Message ID | 1437757979-25889-1-git-send-email-shenwei.wang@freescale.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 24/07/15 10:12, Shenwei Wang wrote: > When system goes into low power states like SUSPEND_MEM and > HIBERNATION, the hardware IP block may be powered off to reduce > the power consumption. This power down will lost all the > data inside the ram. This patch added the dev_pm_ops and > implemented two callbacks: suspend_noirq and resume_noirq, which > will save the data in the on-chip-ram right before power down > and restore it after system resumes. > > A new property string named "can-power-gate" is added to > the devicetree bindings too. > > Signed-off-by: Shenwei Wang <shenwei.wang@freescale.com> > Signed-off-by: Anson Huang <b20788@freescale.com> > --- [snip] > +static int sram_suspend_noirq(struct device *dev) > +{ > + struct platform_device *pdev = to_platform_device(dev); > + struct sram_dev *sram = platform_get_drvdata(pdev); > + > + if (!sram->power_off_save) > + return 0; > + > + /* Save necessary regs */ > + clk_enable(sram->clk); > + memcpy(sram->power_off_save, sram->virt_base,gen_pool_size(sram->pool)); > + clk_disable(sram->clk); memcpy_fromio()? > + > + return 0; > +} > + > +static int sram_resume_noirq(struct device *dev) > +{ > + struct platform_device *pdev = to_platform_device(dev); > + struct sram_dev *sram = platform_get_drvdata(pdev); > + > + if (!sram->power_off_save) > + return 0; > + > + clk_enable(sram->clk); > + memcpy(sram->virt_base, sram->power_off_save, gen_pool_size(sram->pool)); memcpy_toio()? > + clk_disable(sram->clk); > + > + return 0; > +} > + > static int sram_probe(struct platform_device *pdev) > { > struct sram_dev *sram; > @@ -203,6 +235,12 @@ static int sram_probe(struct platform_device *pdev) > > platform_set_drvdata(pdev, sram); > > + if (of_get_property(pdev->dev.of_node, "can-power-gate", NULL)) Since this is a boolean you could use of_property_read_bool() > + { > + sram->power_off_save = devm_kzalloc(&pdev->dev, > + gen_pool_size(sram->pool), GFP_KERNEL); Parenthesis not needed. This will only work with SRAM sizes which are small enough they can be serviced by kzalloc() allocations. The problem is that you might actually need direct-mapped memory but what if you have larger SRAMs that are several MB? Maybe there should be a check here to prevent that. > + } > + > dev_dbg(sram->dev, "SRAM pool: %zu KiB @ 0x%p\n", > gen_pool_size(sram->pool) / 1024, sram->virt_base); > > @@ -229,10 +267,16 @@ static const struct of_device_id sram_dt_ids[] = { > }; > #endif > > +static const struct dev_pm_ops sram_pm_ops = { > + .suspend_noirq = sram_suspend_noirq, > + .resume_noirq = sram_resume_noirq, > +}; You may have to enclose this within an ifdef CONFIG_PM_SLEEP/endif statement not to cause warnings if CONFIG_PM_SLEEP is disabled.
