Re: [PATCH 1/1] PCI: vmd: Do not disable MSI-X remapping if interrupt remapping is enabled by IOMMU

[Jon Derrick <jonathan.derrick@intel.com>]

Thank you Adrian

On 9/1/21 6:40 AM, Adrian Huang wrote:
> From: Adrian Huang <ahuang12@lenovo.com>
> 
> When enabling VMD in BIOS setup (Ice Lake Processor: Whitley platform),
> the host OS cannot boot successfully with the following error message:
> 
>   nvme nvme0: I/O 12 QID 0 timeout, completion polled
>   nvme nvme0: Shutdown timeout set to 6 seconds
>   DMAR: DRHD: handling fault status reg 2
>   DMAR: [INTR-REMAP] Request device [0x00:0x00.5] fault index 0xa00 [fault reason 0x25] Blocked a compatibility format interrupt request

I know we'd really prefer to support interrupt remapping with the VMD feature,
and I'm not certain how EIME differs from the interrupt remapping modes that
were tested while developing the VMD feature.

I think this will have to be acceptable for now.

Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>

> 
> The request device is the VMD controller:
>   # lspci -s 0000:00.5 -nn
>   0000:00:00.5 RAID bus controller [0104]: Intel Corporation Volume
>   Management Device NVMe RAID Controller [8086:28c0] (rev 04)
> 
> `git bisect` points to this offending commit ee81ee84f873 ("PCI:
> vmd: Disable MSI-X remapping when possible"), which disables VMD MSI
> remapping. The IOMMU hardware blocks the compatibility format
> interrupt request because Interrupt Remapping Enable Status (IRES) and
> Extended Interrupt Mode Enable (EIME) are enabled. Please refer to
> section "5.1.4 Interrupt-Remapping Hardware Operation" in Intel VT-d
> spec.
> 
> To fix the issue, VMD driver still enables the interrupt remapping
> irrespective of VMD_FEAT_CAN_BYPASS_MSI_REMAP if the IOMMU subsystem
> enables the interrupt remapping.
> 
> Test configuration is shown as follows:
>   * Two VMD controllers
>     1. 8086:28c0 (Whitley's VMD)
>     2. 8086:201d (Purley's VMD: The issue does not appear in this
>        controller. Just make sure if any side effect occurs.)
>   * w/wo intremap=off
> 
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=214219
> Cc: Jon Derrick <jonathan.derrick@intel.com>
> Cc: Nirmal Patel <nirmal.patel@linux.intel.com>
> Cc: Joerg Roedel <jroedel@suse.de>
> Signed-off-by: Adrian Huang <ahuang12@lenovo.com>
> ---
>  drivers/pci/controller/vmd.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
> index e3fcdfec58b3..db72932d049f 100644
> --- a/drivers/pci/controller/vmd.c
> +++ b/drivers/pci/controller/vmd.c
> @@ -6,6 +6,7 @@
>  
>  #include <linux/device.h>
>  #include <linux/interrupt.h>
> +#include <linux/iommu.h>
>  #include <linux/irq.h>
>  #include <linux/kernel.h>
>  #include <linux/module.h>
> @@ -710,7 +711,8 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
>  	 * acceptable because the guest is usually CPU-limited and MSI
>  	 * remapping doesn't become a performance bottleneck.
>  	 */
> -	if (!(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
> +	if (iommu_capable(vmd->dev->dev.bus, IOMMU_CAP_INTR_REMAP) ||
> +	    !(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
>  	    offset[0] || offset[1]) {
>  		ret = vmd_alloc_irqs(vmd);
>  		if (ret)
>