* [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock.
@ 2018-08-24 14:28 murphyt7
2018-08-24 14:53 ` Christoph Hellwig
2018-09-25 9:09 ` Joerg Roedel
0 siblings, 2 replies; 6+ messages in thread
From: murphyt7 @ 2018-08-24 14:28 UTC (permalink / raw)
To: leo.duran; +Cc: Tom Murphy, Joerg Roedel, iommu, linux-kernel
From: Tom Murphy <murphyt7@tcd.ie>
---
This patch allows the IOMMU API path in the AMD driver to be called from an atomic context.
This is useful for anyone building a driver which needs to call the IOMMU API while holding a spinlock.
drivers/iommu/amd_iommu.c | 14 ++++++++------
drivers/iommu/amd_iommu_types.h | 2 +-
2 files changed, 9 insertions(+), 7 deletions(-)
diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 0c910a8..642afc9 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -2904,7 +2904,7 @@ static void protection_domain_free(struct protection_domain *domain)
static int protection_domain_init(struct protection_domain *domain)
{
spin_lock_init(&domain->lock);
- mutex_init(&domain->api_lock);
+ spin_lock_init(&domain->api_lock);
domain->id = domain_id_alloc();
if (!domain->id)
return -ENOMEM;
@@ -3088,6 +3088,7 @@ static int amd_iommu_map(struct iommu_domain *dom, unsigned long iova,
phys_addr_t paddr, size_t page_size, int iommu_prot)
{
struct protection_domain *domain = to_pdomain(dom);
+ unsigned long flags;
int prot = 0;
int ret;
@@ -3099,9 +3100,9 @@ static int amd_iommu_map(struct iommu_domain *dom, unsigned long iova,
if (iommu_prot & IOMMU_WRITE)
prot |= IOMMU_PROT_IW;
- mutex_lock(&domain->api_lock);
- ret = iommu_map_page(domain, iova, paddr, page_size, prot, GFP_KERNEL);
- mutex_unlock(&domain->api_lock);
+ spin_lock_irqsave(&domain->api_lock, flags);
+ ret = iommu_map_page(domain, iova, paddr, page_size, prot, GFP_ATOMIC);
+ spin_unlock_irqrestore(&domain->api_lock, flags);
return ret;
}
@@ -3110,14 +3111,15 @@ static size_t amd_iommu_unmap(struct iommu_domain *dom, unsigned long iova,
size_t page_size)
{
struct protection_domain *domain = to_pdomain(dom);
+ unsigned long flags;
size_t unmap_size;
if (domain->mode == PAGE_MODE_NONE)
return -EINVAL;
- mutex_lock(&domain->api_lock);
+ spin_lock_irqsave(&domain->api_lock, flags);
unmap_size = iommu_unmap_page(domain, iova, page_size);
- mutex_unlock(&domain->api_lock);
+ spin_unlock_irqrestore(&domain->api_lock, flags);
domain_flush_tlb_pde(domain);
domain_flush_complete(domain);
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index 0d91785..a9077a5 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -444,7 +444,7 @@ struct protection_domain {
struct iommu_domain domain; /* generic domain handle used by
iommu core code */
spinlock_t lock; /* mostly used to lock the page table*/
- struct mutex api_lock; /* protect page tables in the iommu-api path */
+ spinlock_t api_lock; /* protect page tables in the iommu-api path */
u16 id; /* the domain id written to the device table */
int mode; /* paging mode (0-6 levels) */
u64 *pt_root; /* page table root pointer */
--
2.7.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock.
2018-08-24 14:28 [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock murphyt7
@ 2018-08-24 14:53 ` Christoph Hellwig
2018-08-24 15:24 ` Robin Murphy
2018-09-25 9:09 ` Joerg Roedel
1 sibling, 1 reply; 6+ messages in thread
From: Christoph Hellwig @ 2018-08-24 14:53 UTC (permalink / raw)
To: murphyt7; +Cc: leo.duran, iommu, linux-kernel
On Fri, Aug 24, 2018 at 02:28:49PM +0000, murphyt7@tcd.ie wrote:
> From: Tom Murphy <murphyt7@tcd.ie>
>
> ---
>
> This patch allows the IOMMU API path in the AMD driver to be called from an atomic context.
