On 3/22/2024 4:13 AM, Jeff Johnson wrote:
> On 3/21/2024 10:23 AM, Jose Ignacio Tornos Martinez wrote:
>> Currently, this driver is not working when the device is handled in a
>> Virtual Machine (PCI pass-through), as it was already reported here:
>> https://lore.kernel.org/all/[email protected]/T/
>> Baochen Qiang focused the problem and described how to have it working
>> for a specific real MSI vector from host that needs to be used in VM too.
>> And this value, as it was commented, can change.
>>
>> The problem seems complex to me and I don't know if there is any easy way
>> to solve it. Meanwhile and using the information from Baochen Qiang,
>> since the use of VMs is very interesting for testing procedures,
>> I would like to just add this workaround that consists on adding a
>> parameter to pass the real MSI vector from host to the VM. In that way,
>> checking the 'lscpi' command output from host, it could be handled manually
>> or with some user tool in order to have the VM with the driver working.
>> Of course, if this parameter is not configured (zero value and default),
>> we will have the same behavior as always.
>>
>> Signed-off-by: Jose Ignacio Tornos Martinez <[email protected]>
>> ---
>> drivers/net/wireless/ath/ath11k/pci.c | 20 ++++++++++++++++++++
>> 1 file changed, 20 insertions(+)
>>
>> diff --git a/drivers/net/wireless/ath/ath11k/pci.c b/drivers/net/wireless/ath/ath11k/pci.c
>> index be9d2c69cc41..0e322956b10f 100644
>> --- a/drivers/net/wireless/ath/ath11k/pci.c
>> +++ b/drivers/net/wireless/ath/ath11k/pci.c
>> @@ -31,6 +31,11 @@
>>
>> #define TCSR_SOC_HW_SUB_VER 0x1910010
>>
>> +static ulong ath11k_host_msi_addr = 0;
>> +module_param_named(host_msi_addr, ath11k_host_msi_addr, ulong, 0644);
>> +MODULE_PARM_DESC(host_msi_addr,
>> + "Workaround to configure the MSI address that is used from host in order to be used in VM");
>> +
>> static const struct pci_device_id ath11k_pci_id_table[] = {
>> { PCI_VDEVICE(QCOM, QCA6390_DEVICE_ID) },
>> { PCI_VDEVICE(QCOM, WCN6855_DEVICE_ID) },
>> @@ -443,6 +448,18 @@ static int ath11k_pci_alloc_msi(struct ath11k_pci *ab_pci)
>>
>> ath11k_pci_msi_disable(ab_pci);
>>
>> + if (ath11k_host_msi_addr) {
>> + ab_pci->ab->pci.msi.ep_base_data = 0;
>> + ab->pci.msi.addr_hi = (u32)(ath11k_host_msi_addr >> 32);
>> + ab->pci.msi.addr_lo = (u32)(ath11k_host_msi_addr & 0xffffffff);
>> +
>> + ath11k_dbg(ab, ATH11K_DBG_PCI, "msi addr hi 0x%x lo 0x%x base data is %d\n",
>> + ab->pci.msi.addr_hi,
>> + ab->pci.msi.addr_lo,
>> + ab->pci.msi.ep_base_data);
>> + return 0;
>> + }
>> +
>> msi_desc = irq_get_msi_desc(ab_pci->pdev->irq);
>> if (!msi_desc) {
>> ath11k_err(ab, "msi_desc is NULL!\n");
>> @@ -482,6 +499,9 @@ static int ath11k_pci_config_msi_data(struct ath11k_pci *ab_pci)
>> {
>> struct msi_desc *msi_desc;
>>
>> + if (ath11k_host_msi_addr)
>> + return 0;
>> +
>> msi_desc = irq_get_msi_desc(ab_pci->pdev->irq);
>> if (!msi_desc) {
>> ath11k_err(ab_pci->ab, "msi_desc is NULL!\n");
>
> + [email protected] to make sure the Qualcomm VM experts are aware of this
> issue and to see if they have any additional suggestions.
>
If limit the fix/WAR to be within ath11k, not touching vfio or any other
related module, I was imagining that if hardware could map the standard
MSI registers in config space to some private MMIO registers so VM
ath11k could read, or that if firmware could read them and send to VM
ath11k via QMI interface? don't know if it is feasible to hardware/firmware.
>