mirror of
https://github.com/AuxXxilium/linux_dsm_epyc7002.git
synced 2024-11-25 16:40:52 +07:00
x86, AMD IOMMU: add comments to core code
This patch adds comments about how the AMD IOMMU core code works for the DMA remapping functionality. Signed-off-by: Joerg Roedel <joerg.roedel@amd.com> Cc: iommu@lists.linux-foundation.org Cc: bhavna.sarathy@amd.com Cc: robert.richter@amd.com Signed-off-by: Ingo Molnar <mingo@elte.hu>
This commit is contained in:
Joerg Roedel
Ingo Molnar
parent
b65233a9c1
commit
431b2a2015
@ -34,6 +34,9 @@
|
||||
|
||||
static DEFINE_RWLOCK(amd_iommu_devtable_lock);
|
||||
|
||||
/*
|
||||
* general struct to manage commands send to an IOMMU
|
||||
*/
|
||||
struct command {
|
||||
u32 data[4];
|
||||
};
|
||||
@ -41,11 +44,22 @@ struct command {
|
||||
static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
|
||||
struct unity_map_entry *e);
|
||||
|
||||
/* returns !0 if the IOMMU is caching non-present entries in its TLB */
|
||||
static int iommu_has_npcache(struct amd_iommu *iommu)
|
||||
{
|
||||
return iommu->cap & IOMMU_CAP_NPCACHE;
|
||||
}
|
||||
|
||||
/****************************************************************************
|
||||
*
|
||||
* IOMMU command queuing functions
|
||||
*
|
||||
****************************************************************************/
|
||||
|
||||
/*
|
||||
* Writes the command to the IOMMUs command buffer and informs the
|
||||
* hardware about the new command. Must be called with iommu->lock held.
|
||||
*/
|
||||
static int __iommu_queue_command(struct amd_iommu *iommu, struct command *cmd)
|
||||
{
|
||||
u32 tail, head;
|
||||
@ -63,6 +77,10 @@ static int __iommu_queue_command(struct amd_iommu *iommu, struct command *cmd)
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* General queuing function for commands. Takes iommu->lock and calls
|
||||
* __iommu_queue_command().
|
||||
*/
|
||||
static int iommu_queue_command(struct amd_iommu *iommu, struct command *cmd)
|
||||
{
|
||||
unsigned long flags;
|
||||
@ -75,6 +93,13 @@ static int iommu_queue_command(struct amd_iommu *iommu, struct command *cmd)
|
||||
return ret;
|
||||
}
|
||||
|
||||
/*
|
||||
* 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
|
||||
* us about that by writing a value to a physical address we pass with
|
||||
* the command.
|
||||
*/
|
||||
static int iommu_completion_wait(struct amd_iommu *iommu)
|
||||
{
|
||||
int ret;
|
||||
@ -101,6 +126,9 @@ static int iommu_completion_wait(struct amd_iommu *iommu)
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* Command send function for invalidating a device table entry
|
||||
*/
|
||||
static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
|
||||
{
|
||||
struct command cmd;
|
||||
@ -116,6 +144,9 @@ static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
|
||||
return iommu_queue_command(iommu, &cmd);
|
||||
}
|
||||
|
||||
/*
|
||||
* Generic command send function for invalidaing TLB entries
|
||||
*/
|
||||
static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
|
||||
u64 address, u16 domid, int pde, int s)
|
||||
{
|
||||
@ -127,9 +158,9 @@ static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
|
||||
cmd.data[1] |= domid;
|
||||
cmd.data[2] = LOW_U32(address);
|
||||
cmd.data[3] = HIGH_U32(address);
|
||||
if (s)
|
||||
if (s) /* size bit - we flush more than one 4kb page */
|
||||
cmd.data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK;
|
||||
if (pde)
|
||||
if (pde) /* PDE bit - we wan't flush everything not only the PTEs */
|
||||
cmd.data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK;
|
||||
|
||||
iommu->need_sync = 1;
|
||||
@ -137,6 +168,11 @@ static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
|
||||
return iommu_queue_command(iommu, &cmd);
|
||||
}
|
||||
|
||||
/*
|
||||
* TLB invalidation function which is called from the mapping functions.
|
||||
* It invalidates a single PTE if the range to flush is within a single
|
||||
* page. Otherwise it flushes the whole TLB of the IOMMU.
|
||||
*/
|
||||
static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
|
||||
u64 address, size_t size)
|
||||
{
|
||||
@ -159,6 +195,20 @@ static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
|
||||
return 0;
|
||||
}
|
||||
|
||||
/****************************************************************************
|
||||
*
|
||||
* The functions below are used the create the page table mappings for
|
||||
* unity mapped regions.
|
||||
*
|
||||
****************************************************************************/
|
||||
|
||||
/*
|
||||
* Generic mapping functions. It maps a physical address into a DMA
|
||||
* address space. It allocates the page table pages if necessary.
