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]
Message-ID: <c8311933-d129-4618-b81b-aa627b7b6de0@st.com>
Date:   Fri, 15 Nov 2019 15:59:47 +0100
From:   Ludovic BARRE <ludovic.barre@...com>
To:     Ulf Hansson <ulf.hansson@...aro.org>
CC:     Rob Herring <robh+dt@...nel.org>,
        Srinivas Kandagatla <srinivas.kandagatla@...aro.org>,
        Maxime Coquelin <mcoquelin.stm32@...il.com>,
        Alexandre Torgue <alexandre.torgue@...com>,
        Linux ARM <linux-arm-kernel@...ts.infradead.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        DTML <devicetree@...r.kernel.org>,
        "linux-mmc@...r.kernel.org" <linux-mmc@...r.kernel.org>,
        <linux-stm32@...md-mailman.stormreply.com>
Subject: Re: [PATCH 1/1] mmc: mmci: add threaded irq to abort DPSM of
 non-functional state

hi Ulf

Le 11/14/19 à 4:09 PM, Ulf Hansson a écrit :
> On Wed, 13 Nov 2019 at 18:25, Ludovic Barre <ludovic.Barre@...com> wrote:
>>
>> From: Ludovic Barre <ludovic.barre@...com>
>>
>> If datatimeout occurs on R1B request, the Data Path State Machine stays
>> in busy and is non-functional. Only a reset aborts the DPSM.
> 
> Please clarify/extend this information to tell that this is for the
> variant, that keeps DPSM enabled and uses the data timer while sending
> a CMD12. Or something along those lines.
>

yes, of course.

>>
>> Like a reset must be outside of critical section, this patch adds
> 
> /s/critical section/atomic context

