struct sb_report_wd *sbw; >> + >> + dev_dbg(&ctrl->dev, "report:%d for slave:%s\n", report, sb->name); > > This is the only place where sb->name is used in this driver. If you > instead invoke dev_*() on &sb->dev you should get prettier output and > can drop the double storage of the device name. Makes sense, we could get rid of sb->name storage too. > >> + >> + sbw = kmalloc(sizeof(*sbw), GFP_KERNEL); >> + if (!sbw) >> + return; >> + >> + INIT_WORK(&sbw->wd, slim_report); >> + sbw->sbdev = sb; >> + sbw->report = report; >> + if (!queue_work(ctrl->wq, &sbw->wd)) { > > When a controller is torn down destroy_workqueue() is called after all > child devices has been unregistered, so this work might be scheduled > after "sb" is gone, if I get this properly. I agree, That is possible! We should probably flush the workqueue before we start removing the clients. > >> + dev_err(&ctrl->dev, "failed to queue report:%d slave:%s\n", >> + report, sb->name); >> + kfree(sbw); >> + } >> +} >> + >> +static int slim_device_probe(struct device *dev) >> +{ >> + struct slim_device *sbdev; >> + struct slim_driver *sbdrv; >> + int status = 0; >> + >> + sbdev = to_slim_device(dev); >> + sbdrv = to_slim_driver(dev->driver); >> + >> + sbdev->driver = sbdrv; >> + >> + if (sbdrv->probe) >> + status = sbdrv->probe(sbdev); > > So a driver can have a probe() and device_up() or just any one of them? > > And probe() is called when the controller enumerates all devices > mentioned in DT and then device_up() is called at that point in time and > when it's advertised on the bus? > > Is there a reason for this split model? > yes, Some of the devices need to be powered up before they become usable, so probe is used to do the initial power up of the device. >> + >> + if (status) >> + sbdev->driver = NULL; >> + else if (sbdrv->device_up) >> + schedule_slim_report(sbdev->ctrl, sbdev, true); >> + >> + return status; >> +} >> + >> +static int slim_device_remove(struct device *dev) >> +{ >> + struct slim_device *sbdev; >> + struct slim_driver *sbdrv; >> + int status = 0; >> + >> + sbdev = to_slim_device(dev); >> + if (!dev->driver) >> + return 0; >> + >> + sbdrv = to_slim_driver(dev->driver); >> + if (sbdrv->remove) >> + status = sbdrv->remove(sbdev); >> + >> + mutex_lock(&sbdev->report_lock); >> + sbdev->notified = false; >> + if (status == 0) >> + sbdev->driver = NULL; >> + mutex_unlock(&sbdev->report_lock); >> + return status; > > device_unregister() will call device_del() which will end up in > __device_release_driver() which will call this function. Upon returning > from this function the core expect the bus to have cleaned up after the > dev (normally by calling drv->remove(dev)). > > It will completely ignore the return value and continue tearing down the > rest of the core resources, e.g. three lines down it will > devres_release_all(). > > > So you have the option of sleeping, while waiting for stuff to be > aborted/finished and then you need to clean things up. > > The slim_device object itself will stick around until all references are > dropped though. So you are suggesting that we make slim_driver remove not return anything? > >> +} >> + >> +struct bus_type slimbus_type = { >> + .name = "slimbus", >> + .match = slim_device_match, >> + .probe = slim_device_probe, >> + .remove = slim_device_remove, >> +}; >> +EXPORT_SYMBOL_GPL(slimbus_type); >> + >> +/** >> + * slim_driver_register: Client driver registration with slimbus >> + * @drv:Client driver to be associated with client-device. >> + * @owner: owning module/driver >> + * This API will register the client driver with the slimbus >> + * It is called from the driver's module-init function. >> + */ >> +int __slim_driver_register(struct slim_driver *drv, struct module *owner) >> +{ >> + drv->driver.bus = &slimbus_type; >> + drv->driver.owner = owner; >> + return driver_register(&drv->driver); >> +} >> +EXPORT_SYMBOL_GPL(__slim_driver_register); >> + >> +/** >> + * slim_driver_unregister: Undo effect of slim_driver_register >> + * @drv: Client driver to be unregistered >> + */ >> +void slim_driver_unregister(struct slim_driver *drv) >> +{ >> + if (drv) > > A driver invoking slim_driver_unregister(NULL) is broken, drop this > check and let it oops on the dereference instead. Yep. > >> + driver_unregister(&drv->driver); >> +} >> +EXPORT_SYMBOL_GPL(slim_driver_unregister); >> + >> +static struct slim_controller *slim_ctrl_get(struct slim_controller *ctrl) >> +{ >> + if (!ctrl || !get_device(&ctrl->dev)) > > ctrl can't be NULL here. In all code paths leading here it's > dereferenced multiple times already. I agree. > >> + return NULL; >> + >> + return ctrl; >> +} >> + >> +static void slim_ctrl_put(struct slim_controller *ctrl) >> +{ >> + if (ctrl) >> + put_device(&ctrl->dev); >> +} >> + >> +static void slim_dev_release(struct device *dev) >> +{ >> + struct slim_device *sbdev = to_slim_device(dev); >> + >> + slim_ctrl_put(sbdev->ctrl); > > As far as I can see there's no case where sbdev->ctrl will ever be NULL, > so yo can just replace this with > put_device(&ctrl->dev); > > And drop slim_ctrl_put(). Yes.. > >> + kfree(sbdev->name); >> + kfree(sbdev); >> +} >> + >> +static int slim_add_device(struct slim_controller *ctrl, >> + struct slim_device *sbdev) >> +{ >> + sbdev->dev.bus = &slimbus_type; >> + sbdev->dev.parent = &ctrl->dev; >> + sbdev->dev.release = slim_dev_release; >> + sbdev->dev.driver = NULL; >> + sbdev->ctrl = ctrl; >> + >> + slim_ctrl_get(ctrl); > > Unfolding slim_ctrl_get(), with ctrl != NULL, gives us a container_of(), > so this can't fail. Which is good because then an error check would have > been nice. > > But it also means that you can replace this with just: > get_device(&ctrl->dev); > > And drop slim_ctrl_get() > ya. >> + sbdev->name = kasprintf(GFP_KERNEL, "%x:%x:%x:%x", >> + sbdev->e_addr.manf_id, >> + sbdev->e_addr.prod_code, >> + sbdev->e_addr.dev_index, >> + sbdev->e_addr.instance); >> + if (!sbdev->name) >> + return -ENOMEM; >> + >> + dev_set_name(&sbdev->dev, "%s", sbdev->name); > > This will create another copy of the same string and as noted above > there seems to be only one consumer, which could be switched over. So > you can drop above kasprintf() and just use dev_set_name to format the > name. > > An added benefit is that you're not leaking the device reference from > slim_ctrl_get() on the line before. I agree, will fix this in next version. > >> + mutex_init(&sbdev->report_lock); >> + >> + /* probe slave on this controller */ >> + return device_register(&sbdev->dev); >> +} >> + >> +/* Helper to get hex Manufacturer ID and Product id from compatible */ >> +static unsigned long str2hex(unsigned char *str) > > The caller of this passes char *, so you can drop the unsigned. And add > "const" while you're at it. I will be removing this function in next version, replacing it with kstrtoul() > >> +{ >> + int value = 0; >> + >> + while (*str) { >> + char c = *str++; >> + >> + value = value << 4; >> + if (c >= '0' && c <= '9') >> + value |= (c - '0'); >> + if (c >= 'a' && c <= 'f') >> + value |= (c - 'a' + 10); > > At the cost of one more check here you can drop the line in the > documentation about this only working for lower-case hex digits. > >> + >> + } >> + >> + return value; >> +} >> + >> +/* OF helpers for SLIMbus */ >> +static void of_register_slim_devices(struct slim_controller *ctrl) >> +{ >> + struct device *dev = &ctrl->dev; >> + struct device_node *node; >> + >> + if (!ctrl->dev.of_node) >> + return; >> + >> + for_each_child_of_node(ctrl->dev.of_node, node) { >> + struct slim_device *slim; >> + const char *compat = NULL; >> + char *p, *tok; >> + int reg[2], ret; >> + >> + slim = kzalloc(sizeof(*slim), GFP_KERNEL); > > This is leaked in several places below. > >> + if (!slim) >> + continue; >> + >> + slim->dev.of_node = of_node_get(node); > > Dito. > >> + >> + compat = of_get_property(node, "compatible", NULL); >> + if (!compat) >> + continue; >> + >> + p = kasprintf(GFP_KERNEL, "%s", compat + strlen("slim")); > > Allocating a new string using string formatting based on an offset from > a string we don't know the size of, just to tokenize it does not seem > like the most efficient (nor safe) way of doing this. > > > How about: > ret = sscanf(compat, "slim%x,%x", &manf_id, &prod_code); > if (ret != 2) > error(); > Looks much better! >> + >> + tok = strsep(&p, ","); >> + if (!tok) { >> + dev_err(dev, "No valid Manufacturer ID found\n"); >> + kfree(p); >> + continue; >> + } >> + slim->e_addr.manf_id = str2hex(tok); >> + >> + tok = strsep(&p, ","); >> + if (!tok) { >> + dev_err(dev, "No valid Product ID found\n"); >> + kfree(p); >> + continue; >> + } >> + slim->e_addr.prod_code = str2hex(tok); >> + kfree(p); >> + >> + ret = of_property_read_u32_array(node, "reg", reg, 2); >> + if (ret) { >> + dev_err(dev, "Device and Instance id not found:%d\n", >> + ret); >> + continue; >> + } >> + slim->e_addr.dev_index = reg[0]; >> + slim->e_addr.instance = reg[1]; >> + >> + ret = slim_add_device(ctrl, slim); >> + if (ret) >> + dev_err(dev, "of_slim device register err:%d\n", ret); > > Cleanup if this fails. Yes, we are leaking memory here, will fix this in next version. > >> + } >> +} >> + >> +/** >> + * slim_register_controller: Controller bring-up and registration. >> + * @ctrl: Controller to be registered. >> + * A controller is registered with the framework using this API. >> + * If devices on a controller were registered before controller, >> + * this will make sure that they get probed when controller is up >> + */ >> +int slim_register_controller(struct slim_controller *ctrl) >> +{ >> + int id, ret = 0; >> + >> + mutex_lock(&slim_lock); >> + id = idr_alloc(&ctrl_idr, ctrl, ctrl->nr, -1, GFP_KERNEL); > > The purpose of ctrl_idr is to generate unique ids for the name and to > check that slim_del_controller() is only called on valid > slim_controllers. > > The latter is okay to just expect the controller drivers to do right and > the prior is better done with an ida. > I agree, IDA seems to be better fit here. > > Also, the lower boundary should be 0, not ctrl->nr. > >> + mutex_unlock(&slim_lock); >> + >> + if (id < 0) >> + return id; >> + >> + ctrl->nr = id; >> + >> + dev_set_name(&ctrl->dev, "sb-%d", ctrl->nr); > > This name is used in a lot of debug prints, can we do better? > In some of the previous discussions with Vinod, it was suggested that slim_register_controller should not create an additional device here, which is redundant to the actual controller device itself. So this name would actually come from the controller driver in next version of patches. >> + ctrl->num_dev = 0; >> + >> + if (!ctrl->min_cg) >> + ctrl->min_cg = SLIM_MIN_CLK_GEAR; >> + if (!ctrl->max_cg) >> + ctrl->max_cg = SLIM_MAX_CLK_GEAR; >> + >> + mutex_init(&ctrl->m_ctrl); >> + ret = device_register(&ctrl->dev); >> + if (ret) >> + goto dev_reg_failed; >> + >> + dev_dbg(&ctrl->dev, "Bus [%s] registered:dev:%p\n", >> + ctrl->name, &ctrl->dev); > > This is the only place ctrl->name is used. Perhaps it would make more > sense to base the dev_name off this string, to make all these dev_*() > more useful. Agreed. > >> + >> + ctrl->wq = create_singlethread_workqueue(dev_name(&ctrl->dev)); >> + if (!ctrl->wq) >> + goto err_workq_failed; >> + >> + of_register_slim_devices(ctrl); >> + >> + return 0; >> + >> +err_workq_failed: >> + device_unregister(&ctrl->dev); >> +dev_reg_failed: >> + mutex_lock(&slim_lock); >> + idr_remove(&ctrl_idr, ctrl->nr); >> + mutex_unlock(&slim_lock); >> + return ret; >> +} >> +EXPORT_SYMBOL_GPL(slim_register_controller); >> + >> +/* slim_remove_device: Remove the effect of slim_add_device() */ >> +static void slim_remove_device(struct slim_device *sbdev) >> +{ >> + device_unregister(&sbdev->dev); >> +} >> + >> +static int slim_ctrl_remove_device(struct device *dev, void *null) >> +{ >> + slim_remove_device(to_slim_device(dev)); >> + return 0; >> +} >> + >> +/** >> + * slim_del_controller: Controller tear-down. >> + * @ctrl: Controller to tear-down. >> + */ >> +int slim_del_controller(struct slim_controller *ctrl) > > This is the opposite of slim_register_controller() so should it perhaps > be called slim_unregister_controller() ? makes sense.. > >> +{ >> + struct slim_controller *found; >> + >> + /* First make sure that this bus was added */ >> + mutex_lock(&slim_lock); >> + found = idr_find(&ctrl_idr, ctrl->nr); >> + mutex_unlock(&slim_lock); >> + if (found != ctrl) >> + return -EINVAL; > > Just rely on the caller doing the right thing and just > ida_remove(&ctrl_ida, ctrl->nr); Moving to ida is best > >> + >> + /* Remove all clients */ >> + device_for_each_child(&ctrl->dev, NULL, slim_ctrl_remove_device); >> + > > As stated above there might be work items left in flight here, after the > slim_devices are released. > We should flush the workqueue before we remove the devices. >> >> + destroy_workqueue(ctrl->wq); >> + >> + /* free bus id */ >> + mutex_lock(&slim_lock); >> + idr_remove(&ctrl_idr, ctrl->nr); >> + mutex_unlock(&slim_lock); >> + >> + device_unregister(&ctrl->dev); >> + return 0; >> +} >> +EXPORT_SYMBOL_GPL(slim_del_controller); > [..] >> +/** >> + * slim_query_device: Query and get handle to a device. >> + * @ctrl: Controller on which this device will be added/queried >> + * @e_addr: Enumeration address of the device to be queried >> + * Returns pointer to a device if it has already reported. Creates a new >> + * device and returns pointer to it if the device has not yet enumerated. >> + */ >> +struct slim_device *slim_query_device(struct slim_controller *ctrl, >> + struct slim_eaddr *e_addr) >> +{ >> + struct device *dev; >> + struct slim_device *slim = NULL; > > This will be written before read, so please don't initialize. > yep. >> + >> + dev = device_find_child(&ctrl->dev, e_addr, slim_match_dev); >> + if (dev) { >> + slim = to_slim_device(dev); >> + return slim; >> + } >> + >> + slim = kzalloc(sizeof(struct slim_device), GFP_KERNEL); >> + if (IS_ERR(slim)) > > !slim yep. > >> + return NULL; >> + >> + slim->e_addr = *e_addr; >> + if (slim_add_device(ctrl, slim) != 0) { > > The idiomatic way is: > > ret = fn(); > if (ret) { > failure... > } > okay >> + kfree(slim); >> + return NULL; >> + } >> + return slim; >> +} >> +EXPORT_SYMBOL_GPL(slim_query_device); >> + >> +static int ctrl_getaddr_entry(struct slim_controller *ctrl, >> + struct slim_eaddr *eaddr, u8 *entry) >> +{ >> + int i; >> + >> + for (i = 0; i < ctrl->num_dev; i++) { >> + if (ctrl->addrt[i].valid && >> + slim_eaddr_equal(&ctrl->addrt[i].eaddr, eaddr)) { >> + *entry = i; > > i will be >= 0, so it can easily be distinguished from -ENXIO. So you > could return that directly instead of passing "i" as a reference and to > get the value. > > On the other hand the return value is only used as an offset in addrt[] > to read out addrt[i].laddr, so perhaps you should just return that > instead (as an int). > sounds sensible! >> + return 0; >> + } >> + } >> + return -ENXIO; >> +} >> + >> +/** >> + * slim_assign_laddr: Assign logical address to a device enumerated. > > So I presume this will either report a new (not seen before) e_addr > which should cause a new device to be spawned (although it might be > already mentioned in DT) or it might be called for an existing device to > update the logical address. > > Can you describe when the latter is the case? Or is this a side effect > of the code, rather than slimbus? If this function is called for an existing device with logical address, it would give the same logical address, if it finds a vaild one in the table. Other case is when it does not find a valid on then a new address is assigned depending on the valid flag passed to the function. This case is more to do with the Qualcomm B family Slim controller, where linux side can not assign logical address, As DSP would only be able to assign this address. >> + * @ctrl: Controller with which device is enumerated. >> + * @e_addr: Enumeration address of the device. >> + * @laddr: Return logical address (if valid flag is false) >> + * @valid: true if laddr holds a valid address that controller wants to >> + * set for this enumeration address. Otherwise framework sets index into >> + * address table as logical address. > > How do you ensure this laddr is unique? Its index into a table which is based on unique enumeration address. > >> + * Called by controller in response to REPORT_PRESENT. Framework will assign >> + * a logical address to this enumeration address. >> + * Function returns -EXFULL to indicate that all logical addresses are already >> + * taken. >> + */ >> +int slim_assign_laddr(struct slim_controller *ctrl, struct slim_eaddr *e_addr, >> + u8 *laddr, bool valid) >> +{ >> + int ret; >> + u8 i = 0; >> + bool exists = false; >> + struct slim_device *slim; >> + struct slim_addrt *temp; >> + >> + mutex_lock(&ctrl->m_ctrl); >> + /* already assigned */ >> + if (ctrl_getaddr_entry(ctrl, e_addr, &i) == 0) { > > So I presume this updates an existing e_addr -> laddr mapping. But this > should imply that there is an associated slim_device with this e_addr > and laddr. Shouldn't said slim_device have its laddr updated then? If we find a vaild an matching entry in the list, we assume that the slim device is already aware of this laddr. > >> + *laddr = ctrl->addrt[i].laddr; >> + exists = true; >> + } else { >> + if (ctrl->num_dev >= (SLIM_LA_MANAGER - 1)) { >> + ret = -EXFULL; >> + goto ret_assigned_laddr; >> + } >> + for (i = 0; i < ctrl->num_dev; i++) { >> + if (ctrl->addrt[i].valid == false) >> + break; >> + } >> + if (i == ctrl->num_dev) { >> + temp = krealloc(ctrl->addrt, >> + (ctrl->num_dev + 1) * >> + sizeof(struct slim_addrt), >> + GFP_KERNEL); >> + if (!temp) { >> + ret = -ENOMEM; >> + goto ret_assigned_laddr; >> + } >> + ctrl->addrt = temp; >> + ctrl->num_dev++; >> + } > > This seems better handled by a list than an array that we realloc. But > better yet, this array seems to mirror the list of registered devices. > > So why doesn't this function just try to resolve the slim_device with > e_addr and if not allocate a new slim_device with a free laddr. Using an > ida to keep track of used logical addresses would make this much > cleaner. > > Or a idr keyed by the laddr, it would still be O(n) to scan based on > e_addr, but this logic as well as lookups by laddr would be cleaner. > I will give this a try before sending next version and see how it looks! >> + ctrl->addrt[i].eaddr = *e_addr; >> + ctrl->addrt[i].valid = true; >> + >> + /* Preferred address is index into table */ >> + if (!valid) >> + *laddr = i; > > Is this laddr available? this is the new index or first instance of invalid entry in the table, so it should be available. > >> + } >> + >> + ret = ctrl->set_laddr(ctrl, &ctrl->addrt[i].eaddr, *laddr); >> + if (ret) { >> + ctrl->addrt[i].valid = false; >> + goto ret_assigned_laddr; >> + } >> + ctrl->addrt[i].laddr = *laddr; >> + >> +ret_assigned_laddr: >> + mutex_unlock(&ctrl->m_ctrl); >> + if (exists || ret) >> + return ret; >> + >> + dev_info(&ctrl->dev, "setting slimbus l-addr:%x, ea:%x,%x,%x,%x\n", >> + *laddr, e_addr->manf_id, e_addr->prod_code, >> + e_addr->dev_index, e_addr->instance); >> + >> + /** >> + * Add this device to list of devices on this controller if it's >> + * not already present >> + */ >> + slim = slim_query_device(ctrl, e_addr); >> + if (!slim) { >> + ret = -ENODEV; >> + } else { >> + struct slim_driver *sbdrv; >> + >> + slim->laddr = *laddr; >> + mutex_lock(&slim->report_lock); >> + slim->reported = true; >> + if (slim->dev.driver) { >> + sbdrv = to_slim_driver(slim->dev.driver); >> + if (sbdrv->device_up) >> + schedule_slim_report(ctrl, slim, true); >> + } >> + mutex_unlock(&slim->report_lock); > > I can't help feeling that this is the one and only point where you > should call probe() on the slim_device. Are there some funky > dependencies or protocol issues that makes this infeasible? The reason probe is called earlier because the slim device needs power-up or reset sequence to make it discoverable on the bus. > >> + } >> + return ret; >> +} >> +EXPORT_SYMBOL_GPL(slim_assign_laddr); >> + >> +/** >> + * slim_get_logical_addr: Return the logical address of a slimbus device. >> + * @sb: client handle requesting the address. >> + * @e_addr: Enumeration address of the device. >> + * @laddr: output buffer to store the address >> + * context: can sleep >> + * -EINVAL is returned in case of invalid parameters, and -ENXIO is returned if >> + * the device with this enumeration address is not found. >> + */ >> +int slim_get_logical_addr(struct slim_device *sb, struct slim_eaddr *e_addr, >> + u8 *laddr) > > In what case would e_addr != sb->e_addr and why can't this function just > be return sb->laddr? > I don't think e_addr and sb->eaddr should be different in this case. In usecase like B family Qualcomm SOCs, where linux cannot assign logical address, It needs to get logical address from ADSP and then use that. Current model of get_logical_addr provides abstraction so that this ADSP communication to get logical addr is hidden from client. >> +{ >> + int ret; >> + u8 entry; >> + struct slim_controller *ctrl = sb->ctrl; >> + >> + if (!ctrl || !laddr || !e_addr) >> + return -EINVAL; >> + >> + mutex_lock(&ctrl->m_ctrl); >> + ret = ctrl_getaddr_entry(ctrl, e_addr, &entry); >> + if (!ret) >> + *laddr = ctrl->addrt[entry].laddr; >> + mutex_unlock(&ctrl->m_ctrl); >> + >> + if (ret == -ENXIO && ctrl->get_laddr) { >> + ret = ctrl->get_laddr(ctrl, e_addr, laddr); >> + if (!ret) >> + ret = slim_assign_laddr(ctrl, e_addr, laddr, true); >> + } >> + return ret; >> +} >> +EXPORT_SYMBOL_GPL(slim_get_logical_addr); > [..] >> + >> +MODULE_LICENSE("GPL v2"); >> +MODULE_VERSION("0.1"); > > Who will ever update this version number? It's probably better to just > omit it. > Yep. >> +MODULE_DESCRIPTION("Slimbus module"); > > Rather than "Slimbus module", this is actually the "Slimbus core". > Yep. >> diff --git a/include/linux/mod_devicetable.h b/include/linux/mod_devicetable.h > [..] >> +extern struct bus_type slimbus_type; > > The device struct has a "struct device_type *type", which causes this > name to be confusing. Please rename it slimbus_bus instead. > > Why does this bus_type have to be known to the world? May be not. can be removed. > >> + >> +/* Standard values per SLIMbus spec needed by controllers and devices */ >> +#define SLIM_CL_PER_SUPERFRAME 6144 >> +#define SLIM_CL_PER_SUPERFRAME_DIV8 (SLIM_CL_PER_SUPERFRAME >> 3) >> +#define SLIM_MAX_CLK_GEAR 10 >> +#define SLIM_MIN_CLK_GEAR 1 >> +#define SLIM_CL_PER_SL 4 >> +#define SLIM_SL_PER_SUPERFRAME (SLIM_CL_PER_SUPERFRAME >> 2) >> +#define SLIM_FRM_SLOTS_PER_SUPERFRAME 16 >> +#define SLIM_GDE_SLOTS_PER_SUPERFRAME 2 > > Keep the min/max here and move the rest to the patches that introduce > consumers of them. makes sense! > > [..] >> +/* SLIMbus message types. Related to interpretation of message code. */ >> +#define SLIM_MSG_MT_CORE 0x0 >> +#define SLIM_MSG_MT_DEST_REFERRED_CLASS 0x1 >> +#define SLIM_MSG_MT_DEST_REFERRED_USER 0x2 >> +#define SLIM_MSG_MT_SRC_REFERRED_CLASS 0x5 >> +#define SLIM_MSG_MT_SRC_REFERRED_USER 0x6 > > These are not currently used, move them to the patch that actually use > them. > >> + >> +/* SLIMbus core type Message Codes. */ >> +/* Device management messages used by this framework */ >> +#define SLIM_MSG_MC_REPORT_PRESENT 0x1 >> +#define SLIM_MSG_MC_ASSIGN_LOGICAL_ADDRESS 