Re: [PATCH 1/1] i2c: i801: Restore the presence state of P2SB PCI device after reading BAR

[Jean Delvare <jdelvare@suse.de>]

On Tue, 15 Aug 2017 00:04:50 +0800, Qiuxu Zhuo wrote:
> Sun, Yunying reported the following failure on Denverton micro-server:
> 
>  EDAC DEBUG: pnd2_init:
>  EDAC DEBUG: pnd2_probe:
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_tolud_pci=00000000_80000000
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_touud_lo_pci=00000000_80000000
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_touud_hi_pci=00000000_00000004
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_asym_mem_region0_mchbar=00000000_00000000
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_asym_mem_region1_mchbar=00000000_00000000
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_mot_out_base_mchbar=00000000_00000000
>  EDAC DEBUG: dnv_rd_reg: Read b_cr_mot_out_mask_mchbar=00000000_00000000
>  EDAC pnd2: Failed to register device with error -19.
> 
> On Denverton micro-server, the presence of the P2SB bridge PCI device is
> enabled or disabled by the item 'RelaxSecConf' in BIOS setup menu. When
> 'RelaxSecConf' is enabled, the P2SB PCI device is present and the pnd2_edac
> EDAC driver also uses it to get BAR. Hiding the P2SB PCI device caused the
> pnd2_edac EDAC driver failed to get BAR then reported the above failure.
> 
> Therefor, store the presence state of P2SB PCI device before unhiding it
> for reading BAR and restore the presence state after reading BAR.
> 
> Signed-off-by: Qiuxu Zhuo <qiuxu.zhuo@intel.com>
> Reported-and-tested-by: Yunying Sun <yunying.sun@intel.com>
> ---
>  drivers/i2c/busses/i2c-i801.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-i801.c b/drivers/i2c/busses/i2c-i801.c
> index c9536e1..e114e4e 100644
> --- a/drivers/i2c/busses/i2c-i801.c
> +++ b/drivers/i2c/busses/i2c-i801.c
> @@ -1332,6 +1332,7 @@ static void i801_add_tco(struct i801_priv *priv)
>  	u32 tco_base, tco_ctl;
>  	u32 base_addr, ctrl_val;
>  	u64 base64_addr;
> +	u8 hidden;
>  
>  	if (!(priv->features & FEATURE_TCO))
>  		return;
> @@ -1376,8 +1377,10 @@ static void i801_add_tco(struct i801_priv *priv)
>  
>  	devfn = PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 1);
>  
> -	/* Unhide the P2SB device */
> -	pci_bus_write_config_byte(pci_dev->bus, devfn, 0xe1, 0x0);
> +	/* Unhide the P2SB device, if it is hidden */
> +	pci_bus_read_config_byte(pci_dev->bus, devfn, 0xe1, &hidden);
> +	if (hidden)
> +		pci_bus_write_config_byte(pci_dev->bus, devfn, 0xe1, 0x0);
>  
>  	pci_bus_read_config_dword(pci_dev->bus, devfn, SBREG_BAR, &base_addr);
>  	base64_addr = base_addr & 0xfffffff0;
> @@ -1385,8 +1388,9 @@ static void i801_add_tco(struct i801_priv *priv)
>  	pci_bus_read_config_dword(pci_dev->bus, devfn, SBREG_BAR + 0x4, &base_addr);
>  	base64_addr |= (u64)base_addr << 32;
>  
> -	/* Hide the P2SB device */
> -	pci_bus_write_config_byte(pci_dev->bus, devfn, 0xe1, 0x1);
> +	/* Hide the P2SB device, if it was hidden before */
> +	if (hidden)
> +		pci_bus_write_config_byte(pci_dev->bus, devfn, 0xe1, hidden);
>  	spin_unlock(&p2sb_spinlock);
>  
>  	res = &tco_res[ICH_RES_MEM_OFF];

Sorry, this happened while I was on vacation and I missed it.

The change looks sane, so:

Reviewed-by: Jean Delvare <jdelvare@suse.de>

And I approve backporting to older kernels if needed.

That being said... Looking at the code, I have to say I'm surprised
that the whole value of configuration byte 0xe1 is used as true/false.
I'd expect a single bit (bit 0) in this configuration file to decide if
the P2SB device is hidden or visible. So it seems to me that we are
trashing 7 bits in the process. These bits may be unused at the moment,
but could be used in future devices, and then unexpected behavior will
appear.

I'm still approving this patch because the same problem was present
before, but I think this should be checked, and fixed if I'm right.
I'll be happy to do it myself if someone at Intel points me to the
Denverton datasheet - if it is public.

-- 
Jean Delvare
SUSE L3 Support