Re: [PATCH 4/4] mfd: intel-m10-bmc: Manage access to MAX 10 fw handshake registers

[Xu Yilun]

On 2023-04-11 at 14:54:58 +0300, Ilpo Järvinen wrote:
> On Fri, 7 Apr 2023, Xu Yilun wrote:
> 
> > On 2023-04-05 at 11:01:52 +0300, Ilpo Järvinen wrote:
> > > On some MAX 10 cards, the BMC firmware is not available to service
> > > handshake registers during secure update erase and write phases at
> > > normal speeds. This problem affects at least hwmon driver. When the MAX
> > > 10 hwmon driver tries to read the sensor values during a secure update,
> > > the reads are slowed down (e.g., reading all D5005 sensors takes ~24s
> > > which is magnitudes worse than the normal <0.02s).
> > > 
> > > Manage access to the handshake registers using a rw semaphore and a FW
> > > state variable to prevent accesses during those secure update phases
> > > and return -EBUSY instead.
> > > 
> > > Co-developed-by: Russ Weight <russell.h.weight@xxxxxxxxx>
> > > Signed-off-by: Russ Weight <russell.h.weight@xxxxxxxxx>
> > > Co-developed-by: Xu Yilun <yilun.xu@xxxxxxxxx>
> > > Signed-off-by: Xu Yilun <yilun.xu@xxxxxxxxx>
> > > Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx>
> > > ---
> > >  drivers/fpga/intel-m10-bmc-sec-update.c | 17 +++++--
> > >  drivers/mfd/intel-m10-bmc-core.c        | 63 ++++++++++++++++++++++++-
> > >  drivers/mfd/intel-m10-bmc-pmci.c        |  4 ++
> > >  drivers/mfd/intel-m10-bmc-spi.c         | 14 ++++++
> > >  include/linux/mfd/intel-m10-bmc.h       | 27 +++++++++++
> > >  5 files changed, 120 insertions(+), 5 deletions(-)
> > >
> > 
> > [...]
> >  
> > >  
> > > +void m10bmc_fw_state_set(struct intel_m10bmc *m10bmc, enum m10bmc_fw_state new_state)
> > > +{
> > > +	down_write(&m10bmc->bmcfw_lock);
> > > +	m10bmc->bmcfw_state = new_state;
> > > +	up_write(&m10bmc->bmcfw_lock);
> > 
> > Could we also skip this if no handshake is possible like for PMCI?
> 
> Did you mean guarding it with !m10bmc->info->handshake_sys_reg_nranges ?

Yes. My concern is, the handshake_sys_reg_nranges is the constant value
for a device, so if the device doesn't have handshake registers, we
could save the locking costs.

Thanks,
Yilun

> If yes, it's doable (+ I'd add comment mentioning it since it's slightly 
> trappy to not always have that state updated).
> 
> 
> -- 
>  i.