Hi Josh, On 05/05/2014 04:54 PM, Josh Cartwright wrote: > On Mon, May 05, 2014 at 04:44:25PM -0500, Suman Anna wrote: >> Hi Rob, >> >> On 04/30/2014 07:34 PM, Suman Anna wrote: >>> The property 'hwlock-reserved-locks' will be used to represent >>> the number of locks to be reserved for clients that would need >>> to request/operate on specific locks. A new OF helper function, >>> of_hwspin_lock_get_num_reserved_locks(), is added to minimize >>> duplication in different platform implementations. >>> >>> The function will return a value of 0 if the property is not >>> defined, so as to support a default behavior of marking all >>> locks as unused and open for anonymous allocations. >>> >>> Signed-off-by: Suman Anna <s-anna@xxxxxx> >>> --- >>> .../devicetree/bindings/hwlock/hwlock.txt | 3 +++ >>> drivers/hwspinlock/hwspinlock_core.c | 25 ++++++++++++++++++++++ >>> include/linux/hwspinlock.h | 1 + >>> 3 files changed, 29 insertions(+) >>> >>> diff --git a/Documentation/devicetree/bindings/hwlock/hwlock.txt b/Documentation/devicetree/bindings/hwlock/hwlock.txt >>> index d538a9b..88d16d2 100644 >>> --- a/Documentation/devicetree/bindings/hwlock/hwlock.txt >>> +++ b/Documentation/devicetree/bindings/hwlock/hwlock.txt >>> @@ -18,6 +18,9 @@ Common properties: >>> property is needed on hwlock devices, where the number >>> of supported locks within a hwlock device cannot be >>> read from a register. >>> +- hwlock-reserved-locks: Number of locks to reserve for clients requiring >>> + specific locks. This value cannot exceed the value of >>> + hwlock-num-locks. >> >> Any suggestions here on the approach? This property falls into a gray >> area as well, as the current approach is somewhat limiting (it doesn't >> support sparse reserved locks, and expects them at the beginning of the >> lock range). > > Is it possible to implement a pinctrl-like hogging approach, whereby the > specific locks that need to be reserved are consumed by the controller > itself? > Thanks for the suggestion. I did take a look at pinctrl and while it is possible to implement something similar, I feel it is a bit heavy for hwspinlock framework with no added advantages. It requires that the controller and clients always need to be updated together. Ohad had already brought this up [1]. Here's an alternate approach that does not require any additional property to the controller itself, while the client node usage is as before. The logic is based on parsing through the DT blob and marking only those that are used by any clients. The RFC patch below is a replacement for Patches 11 to 15, and do not require any changes to platform implementations or additional DT properties. It currently marks locks as reserved even for disabled client nodes (very easy to change that behavior). It will also impose a standard property name "hwlocks" on the client nodes. What do you think of this approach? regards Suman [1] http://marc.info/?l=linux-omap&m=139514977622964&w=2 ---- >From 4f4cbe91e56c1be8faa6a3ee863add4df6e6714b Mon Sep 17 00:00:00 2001 From: Suman Anna <s-anna@xxxxxx> Date: Fri, 9 May 2014 14:26:54 -0500 Subject: [RFC PATCH] hwspinlock/core: add support for reserved locks The HwSpinlock core allows requesting either a specific lock or an available normal lock. The specific locks are usually reserved during board init time, while the normal available locks are intended to be assigned at runtime. The HwSpinlock core has been enhanced to mark certain locks as 'reserved' by parsing through the DT blob. Thes The HwSpinlock core has been enhanced to: 1. mark certain locks as 'reserved' by parsing the DT blob for any locks used by client nodes. 2. restrict the anonymous hwspin_lock_request() API to allocate only from non-reserved locks for DT boots. 3. limit these reserved locks to be allocated only using the _request_specific() API variants for DT boots. Signed-off-by: Suman Anna <s-anna@xxxxxx> --- drivers/hwspinlock/hwspinlock_core.c | 50 ++++++++++++++++++++++++++++++++++-- 1 file changed, 48 insertions(+), 2 deletions(-) diff --git a/drivers/hwspinlock/hwspinlock_core.c b/drivers/hwspinlock/hwspinlock_core.c index c2063bc..0c924c9 100644 --- a/drivers/hwspinlock/hwspinlock_core.c +++ b/drivers/hwspinlock/hwspinlock_core.c @@ -425,6 +425,42 @@ static int hwspinlock_device_add(struct hwspinlock_device *bank) return ret; } +static void hwspin_mark_reserved_locks(struct hwspinlock_device *bank) +{ + struct device_node *np = bank->dev->of_node; + const char *prop_name = "hwlocks"; + const char *cells_name = "#hwlock-cells"; + struct device_node *node = NULL; + struct of_phandle_args args; + struct hwspinlock *hwlock; + int i, id, count, ret; + + for_each_node_with_property(node, prop_name) { + count = of_count_phandle_with_args(node, prop_name, cells_name); + if (count <= 0) + continue; + + for (i = 0; i < count; i++) { + ret = of_parse_phandle_with_args(node, prop_name, + cells_name, i, &args); + if (ret || np != args.np) + continue; + + id = bank->ops->of_xlate(bank, &args); + if (id < 0 || id >= bank->num_locks) + continue; + + hwlock = &bank->lock[id]; + if (hwlock->type == HWSPINLOCK_RESERVED) { + dev_err(bank->dev, "potential reuse of hwspinlock %d between multiple clients on %s\n", + id, np->full_name); + continue; + } + hwlock->type = HWSPINLOCK_RESERVED; + } + } +} + /** * hwspin_lock_register() - register a new hw spinlock device * @bank: the hwspinlock device, which usually provides numerous hw locks @@ -463,12 +499,16 @@ int hwspin_lock_register(struct hwspinlock_device *bank, struct device *dev, if (ret) return ret; + if (dev->of_node) + hwspin_mark_reserved_locks(bank); + for (i = 0; i < num_locks; i++) { hwlock = &bank->lock[i]; spin_lock_init(&hwlock->lock); hwlock->bank = bank; - hwlock->type = HWSPINLOCK_UNUSED; + if (hwlock->type != HWSPINLOCK_RESERVED) + hwlock->type = HWSPINLOCK_UNUSED; ret = hwspin_lock_register_single(hwlock, base_id + i); if (ret) @@ -651,7 +691,13 @@ struct hwspinlock *hwspin_lock_request_specific(unsigned int id) /* sanity check (this shouldn't happen) */ WARN_ON(hwlock_to_id(hwlock) != id); - /* make sure this hwspinlock is unused */ + if (hwlock->type != HWSPINLOCK_RESERVED) { + pr_warn("hwspinlock %u is not a reserved lock\n", id); + hwlock = NULL; + goto out; + } + + /* make sure this hwspinlock is an unused reserved lock */ ret = radix_tree_tag_get(&hwspinlock_tree, id, hwlock->type); if (ret == 0) { pr_warn("hwspinlock %u is already in use\n", id); -- 1.9.2 -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