> This is useful for anyone building a driver which needs to call the IOMMU API while holding a spinlock.
I don't think that is a good idea. Please point to the code in the
driver and we can't probably find a better solution.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock.
2018-08-24 14:53 ` Christoph Hellwig
@ 2018-08-24 15:24 ` Robin Murphy
2018-08-27 6:58 ` Christoph Hellwig
[not found] ` <CALQxJuuWqiygrGMVjZ0j-AxBZOskoNHfdYoHEwW4Zgigqjf7kg@mail.gmail.com>
0 siblings, 2 replies; 6+ messages in thread
From: Robin Murphy @ 2018-08-24 15:24 UTC (permalink / raw)
To: Christoph Hellwig, murphyt7; +Cc: iommu, linux-kernel
On 24/08/18 15:53, Christoph Hellwig wrote:
> On Fri, Aug 24, 2018 at 02:28:49PM +0000, murphyt7@tcd.ie wrote:
>> From: Tom Murphy <murphyt7@tcd.ie>
>>
>> ---
>>
>> This patch allows the IOMMU API path in the AMD driver to be called from an atomic context.
>> This is useful for anyone building a driver which needs to call the IOMMU API while holding a spinlock.
>
> I don't think that is a good idea. Please point to the code in the
> driver and we can't probably find a better solution.
Although IIRC the AMD driver is in fact the only one whose map/unmap
callbacks aren't already spinlock-safe (or at least it was last time I
was looking). Stuff like iommu-dma is already relying on this in order
to implement streaming DMA API calls (which may be in atomic context) on
top of the corresponding IOMMU API operations.
Robin.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock.
2018-08-24 14:28 [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock murphyt7
2018-08-24 14:53 ` Christoph Hellwig
@ 2018-09-25 9:09 ` Joerg Roedel
1 sibling, 0 replies; 6+ messages in thread
From: Joerg Roedel @ 2018-09-25 9:09 UTC (permalink / raw)
To: murphyt7; +Cc: leo.duran, iommu, linux-kernel
On Fri, Aug 24, 2018 at 02:28:49PM +0000, murphyt7@tcd.ie wrote:
> From: Tom Murphy <murphyt7@tcd.ie>
>
> ---
>
> This patch allows the IOMMU API path in the AMD driver to be called
> from an atomic context. This is useful for anyone building a driver
> which needs to call the IOMMU API while holding a spinlock.
A bit more context would be helpful here. This is a pretty intrusive
change since we will drain the memory allocators atomic pools now while
building page-tables. This needs some more justification than just 'some
drivers might need it'.
Further, if we really address this problem it needs to happen through
the IOMMU-API, because the functions you are patching are called from
iommu_map()/iommu_unmap(), and these functions are allowed to sleep.
So the required approach here would be to introduce something like
iommu_map_atomic()/iommu_unmap_atomic() and implement call-backs for
that in more than just one driver.
Regards,
Joerg
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2018-09-25 9:09 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-08-24 14:28 [PATCH] Changing the AMD IOMMU API path to work in an atomic context which is necessary for any custom drivers using the IOMMU API while holding a spinlock murphyt7
2018-08-24 14:53 ` Christoph Hellwig
2018-08-24 15:24 ` Robin Murphy
2018-08-27 6:58 ` Christoph Hellwig
[not found] ` <CALQxJuuWqiygrGMVjZ0j-AxBZOskoNHfdYoHEwW4Zgigqjf7kg@mail.gmail.com>
[not found] ` <CALQxJutCDrg5A6Mhqq9Tkyd3siGrdY03LdmfEtV9v5vjYO+ZZQ@mail.gmail.com>
2018-08-27 7:07 ` Christoph Hellwig
2018-09-25 9:09 ` Joerg Roedel
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).