#include <linux/pci.h>
#include <linux/gfp.h>
#include <linux/bitops.h>
+#include <linux/debugfs.h>
#include <linux/scatterlist.h>
+#include <linux/dma-mapping.h>
#include <linux/iommu-helper.h>
+#include <linux/iommu.h>
#include <asm/proto.h>
#include <asm/iommu.h>
#include <asm/gart.h>
static LIST_HEAD(iommu_pd_list);
static DEFINE_SPINLOCK(iommu_pd_list_lock);
+#ifdef CONFIG_IOMMU_API
+static struct iommu_ops amd_iommu_ops;
+#endif
+
/*
* general struct to manage commands send to an IOMMU
*/
static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
struct unity_map_entry *e);
+static struct dma_ops_domain *find_protection_domain(u16 devid);
+static u64* alloc_pte(struct protection_domain *dom,
+ unsigned long address, u64
+ **pte_page, gfp_t gfp);
+static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
+ unsigned long start_page,
+ unsigned int pages);
+
+#ifndef BUS_NOTIFY_UNBOUND_DRIVER
+#define BUS_NOTIFY_UNBOUND_DRIVER 0x0005
+#endif
+
+#ifdef CONFIG_AMD_IOMMU_STATS
+
+/*
+ * Initialization code for statistics collection
+ */
+
+DECLARE_STATS_COUNTER(compl_wait);
+DECLARE_STATS_COUNTER(cnt_map_single);
+DECLARE_STATS_COUNTER(cnt_unmap_single);
+DECLARE_STATS_COUNTER(cnt_map_sg);
+DECLARE_STATS_COUNTER(cnt_unmap_sg);
+DECLARE_STATS_COUNTER(cnt_alloc_coherent);
+DECLARE_STATS_COUNTER(cnt_free_coherent);
+DECLARE_STATS_COUNTER(cross_page);
+DECLARE_STATS_COUNTER(domain_flush_single);
+DECLARE_STATS_COUNTER(domain_flush_all);
+DECLARE_STATS_COUNTER(alloced_io_mem);
+DECLARE_STATS_COUNTER(total_map_requests);
+
+static struct dentry *stats_dir;
+static struct dentry *de_isolate;
+static struct dentry *de_fflush;
+
+static void amd_iommu_stats_add(struct __iommu_counter *cnt)
+{
+ if (stats_dir == NULL)
+ return;
+
+ cnt->dent = debugfs_create_u64(cnt->name, 0444, stats_dir,
+ &cnt->value);
+}
+
+static void amd_iommu_stats_init(void)
+{
+ stats_dir = debugfs_create_dir("amd-iommu", NULL);
+ if (stats_dir == NULL)
+ return;
+
+ de_isolate = debugfs_create_bool("isolation", 0444, stats_dir,
+ (u32 *)&amd_iommu_isolate);
+
+ de_fflush = debugfs_create_bool("fullflush", 0444, stats_dir,
+ (u32 *)&amd_iommu_unmap_flush);
+
+ amd_iommu_stats_add(&compl_wait);
+ amd_iommu_stats_add(&cnt_map_single);
+ amd_iommu_stats_add(&cnt_unmap_single);
+ amd_iommu_stats_add(&cnt_map_sg);
+ amd_iommu_stats_add(&cnt_unmap_sg);
+ amd_iommu_stats_add(&cnt_alloc_coherent);
+ amd_iommu_stats_add(&cnt_free_coherent);
+ amd_iommu_stats_add(&cross_page);
+ amd_iommu_stats_add(&domain_flush_single);
+ amd_iommu_stats_add(&domain_flush_all);
+ amd_iommu_stats_add(&alloced_io_mem);
+ amd_iommu_stats_add(&total_map_requests);
+}
+
+#endif
/* returns !0 if the IOMMU is caching non-present entries in its TLB */
static int iommu_has_npcache(struct amd_iommu *iommu)
{
struct amd_iommu *iommu;
- list_for_each_entry(iommu, &amd_iommu_list, list)
+ for_each_iommu(iommu)
iommu_poll_events(iommu);
return IRQ_HANDLED;
spin_lock_irqsave(&iommu->lock, flags);
ret = __iommu_queue_command(iommu, cmd);
+ if (!ret)
+ iommu->need_sync = true;
spin_unlock_irqrestore(&iommu->lock, flags);
return ret;
}
/*
+ * This function waits until an IOMMU has completed a completion
+ * wait command
+ */
+static void __iommu_wait_for_completion(struct amd_iommu *iommu)
+{
+ int ready = 0;
+ unsigned status = 0;
+ unsigned long i = 0;
+
+ INC_STATS_COUNTER(compl_wait);
+
+ while (!ready && (i < EXIT_LOOP_COUNT)) {
+ ++i;
+ /* wait for the bit to become one */
+ status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
+ ready = status & MMIO_STATUS_COM_WAIT_INT_MASK;
+ }
+
+ /* set bit back to zero */
+ status &= ~MMIO_STATUS_COM_WAIT_INT_MASK;
+ writel(status, iommu->mmio_base + MMIO_STATUS_OFFSET);
+
+ if (unlikely(i == EXIT_LOOP_COUNT))
+ panic("AMD IOMMU: Completion wait loop failed\n");
+}
+
+/*
+ * This function queues a completion wait command into the command
+ * buffer of an IOMMU
+ */
+static int __iommu_completion_wait(struct amd_iommu *iommu)
+{
+ struct iommu_cmd cmd;
+
+ memset(&cmd, 0, sizeof(cmd));
+ cmd.data[0] = CMD_COMPL_WAIT_INT_MASK;
+ CMD_SET_TYPE(&cmd, CMD_COMPL_WAIT);
+
+ return __iommu_queue_command(iommu, &cmd);
+}
+
+/*
* This function is called whenever we need to ensure that the IOMMU has
* completed execution of all commands we sent. It sends a
* COMPLETION_WAIT command and waits for it to finish. The IOMMU informs
*/
static int iommu_completion_wait(struct amd_iommu *iommu)
{
- int ret = 0, ready = 0;
- unsigned status = 0;
- struct iommu_cmd cmd;
- unsigned long flags, i = 0;
+ int ret = 0;
+ unsigned long flags;
- memset(&cmd, 0, sizeof(cmd));
- cmd.data[0] = CMD_COMPL_WAIT_INT_MASK;
- CMD_SET_TYPE(&cmd, CMD_COMPL_WAIT);
+ spin_lock_irqsave(&iommu->lock, flags);
- iommu->need_sync = 0;
+ if (!iommu->need_sync)
+ goto out;
- spin_lock_irqsave(&iommu->lock, flags);
+ ret = __iommu_completion_wait(iommu);
- ret = __iommu_queue_command(iommu, &cmd);
+ iommu->need_sync = false;
if (ret)
goto out;
- while (!ready && (i < EXIT_LOOP_COUNT)) {
- ++i;
- /* wait for the bit to become one */
- status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
- ready = status & MMIO_STATUS_COM_WAIT_INT_MASK;
- }
+ __iommu_wait_for_completion(iommu);
- /* set bit back to zero */
- status &= ~MMIO_STATUS_COM_WAIT_INT_MASK;
- writel(status, iommu->mmio_base + MMIO_STATUS_OFFSET);
-
- if (unlikely((i == EXIT_LOOP_COUNT) && printk_ratelimit()))
- printk(KERN_WARNING "AMD IOMMU: Completion wait loop failed\n");