0x2 >> +#define SLIM_MSG_MC_REPORT_ABSENT 0xF > > Dito > >> + >> +/* Destination type Values */ >> +#define SLIM_MSG_DEST_LOGICALADDR 0 >> +#define SLIM_MSG_DEST_ENUMADDR 1 >> +#define SLIM_MSG_DEST_BROADCAST 3 > > Dito will do it in next version. > >> + >> +/** >> + * struct slim_controller: Controls every instance of SLIMbus >> + * (similar to 'master' on SPI) >> + * 'Manager device' is responsible for device management, bandwidth >> + * allocation, channel setup, and port associations per channel. >> + * Device management means Logical address assignment/removal based on >> + * enumeration (report-present, report-absent) if a device. >> + * Bandwidth allocation is done dynamically by the manager based on active >> + * channels on the bus, message-bandwidth requests made by slimbus devices. >> + * Based on current bandwidth usage, manager chooses a frequency to run >> + * the bus at (in steps of 'clock-gear', 1 through 10, each clock gear >> + * representing twice the frequency than the previous gear). >> + * Manager is also responsible for entering (and exiting) low-power-mode >> + * (known as 'clock pause'). >> + * Manager can do handover of framer if there are multiple framers on the >> + * bus and a certain usecase warrants using certain framer to avoid keeping >> + * previous framer being powered-on. >> + * >> + * Controller here performs duties of the manager device, and 'interface >> + * device'. Interface device is responsible for monitoring the bus and >> + * reporting information such as loss-of-synchronization, data >> + * slot-collision. >> + * @dev: Device interface to this driver >> + * @nr: Board-specific number identifier for this controller/bus >> + * @list: Link with other slimbus controllers >> + * @name: Name for this controller >> + * @min_cg: Minimum clock gear supported by this controller (default value: 1) >> + * @max_cg: Maximum clock gear supported by this controller (default value: 10) >> + * @clkgear: Current clock gear in which this bus is running >> + * @a_framer: Active framer which is clocking the bus managed by this controller >> + * @m_ctrl: Mutex protecting controller data structures > > This mutex protects operations on the addrt array, so both name and > documentation can be improved. Agreed, will address this in next version. > >> + * @addrt: Logical address table > > Consider replacing with a idr, keyed by laddr. If there's actually a > point in having this list... > >> + * @num_dev: Number of active slimbus slaves on this bus > > This is not so much "number of devices", but rather the length of @addrt. lenght of addrt might be more than active devices as some of the entries in addrt are not valid after the device goes down. > >> + * @wq: Workqueue per controller used to notify devices when they report present >> + * @xfer_msg: Transfer a message on this controller (this can be a broadcast >> + * control/status message like data channel setup, or a unicast message >> + * like value element read/write. >> + * @set_laddr: Setup logical address at laddr for the slave with elemental >> + * address e_addr. Drivers implementing controller will be expected to >> + * send unicast message to this device with its logical address. >> + * @get_laddr: It is possible that controller needs to set fixed logical >> + * address table and get_laddr can be used in that case so that controller >> + * can do this assignment. > > Can you describe the use case for get_laddr() a little bit more? Usecase is for B family Qualcomm SOCs, where linux cannot assign logical address, It needs to get logical address from ADSP and then use that. Current model of get_logical_addr provides abstraction so that this ADSP communication to get logical addr is hidden from client. > >> + */ >> +struct slim_controller { >> + struct device dev; >> + unsigned int nr; >> + char name[SLIMBUS_NAME_SIZE]; >> + int min_cg; >> + int max_cg; >> + int clkgear; >> + struct slim_framer *a_framer; >> + struct mutex m_ctrl; >> + struct slim_addrt *addrt; >> + u8 num_dev; >> + struct workqueue_struct *wq; >> + int (*set_laddr)(struct slim_controller *ctrl, >> + struct slim_eaddr *ea, u8 laddr); >> + int (*get_laddr)(struct slim_controller *ctrl, >> + struct slim_eaddr *ea, u8 *laddr); > > If nothing else I think this should return the laddr, rather than pass > it back into the referenced u8. I will give it a go and see how it fits with every thing around it. > >> +}; >> + > > Regards, > Bjorn > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Thanks for Review Comments, On 18/10/17 07:15, Bjorn Andersson wrote: > On Fri 06 Oct 08:51 PDT 2017, srinivas.kandagatla@linaro.org wrote: > >> From: Sagar Dharia <sdharia@codeaurora.org> >> >> Slimbus devices use value-element, and information elements to >> control device parameters (e.g. value element is used to represent >> gain for codec, information element is used to represent interrupt >> status for codec when codec interrupt fires). >> Messaging APIs are used to set/get these value and information >> elements. Slimbus specification uses 8-bit "transaction IDs" for >> messages where a read-value is anticipated. Framework uses a table >> of pointers to store those TIDs and responds back to the caller in >> O(1). > > I think we can implement this "optimization" with less complex code, > regardless I don't think we need to mention this in the commit > message... > > [..] >> diff --git a/drivers/slimbus/slim-messaging.c b/drivers/slimbus/slim-messaging.c > [..] >> +/** >> + * slim_msg_response: Deliver Message response received from a device to the >> + * framework. >> + * @ctrl: Controller handle >> + * @reply: Reply received from the device >> + * @len: Length of the reply >> + * @tid: Transaction ID received with which framework can associate reply. >> + * Called by controller to inform framework about the