Re: [PATCH net-next 1/2] gve: Update QPL page registration logic

[Paolo Abeni]

On 2/7/26 2:17 AM, Joshua Washington wrote:
> +void gve_update_num_qpl_pages(struct gve_priv *priv,
> +			      struct gve_rx_alloc_rings_cfg *rx_alloc_cfg,
> +			      struct gve_tx_alloc_rings_cfg *tx_alloc_cfg)
> +{
> +	u64 ideal_tx_pages, ideal_rx_pages;
> +	u16 tx_num_queues, rx_num_queues;
> +	u64 max_pages, tx_pages;
> +
> +	if (priv->queue_format == GVE_GQI_QPL_FORMAT) {
> +		rx_alloc_cfg->pages_per_qpl = rx_alloc_cfg->ring_size;
> +	} else if (priv->queue_format == GVE_DQO_QPL_FORMAT) {
> +		/*
> +		 * We want 2 pages per RX descriptor and half a page per TX
> +		 * descriptor, which means the fraction ideal_tx_pages /
> +		 * (ideal_tx_pages + ideal_rx_pages) of the pages we allocate
> +		 * should be for TX. Shrink proportionally as necessary to avoid
> +		 * allocating more than max_registered_pages total pages.
> +		 */
> +		tx_num_queues = tx_alloc_cfg->qcfg->num_queues;
> +		rx_num_queues = rx_alloc_cfg->qcfg_rx->num_queues;
> +
> +		ideal_tx_pages = tx_alloc_cfg->ring_size * tx_num_queues / 2;
> +		ideal_rx_pages = rx_alloc_cfg->ring_size * rx_num_queues * 2;
> +		max_pages = min(priv->max_registered_pages,
> +				ideal_tx_pages + ideal_rx_pages);
> +
> +		tx_pages = (max_pages * ideal_tx_pages) /
> +			   (ideal_tx_pages + ideal_rx_pages);
> +		tx_alloc_cfg->pages_per_qpl = tx_pages / tx_num_queues;
> +		rx_alloc_cfg->pages_per_qpl = (max_pages - tx_pages) /
> +					      rx_num_queues;

Does not build on 32 bits systems:

RROR: modpost: "__udivdi3" [drivers/net/ethernet/google/gve/gve.ko]
undefined!
make[3]: *** [../scripts/Makefile.modpost:147: Module.symvers] Error 1
make[2]: *** [/srv/nipa-poller/net-next/wt-1/Makefile:2005: modpost] Error 2
make[1]: *** [/srv/nipa-poller/net-next/wt-1/Makefile:248: __sub-make]
Error 2
make: *** [Makefile:248: __sub-make] Error 2

AFAICS because above 'rx_num_queues' is implicitly promoted to u64, and
the statement yield a 64 bits division. You need to use a div64* variant
here (and possibly somewhere else, too).

/P