out:
spin_unlock_irqrestore(&iommu->lock, flags);
ret = iommu_queue_command(iommu, &cmd);
- iommu->need_sync = 1;
-
return ret;
}
+static void __iommu_build_inv_iommu_pages(struct iommu_cmd *cmd, u64 address,
+ u16 domid, int pde, int s)
+{
+ memset(cmd, 0, sizeof(*cmd));
+ address &= PAGE_MASK;
+ CMD_SET_TYPE(cmd, CMD_INV_IOMMU_PAGES);
+ cmd->data[1] |= domid;
+ cmd->data[2] = lower_32_bits(address);
+ cmd->data[3] = upper_32_bits(address);
+ if (s) /* size bit - we flush more than one 4kb page */
+ cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK;
+ if (pde) /* PDE bit - we wan't flush everything not only the PTEs */
+ cmd->data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK;
+}
+
/*
* Generic command send function for invalidaing TLB entries
*/
struct iommu_cmd cmd;
int ret;
- memset(&cmd, 0, sizeof(cmd));
- address &= PAGE_MASK;
- CMD_SET_TYPE(&cmd, CMD_INV_IOMMU_PAGES);
- cmd.data[1] |= domid;
- cmd.data[2] = lower_32_bits(address);
- cmd.data[3] = upper_32_bits(address);
- if (s) /* size bit - we flush more than one 4kb page */
- cmd.data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK;
- if (pde) /* PDE bit - we wan't flush everything not only the PTEs */
- cmd.data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK;
+ __iommu_build_inv_iommu_pages(&cmd, address, domid, pde, s);
ret = iommu_queue_command(iommu, &cmd);
- iommu->need_sync = 1;
-
return ret;
}
{
u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
+ INC_STATS_COUNTER(domain_flush_single);
+
iommu_queue_inv_iommu_pages(iommu, address, domid, 0, 1);
}
+/* Flush the whole IO/TLB for a given protection domain - including PDE */
+static void iommu_flush_tlb_pde(struct amd_iommu *iommu, u16 domid)
+{
+ u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
+
+ INC_STATS_COUNTER(domain_flush_single);
+
+ iommu_queue_inv_iommu_pages(iommu, address, domid, 1, 1);
+}
+
+/*
+ * This function is used to flush the IO/TLB for a given protection domain
+ * on every IOMMU in the system
+ */
+static void iommu_flush_domain(u16 domid)
+{
+ unsigned long flags;
+ struct amd_iommu *iommu;
+ struct iommu_cmd cmd;
+
+ INC_STATS_COUNTER(domain_flush_all);
+
+ __iommu_build_inv_iommu_pages(&cmd, CMD_INV_IOMMU_ALL_PAGES_ADDRESS,
+ domid, 1, 1);
+
+ for_each_iommu(iommu) {
+ spin_lock_irqsave(&iommu->lock, flags);
+ __iommu_queue_command(iommu, &cmd);
+ __iommu_completion_wait(iommu);
+ __iommu_wait_for_completion(iommu);
+ spin_unlock_irqrestore(&iommu->lock, flags);
+ }
+}
+
+void amd_iommu_flush_all_domains(void)
+{
+ int i;
+
+ for (i = 1; i < MAX_DOMAIN_ID; ++i) {
+ if (!test_bit(i, amd_iommu_pd_alloc_bitmap))
+ continue;
+ iommu_flush_domain(i);
+ }
+}
+
+void amd_iommu_flush_all_devices(void)
+{
+ struct amd_iommu *iommu;
+ int i;
+
+ for (i = 0; i <= amd_iommu_last_bdf; ++i) {
+ if (amd_iommu_pd_table[i] == NULL)
+ continue;
+
+ iommu = amd_iommu_rlookup_table[i];
+ if (!iommu)
+ continue;
+
+ iommu_queue_inv_dev_entry(iommu, i);
+ iommu_completion_wait(iommu);
+ }
+}
+
/****************************************************************************
*
* The functions below are used the create the page table mappings for
* supporting all features of AMD IOMMU page tables like level skipping
* and full 64 bit address spaces.
*/
-static int iommu_map(struct protection_domain *dom,
- unsigned long bus_addr,
- unsigned long phys_addr,
- int prot)
+static int iommu_map_page(struct protection_domain *dom,
+ unsigned long bus_addr,
+ unsigned long phys_addr,
+ int prot)
{
- u64 __pte, *pte, *page;
+ u64 __pte, *pte;
bus_addr = PAGE_ALIGN(bus_addr);
- phys_addr = PAGE_ALIGN(bus_addr);
+ phys_addr = PAGE_ALIGN(phys_addr);
/* only support 512GB address spaces for now */
if (bus_addr > IOMMU_MAP_SIZE_L3 || !(prot & IOMMU_PROT_MASK))
return -EINVAL;
- pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(bus_addr)];
-
- if (!IOMMU_PTE_PRESENT(*pte)) {
- page = (u64 *)get_zeroed_page(GFP_KERNEL);
- if (!page)
- return -ENOMEM;
- *pte = IOMMU_L2_PDE(virt_to_phys(page));
- }
-
- pte = IOMMU_PTE_PAGE(*pte);
- pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)];
-
- if (!IOMMU_PTE_PRESENT(*pte)) {
- page = (u64 *)get_zeroed_page(GFP_KERNEL);
- if (!page)
- return -ENOMEM;
- *pte = IOMMU_L1_PDE(virt_to_phys(page));
- }
-
- pte = IOMMU_PTE_PAGE(*pte);
- pte = &pte[IOMMU_PTE_L0_INDEX(bus_addr)];
+ pte = alloc_pte(dom, bus_addr, NULL, GFP_KERNEL);
if (IOMMU_PTE_PRESENT(*pte))
return -EBUSY;
return 0;
}
+static void iommu_unmap_page(struct protection_domain *dom,
+ unsigned long bus_addr)
+{
+ u64 *pte;
+
+ pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(bus_addr)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return;
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return;
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)];
+
+ *pte = 0;
+}
+
/*
* This function checks if a specific unity mapping entry is needed for
* this specific IOMMU.
for (addr = e->address_start; addr < e->address_end;
addr += PAGE_SIZE) {
- ret = iommu_map(&dma_dom->domain, addr, addr, e->prot);
+ ret = iommu_map_page(&dma_dom->domain, addr, addr, e->prot);
if (ret)
return ret;
/*
* as allocated in the aperture
*/
if (addr < dma_dom->aperture_size)
- __set_bit(addr >> PAGE_SHIFT, dma_dom->bitmap);
+ __set_bit(addr >> PAGE_SHIFT,
+ dma_dom->aperture[0]->bitmap);
}
return 0;
****************************************************************************/
/*
- * The address allocator core function.
