On 4/2/2013 4:08 PM, Ulf Hansson wrote:
On 1 April 2013 10:28, Asutosh Das <asutoshd@xxxxxxxxxxxxxx> wrote:On 3/6/2013 12:27 PM, Ulf Hansson wrote:On 6 March 2013 02:04, Asutosh Das <asutoshd@xxxxxxxxxxxxxx> wrote:On 3/5/2013 7:09 AM, Ulf Hansson wrote:On 4 March 2013 21:48, Asutosh Das <asutoshd@xxxxxxxxxxxxxx> wrote:On 3/1/2013 6:17 PM, Ulf Hansson wrote:From: Ulf Hansson <ulf.hansson@xxxxxxxxxx> Once the mmc blkdevice is being probed, runtime pm will be enabled. By using runtime autosuspend, the power save operations can be done when request inactivity occurs for a certain time. Right now the selected timeout value is set to 3 s. Moreover, when the blk device is being suspended, we make sure the device will be runtime resumed. The reason for doing this is that we what the host suspend sequence to be unaware of any runtime power save operations, so it can just handle the suspend as the device is fully powerered from a runtime perspective. This patch is preparing to make it possible to move BKOPS handling into the runtime callbacks for the mmc bus_ops. Thus IDLE BKOPS can be accomplished. Signed-off-by: Ulf Hansson <ulf.hansson@xxxxxxxxxx> --- drivers/mmc/card/block.c | 28 ++++++++++++++++++++++++++-- 1 file changed, 26 insertions(+), 2 deletions(-) diff --git a/drivers/mmc/card/block.c b/drivers/mmc/card/block.c index 5bab73b..89d1c39 100644 --- a/drivers/mmc/card/block.c +++ b/drivers/mmc/card/block.c @@ -34,6 +34,7 @@ #include <linux/delay.h> #include <linux/capability.h> #include <linux/compat.h> +#include <linux/pm_runtime.h> #include <linux/mmc/ioctl.h> #include <linux/mmc/card.h> @@ -222,6 +223,7 @@ static ssize_t power_ro_lock_store(struct device *dev, md = mmc_blk_get(dev_to_disk(dev)); card = md->queue.card; + pm_runtime_get_sync(&card->dev); mmc_claim_host(card->host); ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL, EXT_CSD_BOOT_WP, @@ -234,6 +236,8 @@ static ssize_t power_ro_lock_store(struct device *dev, card->ext_csd.boot_ro_lock |= EXT_CSD_BOOT_WP_B_PWR_WP_EN; mmc_release_host(card->host); + pm_runtime_mark_last_busy(&card->dev); + pm_runtime_put_autosuspend(&card->dev); if (!ret) { pr_info("%s: Locking boot partition ro until next power on\n", @@ -492,6 +496,7 @@ static int mmc_blk_ioctl_cmd(struct block_device *bdev, mrq.cmd = &cmd; + pm_runtime_get_sync(&card->dev); mmc_claim_host(card->host); err = mmc_blk_part_switch(card, md); @@ -560,6 +565,8 @@ static int mmc_blk_ioctl_cmd(struct block_device *bdev, cmd_rel_host: mmc_release_host(card->host); + pm_runtime_mark_last_busy(&card->dev); + pm_runtime_put_autosuspend(&card->dev); cmd_done: mmc_blk_put(md); @@ -1894,9 +1901,11 @@ static int mmc_blk_issue_rq(struct mmc_queue *mq, struct request *req) struct mmc_host *host = card->host; unsigned long flags; - if (req && !mq->mqrq_prev->req) + if (req && !mq->mqrq_prev->req) { + pm_runtime_get_sync(&card->dev); /* claim host only for the first request */ mmc_claim_host(card->host); + } ret = mmc_blk_part_switch(card, md); if (ret) { @@ -1932,9 +1941,13 @@ static int mmc_blk_issue_rq(struct mmc_queue *mq, struct request *req) } out: - if (!req && !(mq->flags & MMC_QUEUE_NEW_REQUEST)) + if (!req && !(mq->flags & MMC_QUEUE_NEW_REQUEST)) { /* release host only when there are no more requests */ mmc_release_host(card->host); + pm_runtime_mark_last_busy(&card->dev); + pm_runtime_put_autosuspend(&card->dev); + } + return ret; } @@ -2331,6 +2344,12 @@ static int mmc_blk_probe(struct mmc_card *card) if (mmc_add_disk(part_md)) goto out; } + + pm_runtime_set_active(&card->dev); + pm_runtime_set_autosuspend_delay(&card->dev, 3000); + pm_runtime_use_autosuspend(&card->dev); + pm_runtime_enable(&card->dev); + return 0; out: @@ -2344,9 +2363,12 @@ static void mmc_blk_remove(struct mmc_card *card) struct mmc_blk_data *md = mmc_get_drvdata(card); mmc_blk_remove_parts(card, md); + pm_runtime_get_sync(&card->dev); mmc_claim_host(card->host); mmc_blk_part_switch(card, md); mmc_release_host(card->host); + pm_runtime_disable(&card->dev); + pm_runtime_put_noidle(&card->dev); mmc_blk_remove_req(md); mmc_set_drvdata(card, NULL); } @@ -2358,6 +2380,7 @@ static int mmc_blk_suspend(struct mmc_card *card) struct mmc_blk_data *md = mmc_get_drvdata(card); if (md) { + pm_runtime_get_sync(&card->dev); mmc_queue_suspend(&md->queue); list_for_each_entry(part_md, &md->part, part) { mmc_queue_suspend(&part_md->queue); @@ -2381,6 +2404,7 @@ static int mmc_blk_resume(struct mmc_card *card) list_for_each_entry(part_md, &md->part, part) { mmc_queue_resume(&part_md->queue); } + pm_runtime_put(&card->dev); } return 0; }Hi Asutosh,Hi Ulf Currently, I am implementing a patch that facilitates each device to manage is runtime pm on its own. I am using the parent-child relationship that is already established in the mmc stack for this implementation. In this case, mmc_card is a child of mmc_host which is in turn is the child of platform-device. The following contexts exist which would have to invoke get_sync and put_sync on the mmc_card device: 1. mmcqd 2. bkops 3. mmc_rescan The get_sync on card device would resume all the 3 devices starting from the platform-device and the put-sync would suspend all the 3 devices starting from the card-device. pm_auto_suspend/pm_schedule_suspend may be used to trigger the suspend from the above contexts. I believe this would simplify the runtime pm functionality. I can put up the patch for review in a couple of days. Please let me know your opinion about this approach.No, I don't think this is way of doing it. I will try to elaborate a bit with the approach I took in my patchset and why. First of all, the host platform device must be kept separate (no parent/child and not the same bus) from the card device. There is many reason to why, but within this context (runtime pm point of view), the host must be able to handle it's runtime pm resources completely separate from the blk device (card device) runtime resources. More precisely and from a BKOPS point of view; while doing BKOPS the host platform device must still be able to enter runtime power save states. I realize that in your case, you are doing mmc_suspend|resume_host in your host drivers runtime callbacks, which is kind of a very special case, though worth to consider of course. There are two solutions to enable the option for this functionality. In both cases a host caps flag will be needed to enable this. 1. In mmc bus_ops runtime callback, do a pm_runtime_get_sync("host plf device"), and vice verse in the runtime resume callback. This will prevent the host driver from entering runtime power save sate while for example doing BKOPS, thus preventing your host driver from doing mmc_suspend_host while BKOPS is running. 2. Move mmc_suspend|resume_host from your host runtime callbacks, into the bus_ops runtime callbacks. Typically, when no BKOPS is needed mmc_suspend_host can be done. What are you thoughts around this? Kind regards Ulf Hansson-- Thanks Asutosh -- Sent by a consultant of the Qualcomm Innovation Center, Inc. The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.Hi Ulf, Typically, when the pltfm device enters power-save during bkops, it can only shut-off the clocks to the card (?), the power would still be on. With clock-gating in place, this would be done anyhow.Clock gating is only one of the things you might want to do. For example; Your host plf device can reside in a power domain. You might want to save power by doing pinctrl. Disable/enable irqs. Etc. So, to conclude it's is realy important runtime pm for the block device (card device) is kept separate from the host plf device. They do handle completly different stuff.I wanted to separate the functionality as detailed below: Say we do aggressive power management: (invoke mmc_[suspend/resume]_host in runtime pm as well), idle-timeout of 10 s (configurable though) during suspend of mmc_card