response received. >> + * This helps in making the API asynchronous, and controller-driver doesn't need >> + * to manage 1 more table other than the one managed by framework mapping TID >> + * with buffers >> + */ >> +void slim_msg_response(struct slim_controller *ctrl, u8 *reply, u8 tid, u8 len) > > Even if tid and len comes from the spec I recommend you making them int > and size_t. okay, will give that a go. > >> +{ >> + struct slim_val_inf *msg; >> + unsigned long flags; >> + >> + spin_lock_irqsave(&ctrl->txn_lock, flags); >> + msg = ctrl->tid_tbl[tid]; >> + if (msg == NULL || msg->rbuf == NULL) { > > if (!msg || !msg->rbuf) > > > When is it valid to add a transaction to tid_tbl with msg->rbuf = NULL? > Should we reject it earlier? We do sanity checks before posting the request, however there are cases where this checks are not in place like calling slim_processtxn() directly. May be we should add this check all the case > >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + dev_err(&ctrl->dev, "Got response to invalid TID:%d, len:%d\n", >> + tid, len); >> + return; >> + } >> + ctrl->tid_tbl[tid] = NULL; >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + >> + memcpy(msg->rbuf, reply, len); >> + if (msg->comp_cb) >> + msg->comp_cb(msg->ctx, 0); >> +} >> +EXPORT_SYMBOL_GPL(slim_msg_response); > [..] >> +int slim_processtxn(struct slim_controller *ctrl, >> + struct slim_msg_txn *txn) >> +{ >> + int ret, i = 0; >> + unsigned long flags; >> + u8 *buf; >> + bool async = false; >> + struct slim_cb_data cbd; >> + DECLARE_COMPLETION_ONSTACK(done); >> + bool need_tid = slim_tid_txn(txn->mt, txn->mc); >> + >> + if (!txn->msg->comp_cb) { >> + txn->msg->comp_cb = slim_sync_default_cb; >> + cbd.comp = &done; >> + txn->msg->ctx = &cbd; >> + } else { >> + async = true; >> + } >> + >> + buf = slim_get_tx(ctrl, txn, need_tid); >> + if (!buf) >> + return -ENOMEM; >> + >> + if (need_tid) { >> + spin_lock_irqsave(&ctrl->txn_lock, flags); >> + for (i = 0; i < ctrl->last_tid; i++) { >> + if (ctrl->tid_tbl[i] == NULL) >> + break; >> + } >> + if (i >= ctrl->last_tid) { >> + if (ctrl->last_tid == (SLIM_MAX_TIDS - 1)) { >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + slim_return_tx(ctrl, -ENOMEM); >> + return -ENOMEM; >> + } >> + ctrl->last_tid++; >> + } >> + ctrl->tid_tbl[i] = txn->msg; >> + txn->tid = i; >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + } >> + >> + ret = ctrl->xfer_msg(ctrl, txn, buf); >> + >> + if (!ret && !async) { /* sync transaction */ >> + /* Fine-tune calculation after bandwidth management */ >> + unsigned long ms = txn->rl + 100; >> + >> + ret = wait_for_completion_timeout(&done, >> + msecs_to_jiffies(ms)); >> + if (!ret) >> + slim_return_tx(ctrl, -ETIMEDOUT); >> + >> + ret = cbd.ret; >> + } >> + >> + if (ret && need_tid) { >> + spin_lock_irqsave(&ctrl->txn_lock, flags); >> + /* Invalidate the transaction */ >> + ctrl->tid_tbl[txn->tid] = NULL; >> + spin_unlock_irqrestore(&ctrl->txn_lock, flags); >> + } >> + if (ret) >> + dev_err(&ctrl->dev, "Tx:MT:0x%x, MC:0x%x, LA:0x%x failed:%d\n", >> + txn->mt, txn->mc, txn->la, ret); > > if (ret) { > if (need_tid) > drop(); > > dev_err(); > } > > Would probably make this a little bit cleaner... I agree. > >> + if (!async) { >> + txn->msg->comp_cb = NULL; >> + txn->msg->ctx = NULL; > > I believe txn->msg is always required, so you don't need to do this > contidionally. I don't get this, why do you want to set comp_cb to NULL unconditionally? > >> + } >> + return ret; >> +} >> +EXPORT_SYMBOL_GPL(slim_processtxn); > [..] >> +int slim_request_val_element(struct slim_device *sb, >> + struct slim_val_inf *msg) >> +{ >> + struct slim_controller *ctrl = sb->ctrl; >> + >> + if (!ctrl) >> + return -EINVAL; > > From patch 1 I believe it's invalid for sb->ctrl to be NULL, so there > shouldn't be a need to check this. > yep. >> + >> + return slim_xfer_msg(ctrl, sb, msg, SLIM_MSG_MC_REQUEST_VALUE); >> +} >> +EXPORT_SYMBOL_GPL(slim_request_val_element); > [..] >> +int slim_return_rx(struct slim_controller *ctrl, void *buf) >> +{ >> + unsigned long flags; >> + >> + spin_lock_irqsave(&ctrl->rx.lock, flags); >> + if (ctrl->rx.tail == ctrl->rx.head) { >> + spin_unlock_irqrestore(&ctrl->rx.lock, flags); >> + return -ENODATA; >> + } >> + memcpy(buf, ctrl->rx.base + (ctrl->rx.head * ctrl->rx.sl_sz), >> + ctrl->rx.sl_sz); >> + ctrl->rx.head = (ctrl->rx.head + 1) % ctrl->rx.n; >> + spin_unlock_irqrestore(&ctrl->rx.lock, flags); >> + >> + return 0; >> +} >> +EXPORT_SYMBOL_GPL(slim_return_rx); >> + > > Please provide kerneldoc for exported symbols. > Yes, I will fix this in next version. >> +void slim_return_tx(struct slim_controller *ctrl, int err) >> +{ >> + unsigned long flags; >> + int idx; >> + struct slim_pending cur; >> + >> + spin_lock_irqsave(&ctrl->tx.lock, flags); >> + idx = ctrl->tx.head; >> + ctrl->tx.head = (ctrl->tx.head + 1) % ctrl->tx.n; >> + cur = ctrl->pending_wr[idx]; > > Why is this doing struct copy? > Not sure, do you see any issue with this? >> + spin_unlock_irqrestore(&ctrl->tx.lock, flags); >> + >> + if (!cur.cb) >> + dev_err(&ctrl->dev, "NULL Transaction or completion"); >> + else >> + cur.cb(cur.ctx, err); >> + >> + up(&ctrl->tx_sem); >> +} >> +EXPORT_SYMBOL_GPL(slim_return_tx); > [..] >> /** >> + * struct slim_val_inf: Slimbus value or information element >> + * @start_offset: Specifies starting offset in information/value element map >> + * @num_bytes: upto 16. This ensures that the message will fit the slicesize >> + * per slimbus spec >> + * @comp_cb: Callback if this read/write is asynchronous >> + * @ctx: Argument for comp_cb >> + */ >> +struct slim_val_inf { >> + u16 start_offset; >> + u8 num_bytes; >> + u8 *rbuf; > > This is not mentioned in the kerneldoc. Use void * for data buffers. > >> + const u8 *wbuf; > > Can a message ever be read and write? Otherwise it should