|
||||
* In the future it can be extended to a generic mapping function
|
||||
* 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,
|
||||
@ -209,6 +259,10 @@ static int iommu_map(struct protection_domain *dom,
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* This function checks if a specific unity mapping entry is needed for
|
||||
* this specific IOMMU.
|
||||
*/
|
||||
static int iommu_for_unity_map(struct amd_iommu *iommu,
|
||||
struct unity_map_entry *entry)
|
||||
{
|
||||
@ -223,6 +277,12 @@ static int iommu_for_unity_map(struct amd_iommu *iommu,
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* Init the unity mappings for a specific IOMMU in the system
|
||||
*
|
||||
* Basically iterates over all unity mapping entries and applies them to
|
||||
* the default domain DMA of that IOMMU if necessary.
|
||||
*/
|
||||
static int iommu_init_unity_mappings(struct amd_iommu *iommu)
|
||||
{
|
||||
struct unity_map_entry *entry;
|
||||
@ -239,6 +299,10 @@ static int iommu_init_unity_mappings(struct amd_iommu *iommu)
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* This function actually applies the mapping to the page table of the
|
||||
* dma_ops domain.
|
||||
*/
|
||||
static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
|
||||
struct unity_map_entry *e)
|
||||
{
|
||||
@ -261,6 +325,9 @@ static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* Inits the unity mappings required for a specific device
|
||||
*/
|
||||
static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
|
||||
u16 devid)
|
||||
{
|
||||
@ -278,12 +345,26 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
|
||||
return 0;
|
||||
}
|
||||
|
||||
/****************************************************************************
|
||||
*
|
||||
* The next functions belong to the address allocator for the dma_ops
|
||||
* interface functions. They work like the allocators in the other IOMMU
|
||||
* drivers. Its basically a bitmap which marks the allocated pages in
|
||||
* the aperture. Maybe it could be enhanced in the future to a more
|
||||
* efficient allocator.
|
||||
*
|
||||
****************************************************************************/
|
||||
static unsigned long dma_mask_to_pages(unsigned long mask)
|
||||
{
|
||||
return (mask >> PAGE_SHIFT) +
|
||||
(PAGE_ALIGN(mask & ~PAGE_MASK) >> PAGE_SHIFT);
|
||||
}
|
||||
|
||||
/*
|
||||
* The address allocator core function.
|
||||
*
|
||||
* called with domain->lock held
|
||||
*/
|
||||
static unsigned long dma_ops_alloc_addresses(struct device *dev,
|
||||
struct dma_ops_domain *dom,
|
||||
unsigned int pages)
|
||||
@ -317,6 +398,11 @@ static unsigned long dma_ops_alloc_addresses(struct device *dev,
|
||||
return address;
|
||||
}
|
||||
|
||||
/*
|
||||
* The address free function.
|
||||
*
|
||||
* called with domain->lock held
|
||||
*/
|
||||
static void dma_ops_free_addresses(struct dma_ops_domain *dom,
|
||||
unsigned long address,
|
||||
unsigned int pages)
|
||||
@ -325,6 +411,16 @@ static void dma_ops_free_addresses(struct dma_ops_domain *dom,
|
||||
iommu_area_free(dom->bitmap, address, pages);
|
||||
}
|
||||
|
||||
/****************************************************************************
|
||||
*
|
||||
* The next functions belong to the domain allocation. A domain is
|
||||
* allocated for every IOMMU as the default domain. If device isolation
|
||||
* is enabled, every device get its own domain. The most important thing
|
||||
* about domains is the page table mapping the DMA address space they
|
||||
* contain.
|
||||
*
|
||||
****************************************************************************/
|
||||
|
||||
static u16 domain_id_alloc(void)
|
||||
{
|
||||
unsigned long flags;
|
||||
@ -342,6 +438,10 @@ static u16 domain_id_alloc(void)
|
||||
return id;
|
||||
}
|
||||
|
||||
/*
|
||||
* Used to reserve address ranges in the aperture (e.g. for exclusion
|
||||
* ranges.
|
||||
*/
|
||||
static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
|
||||
unsigned long start_page,
|
||||
unsigned int pages)
|
||||
@ -382,6 +482,10 @@ static void dma_ops_free_pagetable(struct dma_ops_domain *dma_dom)
|
||||
free_page((unsigned long)p1);
|
||||
}
|
||||
|
||||
/*
|
||||
* Free a domain, only used if something went wrong in the
|
||||
* allocation path and we need to free an already allocated page table
|
||||
*/
|
||||
static void dma_ops_domain_free(struct dma_ops_domain *dom)
|
||||
{
|
||||
if (!dom)
|
||||
@ -396,6 +500,11 @@ static void dma_ops_domain_free(struct dma_ops_domain *dom)
|
||||
kfree(dom);
|
||||
}
|
||||
|
||||
/*
|
||||
* Allocates a new protection domain usable for the dma_ops functions.
