On Mon, Nov 01, 2021 at 09:36:59AM +0200, Dmitry Kozlyuk wrote:
> FreeBSD EAL selected IOVA mode PA even in --no-huge mode
> where PA are not available. Memory zones were created with IOVA
> equal to RTE_BAD_IOVA with no indication this field is not usable.
>
> Change IOVA mode detection:
> 1. Always allow to force --iova-mode=va.
> 2. In --no-huge mode, disallow forcing --iova-mode=pa, and select VA.
> 3. Otherwise select IOVA mode according to bus requests, default to PA.
> In case contigmem is inaccessible, memory initialization will fail
> with a message indicating the cause.
>
> Fixes: c2361bab70c5 ("eal: compute IOVA mode based on PA availability")
> Cc: benjamin.wal...@intel.com
> Cc: sta...@dpdk.org
>
> Signed-off-by: Dmitry Kozlyuk <dkozl...@nvidia.com>
> ---
> lib/eal/freebsd/eal.c | 37 +++++++++++++++++++++++++------------
> 1 file changed, 25 insertions(+), 12 deletions(-)
>
> diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
> index 9935356ed4..519a645344 100644
> --- a/lib/eal/freebsd/eal.c
> +++ b/lib/eal/freebsd/eal.c
> @@ -677,6 +677,8 @@ rte_eal_init(int argc, char **argv)
> const struct rte_config *config = rte_eal_get_configuration();
> struct internal_config *internal_conf =
> eal_get_internal_configuration();
> + bool has_phys_addr;
> + enum rte_iova_mode iova_mode;
>
> /* checks if the machine is adequate */
> if (!rte_cpu_is_supported()) {
> @@ -777,21 +779,32 @@ rte_eal_init(int argc, char **argv)
> return -1;
> }
>
> - /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
> - if (internal_conf->iova_mode == RTE_IOVA_DC) {
> - /* autodetect the IOVA mapping mode (default is RTE_IOVA_PA) */
> - enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
> -
> - if (iova_mode == RTE_IOVA_DC)
> - iova_mode = RTE_IOVA_PA;
> - rte_eal_get_configuration()->iova_mode = iova_mode;
> - } else {
> - rte_eal_get_configuration()->iova_mode =
> - internal_conf->iova_mode;
> + /*
> + * PA are only available for hugepages via contigmem.
> + * If contigmem is inaccessible, rte_eal_hugepage_init() will fail
> + * with a message describing the cause.
> + */
> + has_phys_addr = internal_conf->no_hugetlbfs == 0;
> + iova_mode = internal_conf->iova_mode;
> + if (iova_mode == RTE_IOVA_PA && !has_phys_addr) {
> + rte_eal_init_alert("Cannot use IOVA as 'PA' since physical
> addresses are not available");
> + rte_errno = EINVAL;
> + return -1;
> + }
> + if (iova_mode == RTE_IOVA_DC) {
> + RTE_LOG(DEBUG, EAL, "Specific IOVA mode is not requested,
> autodetecting\n");
> + if (has_phys_addr) {
> + RTE_LOG(DEBUG, EAL, "Selecting IOVA mode according to
> bus requests\n");
> + iova_mode = rte_bus_get_iommu_class();
> + if (iova_mode == RTE_IOVA_DC)
> + iova_mode = RTE_IOVA_PA;
> + } else {
> + iova_mode = RTE_IOVA_VA;
> + }
> }
> -
> RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
> rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
> + rte_eal_get_configuration()->iova_mode = iova_mode;
>
This line needs to come before the log, or else the log statement needs to
be changed to use iova_mode rather than rte_eal_iova_mode(). I'd suggest
the former option.
With that change, you can add my ack to v2:
Acked-by: Bruce Richardson <bruce.richard...@intel.com>
