Date:   Fri, 17 Mar 2017 11:39:59 -0700
From:   sathyanarayanan kuppuswamy 
        <sathyanarayanan.kuppuswamy@...ux.intel.com>
To:     Guenter Roeck <linux@...ck-us.net>
Cc:     Rajneesh Bhardwaj <rajneesh.bhardwaj@...el.com>,
        andy@...radead.org, qipeng.zha@...el.com, dvhart@...radead.org,
        david.e.box@...ux.intel.com, platform-driver-x86@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-watchdog@...r.kernel.org,
        wim@...ana.be
Subject: Re: [PATCH v2 3/4] watchdog: iTCO_wdt: Fix PMC GCR memory mapping
 failure



On 03/17/2017 10:50 AM, Guenter Roeck wrote:
> On Fri, Mar 17, 2017 at 10:24:35AM -0700, sathyanarayanan kuppuswamy wrote:
>>
>> On 03/17/2017 06:40 AM, Guenter Roeck wrote:
>>> On 03/17/2017 04:43 AM, Rajneesh Bhardwaj wrote:
>>>> On Thu, Mar 16, 2017 at 05:41:35PM -0700, Kuppuswamy Sathyanarayanan
>>>> wrote:
>>>>> Currently, iTCO watchdog driver uses memory map to access
>>>>> PMC_CFG GCR register. But the entire GCR address space is
>>>>> already mapped in intel_scu_ipc driver. So remapping the
>>>> intel_pmc_ipc driver.
>>>>
>>>>> GCR register in this driver causes the mem request failure in
>>>>> iTCO_wdt probe function. This patch fixes this issue by
>>>>> using PMC GCR read/write API's to access PMC_CFG register.
>>>>>
>>>>> Signed-off-by: Kuppuswamy Sathyanarayanan
>>>>> <sathyanarayanan.kuppuswamy@...ux.intel.com>
>>>>> ---
>>>>> drivers/watchdog/iTCO_wdt.c | 31 +++++++------------------------
>>>>> 1 file changed, 7 insertions(+), 24 deletions(-)
>>>>>
>>>>> diff --git a/drivers/watchdog/iTCO_wdt.c b/drivers/watchdog/iTCO_wdt.c
>>>>> index 3d0abc0..31abfc5 100644
>>>>> --- a/drivers/watchdog/iTCO_wdt.c
>>>>> +++ b/drivers/watchdog/iTCO_wdt.c
>>>>> @@ -68,6 +68,8 @@
>>>>> #include <linux/io.h>            /* For inb/outb/... */
>>>>> #include <linux/platform_data/itco_wdt.h>
>>>>>
>>>>> +#include <asm/intel_pmc_ipc.h>
>>>>> +
>>>>> #include "iTCO_vendor.h"
>>>>>
>>>>> /* Address definitions for the TCO */
>>>>> @@ -94,12 +96,6 @@ struct iTCO_wdt_private {
>>>>>      unsigned int iTCO_version;
>>>>>      struct resource *tco_res;
>>>>>      struct resource *smi_res;
>>>>> -    /*
>>>>> -     * NO_REBOOT flag is Memory-Mapped GCS register bit 5 (TCO
>>>>> version 2),
>>>>> -     * or memory-mapped PMC register bit 4 (TCO version 3).
>>>>> -     */
>>>> Better to retain this comment elsewhere.
>>>>
>>>>> -    struct resource *gcs_pmc_res;
>>>>> -    unsigned long __iomem *gcs_pmc;
>>>>>      /* the lock for io operations */
>>>>>      spinlock_t io_lock;
>>>>>      /* the PCI-device */
>>>>> @@ -176,9 +172,9 @@ static void iTCO_wdt_set_NO_REBOOT_bit(struct
>>>>> iTCO_wdt_private *p)
>>>>>
>>>>>      /* Set the NO_REBOOT bit: this disables reboots */
>>>>>      if (p->iTCO_version >= 2) {
>>>>> -        val32 = readl(p->gcs_pmc);
>>>>> +        val32 = intel_pmc_gcr_read(PMC_GCR_PMC_CFG_REG);
>>>> better to have protection and error handling, discussed in v2, 2/4.
>>>>
>>>> compiled and tested this on APL and i see iTCO_WDT driver loads fine.
>>>> Since
>>>> it impacts core WDT functionality, need to be thoroughly tested on
>>>> various
>>>> platforms.
>>>>
>>> I don't think I (or the watchdog mailing list) was copied on the original
>>> patch.
>> Sorry. Its my mistake. I will fix it in next series update.
>>> Major immediate concern is that this introduces a dependency on external
>>> code.
>>> The pmc_ipc driver's Kconfig entry states "This is not needed for PC-type
>>> machines". I don't know where the function is introduced, but I hope this
>>> change
>>> does not require the pmc_ipc code to be present on such machines for the
>>> watchdog
>>> to work. It would be bad if it does. If it doesn't, it appears that the
>>> function
>>> should not be declared in asm/intel_pmc_ipc.h.
>> It should not create any compile time dependency with INTEL_PMC_IPC config
>> option. If INTEL_PMC_IPC_CONFIG is disabled, we use
>> empty definitions for these calls defined in asm/intel_pmc_ipc.h
>>
> So the watchdog driver would get an error if CONFIG_INTEL_PMC_IPC
> is not defined ? And that is supposed to be acceptable ?
Sorry, It looks like I missed your point in your previous email.  I 
thought that gcs_pmc mem resource will be used only if watchdog device 
is enumerated by intel_pmc_ipc.c

After reviewing the code again, I found out this iTCO_wdt driver can 
also be enumerated by drivers/mfd/lpc_ich.c and 
drivers/i2c/busses/i2c-i801.c. Both these drivers pass the GCS as memory 
resource. So using intel_pmc_ipc specific calls will break watchdog 
functionality if its enumerated by any device other than intel_pmic_ipc.c

May be I should add a flag for ipc case in itco_wdt_platform_data and 
handle this as a special case.
>
>> But iTCO_wdt driver already has runtime dependency with INTEL_PMIC_IPC if
>> its version iTCO_version >= 2.
>>
> Unless I am missing something, there is no explicit dependency. AFAICS
> the watchdog driver works just fine if INTEL_PMIC_IPC is not enabled,
> and/or if it is built as module and the module is not loaded.
>
> Maybe you mean that the watchdog driver doesn't load if the INTEL_PMIC_IPC
> driver is loaded. That would be a bug, not a dependency.
>
> Guenter
>

-- 
Sathyanarayanan Kuppuswamy
Android kernel developer

Powered by blists - more mailing lists