+ * The address allocator core functions.
*
* called with domain->lock held
*/
+
+/*
+ * This function checks if there is a PTE for a given dma address. If
+ * there is one, it returns the pointer to it.
+ */
+static u64* fetch_pte(struct protection_domain *domain,
+ unsigned long address)
+{
+ u64 *pte;
+
+ pte = &domain->pt_root[IOMMU_PTE_L2_INDEX(address)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return NULL;
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L1_INDEX(address)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return NULL;
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L0_INDEX(address)];
+
+ return pte;
+}
+
+/*
+ * This function is used to add a new aperture range to an existing
+ * aperture in case of dma_ops domain allocation or address allocation
+ * failure.
+ */
+static int alloc_new_range(struct amd_iommu *iommu,
+ struct dma_ops_domain *dma_dom,
+ bool populate, gfp_t gfp)
+{
+ int index = dma_dom->aperture_size >> APERTURE_RANGE_SHIFT;
+ int i;
+
+#ifdef CONFIG_IOMMU_STRESS
+ populate = false;
+#endif
+
+ if (index >= APERTURE_MAX_RANGES)
+ return -ENOMEM;
+
+ dma_dom->aperture[index] = kzalloc(sizeof(struct aperture_range), gfp);
+ if (!dma_dom->aperture[index])
+ return -ENOMEM;
+
+ dma_dom->aperture[index]->bitmap = (void *)get_zeroed_page(gfp);
+ if (!dma_dom->aperture[index]->bitmap)
+ goto out_free;
+
+ dma_dom->aperture[index]->offset = dma_dom->aperture_size;
+
+ if (populate) {
+ unsigned long address = dma_dom->aperture_size;
+ int i, num_ptes = APERTURE_RANGE_PAGES / 512;
+ u64 *pte, *pte_page;
+
+ for (i = 0; i < num_ptes; ++i) {
+ pte = alloc_pte(&dma_dom->domain, address,
+ &pte_page, gfp);
+ if (!pte)
+ goto out_free;
+
+ dma_dom->aperture[index]->pte_pages[i] = pte_page;
+
+ address += APERTURE_RANGE_SIZE / 64;
+ }
+ }
+
+ dma_dom->aperture_size += APERTURE_RANGE_SIZE;
+
+ /* Intialize the exclusion range if necessary */
+ if (iommu->exclusion_start &&
+ iommu->exclusion_start >= dma_dom->aperture[index]->offset &&
+ iommu->exclusion_start < dma_dom->aperture_size) {
+ unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
+ int pages = iommu_num_pages(iommu->exclusion_start,
+ iommu->exclusion_length,
+ PAGE_SIZE);
+ dma_ops_reserve_addresses(dma_dom, startpage, pages);
+ }
+
+ /*
+ * Check for areas already mapped as present in the new aperture
+ * range and mark those pages as reserved in the allocator. Such
+ * mappings may already exist as a result of requested unity
+ * mappings for devices.
+ */
+ for (i = dma_dom->aperture[index]->offset;
+ i < dma_dom->aperture_size;
+ i += PAGE_SIZE) {
+ u64 *pte = fetch_pte(&dma_dom->domain, i);
+ if (!pte || !IOMMU_PTE_PRESENT(*pte))
+ continue;
+
+ dma_ops_reserve_addresses(dma_dom, i << PAGE_SHIFT, 1);
+ }
+
+ return 0;
+
+out_free:
+ free_page((unsigned long)dma_dom->aperture[index]->bitmap);
+
+ kfree(dma_dom->aperture[index]);
+ dma_dom->aperture[index] = NULL;
+
+ return -ENOMEM;
+}
+
+static unsigned long dma_ops_area_alloc(struct device *dev,
+ struct dma_ops_domain *dom,
+ unsigned int pages,
+ unsigned long align_mask,
+ u64 dma_mask,
+ unsigned long start)
+{
+ unsigned long next_bit = dom->next_address % APERTURE_RANGE_SIZE;
+ int max_index = dom->aperture_size >> APERTURE_RANGE_SHIFT;
+ int i = start >> APERTURE_RANGE_SHIFT;
+ unsigned long boundary_size;
+ unsigned long address = -1;
+ unsigned long limit;
+
+ next_bit >>= PAGE_SHIFT;
+
+ boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
+ PAGE_SIZE) >> PAGE_SHIFT;
+
+ for (;i < max_index; ++i) {
+ unsigned long offset = dom->aperture[i]->offset >> PAGE_SHIFT;
+
+ if (dom->aperture[i]->offset >= dma_mask)
+ break;
+
+ limit = iommu_device_max_index(APERTURE_RANGE_PAGES, offset,
+ dma_mask >> PAGE_SHIFT);
+
+ address = iommu_area_alloc(dom->aperture[i]->bitmap,
+ limit, next_bit, pages, 0,
+ boundary_size, align_mask);
+ if (address != -1) {
+ address = dom->aperture[i]->offset +
+ (address << PAGE_SHIFT);
+ dom->next_address = address + (pages << PAGE_SHIFT);
+ break;
+ }
+
+ next_bit = 0;
+ }
+
+ return address;
+}
+
static unsigned long dma_ops_alloc_addresses(struct device *dev,
struct dma_ops_domain *dom,
unsigned int pages,
unsigned long align_mask,
u64 dma_mask)
{
- unsigned long limit;
unsigned long address;
- unsigned long boundary_size;
- boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
- PAGE_SIZE) >> PAGE_SHIFT;
- limit = iommu_device_max_index(dom->aperture_size >> PAGE_SHIFT, 0,
- dma_mask >> PAGE_SHIFT);
+#ifdef CONFIG_IOMMU_STRESS
+ dom->next_address = 0;
+ dom->need_flush = true;
+#endif
- if (dom->next_bit >= limit) {
- dom->next_bit = 0;
- dom->need_flush = true;
- }
+ address = dma_ops_area_alloc(dev, dom, pages, align_mask,
+ dma_mask, dom->next_address);
- address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages,
- 0 , boundary_size, align_mask);
if (address == -1) {
- address = iommu_area_alloc(dom->bitmap, limit, 0, pages,
- 0, boundary_size, align_mask);
+ dom->next_address = 0;
+ address = dma_ops_area_alloc(dev, dom, pages, align_mask,
+ dma_mask, 0);
dom->need_flush = true;
}
- if (likely(address != -1)) {
- dom->next_bit = address + pages;
- address <<= PAGE_SHIFT;
- } else
+ if (unlikely(address == -1))
address = bad_dma_address;
WARN_ON((address + (PAGE_SIZE*pages)) > dom->aperture_size);
unsigned long address,
unsigned int pages)
{
- address >>= PAGE_SHIFT;
- iommu_area_free(dom->bitmap, address, pages);
+ unsigned i = address >> APERTURE_RANGE_SHIFT;
+ struct aperture_range *range = dom->aperture[i];
+
+ BUG_ON(i >= APERTURE_MAX_RANGES || range == NULL);
+
+#ifdef CONFIG_IOMMU_STRESS
+ if (i < 4)
+ return;
+#endif
- if (address >= dom->next_bit)
+ if (address >= dom->next_address)
dom->need_flush = true;
+
+ address = (address % APERTURE_RANGE_SIZE) >> PAGE_SHIFT;
+
+ iommu_area_free(range->bitmap, address, pages);
+
}
/****************************************************************************
return id;
}
+static void domain_id_free(int id)
+{
+ unsigned long flags;
+
+ write_lock_irqsave(&amd_iommu_devtable_lock, flags);
+ if (id > 0 && id < MAX_DOMAIN_ID)
+ __clear_bit(id, amd_iommu_pd_alloc_bitmap);
+ write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+}
+
/*
* Used to reserve address ranges in the aperture (e.g. for exclusion
* ranges.