|
||||
* 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)
|
||||
{
|
||||
@ -436,6 +545,7 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
|
||||
dma_dom->bitmap[0] = 1;
|
||||
dma_dom->next_bit = 0;
|
||||
|
||||
/* 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;
|
||||
@ -444,6 +554,11 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
|
||||
dma_ops_reserve_addresses(dma_dom, startpage, pages);
|
||||
}
|
||||
|
||||
/*
|
||||
* 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.
|
||||
*/
|
||||
num_pte_pages = dma_dom->aperture_size / (PAGE_SIZE * 512);
|
||||
dma_dom->pte_pages = kzalloc(num_pte_pages * sizeof(void *),
|
||||
GFP_KERNEL);
|
||||
@ -472,6 +587,10 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/*
|
||||
* 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;
|
||||
@ -484,6 +603,10 @@ static struct protection_domain *domain_for_device(u16 devid)
|
||||
return dom;
|
||||
}
|
||||
|
||||
/*
|
||||
* 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)
|
||||
@ -508,6 +631,19 @@ static void set_device_domain(struct amd_iommu *iommu,
|
||||
iommu->need_sync = 1;
|
||||
}
|
||||
|
||||
/*****************************************************************************
|
||||
*
|
||||
* The next functions belong to the dma_ops mapping/unmapping code.
|
||||
*
|
||||
*****************************************************************************/
|
||||
|
||||
/*
|
||||
* In the dma_ops path we only have the struct device. This function
|
||||
* finds the corresponding IOMMU, the protection domain and the
|
||||
* requestor id for a given device.
|
||||
* If the device is not yet associated with a domain this is also done
|
||||
* in this function.
|
||||
*/
|
||||
static int get_device_resources(struct device *dev,
|
||||
struct amd_iommu **iommu,
|
||||
struct protection_domain **domain,
|
||||
@ -522,6 +658,7 @@ static int get_device_resources(struct device *dev,
|
||||
pcidev = to_pci_dev(dev);
|
||||
_bdf = (pcidev->bus->number << 8) | pcidev->devfn;
|
||||
|
||||
/* device not translated by any IOMMU in the system? */
|
||||
if (_bdf >= amd_iommu_last_bdf) {
|
||||
*iommu = NULL;
|
||||
*domain = NULL;
|
||||
@ -547,6 +684,10 @@ static int get_device_resources(struct device *dev,
|
||||
return 1;
|
||||
}
|
||||
|
||||
/*
|
||||
* 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.
|
||||
*/
|
||||
static dma_addr_t dma_ops_domain_map(struct amd_iommu *iommu,
|
||||
struct dma_ops_domain *dom,
|
||||
unsigned long address,
|
||||
@ -578,6 +719,9 @@ static dma_addr_t dma_ops_domain_map(struct amd_iommu *iommu,
|
||||
return (dma_addr_t)address;
|
||||
}
|
||||
|
||||
/*
|
||||
* The generic unmapping function for on page in the DMA address space.
|
||||
*/
|
||||
static void dma_ops_domain_unmap(struct amd_iommu *iommu,
|
||||
struct dma_ops_domain *dom,
|
||||
unsigned long address)
|
||||
@ -597,6 +741,12 @@ static void dma_ops_domain_unmap(struct amd_iommu *iommu,
|
||||
*pte = 0ULL;
|
||||
}
|
||||
|
||||
/*
|
||||
* 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.
|
||||
* Must be called with the domain lock held.
|
||||
*/
|
||||
static dma_addr_t __map_single(struct device *dev,
|
||||
struct amd_iommu *iommu,
|
||||
struct dma_ops_domain *dma_dom,
|
||||
@ -628,6 +778,10 @@ static dma_addr_t __map_single(struct device *dev,
|
||||
return address;
|
||||
}
|
||||
|
||||
/*
|
||||
* Does the reverse of the __map_single function. Must be called with
|
||||
* the domain lock held too
|
||||
*/
|
||||
static void __unmap_single(struct amd_iommu *iommu,
|
||||
struct dma_ops_domain *dma_dom,
|
||||
dma_addr_t dma_addr,
|
||||
@ -652,6 +806,9 @@ static void __unmap_single(struct amd_iommu *iommu,
|
||||
dma_ops_free_addresses(dma_dom, dma_addr, pages);
|
||||
}
|
||||
|
||||
/*
|
||||
* 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)
|
||||
{
|
||||
@ -664,6 +821,7 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
|
||||
get_device_resources(dev, &iommu, &domain, &devid);
|
||||
|
||||
if (iommu == NULL || domain == NULL)
|
||||
/* device not handled by any AMD IOMMU */
|
||||
return (dma_addr_t)paddr;
|
||||
|
||||
spin_lock_irqsave(&domain->lock, flags);
|
||||
@ -683,6 +841,9 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
|
||||
return addr;
|
||||
}
|
||||
|
||||
/*
|
||||
* 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)
|
||||
{
|
||||
@ -692,6 +853,7 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,
|
||||
u16 devid;
|
||||
|
||||
if (!get_device_resources(dev, &iommu, &domain, &devid))
|
||||
/* device not handled by any AMD IOMMU */
|
||||
return;
|
||||
|
||||
spin_lock_irqsave(&domain->lock, flags);
|
||||
@ -706,6 +868,10 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,
|
||||
spin_unlock_irqrestore(&domain->lock, flags);
|
||||
}
|
||||
|
||||
/*
|
||||
* This is a special map_sg function which is used if we should map a
|
||||
* device which is not handled by an AMD IOMMU in the system.