device: - do all card related power management, like power-off notifications etc. during suspend of mmc_host device: - do all the host related power management, like power-off host, shut-off clocks etc. during suspend of pltfm device: - do all the pltfm specific power management, like disable irqs, configure wake-ups etc. During system-suspend, it can be checked if the device is runtime-suspended, if yes, return. On resume, the above path would be retraced in the reverse order. I think each bus can be responsible for suspending its devices during system suspend.According to my comment above, I don't think this sequence will be possible.First of all, the host platform device must be kept separate (no parent/child and not the same bus) from the card device.Can you please elaborate on this point a bit.See above comment for what actions a host plf device can do at runtime power management.I guess you would be joining the IRC chat tomorrow, if yes, we can discuss there itself.It wont be possible for me to joing IRC today, I am at Hongkong with Linaro Connect this week. Although it seems like a good discussion on IRC, I suppose it would help to clarify on this topic.-- Sent by a consultant of the Qualcomm Innovation Center, Inc. The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.Kind regards Ulf HanssonHi Ulf Sorry for the delayed response. The card and host are anyhow coupled except in the idle-time bkops case, which is kind of a special case. I think this can be handled in the approach I am suggesting by not invoking mmc_suspend_host if bkops is running.In this way, the pltfm device can still be put in low-power mode (like you suggested), when card is doing bkops.A running BKOPS must not prevent the system from suspend (host driver calls mmc_suspend_host). Thus if we are running a BKOPS when system suspend occurs, we need to interrupt the BKOPS (send HPI). So I think this will not work properly, unless I missed something here. Moreover as I also stated earlier, the pm_runtime_get_sync of the card device when the card device enters suspend state, will accomplish just that if BKOPS is adapted properly.Moreover, the parent-child relationship is already established (in mmc_alloc_host and mmc_alloc_card) and runtime-pm framework only uses it if we enable runtime-pm of the particular device. Even now, we first put the card to sleep and then the host.I am not sure I understand what you are proposing. In what way should should I change my patch?What are your thoughts on this ? -- Sent by a consultant of the Qualcomm Innovation Center, Inc. The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.Some final thoughts, this patchset has been around for some time now. We have mainly been discussing hypothetical issues which relates to "aggresive card suspend", for which code right now not even exists in the mainline kernel. Nevertheless very good discussions, but I think we shoud be able to handle all these additional features as new patches build on top of this patchset, don't you think?
Thanks Ulf, we got your point.I guess you may agree that even if none of the in-tree drivers implementaing "agreesive card suspend" during runtime suspend, its worth considering it and lets atleast give an option (defining new cap for agressive RPM) to choose to different flavours of host controller drivers. I guess other host controllers will also benefit from this. If you are interested to look at the power savings achieved with this agreesive RPM, we may post the same.
But main point is to keep this aggressive power management requirement in mind, when reviewing this patch or any additional enhacements coming in on top of this patch.
Regards, Subhash
It is always easier to disuss around real patches, but discussing hypothetical issues, if you see what mean. Kind regards Ulf Hansson -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
-- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