unsigned long start_page,
unsigned int pages)
{
- unsigned int last_page = dom->aperture_size >> PAGE_SHIFT;
+ unsigned int i, last_page = dom->aperture_size >> PAGE_SHIFT;
if (start_page + pages > last_page)
pages = last_page - start_page;
- iommu_area_reserve(dom->bitmap, start_page, pages);
+ for (i = start_page; i < start_page + pages; ++i) {
+ int index = i / APERTURE_RANGE_PAGES;
+ int page = i % APERTURE_RANGE_PAGES;
+ __set_bit(page, dom->aperture[index]->bitmap);
+ }
}
-static void dma_ops_free_pagetable(struct dma_ops_domain *dma_dom)
+static void free_pagetable(struct protection_domain *domain)
{
int i, j;
u64 *p1, *p2, *p3;
- p1 = dma_dom->domain.pt_root;
+ p1 = domain->pt_root;
if (!p1)
return;
continue;
p2 = IOMMU_PTE_PAGE(p1[i]);
- for (j = 0; j < 512; ++i) {
+ for (j = 0; j < 512; ++j) {
if (!IOMMU_PTE_PRESENT(p2[j]))
continue;
p3 = IOMMU_PTE_PAGE(p2[j]);
}
free_page((unsigned long)p1);
+
+ domain->pt_root = NULL;
}
/*
*/
static void dma_ops_domain_free(struct dma_ops_domain *dom)
{
+ int i;
+
if (!dom)
return;
- dma_ops_free_pagetable(dom);
+ free_pagetable(&dom->domain);
- kfree(dom->pte_pages);
-
- kfree(dom->bitmap);
+ for (i = 0; i < APERTURE_MAX_RANGES; ++i) {
+ if (!dom->aperture[i])
+ continue;
+ free_page((unsigned long)dom->aperture[i]->bitmap);
+ kfree(dom->aperture[i]);
+ }
kfree(dom);
}
* It also intializes the page table and the address allocator data
* structures required for the dma_ops interface
*/
-static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
- unsigned order)
+static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu)
{
struct dma_ops_domain *dma_dom;
- unsigned i, num_pte_pages;
- u64 *l2_pde;
- u64 address;
-
- /*
- * Currently the DMA aperture must be between 32 MB and 1GB in size
- */
- if ((order < 25) || (order > 30))
- return NULL;
dma_dom = kzalloc(sizeof(struct dma_ops_domain), GFP_KERNEL);
if (!dma_dom)
goto free_dma_dom;
dma_dom->domain.mode = PAGE_MODE_3_LEVEL;
dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL);
+ dma_dom->domain.flags = PD_DMA_OPS_MASK;
dma_dom->domain.priv = dma_dom;
if (!dma_dom->domain.pt_root)
goto free_dma_dom;
- dma_dom->aperture_size = (1ULL << order);
- dma_dom->bitmap = kzalloc(dma_dom->aperture_size / (PAGE_SIZE * 8),
- GFP_KERNEL);
- if (!dma_dom->bitmap)
- goto free_dma_dom;
- /*
- * mark the first page as allocated so we never return 0 as
- * a valid dma-address. So we can use 0 as error value
- */
- dma_dom->bitmap[0] = 1;
- dma_dom->next_bit = 0;
dma_dom->need_flush = false;
dma_dom->target_dev = 0xffff;
- /* Intialize the exclusion range if necessary */
- if (iommu->exclusion_start &&
- iommu->exclusion_start < dma_dom->aperture_size) {
- unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
- int pages = iommu_num_pages(iommu->exclusion_start,
- iommu->exclusion_length,
- PAGE_SIZE);
- dma_ops_reserve_addresses(dma_dom, startpage, pages);
- }
+ if (alloc_new_range(iommu, dma_dom, true, GFP_KERNEL))
+ goto free_dma_dom;
/*
- * At the last step, build the page tables so we don't need to
- * allocate page table pages in the dma_ops mapping/unmapping
- * path.
+ * mark the first page as allocated so we never return 0 as
+ * a valid dma-address. So we can use 0 as error value
*/
- num_pte_pages = dma_dom->aperture_size / (PAGE_SIZE * 512);
- dma_dom->pte_pages = kzalloc(num_pte_pages * sizeof(void *),
- GFP_KERNEL);
- if (!dma_dom->pte_pages)
- goto free_dma_dom;
+ dma_dom->aperture[0]->bitmap[0] = 1;
+ dma_dom->next_address = 0;
- l2_pde = (u64 *)get_zeroed_page(GFP_KERNEL);
- if (l2_pde == NULL)
- goto free_dma_dom;
-
- dma_dom->domain.pt_root[0] = IOMMU_L2_PDE(virt_to_phys(l2_pde));
-
- for (i = 0; i < num_pte_pages; ++i) {
- dma_dom->pte_pages[i] = (u64 *)get_zeroed_page(GFP_KERNEL);
- if (!dma_dom->pte_pages[i])
- goto free_dma_dom;
- address = virt_to_phys(dma_dom->pte_pages[i]);
- l2_pde[i] = IOMMU_L1_PDE(address);
- }
return dma_dom;
free_dma_dom:
dma_ops_domain_free(dma_dom);
- return NULL;
+ return NULL;
+}
+
+/*
+ * little helper function to check whether a given protection domain is a
+ * dma_ops domain
+ */
+static bool dma_ops_domain(struct protection_domain *domain)
+{
+ return domain->flags & PD_DMA_OPS_MASK;
+}
+
+/*
+ * Find out the protection domain structure for a given PCI device. This
+ * will give us the pointer to the page table root for example.