OK, thanks

> 
>> threaded irq function to release state machine. In this case,
>> the mmc_request_done is called at the end of threaded irq and
>> skipped into irq handler.
>>
>> Signed-off-by: Ludovic Barre <ludovic.barre@...com>
>> ---
>>   drivers/mmc/host/mmci.c | 44 ++++++++++++++++++++++++++++++++++++-----
>>   drivers/mmc/host/mmci.h |  1 +
>>   2 files changed, 40 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
>> index 40e72c30ea84..ec6e249c87ca 100644
>> --- a/drivers/mmc/host/mmci.c
>> +++ b/drivers/mmc/host/mmci.c
>> @@ -556,6 +556,9 @@ static void mmci_dma_error(struct mmci_host *host)
>>   static void
>>   mmci_request_end(struct mmci_host *host, struct mmc_request *mrq)
>>   {
>> +       if (host->irq_action == IRQ_WAKE_THREAD)
>> +               return;
>> +
>>          writel(0, host->base + MMCICOMMAND);
>>
>>          BUG_ON(host->data);
>> @@ -1321,6 +1324,7 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
>>          } else if (host->variant->busy_timeout && busy_resp &&
>>                     status & MCI_DATATIMEOUT) {
>>                  cmd->error = -ETIMEDOUT;
>> +               host->irq_action = IRQ_WAKE_THREAD;
>>          } else {
>>                  cmd->resp[0] = readl(base + MMCIRESPONSE0);
>>                  cmd->resp[1] = readl(base + MMCIRESPONSE1);
>> @@ -1532,9 +1536,9 @@ static irqreturn_t mmci_irq(int irq, void *dev_id)
>>   {
>>          struct mmci_host *host = dev_id;
>>          u32 status;
>> -       int ret = 0;
>>
>>          spin_lock(&host->lock);
>> +       host->irq_action = IRQ_HANDLED;
>>
>>          do {
>>                  status = readl(host->base + MMCISTATUS);
>> @@ -1574,12 +1578,41 @@ static irqreturn_t mmci_irq(int irq, void *dev_id)
>>                  if (host->variant->busy_detect_flag)
>>                          status &= ~host->variant->busy_detect_flag;
>>
>> -               ret = 1;
>>          } while (status);
>>
>>          spin_unlock(&host->lock);
>>
>> -       return IRQ_RETVAL(ret);
>> +       return host->irq_action;
>> +}
>> +
>> +/*
>> + * mmci_irq_threaded is call if the mmci host need to release state machines
>> + * before to terminate the request.
>> + * If datatimeout occurs on R1B request, the Data Path State Machine stays
>> + * in busy and is non-functional. Only a reset can to abort the DPSM.
>> + */
>> +static irqreturn_t mmci_irq_threaded(int irq, void *dev_id)
>> +{
>> +       struct mmci_host *host = dev_id;
>> +       unsigned long flags;
>> +
>> +       if (host->rst) {
>> +               reset_control_assert(host->rst);
>> +               udelay(2);
>> +               reset_control_deassert(host->rst);
>> +       }
>> +
>> +       spin_lock_irqsave(&host->lock, flags);
>> +       writel(host->clk_reg, host->base + MMCICLOCK);
>> +       writel(host->pwr_reg, host->base + MMCIPOWER);
>> +       writel(MCI_IRQENABLE | host->variant->start_err,
>> +              host->base + MMCIMASK0);
>> +
>> +       host->irq_action = IRQ_HANDLED;
>> +       mmci_request_end(host, host->mrq);
>> +       spin_unlock_irqrestore(&host->lock, flags);
>> +
>> +       return host->irq_action;
>>   }
>>
>>   static void mmci_request(struct mmc_host *mmc, struct mmc_request *mrq)
>> @@ -2071,8 +2104,9 @@ static int mmci_probe(struct amba_device *dev,
>>                          goto clk_disable;
>>          }
>>
>> -       ret = devm_request_irq(&dev->dev, dev->irq[0], mmci_irq, IRQF_SHARED,
>> -                       DRIVER_NAME " (cmd)", host);
>> +       ret = devm_request_threaded_irq(&dev->dev, dev->irq[0], mmci_irq,
>> +                                       mmci_irq_threaded, IRQF_SHARED,
>> +                                       DRIVER_NAME " (cmd)", host);
> 
> In general it's a good idea to move drivers into using a threaded IRQ handler.
> 
> However, the reason this hasn't been done for mmci before, is because
> there are some legacy variants, that doesn't support HW flow control.

With this patch, the mmci_irq_threaded is called only when
a datatimeout occurs on R1B request else the mmci_irq return IRQ_HANDLED
and the threaded irq is not called.
So normally, there is no impact for legacy variants.

> 
> Unless I am mistaken, that means when the fifo gets full during data
> transfers - it's too late to act. In other words, running the handler
> in hard IRQ context, should increase the probability of not missing
> the deadline.
> 
> If a threaded IRQ handler also is sufficient for these legacy
> variants, only tests can tell.
> 
> An option, would be to use a threaded handler for those variants that
> supports HW flow control. Not sure how messy the code would be with
> this option, perhaps you can give this a try?
For all these reasons you mentioned above, I'm not sure it's safe to 
extend the thread manager to anything other than "aborting the DPSM" on
R1B datatiemout ? what do you think about that?

I prefer just used the irq threaded for specific errors case which need
be outside of atomic context.

> 
> 
>>          if (ret)
>>                  goto clk_disable;
>>
>> diff --git a/drivers/mmc/host/mmci.h b/drivers/mmc/host/mmci.h
>> index 158e1231aa23..5e63c0596364 100644
>> --- a/drivers/mmc/host/mmci.h
>> +++ b/drivers/mmc/host/mmci.h
>> @@ -412,6 +412,7 @@ struct mmci_host {
>>
>>          struct timer_list       timer;
>>          unsigned int            oldstat;
>> +       u32                     irq_action;
>>
>>          /* pio stuff */
>>          struct sg_mapping_iter  sg_miter;
>> --
>> 2.17.1
>>
> 
> Kind regards
> Uffe
> 

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