|
||||
*/
|
||||
static int map_sg_no_iommu(struct device *dev, struct scatterlist *sglist,
|
||||
int nelems, int dir)
|
||||
{
|
||||
@ -720,6 +886,10 @@ static int map_sg_no_iommu(struct device *dev, struct scatterlist *sglist,
|
||||
return nelems;
|
||||
}
|
||||
|
||||
/*
|
||||
* The exported map_sg function for dma_ops (handles scatter-gather
|
||||
* lists).
|
||||
*/
|
||||
static int map_sg(struct device *dev, struct scatterlist *sglist,
|
||||
int nelems, int dir)
|
||||
{
|
||||
@ -775,6 +945,10 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
|
||||
goto out;
|
||||
}
|
||||
|
||||
/*
|
||||
* The exported map_sg function for dma_ops (handles scatter-gather
|
||||
* lists).
|
||||
*/
|
||||
static void unmap_sg(struct device *dev, struct scatterlist *sglist,
|
||||
int nelems, int dir)
|
||||
{
|
||||
@ -804,6 +978,9 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
|
||||
spin_unlock_irqrestore(&domain->lock, flags);
|
||||
}
|
||||
|
||||
/*
|
||||
* The exported alloc_coherent function for dma_ops.
|
||||
*/
|
||||
static void *alloc_coherent(struct device *dev, size_t size,
|
||||
dma_addr_t *dma_addr, gfp_t flag)
|
||||
{
|
||||
@ -851,6 +1028,11 @@ static void *alloc_coherent(struct device *dev, size_t size,
|
||||
return virt_addr;
|
||||
}
|
||||
|
||||
/*
|
||||
* The exported free_coherent function for dma_ops.
|
||||
* FIXME: fix the generic x86 DMA layer so that it actually calls that
|
||||
* function.
|
||||
*/
|
||||
static void free_coherent(struct device *dev, size_t size,
|
||||
void *virt_addr, dma_addr_t dma_addr)
|
||||
{
|
||||
@ -879,6 +1061,8 @@ static void free_coherent(struct device *dev, size_t size,
|
||||
}
|
||||
|
||||
/*
|
||||
* The function for pre-allocating protection domains.
|
||||
*
|
||||
* If the driver core informs the DMA layer if a driver grabs a device
|
||||
* we don't need to preallocate the protection domains anymore.
|
||||
* For now we have to.
|
||||
@ -921,12 +1105,20 @@ static struct dma_mapping_ops amd_iommu_dma_ops = {
|
||||
.unmap_sg = unmap_sg,
|
||||
};
|
||||
|
||||
/*
|
||||
* The function which clues the AMD IOMMU driver into dma_ops.
|
||||
*/
|
||||
int __init amd_iommu_init_dma_ops(void)
|
||||
{
|
||||
struct amd_iommu *iommu;
|
||||
int order = amd_iommu_aperture_order;
|
||||
int ret;
|
||||
|
||||
/*
|
||||
* first allocate a default protection domain for every IOMMU we
|
||||
* 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);
|
||||
if (iommu->default_dom == NULL)
|
||||
@ -936,6 +1128,10 @@ int __init amd_iommu_init_dma_ops(void)
|
||||
goto free_domains;
|
||||
}
|
||||
|
||||
/*
|
||||
* If device isolation is enabled, pre-allocate the protection
|
||||
* domains for each device.
|
||||
*/
|
||||
if (amd_iommu_isolate)
|
||||
prealloc_protection_domains();
|
||||
|
||||
@ -947,6 +1143,7 @@ int __init amd_iommu_init_dma_ops(void)
|
||||
gart_iommu_aperture = 0;
|
||||
#endif
|
||||
|
||||
/* Make the driver finally visible to the drivers */
|
||||
dma_ops = &amd_iommu_dma_ops;
|
||||
|
||||
return 0;
|
||||
|
||||
Loading…
Reference in New Issue
Block a user