+ */
+static struct protection_domain *domain_for_device(u16 devid)
+{
+ struct protection_domain *dom;
+ unsigned long flags;
+
+ read_lock_irqsave(&amd_iommu_devtable_lock, flags);
+ dom = amd_iommu_pd_table[devid];
+ read_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+
+ return dom;
+}
+
+/*
+ * If a device is not yet associated with a domain, this function does
+ * assigns it visible for the hardware
+ */
+static void attach_device(struct amd_iommu *iommu,
+ struct protection_domain *domain,
+ u16 devid)
+{
+ unsigned long flags;
+ u64 pte_root = virt_to_phys(domain->pt_root);
+
+ domain->dev_cnt += 1;
+
+ pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK)
+ << DEV_ENTRY_MODE_SHIFT;
+ pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV;
+
+ write_lock_irqsave(&amd_iommu_devtable_lock, flags);
+ amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root);
+ amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root);
+ amd_iommu_dev_table[devid].data[2] = domain->id;
+
+ amd_iommu_pd_table[devid] = domain;
+ write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+
+ /*
+ * We might boot into a crash-kernel here. The crashed kernel
+ * left the caches in the IOMMU dirty. So we have to flush
+ * here to evict all dirty stuff.
+ */
+ iommu_queue_inv_dev_entry(iommu, devid);
+ iommu_flush_tlb_pde(iommu, domain->id);
+}
+
+/*
+ * Removes a device from a protection domain (unlocked)
+ */
+static void __detach_device(struct protection_domain *domain, u16 devid)
+{
+
+ /* lock domain */
+ spin_lock(&domain->lock);
+
+ /* remove domain from the lookup table */
+ amd_iommu_pd_table[devid] = NULL;
+
+ /* remove entry from the device table seen by the hardware */
+ amd_iommu_dev_table[devid].data[0] = IOMMU_PTE_P | IOMMU_PTE_TV;
+ amd_iommu_dev_table[devid].data[1] = 0;
+ amd_iommu_dev_table[devid].data[2] = 0;
+
+ amd_iommu_apply_erratum_63(devid);
+
+ /* decrease reference counter */
+ domain->dev_cnt -= 1;
+
+ /* ready */
+ spin_unlock(&domain->lock);
}
/*
- * Find out the protection domain structure for a given PCI device. This
- * will give us the pointer to the page table root for example.
+ * Removes a device from a protection domain (with devtable_lock held)
*/
-static struct protection_domain *domain_for_device(u16 devid)
+static void detach_device(struct protection_domain *domain, u16 devid)
{
- struct protection_domain *dom;
unsigned long flags;
- read_lock_irqsave(&amd_iommu_devtable_lock, flags);
- dom = amd_iommu_pd_table[devid];
- read_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
-
- return dom;
+ /* lock device table */
+ write_lock_irqsave(&amd_iommu_devtable_lock, flags);
+ __detach_device(domain, devid);
+ write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
}
-/*
- * If a device is not yet associated with a domain, this function does
- * assigns it visible for the hardware
- */
-static void set_device_domain(struct amd_iommu *iommu,
- struct protection_domain *domain,
- u16 devid)
+static int device_change_notifier(struct notifier_block *nb,
+ unsigned long action, void *data)
{
+ struct device *dev = data;
+ struct pci_dev *pdev = to_pci_dev(dev);
+ u16 devid = calc_devid(pdev->bus->number, pdev->devfn);
+ struct protection_domain *domain;
+ struct dma_ops_domain *dma_domain;
+ struct amd_iommu *iommu;
unsigned long flags;
- u64 pte_root = virt_to_phys(domain->pt_root);
+ if (devid > amd_iommu_last_bdf)
+ goto out;
- pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK)
- << DEV_ENTRY_MODE_SHIFT;
- pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV;
+ devid = amd_iommu_alias_table[devid];
- write_lock_irqsave(&amd_iommu_devtable_lock, flags);
- amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root);
- amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root);
- amd_iommu_dev_table[devid].data[2] = domain->id;
+ iommu = amd_iommu_rlookup_table[devid];
+ if (iommu == NULL)
+ goto out;
- amd_iommu_pd_table[devid] = domain;
- write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+ domain = domain_for_device(devid);
+
+ if (domain && !dma_ops_domain(domain))
+ WARN_ONCE(1, "AMD IOMMU WARNING: device %s already bound "
+ "to a non-dma-ops domain\n", dev_name(dev));
+
+ switch (action) {
+ case BUS_NOTIFY_UNBOUND_DRIVER:
+ if (!domain)
+ goto out;
+ detach_device(domain, devid);
+ break;
+ case BUS_NOTIFY_ADD_DEVICE:
+ /* allocate a protection domain if a device is added */
+ dma_domain = find_protection_domain(devid);
+ if (dma_domain)
+ goto out;
+ dma_domain = dma_ops_domain_alloc(iommu);
+ if (!dma_domain)
+ goto out;
+ dma_domain->target_dev = devid;
+
+ spin_lock_irqsave(&iommu_pd_list_lock, flags);
+ list_add_tail(&dma_domain->list, &iommu_pd_list);
+ spin_unlock_irqrestore(&iommu_pd_list_lock, flags);
+
+ break;
+ default:
+ goto out;
+ }
iommu_queue_inv_dev_entry(iommu, devid);
+ iommu_completion_wait(iommu);
- iommu->need_sync = 1;
+out:
+ return 0;
}
+static struct notifier_block device_nb = {
+ .notifier_call = device_change_notifier,
+};
+
/*****************************************************************************
*
* The next functions belong to the dma_ops mapping/unmapping code.
