lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Tue, 25 Aug 2020 21:30:09 +0300
From:   Abel Vesa <abel.vesa@....com>
To:     Philipp Zabel <p.zabel@...gutronix.de>
Cc:     Dong Aisheng <aisheng.dong@....com>, Rob Herring <robh@...nel.org>,
        Peng Fan <peng.fan@....com>, Fugang Duan <fugang.duan@....com>,
        Jacky Bai <ping.bai@....com>,
        Anson Huang <anson.huang@....com>, devicetree@...r.kernel.org,
        Stephen Boyd <sboyd@...nel.org>,
        Mike Turquette <mturquette@...libre.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        NXP Linux Team <linux-imx@....com>,
        Sascha Hauer <kernel@...gutronix.de>,
        Fabio Estevam <fabio.estevam@....com>,
        Shawn Guo <shawnguo@...nel.org>, linux-clk@...r.kernel.org,
        linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH v2 11/17] clk: imx: Add blk_ctrl combo driver

On 20-08-25 14:07:29, Philipp Zabel wrote:
> On Tue, 2020-08-25 at 14:24 +0300, Abel Vesa wrote:
> [...]
> > > > +static int imx_blk_ctrl_reset_set(struct reset_controller_dev *rcdev,
> > > > +				  unsigned long id, bool assert)
> > > > +{
> > > > +	struct imx_blk_ctrl_drvdata *drvdata = container_of(rcdev,
> > > > +			struct imx_blk_ctrl_drvdata, rcdev);
> > > > +	unsigned int offset = drvdata->rst_hws[id].offset;
> > > > +	unsigned int shift = drvdata->rst_hws[id].shift;
> > > > +	unsigned int mask = drvdata->rst_hws[id].mask;
> > > > +	void __iomem *reg_addr = drvdata->base + offset;
> > > > +	unsigned long flags;
> > > > +	unsigned int asserted_before = 0, asserted_after = 0;
> > > > +	u32 reg;
> > > > +	int i;
> > > > +
> > > > +	spin_lock_irqsave(&drvdata->lock, flags);
> > > > +
> > > > +	for (i = 0; i < drvdata->rcdev.nr_resets; i++)
> > > > +		if (drvdata->rst_hws[i].asserted)
> > > > +			asserted_before++;
> > > > +
> > > > +	if (asserted_before == 0 && assert)
> > > > +		pm_runtime_get(rcdev->dev);
> > > 
> > > Shouldn't that be pm_runtime_get_sync() ?
> > > 
> > > I would do that unconditionally before locking drvdata->lock and then
> > > drop unnecessary refcounts afterwards.
> > > 
> > 
> > I thought we already discussed this on the last's version thread.
> 
> This is about something different. pm_runtime_get() just queues the
> device to be enabled at a later point, but I presume you want to have it
> enabled before writing to its registers. (The question here is can you
> write to the registers, and have the device update its internal state,
> while the power domain is disabled?)
> Either way, if you want the reset to be asserted after the function
> returns (as is required by the reset API), as I understand it, you have
> to make sure that the power domain is activated before the function
> returns.
> Therefore pm_runtime_get_sync() is required instead of pm_runtime_get(),
> and that must be called outside of the spin locked section. My
> suggestion would be:
> 
> 	if (assert)
> 		pm_runtime_get_sync();
> 	spin_lock_irqsave();
> 	/* ... */
> 	spin_unlock_irqrestore();
> 	if (assert && asserted_before)
> 		pm_runtime_put();
> 

On a second thought this doesn't work because, for the first assertion,
the runtime put will never be called, if the asserted_before does not count
the current assertion. If it counts the current assertion, then every assertion
will end with runtime put. None of these options work here.

How about the following:

	if (assert && !test_and_set_bit(1, &drvdata->rst_hws[id].asserted))     
		pm_runtime_get_sync(rcdev->dev);                                
                                                                        
	spin_lock_irqsave(&drvdata->lock, flags);                               
                                                                        
	reg = readl(reg_addr);                                                  
	if (assert)                                                             
		writel(reg & ~(mask << shift), reg_addr);                       
	else                                                                    
		writel(reg | (mask << shift), reg_addr);                        
                                                                        
	spin_unlock_irqrestore(&drvdata->lock, flags);                          
                                                                        
	if (!assert && test_and_clear_bit(1, &drvdata->rst_hws[id].asserted))   
		pm_runtime_put(rcdev->dev);                                     
                                                                        
This would only call the get_sync/put once for each reset bit.

> unless the following might be an issue:
> 
> > > > +
> > > > +	if (assert) {
> > > > +		reg = readl(reg_addr);
> > > > +		writel(reg & ~(mask << shift), reg_addr);
> > > > +		drvdata->rst_hws[id].asserted = true;
> > > > +	} else {
> > > > +		reg = readl(reg_addr);
> > > > +		writel(reg | (mask << shift), reg_addr);
> 
> Could this cause problems if the power domain is already disabled? If
> so, it would be best to either temporarily enable power, or to skip the
> register writes if asserted_before == 0 && !assert.
> 
> > > > +		drvdata->rst_hws[id].asserted = false;
> > > > +	}
> > > > +
> > > > +	for (i = 0; i < drvdata->rcdev.nr_resets; i++)
> > > > +		if (drvdata->rst_hws[i].asserted)
> > > > +			asserted_after++;
> > > > +
> > > > +	if (asserted_before == 1 && asserted_after == 0)
> > > > +		pm_runtime_put(rcdev->dev);
> > > > +
> > > > +	spin_unlock_irqrestore(&drvdata->lock, flags);
> > > > +
> > > > +	return 0;
> > > > +}
> 
> regards
> Philipp

Powered by blists - more mailing lists