Re: [PATCH libpciaccess] hurd: Don't read from _SERVERS_BUS_PCI necessarily

[Samuel Thibault]

Damien Zammit, le lun. 14 févr. 2022 09:23:33 +0000, a ecrit:
> This allows the fallback mechanism in the hurd create method to be
> used in the map range method by reusing the mach port that
> corresponds to the root of the pci filesystem.

> @@ -168,11 +178,11 @@ pci_device_hurd_map_range(struct pci_device *dev,
>          flags = O_RDWR;
>      }
> 
> -    snprintf(server, NAME_MAX, "%s/%04x/%02x/%02x/%01u/%s%01u",
> -            _SERVERS_BUS_PCI, dev->domain, dev->bus, dev->dev, dev->func,
> -            FILE_REGION_NAME, map->region);
> +    snprintf(server, NAME_MAX, "./%04x/%02x/%02x/%01u/%s%01u",

Does "%04x/%02x/%02x/%01u/%s%01u" not work?

> Also, the offset parameter to vm_map was not needed as the
> whole range is always mapped from the beginning.

Please avoid stuffing (hiding?) together unrelated concerns.

>          return EINVAL;
>      }
> 
> -    offset = map->base - region->base_addr;
>      err = vm_map (mach_task_self (), (vm_address_t *)&map->memory, map->size,
>                    0, 1,
>                    pager, /* a memory object proxy containing only the region 
> */
> -                  offset, /* offset from region start */
> +                  0, /* offset is beginning of region */

This looks odd, can't map->base be different from region->base_addr?

Applications can call pci_device_map_legacy any way they'd like.

Samuel