list_for_each_entry(entry, &iommu_pd_list, list) {
if (entry->target_dev == devid) {
ret = entry;
- list_del(&ret->list);
break;
}
}
if (!dma_dom)
dma_dom = (*iommu)->default_dom;
*domain = &dma_dom->domain;
- set_device_domain(*iommu, *domain, *bdf);
- printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
- "device ", (*domain)->id);
- print_devid(_bdf, 1);
+ attach_device(*iommu, *domain, *bdf);
+ DUMP_printk("Using protection domain %d for device %s\n",
+ (*domain)->id, dev_name(dev));
}
+ if (domain_for_device(_bdf) == NULL)
+ attach_device(*iommu, *domain, _bdf);
+
return 1;
}
/*
+ * If the pte_page is not yet allocated this function is called
+ */
+static u64* alloc_pte(struct protection_domain *dom,
+ unsigned long address, u64 **pte_page, gfp_t gfp)
+{
+ u64 *pte, *page;
+
+ pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(address)];
+
+ if (!IOMMU_PTE_PRESENT(*pte)) {
+ page = (u64 *)get_zeroed_page(gfp);
+ if (!page)
+ return NULL;
+ *pte = IOMMU_L2_PDE(virt_to_phys(page));
+ }
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L1_INDEX(address)];
+
+ if (!IOMMU_PTE_PRESENT(*pte)) {
+ page = (u64 *)get_zeroed_page(gfp);
+ if (!page)
+ return NULL;
+ *pte = IOMMU_L1_PDE(virt_to_phys(page));
+ }
+
+ pte = IOMMU_PTE_PAGE(*pte);
+
+ if (pte_page)
+ *pte_page = pte;
+
+ pte = &pte[IOMMU_PTE_L0_INDEX(address)];
+
+ return pte;
+}
+
+/*
+ * This function fetches the PTE for a given address in the aperture
+ */
+static u64* dma_ops_get_pte(struct dma_ops_domain *dom,
+ unsigned long address)
+{
+ struct aperture_range *aperture;
+ u64 *pte, *pte_page;
+
+ aperture = dom->aperture[APERTURE_RANGE_INDEX(address)];
+ if (!aperture)
+ return NULL;
+
+ pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)];
+ if (!pte) {
+ pte = alloc_pte(&dom->domain, address, &pte_page, GFP_ATOMIC);
+ aperture->pte_pages[APERTURE_PAGE_INDEX(address)] = pte_page;
+ } else
+ pte += IOMMU_PTE_L0_INDEX(address);
+
+ return pte;
+}
+
+/*
* This is the generic map function. It maps one 4kb page at paddr to
* the given address in the DMA address space for the domain.
*/
paddr &= PAGE_MASK;
- pte = dom->pte_pages[IOMMU_PTE_L1_INDEX(address)];
- pte += IOMMU_PTE_L0_INDEX(address);
+ pte = dma_ops_get_pte(dom, address);
+ if (!pte)
+ return bad_dma_address;
__pte = paddr | IOMMU_PTE_P | IOMMU_PTE_FC;
struct dma_ops_domain *dom,
unsigned long address)
{
+ struct aperture_range *aperture;
u64 *pte;
if (address >= dom->aperture_size)
return;
- WARN_ON(address & 0xfffULL || address > dom->aperture_size);
+ aperture = dom->aperture[APERTURE_RANGE_INDEX(address)];
+ if (!aperture)
+ return;
+
+ pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)];
+ if (!pte)
+ return;
- pte = dom->pte_pages[IOMMU_PTE_L1_INDEX(address)];
pte += IOMMU_PTE_L0_INDEX(address);
WARN_ON(!*pte);
/*
* This function contains common code for mapping of a physically
- * contiguous memory region into DMA address space. It is uses by all
- * mapping functions provided by this IOMMU driver.
+ * contiguous memory region into DMA address space. It is used by all
+ * mapping functions provided with this IOMMU driver.
* Must be called with the domain lock held.
*/
static dma_addr_t __map_single(struct device *dev,
u64 dma_mask)
{
dma_addr_t offset = paddr & ~PAGE_MASK;
- dma_addr_t address, start;
+ dma_addr_t address, start, ret;
unsigned int pages;
unsigned long align_mask = 0;
int i;
pages = iommu_num_pages(paddr, size, PAGE_SIZE);
paddr &= PAGE_MASK;
+ INC_STATS_COUNTER(total_map_requests);
+
+ if (pages > 1)
+ INC_STATS_COUNTER(cross_page);
+
if (align)
align_mask = (1UL << get_order(size)) - 1;
+retry:
address = dma_ops_alloc_addresses(dev, dma_dom, pages, align_mask,
dma_mask);
- if (unlikely(address == bad_dma_address))
- goto out;
+ if (unlikely(address == bad_dma_address)) {
+ /*
+ * setting next_address here will let the address
+ * allocator only scan the new allocated range in the
+ * first run. This is a small optimization.
+ */
+ dma_dom->next_address = dma_dom->aperture_size;
+
+ if (alloc_new_range(iommu, dma_dom, false, GFP_ATOMIC))
+ goto out;
+
+ /*
+ * aperture was sucessfully enlarged by 128 MB, try
+ * allocation again
+ */
+ goto retry;
+ }
start = address;
for (i = 0; i < pages; ++i) {
- dma_ops_domain_map(iommu, dma_dom, start, paddr, dir);
+ ret = dma_ops_domain_map(iommu, dma_dom, start, paddr, dir);
+ if (ret == bad_dma_address)
+ goto out_unmap;
+
paddr += PAGE_SIZE;
start += PAGE_SIZE;
}
address += offset;
+ ADD_STATS_COUNTER(alloced_io_mem, size);
+
if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) {
iommu_flush_tlb(iommu, dma_dom->domain.id);
dma_dom->need_flush = false;
out:
return address;
+
+out_unmap:
+
+ for (--i; i >= 0; --i) {
+ start -= PAGE_SIZE;
+ dma_ops_domain_unmap(iommu, dma_dom, start);
+ }
+
+ dma_ops_free_addresses(dma_dom, address, pages);
+
+ return bad_dma_address;
}
/*
dma_addr_t i, start;
unsigned int pages;
- if ((dma_addr == 0) || (dma_addr + size > dma_dom->aperture_size))
+ if ((dma_addr == bad_dma_address) ||
+ (dma_addr + size > dma_dom->aperture_size))
return;
pages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
start += PAGE_SIZE;
}
+ SUB_STATS_COUNTER(alloced_io_mem, size);
+
dma_ops_free_addresses(dma_dom, dma_addr, pages);
if (amd_iommu_unmap_flush || dma_dom->need_flush) {
/*
* The exported map_single function for dma_ops.
*/
-static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
- size_t size, int dir)
+static dma_addr_t map_page(struct device *dev, struct page *page,
+ unsigned long offset, size_t size,
+ enum dma_data_direction dir,
+ struct dma_attrs *attrs)
{
unsigned long flags;
struct amd_iommu *iommu;
u16 devid;
dma_addr_t addr;
u64 dma_mask;
+ phys_addr_t paddr = page_to_phys(page) + offset;
+
+ INC_STATS_COUNTER(cnt_map_single);
if (!check_device(dev))
return bad_dma_address;
/* device not handled by any AMD IOMMU */
return (dma_addr_t)paddr;
+ if (!dma_ops_domain(domain))
+ return bad_dma_address;
+
spin_lock_irqsave(&domain->lock, flags);
addr = __map_single(dev, iommu, domain->priv, paddr, size, dir, false,
dma_mask);
if (addr == bad_dma_address)
goto out;
- if (unlikely(iommu->need_sync))
- iommu_completion_wait(iommu);
+ iommu_completion_wait(iommu);
out:
spin_unlock_irqrestore(&domain->lock, flags);
/*
* The exported unmap_single function for dma_ops.
