Messages in this thread |  | | | Date | Thu, 10 Jul 2008 14:37:47 +0200 | | From | Joerg Roedel <> | | Subject | Re: [PATCH 14/34] AMD IOMMU: clue initialization code together |
| |
On Wed, Jul 09, 2008 at 06:55:12PM -0700, Andrew Morton wrote: > On Thu, 26 Jun 2008 21:27:50 +0200 Joerg Roedel <joerg.roedel@amd.com> wrote: > > > This patch puts the AMD IOMMU ACPI table parsing and hardware initialization > > functions together to the main intialization routine. > > > > index 555fcc9..c792ddc 100644 > > --- a/arch/x86/kernel/amd_iommu_init.c > > +++ b/arch/x86/kernel/amd_iommu_init.c > > @@ -643,3 +643,129 @@ static int __init init_memory_definitions(struct acpi_table_header *table) > > return 0; > > } > > > > +int __init amd_iommu_init(void) > > +{ > > + int i, ret = 0; > > + > > + > > + if (amd_iommu_disable) { > > + printk(KERN_INFO "AMD IOMMU disabled by kernel command line\n"); > > + return 0; > > + } > > + > > + /* > > + * First parse ACPI tables to find the largest Bus/Dev/Func > > + * we need to handle. Upon this information the shared data > > + * structures for the IOMMUs in the system will be allocated > > + */ > > + if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0) > > + return -ENODEV; > > + > > + dev_table_size = TBL_SIZE(DEV_TABLE_ENTRY_SIZE); > > + alias_table_size = TBL_SIZE(ALIAS_TABLE_ENTRY_SIZE); > > + rlookup_table_size = TBL_SIZE(RLOOKUP_TABLE_ENTRY_SIZE); > > + > > + ret = -ENOMEM; > > + > > + /* Device table - directly used by all IOMMUs */ > > + amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL, > > + get_order(dev_table_size)); > > Is there a reason for using __get_free_pages() in this manner everywhere? > kmalloc() or kzalloc() are unsuitable?
Yes. The device table has to be page aligned and has a maximum size of 2 MB.
> > + if (amd_iommu_dev_table == NULL) > > + goto out; > > + > > + /* > > + * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the > > + * IOMMU see for that device > > + */ > > + amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL, > > + get_order(alias_table_size));
Hmm, maybe this table will fit into a kmalloc allocation. Its maximum size is 128 kb.
> > + if (amd_iommu_alias_table == NULL) > > + goto free; > > + > > + /* IOMMU rlookup table - find the IOMMU for a specific device */ > > + amd_iommu_rlookup_table = (void *)__get_free_pages(GFP_KERNEL, > > + get_order(rlookup_table_size)); > > + if (amd_iommu_rlookup_table == NULL) > > + goto free; > > + > > + /* > > + * Protection Domain table - maps devices to protection domains > > + * This table has the same size as the rlookup_table > > + */ > > + amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL, > > + get_order(rlookup_table_size));
The rlookup and the pd table have a maximum size of 1 MB.
> > + if (amd_iommu_pd_table == NULL) > > + goto free; > > + > > + amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(GFP_KERNEL, > > + get_order(MAX_DOMAIN_ID/8)); > > + if (amd_iommu_pd_alloc_bitmap == NULL) > > + goto free; > > + > > + /* > > + * memory is allocated now; initialize the device table with all zeroes > > + * and let all alias entries point to itself > > + */ > > + memset(amd_iommu_dev_table, 0, dev_table_size); > > + for (i = 0; i < amd_iommu_last_bdf; ++i) > > + amd_iommu_alias_table[i] = i; > > + > > + memset(amd_iommu_pd_table, 0, rlookup_table_size); > > + memset(amd_iommu_pd_alloc_bitmap, 0, MAX_DOMAIN_ID / 8); > > Use of __GFP_ZERO in here would relieve us of a lot of memsets. > > > + /* > > + * never allocate domain 0 because its used as the non-allocated and > > + * error value placeholder > > + */ > > + amd_iommu_pd_alloc_bitmap[0] = 1; > > + > > + /* > > + * now the data structures are allocated and basically initialized > > + * start the real acpi table scan > > + */ > > + ret = -ENODEV; > > + if (acpi_table_parse("IVRS", init_iommu_all) != 0) > > + goto free; > > + > > + if (acpi_table_parse("IVRS", init_memory_definitions) != 0) > > + goto free; > > + > > + printk(KERN_INFO "AMD IOMMU: aperture size is %d MB\n", > > + (1 << (amd_iommu_aperture_order-20))); > > + > > + printk(KERN_INFO "AMD IOMMU: device isolation "); > > + if (amd_iommu_isolate) > > + printk("enabled\n"); > > + else > > + printk("disabled\n"); > > + > > +out: > > + return ret; > > + > > +free: > > + if (amd_iommu_pd_alloc_bitmap) > > + free_pages((unsigned long)amd_iommu_pd_alloc_bitmap, 1); > > + > > + if (amd_iommu_pd_table) > > + free_pages((unsigned long)amd_iommu_pd_table, > > + get_order(rlookup_table_size)); > > + > > + if (amd_iommu_rlookup_table) > > + free_pages((unsigned long)amd_iommu_rlookup_table, > > + get_order(rlookup_table_size)); > > + > > + if (amd_iommu_alias_table) > > + free_pages((unsigned long)amd_iommu_alias_table, > > + get_order(alias_table_size)); > > + > > + if (amd_iommu_dev_table) > > + free_pages((unsigned long)amd_iommu_dev_table, > > + get_order(dev_table_size)); > > free_pages(0) is legal, so all these tests are unneeded, I expect. > > This remains true if you decide to switch to kzalloc().
Thanks for that hint. I will remove the checks.
> > + free_iommu_all(); > > + > > + free_unity_maps(); > > + > > + goto out; > > +} > >
-- | AMD Saxony Limited Liability Company & Co. KG Operating | Wilschdorfer Landstr. 101, 01109 Dresden, Germany System | Register Court Dresden: HRA 4896 Research | General Partner authorized to represent: Center | AMD Saxony LLC (Wilmington, Delaware, US) | General Manager of AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy
|  |