be sufficient > to only have one data pointer. some of the SLIMBus commands are request_response type, meaning the old value is returned and at the same time the new value is updated. > >> + void (*comp_cb)(void *ctx, int err); >> + void *ctx; >> +}; >> + > [..] >> +/** >> + * struct slim_ctrl_buf: circular buffer used by contoller for TX, RX >> + * @base: virtual base address for this buffer >> + * @phy: physical address for this buffer (this is useful if controller can >> + * DMA the buffers for TX and RX to/from controller hardware >> + * @lock: lock protecting head and tail >> + * @head: index where buffer is returned back >> + * @tail: index from where buffer is consumed >> + * @sl_sz: byte-size of each slot in this buffer >> + * @n: number of elements in this circular ring, note that this needs to be >> + * 1 more than actual buffers to allow for one open slot >> + */ > > Is this ringbuffer mechanism defined in the slimbus specification? Looks > like something specific to the Qualcomm controller, rather than > something that should be enforced in the framework. > Yes, this is not part of the slimbus specs, but Qcom SOCs have concept of Message Queues. Are you suggesting that this buffer handling has to be moved out of core into controller driver? >> +struct slim_ctrl_buf { >> + void *base; >> + phys_addr_t phy; >> + spinlock_t lock; >> + int head; >> + int tail; >> + int sl_sz; >> + int n; >> +}; > [..] >> +/** >> * struct slim_controller: Controls every instance of SLIMbus >> * (similar to 'master' on SPI) >> * 'Manager device' is responsible for device management, bandwidth >> @@ -139,6 +246,16 @@ struct slim_addrt { >> * @addrt: Logical address table >> * @num_dev: Number of active slimbus slaves on this bus >> * @wq: Workqueue per controller used to notify devices when they report present >> + * @tid_tbl: Table of transactions having transaction ID >> + * @txn_lock: Lock to protect table of transactions >> + * @rx: RX buffers used by controller to receive messages. Ctrl may receive more >> + * than 1 message (e.g. multiple report-present messages or messages from >> + * multiple slaves). >> + * @tx: TX buffers used by controller to transmit messages. Ctrl may have >> + * ability to send/queue multiple messages to HW at once. >> + * @pending_wr: Pending write transactions to be acknowledged by controller > > This is out list of pending write requests, yet it's implemented as an > array used in a complex ring buffer fashion. Wouldn't it be easier to > just have this as a linked list of slim_pending struct? Yes, its possible to implement this as list, i will give that a try. > >> + * @tx_sem: Semaphore for available TX buffers for this controller >> + * @last_tid: Last used entry for TID transactions >> * @xfer_msg: Transfer a message on this controller (this can be a broadcast >> * control/status message like data channel setup, or a unicast message >> * like value element read/write. >> @@ -161,6 +278,15 @@ struct slim_controller { >> struct slim_addrt *addrt; >> u8 num_dev; >> struct workqueue_struct *wq; >> + struct slim_val_inf *tid_tbl[SLIM_MAX_TIDS]; >> + u8 last_tid; > > I suggest that you replace these two with an idr, rather than having a > fixed size array and then last_tid as an optimization to limit how far > you linear search for an empty space. Will try that and see how it looks! > >> + spinlock_t txn_lock; >> + struct slim_ctrl_buf tx; >> + struct slim_ctrl_buf rx; >> + struct slim_pending *pending_wr; >> + struct semaphore tx_sem; > > Please don't use semaphores. If you keep pending_wr as a list you can > use list_empty() instead... will give that a go. > >> + int (*xfer_msg)(struct slim_controller *ctrl, >> + struct slim_msg_txn *tx, void *buf); > > I believe buf has fixed size, so please document this. Yep. Will do that in next version. > >> int (*set_laddr)(struct slim_controller *ctrl, >> struct slim_eaddr *ea, u8 laddr); >> int (*get_laddr)(struct slim_controller *ctrl, >> @@ -295,5 +421,40 @@ static inline void slim_set_devicedata(struct slim_device *dev, void *data) >> { >> dev_set_drvdata(&dev->dev, data); >> } > > Regards, > Bjorn > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed 18 Oct 09:39 PDT 2017, Srinivas Kandagatla wrote: > Thanks for Review Comments, > > > On 18/10/17 07:15, Bjorn Andersson wrote: > > On Fri 06 Oct 08:51 PDT 2017, srinivas.kandagatla@linaro.org wrote: [..] > > > > > + if (!async) { > > > + txn->msg->comp_cb = NULL; > > > + txn->msg->ctx = NULL; > > > > I believe txn->msg is always required, so you don't need to do this > > contidionally. > > I don't get this, why do you want to set comp_cb to NULL unconditionally? > I'm just not happy about the complexity of this function, but perhaps it's confusing to always set them, regardless of them being used. Feel free to keep it. [..] > > > +void slim_return_tx(struct slim_controller *ctrl, int err) > > > +{ > > > + unsigned long flags; > > > + int idx; > > > + struct slim_pending cur; > > > + > > > + spin_lock_irqsave(&ctrl->tx.lock, flags); > > > + idx = ctrl->tx.head; > > > + ctrl->tx.head = (ctrl->tx.head + 1) % ctrl->tx.n; > > > + cur = ctrl->pending_wr[idx]; > > > > Why is this doing struct copy? > > > Not sure, do you see any issue with this? > It's a rarely used feature and I don't see a reason for using it here. It's probably better to make a copy of cur.cb and cur.ctx to make their use after the spin-unlock more obvious (but should be fine as the spinlock is for the pending_wr array. > > > + spin_unlock_irqrestore(&ctrl->tx.lock, flags); > > > + > > > + if (!cur.cb) > > > + dev_err(&ctrl->dev, "NULL Transaction or completion"); > > > + else > > > + cur.cb(cur.ctx, err); > > > + > > > + up(&ctrl->tx_sem); > > > +} > > > +EXPORT_SYMBOL_GPL(slim_return_tx); [..] > > > +/** > > > + * struct slim_ctrl_buf: circular buffer used by contoller for TX, RX > > > + * @base: virtual base address for this buffer > > > + * @phy: physical address for this buffer (this is useful if controller can > > > + * DMA the buffers for TX and RX to/from controller hardware > > > + * @lock: lock protecting head and tail > > > + * @head: index where buffer is returned back > > > + * @tail: index from where buffer is consumed > > > + * @sl_sz: byte-size of each slot in this buffer > > > + * @n: number of elements in this circular ring, note that this needs to be > > > + * 1 more than actual buffers to allow for one open slot > > > + */ > > > > Is this ringbuffer mechanism defined in the slimbus specification? Looks > > like something specific to the Qualcomm controller, rather than > > something that should be enforced in the framework. > > > > Yes, this is not part of the slimbus specs, but Qcom SOCs have concept of > Message Queues. > > Are you suggesting that this buffer handling has to be moved out of core > into controller driver? > The fact that this seems to describe a physical ring buffer, with some set of properties that are related to how a ring buffer works in the Qualcomm hardware and it carries a notion of physical mapping, all indicates to me that this describes some Qualcomm hardware interface. I believe this is a hardware implementation detail that should reside in the hardware part of the implementation (i.e. the Qualcomm driver). > > > > +struct slim_ctrl_buf { > > > + void *base; > > > + phys_addr_t phy; > > > + spinlock_t lock; > > > + int head; > > > + int tail; > > > + int sl_sz; > > > + int n; > > > +}; Regards, Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
From: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> It's been very long time since there was any activity on the slimbus patches, Am currently working on getting Qualcomm DSP based audio working on top of mainline. Slimbus is one of the major component for getting any analog audio on QCOM SoC's. So am taking intiative to address the review comments on the older patchset and send it. I have tested this patch on IFC6410 board with wcd9310 codec. SLIMbus (Serial Low Power Interchip Media Bus) is a specification developed by MIPI (Mobile Industry Processor Interface) alliance. SLIMbus is a 2-wire implementation, which is used to communicate with peripheral components like audio-codec. SLIMbus uses Time-Division-Multiplexing to accommodate multiple data channels, and control channel. Control channel has messages to do device-enumeration, messages to send/receive control-data to/from slimbus devices, messages for port/channel management, and messages to do bandwidth allocation. Framework is introduced to support multiple instances of the bus (1 controller per bus), and multiple slave devices per controller. SPI and I2C frameworks, and comments from last time when I submitted the patches were referred-to while working on this framework. These patchsets introduce device-management, OF helpers, and messaging APIs, controller driver for Qualcomm's slimbus controller, and clock-pause feature for entering/exiting low-power mode for SLIMbus. Framework patches to do channel, port and bandwidth management are work-in-progress and will be sent out once these initial patches are accepted. These patchsets were tested on Qualcomm Snapdragon processor board using the controller driver, and a test slave device. Changes from V5 to V6: * aligned slim_driver_register more like other buses, suggested by Arnd. * removed boardinfo and add_device apis for now, suggested by Arnd * Few namespace cleanups suggested by Masami * merged of apis in to first patch as suggested by Arnd. * slimbus clients "compatible" name space made much inline with USB and PCIE, suggested by Rob and Arnd. * Removed memory allocations to controller drivers, as suggested by Arnd. * Various bindings comments addressed as suggested by Mark and others. * Added regmap interface so that codecs can write more generic code. * Added MAINTAINER file. Sagar Dharia (5): slimbus: Device management on SLIMbus slimbus: Add messaging APIs to slimbus framework slimbus: qcom: Add Qualcomm Slimbus controller driver slimbus: Add support for 'clock-pause' feature slimbus: qcom: Add runtime-pm support using clock-pause feature Srinivas Kandagatla (2): regmap: add SLIMBUS support MAINTAINERS: Add SLIMbus maintainer Documentation/devicetree/bindings/slimbus/bus.txt | 57 ++ .../devicetree/bindings/slimbus/slim-qcom-ctrl.txt | 43 ++ Documentation/slimbus/summary | 109 ++++ MAINTAINERS | 8 + drivers/Kconfig | 2 + drivers/Makefile | 1 + drivers/base/regmap/Kconfig | 4 + drivers/base/regmap/Makefile | 1 + drivers/base/regmap/regmap-slimbus.c | 89 +++ drivers/slimbus/Kconfig | 20 + drivers/slimbus/Makefile | 8 + drivers/slimbus/slim-core.c | 725 +++++++++++++++++++++ drivers/slimbus/slim-messaging.c | 509 +++++++++++++++ drivers/slimbus/slim-qcom-ctrl.c | 714 ++++++++++++++++++++ drivers/slimbus/slim-qcom.h | 64 ++ drivers/slimbus/slim-sched.c | 126 ++++ include/linux/mod_devicetable.h | 13 + include/linux/regmap.h | 18 + include/linux/slimbus.h | 512 +++++++++++++++ 19 files changed, 3023 insertions(+) create mode 100644 Documentation/devicetree/bindings/slimbus/bus.txt create mode 100644 Documentation/devicetree/bindings/slimbus/slim-qcom-ctrl.txt create mode 100644 Documentation/slimbus/summary create mode 100644 drivers/base/regmap/regmap-slimbus.c create mode 100644 drivers/slimbus/Kconfig create mode 100644 drivers/slimbus/Makefile create mode 100644 drivers/slimbus/slim-core.c create mode 100644 drivers/slimbus/slim-messaging.c create mode 100644 drivers/slimbus/slim-qcom-ctrl.c create mode 100644 drivers/slimbus/slim-qcom.h create mode 100644 drivers/slimbus/slim-sched.c create mode 100644 include/linux/slimbus.h -- 2.9.3 -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html