*/
-static void unmap_single(struct device *dev, dma_addr_t dma_addr,
- size_t size, int dir)
+static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
+ enum dma_data_direction dir, struct dma_attrs *attrs)
{
unsigned long flags;
struct amd_iommu *iommu;
struct protection_domain *domain;
u16 devid;
+ INC_STATS_COUNTER(cnt_unmap_single);
+
if (!check_device(dev) ||
!get_device_resources(dev, &iommu, &domain, &devid))
/* device not handled by any AMD IOMMU */
return;
+ if (!dma_ops_domain(domain))
+ return;
+
spin_lock_irqsave(&domain->lock, flags);
__unmap_single(iommu, domain->priv, dma_addr, size, dir);
- if (unlikely(iommu->need_sync))
- iommu_completion_wait(iommu);
+ iommu_completion_wait(iommu);
spin_unlock_irqrestore(&domain->lock, flags);
}
* lists).
*/
static int map_sg(struct device *dev, struct scatterlist *sglist,
- int nelems, int dir)
+ int nelems, enum dma_data_direction dir,
+ struct dma_attrs *attrs)
{
unsigned long flags;
struct amd_iommu *iommu;
int mapped_elems = 0;
u64 dma_mask;
+ INC_STATS_COUNTER(cnt_map_sg);
+
if (!check_device(dev))
return 0;
if (!iommu || !domain)
return map_sg_no_iommu(dev, sglist, nelems, dir);
+ if (!dma_ops_domain(domain))
+ return 0;
+
spin_lock_irqsave(&domain->lock, flags);
for_each_sg(sglist, s, nelems, i) {
goto unmap;
}
- if (unlikely(iommu->need_sync))
- iommu_completion_wait(iommu);
+ iommu_completion_wait(iommu);
out:
spin_unlock_irqrestore(&domain->lock, flags);
* lists).
*/
static void unmap_sg(struct device *dev, struct scatterlist *sglist,
- int nelems, int dir)
+ int nelems, enum dma_data_direction dir,
+ struct dma_attrs *attrs)
{
unsigned long flags;
struct amd_iommu *iommu;
u16 devid;
int i;
+ INC_STATS_COUNTER(cnt_unmap_sg);
+
if (!check_device(dev) ||
!get_device_resources(dev, &iommu, &domain, &devid))
return;
+ if (!dma_ops_domain(domain))
+ return;
+
spin_lock_irqsave(&domain->lock, flags);
for_each_sg(sglist, s, nelems, i) {
s->dma_address = s->dma_length = 0;
}
- if (unlikely(iommu->need_sync))
- iommu_completion_wait(iommu);
+ iommu_completion_wait(iommu);
spin_unlock_irqrestore(&domain->lock, flags);
}
phys_addr_t paddr;
u64 dma_mask = dev->coherent_dma_mask;
+ INC_STATS_COUNTER(cnt_alloc_coherent);
+
if (!check_device(dev))
return NULL;
flag |= __GFP_ZERO;
virt_addr = (void *)__get_free_pages(flag, get_order(size));
if (!virt_addr)
- return 0;
+ return NULL;
paddr = virt_to_phys(virt_addr);
return virt_addr;
}
+ if (!dma_ops_domain(domain))
+ goto out_free;
+
if (!dma_mask)
dma_mask = *dev->dma_mask;
size, DMA_BIDIRECTIONAL, true, dma_mask);
if (*dma_addr == bad_dma_address) {
- free_pages((unsigned long)virt_addr, get_order(size));
- virt_addr = NULL;
- goto out;
+ spin_unlock_irqrestore(&domain->lock, flags);
+ goto out_free;
}
- if (unlikely(iommu->need_sync))
- iommu_completion_wait(iommu);
+ iommu_completion_wait(iommu);
-out:
spin_unlock_irqrestore(&domain->lock, flags);
return virt_addr;
+
+out_free:
+
+ free_pages((unsigned long)virt_addr, get_order(size));
+
+ return NULL;
}
/*
struct protection_domain *domain;
u16 devid;
+ INC_STATS_COUNTER(cnt_free_coherent);
+
if (!check_device(dev))
return;
if (!iommu || !domain)
goto free_mem;
+ if (!dma_ops_domain(domain))
+ goto free_mem;
+
spin_lock_irqsave(&domain->lock, flags);
__unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
- if (unlikely(iommu->need_sync))
- iommu_completion_wait(iommu);
+ iommu_completion_wait(iommu);
spin_unlock_irqrestore(&domain->lock, flags);
* we don't need to preallocate the protection domains anymore.
* For now we have to.
*/
-void prealloc_protection_domains(void)
+static void prealloc_protection_domains(void)
{
struct pci_dev *dev = NULL;
struct dma_ops_domain *dma_dom;
struct amd_iommu *iommu;
- int order = amd_iommu_aperture_order;
u16 devid;
while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
- devid = (dev->bus->number << 8) | dev->devfn;
+ devid = calc_devid(dev->bus->number, dev->devfn);
if (devid > amd_iommu_last_bdf)
continue;
devid = amd_iommu_alias_table[devid];
iommu = amd_iommu_rlookup_table[devid];
if (!iommu)
continue;
- dma_dom = dma_ops_domain_alloc(iommu, order);
+ dma_dom = dma_ops_domain_alloc(iommu);
if (!dma_dom)
continue;
init_unity_mappings_for_device(dma_dom, devid);
}
}
-static struct dma_mapping_ops amd_iommu_dma_ops = {
+static struct dma_map_ops amd_iommu_dma_ops = {
.alloc_coherent = alloc_coherent,
.free_coherent = free_coherent,
- .map_single = map_single,
- .unmap_single = unmap_single,
+ .map_page = map_page,
+ .unmap_page = unmap_page,
.map_sg = map_sg,
.unmap_sg = unmap_sg,
.dma_supported = amd_iommu_dma_supported,
int __init amd_iommu_init_dma_ops(void)
{
struct amd_iommu *iommu;
- int order = amd_iommu_aperture_order;
int ret;
/*
* found in the system. Devices not assigned to any other
* protection domain will be assigned to the default one.
*/
- list_for_each_entry(iommu, &amd_iommu_list, list) {
- iommu->default_dom = dma_ops_domain_alloc(iommu, order);
+ for_each_iommu(iommu) {
+ iommu->default_dom = dma_ops_domain_alloc(iommu);
if (iommu->default_dom == NULL)
return -ENOMEM;
+ iommu->default_dom->domain.flags |= PD_DEFAULT_MASK;
ret = iommu_init_unity_mappings(iommu);
if (ret)
goto free_domains;
/* Make the driver finally visible to the drivers */
dma_ops = &amd_iommu_dma_ops;
+ register_iommu(&amd_iommu_ops);
+
+ bus_register_notifier(&pci_bus_type, &device_nb);
+
+ amd_iommu_stats_init();
+
return 0;
free_domains:
- list_for_each_entry(iommu, &amd_iommu_list, list) {
+ for_each_iommu(iommu) {
if (iommu->default_dom)
dma_ops_domain_free(iommu->default_dom);
}
return ret;
}
+
+/*****************************************************************************
+ *
+ * The following functions belong to the exported interface of AMD IOMMU
+ *
+ * This interface allows access to lower level functions of the IOMMU
+ * like protection domain handling and assignement of devices to domains
+ * which is not possible with the dma_ops interface.
