Re: [dpdk-dev] [PATCH v3] bus/pci: forbid VA as IOVA mode if IOMMU address width too small

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi Anatoly,

On 01/12/2018 12:10 PM, Burakov, Anatoly wrote:
> <snip>
> 
>> +#if defined(RTE_ARCH_X86)
>> +static bool
>> +pci_one_device_iommu_support_va(struct rte_pci_device *dev)
>> +{
>> +#define VTD_CAP_MGAW_SHIFT    16
>> +#define VTD_CAP_MGAW_MASK    (0x3fULL << VTD_CAP_MGAW_SHIFT)
>> +#define X86_VA_WIDTH 47 /* From Documentation/x86/x86_64/mm.txt */
>> +    struct rte_pci_addr *addr = &dev->addr;
>> +    char filename[PATH_MAX];
>> +    FILE *fp;
>> +    uint64_t mgaw, vtd_cap_reg = 0;
>> +
>> +    snprintf(filename, sizeof(filename),
>> +         "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
>> +         rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
>> +         addr->function);
>> +    if (access(filename, F_OK) == -1) {
>> +        /* We don't have an Intel IOMMU, assume VA supported*/
>> +        return true;
>> +    }
>> +
>> +    /* We have an intel IOMMU */
>> +    fp = fopen(filename, "r");
>> +    if (fp == NULL) {
>> +        RTE_LOG(ERR, EAL, "%s(): can't open %s\n", __func__, filename);
>> +        return false;
>> +    }
>> +
>> +    if (fscanf(fp, "%lx", &vtd_cap_reg) != 1) {
>> +        RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
>> +        fclose(fp);
>> +        return false;
>> +    }
>> +
>> +    fclose(fp);
> 
> Hi Maxime,
> 
> You probably want to use eal_parse_sysfs_value() for this.

I initially planned to use it, but the sysfs value is hexadecimal, but
not prefixed with "0x". For example:

# cat /sys/devices/pci0000\:00/0000\:00\:02.0/iommu/intel-iommu/cap
1c0000c40660462

So strtoul() assumes the value is decimal in this case, as explained in
its man page:

"
DESCRIPTION
<snip\>
If base is zero or 16, the string may then include a "0x" prefix, and
the number will be read in base 16; otherwise, a zero base is taken as
10 (decimal)
"

Thanks,
Maxime