> -----Original Message----- > From: Florian Fainelli [mailto:f.fainelli@gmail.com] > Sent: 2015?7?24? 15:17 > To: Wang Shenwei-B38339; gregkh@linuxfoundation.org; arnd@arndb.de > Cc: Huang Yongcai-B20788; linux-arm-kernel@lists.infradead.org; > computersforpeace@gmail.com > Subject: Re: [PATCH 1/1] misc: sram: add dev_pm_ops to support module power > > + /* Save necessary regs */ > > + clk_enable(sram->clk); > > + memcpy(sram->power_off_save, > sram->virt_base,gen_pool_size(sram->pool)); > > + clk_disable(sram->clk); > > memcpy_fromio()? > memcpy is safe to call here, and it should be more efficient. > > + > > + clk_enable(sram->clk); > > + memcpy(sram->virt_base, sram->power_off_save, > > +gen_pool_size(sram->pool)); > > memcpy_toio()? > When system goes to suspend state, it will flush the cache in the end. Memcpy here is more efficient. > > struct sram_dev *sram; > > @@ -203,6 +235,12 @@ static int sram_probe(struct platform_device > > *pdev) > > > > platform_set_drvdata(pdev, sram); > > > > + if (of_get_property(pdev->dev.of_node, "can-power-gate", NULL)) > > Since this is a boolean you could use of_property_read_bool() > Yes, of_property_read_bool is better. Will adopt it. > > + { > > + sram->power_off_save = devm_kzalloc(&pdev->dev, > > + gen_pool_size(sram->pool), GFP_KERNEL); > > Parenthesis not needed. Will remove it. > > This will only work with SRAM sizes which are small enough they can be serviced > by kzalloc() allocations. The problem is that you might actually need > direct-mapped memory but what if you have larger SRAMs that are several MB? > Maybe there should be a check here to prevent that. > Yes, it will be a problem. I think vmalloc should be used here. > > +static const struct dev_pm_ops sram_pm_ops = { > > + .suspend_noirq = sram_suspend_noirq, > > + .resume_noirq = sram_resume_noirq, > > +}; > > You may have to enclose this within an ifdef CONFIG_PM_SLEEP/endif statement > not to cause warnings if CONFIG_PM_SLEEP is disabled. > -- The "struct dev_pm_ops *pm" is always a member of device_driver, and it has no relation with the CONFIG_PM_XXX. Thanks, Shenwei > Florian
diff --git a/Documentation/devicetree/bindings/misc/sram.txt b/Documentation/devicetree/bindings/misc/sram.txt index 36cbe5a..1170086 100644 --- a/Documentation/devicetree/bindings/misc/sram.txt +++ b/Documentation/devicetree/bindings/misc/sram.txt @@ -33,6 +33,8 @@ Optional properties in the area nodes: - compatible : standard definition, should contain a vendor specific string in the form <vendor>,[<device>-]<usage> +- can-power-gate: a property to tell the driver that the sram can support + power gate Example: diff --git a/drivers/misc/sram.c b/drivers/misc/sram.c index 15c33cc..924692c 100644 --- a/drivers/misc/sram.c +++ b/drivers/misc/sram.c @@ -30,6 +30,7 @@ struct sram_dev { struct device *dev; + void *power_off_save; void __iomem *virt_base; struct gen_pool *pool; @@ -156,6 +157,37 @@ static int sram_reserve_regions(struct sram_dev *sram, struct resource *res) return ret; } +static int sram_suspend_noirq(struct device *dev) +{ + struct platform_device *pdev = to_platform_device(dev); + struct sram_dev *sram = platform_get_drvdata(pdev); + + if (!sram->power_off_save) + return 0; + + /* Save necessary regs */ + clk_enable(sram->clk); + memcpy(sram->power_off_save, sram->virt_base,gen_pool_size(sram->pool)); + clk_disable(sram->clk); + + return 0; +} + +static int sram_resume_noirq(struct device *dev) +{ + struct platform_device *pdev = to_platform_device(dev); + struct sram_dev *sram = platform_get_drvdata(pdev); + + if (!sram->power_off_save) + return 0; + + clk_enable(sram->clk); + memcpy(sram->virt_base, sram->power_off_save, gen_pool_size(sram->pool)); + clk_disable(sram->clk); + + return 0; +} + static int sram_probe(struct platform_device *pdev) { struct sram_dev *sram; @@ -203,6 +235,12 @@ static int sram_probe(struct platform_device *pdev) platform_set_drvdata(pdev, sram); + if (of_get_property(pdev->dev.of_node, "can-power-gate", NULL)) + { + sram->power_off_save = devm_kzalloc(&pdev->dev, + gen_pool_size(sram->pool), GFP_KERNEL); + } + dev_dbg(sram->dev, "SRAM pool: %zu KiB @ 0x%p\n", gen_pool_size(sram->pool) / 1024, sram->virt_base); @@ -229,10 +267,16 @@ static const struct of_device_id sram_dt_ids[] = { }; #endif +static const struct dev_pm_ops sram_pm_ops = { + .suspend_noirq = sram_suspend_noirq, + .resume_noirq = sram_resume_noirq, +}; + static struct platform_driver sram_driver = { .driver = { .name = "sram", .of_match_table = of_match_ptr(sram_dt_ids), + .pm = &sram_pm_ops, }, .probe = sram_probe, .remove = sram_remove,