+ *
+ *****************************************************************************/
+
+static void cleanup_domain(struct protection_domain *domain)
+{
+ unsigned long flags;
+ u16 devid;
+
+ write_lock_irqsave(&amd_iommu_devtable_lock, flags);
+
+ for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
+ if (amd_iommu_pd_table[devid] == domain)
+ __detach_device(domain, devid);
+
+ write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+}
+
+static int amd_iommu_domain_init(struct iommu_domain *dom)
+{
+ struct protection_domain *domain;
+
+ domain = kzalloc(sizeof(*domain), GFP_KERNEL);
+ if (!domain)
+ return -ENOMEM;
+
+ spin_lock_init(&domain->lock);
+ domain->mode = PAGE_MODE_3_LEVEL;
+ domain->id = domain_id_alloc();
+ if (!domain->id)
+ goto out_free;
+ domain->pt_root = (void *)get_zeroed_page(GFP_KERNEL);
+ if (!domain->pt_root)
+ goto out_free;
+
+ dom->priv = domain;
+
+ return 0;
+
+out_free:
+ kfree(domain);
+
+ return -ENOMEM;
+}
+
+static void amd_iommu_domain_destroy(struct iommu_domain *dom)
+{
+ struct protection_domain *domain = dom->priv;
+
+ if (!domain)
+ return;
+
+ if (domain->dev_cnt > 0)
+ cleanup_domain(domain);
+
+ BUG_ON(domain->dev_cnt != 0);
+
+ free_pagetable(domain);
+
+ domain_id_free(domain->id);
+
+ kfree(domain);
+
+ dom->priv = NULL;
+}
+
+static void amd_iommu_detach_device(struct iommu_domain *dom,
+ struct device *dev)
+{
+ struct protection_domain *domain = dom->priv;
+ struct amd_iommu *iommu;
+ struct pci_dev *pdev;
+ u16 devid;
+
+ if (dev->bus != &pci_bus_type)
+ return;
+
+ pdev = to_pci_dev(dev);
+
+ devid = calc_devid(pdev->bus->number, pdev->devfn);
+
+ if (devid > 0)
+ detach_device(domain, devid);
+
+ iommu = amd_iommu_rlookup_table[devid];
+ if (!iommu)
+ return;
+
+ iommu_queue_inv_dev_entry(iommu, devid);
+ iommu_completion_wait(iommu);
+}
+
+static int amd_iommu_attach_device(struct iommu_domain *dom,
+ struct device *dev)
+{
+ struct protection_domain *domain = dom->priv;
+ struct protection_domain *old_domain;
+ struct amd_iommu *iommu;
+ struct pci_dev *pdev;
+ u16 devid;
+
+ if (dev->bus != &pci_bus_type)
+ return -EINVAL;
+
+ pdev = to_pci_dev(dev);
+
+ devid = calc_devid(pdev->bus->number, pdev->devfn);
+
+ if (devid >= amd_iommu_last_bdf ||
+ devid != amd_iommu_alias_table[devid])
+ return -EINVAL;
+
+ iommu = amd_iommu_rlookup_table[devid];
+ if (!iommu)
+ return -EINVAL;
+
+ old_domain = domain_for_device(devid);
+ if (old_domain)
+ detach_device(old_domain, devid);
+
+ attach_device(iommu, domain, devid);
+
+ iommu_completion_wait(iommu);
+
+ return 0;
+}
+
+static int amd_iommu_map_range(struct iommu_domain *dom,
+ unsigned long iova, phys_addr_t paddr,
+ size_t size, int iommu_prot)
+{
+ struct protection_domain *domain = dom->priv;
+ unsigned long i, npages = iommu_num_pages(paddr, size, PAGE_SIZE);
+ int prot = 0;
+ int ret;
+
+ if (iommu_prot & IOMMU_READ)
+ prot |= IOMMU_PROT_IR;
+ if (iommu_prot & IOMMU_WRITE)
+ prot |= IOMMU_PROT_IW;
+
+ iova &= PAGE_MASK;
+ paddr &= PAGE_MASK;
+
+ for (i = 0; i < npages; ++i) {
+ ret = iommu_map_page(domain, iova, paddr, prot);
+ if (ret)
+ return ret;
+
+ iova += PAGE_SIZE;
+ paddr += PAGE_SIZE;
+ }
+
+ return 0;
+}
+
+static void amd_iommu_unmap_range(struct iommu_domain *dom,
+ unsigned long iova, size_t size)
+{
+
+ struct protection_domain *domain = dom->priv;
+ unsigned long i, npages = iommu_num_pages(iova, size, PAGE_SIZE);
+
+ iova &= PAGE_MASK;
+
+ for (i = 0; i < npages; ++i) {
+ iommu_unmap_page(domain, iova);
+ iova += PAGE_SIZE;
+ }
+
+ iommu_flush_domain(domain->id);
+}
+
+static phys_addr_t amd_iommu_iova_to_phys(struct iommu_domain *dom,
+ unsigned long iova)
+{
+ struct protection_domain *domain = dom->priv;
+ unsigned long offset = iova & ~PAGE_MASK;
+ phys_addr_t paddr;
+ u64 *pte;
+
+ pte = &domain->pt_root[IOMMU_PTE_L2_INDEX(iova)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return 0;
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L1_INDEX(iova)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return 0;
+
+ pte = IOMMU_PTE_PAGE(*pte);
+ pte = &pte[IOMMU_PTE_L0_INDEX(iova)];
+
+ if (!IOMMU_PTE_PRESENT(*pte))
+ return 0;
+
+ paddr = *pte & IOMMU_PAGE_MASK;
+ paddr |= offset;
+
+ return paddr;
+}
+
+static int amd_iommu_domain_has_cap(struct iommu_domain *domain,
+ unsigned long cap)
+{
+ return 0;
+}
+
+static struct iommu_ops amd_iommu_ops = {
+ .domain_init = amd_iommu_domain_init,
+ .domain_destroy = amd_iommu_domain_destroy,
+ .attach_dev = amd_iommu_attach_device,
+ .detach_dev = amd_iommu_detach_device,
+ .map = amd_iommu_map_range,
+ .unmap = amd_iommu_unmap_range,
+ .iova_to_phys = amd_iommu_iova_to_phys,
+ .domain_has_cap = amd_iommu_domain_has_cap,
+};
+
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff