iommu/iova: Solve longterm IOVA issue
Subsystems (3)
v3 cover iommu/iova: Solve longterm IOVA issue
This series contains a patch to solve the longterm IOVA issue which leizhen originally tried to address at [0].
A sieved kernel log is at the following, showing periodic dumps of IOVA sizes, per CPU and per depot bin, per IOVA size granule:
https://raw.githubusercontent.com/hisilicon/kernel-dev/topic-iommu-5.10-iova-debug-v3/aging_test
Notice, for example, the following logs:
[13175.355584] print_iova1 cpu_total=40135 depot_total=3866 total=44001 [83483.457858] print_iova1 cpu_total=62532 depot_total=24476 total=87008
Where total IOVA rcache size has grown from 44K->87K over a long time.
Along with this patch, I included the following:
- A smaller helper to clear all IOVAs for a domain
- Change polarity of the IOVA magazine helpers
- Small optimisation from Cong Wang included, which was never applied [1]. There was some debate of the other patches in that series, but this one is quite straightforward.
Differnces to v2:
- Update commit message for patch 3/4
Differences to v1:
- Add IOVA clearing helper
- Add patch to change polarity of mag helpers
- Avoid logically-redundant extra variable in __iova_rcache_insert()
[0] https://lore.kernel.org/linux-iommu/[email protected]/
[1] https://lore.kernel.org/linux-iommu/[email protected]/
Cong Wang (1):
iommu: avoid taking iova_rbtree_lock twice
John Garry (3):
iommu/iova: Add free_all_cpu_cached_iovas() iommu/iova: Avoid double-negatives in magazine helpers iommu/iova: Flush CPU rcache for when a depot fills
drivers/iommu/iova.c | 66 +++++++++++++++++++++++++------------------- 1 file changed, 38 insertions(+), 28 deletions(-)
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
PATCH RESEND v3 1/4 iommu/iova: Add free_all_cpu_cached_iovas()
Add a helper function to free the CPU rcache for all online CPUs.
There also exists a function of the same name in drivers/iommu/intel/iommu.c, but the parameters are different, and there should be no conflict.
--- drivers/iommu/iova.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/iommu/iova.c +9 -4
@@ -227,6 +227,14 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,return -ENOMEM ; } +static void free_all_cpu_cached_iovas (struct iova_domain *iovad) +{ +unsigned int cpu; + +for_each_online_cpu (cpu) +free_cpu_cached_iovas (cpu, iovad); +} +static struct kmem_cache *iova_cache;static unsigned int iova_cache_users;static DEFINE_MUTEX (iova_cache_mutex);
@@ -422,15 +430,12 @@ alloc_iova_fast(struct iova_domain *iovad, unsigned long size, retry: new_iova =alloc_iova (iovad, size, limit_pfn, true);if (!new_iova) { -unsigned int cpu; -if (!flush_rcache)return 0 ;/* Try replenishing IOVAs by flushing rcache. */ flush_rcache = false; -for_each_online_cpu (cpu) -free_cpu_cached_iovas (cpu, iovad); +free_all_cpu_cached_iovas (iovad); goto retry; }
Signature · 1 line
2.26.2
Re: [RESEND PATCH v3 1/4] iommu/iova: Add free_all_cpu_cached_iovas()
Quoted 37 lines · John Garry
Add a helper function to free the CPU rcache for all online CPUs.
There also exists a function of the same name in
drivers/iommu/intel/iommu.c, but the parameters are different, and there should be no conflict.Signed-off-by John Garry <[email protected]>--- drivers/iommu/iova.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)drivers/iommu/iova.c +8
-2@@ -227,6 +227,14 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,return -ENOMEM ; } +static void free_all_cpu_cached_iovas (struct iova_domain *iovad) +{ +unsigned int cpu; + +for_each_online_cpu (cpu) +free_cpu_cached_iovas (cpu, iovad); +} +static struct kmem_cache *iova_cache;static unsigned int iova_cache_users;static DEFINE_MUTEX (iova_cache_mutex);@@ -422,15 +430,12 @@ alloc_iova_fast(struct iova_domain *iovad, unsigned long size, retry: new_iova =alloc_iova (iovad, size, limit_pfn, true);if (!new_iova) { -unsigned int cpu; -if (!flush_rcache)
drivers/iommu/iova.c alloc_iova_fast(struct iova_domain *iovad, unsigned long size, · line 441
Quoted 10 lines
drivers/iommu/iova.c +1
-2return 0 ;/* Try replenishing IOVAs by flushing rcache. */ flush_rcache = false; -for_each_online_cpu (cpu) -free_cpu_cached_iovas (cpu, iovad); +free_all_cpu_cached_iovas (iovad); goto retry; }
Quoted 1 line
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RESEND PATCH v3 1/4] iommu/iova: Add free_all_cpu_cached_iovas()
Quoted 5 lines · Leizhen (ThunderTown)
Quoted 3 lines
goto retry; }Reviewed-by Zhen Lei<[email protected]>
Thanks, incidentally this needs to be rebased, which I'll do now.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
PATCH RESEND v3 2/4 iommu/iova: Avoid double-negatives in magazine helpers
In reply to John GarryA similar crash to the following could be observed if initial CPU rcache magazine allocations fail in init_iova_rcaches():
Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
Mem abort info:
ESR = 0x96000004
EC = 0x25: DABT (current EL), IL = 32 bits SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS = 0x00000004
CM = 0, WnR = 0
[0000000000000000] user address but active_mm is swapper
Internal error: Oops: 96000004 [#1] PREEMPT SMP Modules linked in: CPU: 11 PID: 696 Comm: irq/40-hisi_sas Not tainted 5.9.0-rc7-dirty #109 Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 Call trace: free_iova_fast+0xfc/0x280 iommu_dma_free_iova+0x64/0x70 __iommu_dma_unmap+0x9c/0xf8 iommu_dma_unmap_sg+0xa8/0xc8 dma_unmap_sg_attrs+0x28/0x50 cq_thread_v3_hw+0x2dc/0x528 irq_thread_fn+0x2c/0xa0 irq_thread+0x130/0x1e0 kthread+0x154/0x158 ret_from_fork+0x10/0x34 Code: f9400060 f102001f 54000981 d4210000 (f9400043) ---[ end trace 4afcbdfc61b60467 ]---
The issue is that expression !iova_magazine_full(NULL) evaluates true; this falls over in in __iova_rcache_insert() when we attempt to cache a mag and cpu_rcache->loaded == NULL:
if (!iova_magazine_full(cpu_rcache->loaded)) {
can_insert = true;...
if (can_insert)
iova_magazine_push(cpu_rcache->loaded, iova_pfn);
As above, can_insert is evaluated true, which it shouldn't be, and we try to insert pfns in a NULL mag, which is not safe.
To avoid this, stop using double-negatives, like !iova_magazine_full() and !iova_magazine_empty(), and use positive tests, like iova_magazine_has_space() and iova_magazine_has_pfns(), respectively; these can safely deal with cpu_rcache->{loaded, prev} = NULL.
--- drivers/iommu/iova.c | 29 +++++++++++++++++------------ 1 file changed, 17 insertions(+), 12 deletions(-)
drivers/iommu/iova.c +17 -12
@@ -827,14 +827,18 @@ iova_magazine_free_pfns(struct iova_magazine *mag, struct iova_domain *iovad) mag->size =0 ; } -static bool iova_magazine_full (struct iova_magazine *mag) +static bool iova_magazine_has_space (struct iova_magazine *mag) { -return (mag && mag->size ==IOVA_MAG_SIZE ); +if (!mag) +return false; +return mag->size <IOVA_MAG_SIZE ; } -static bool iova_magazine_empty (struct iova_magazine *mag) +static bool iova_magazine_has_pfns (struct iova_magazine *mag) { -return (!mag || mag->size ==0 ); +if (!mag) +return false; +return mag->size; } static unsigned long iova_magazine_pop(struct iova_magazine *mag,
@@ -843,7 +847,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag,int i;unsigned long pfn; -BUG_ON (iova_magazine_empty (mag)); +BUG_ON (!iova_magazine_has_pfns (mag));/* Only fall back to the rbtree if we have no suitable pfns at all */ for (i = mag->size -1 ; mag->pfns[i] > limit_pfn; i--)
@@ -859,7 +863,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag,static void iova_magazine_push (struct iova_magazine *mag,unsigned long pfn) { -BUG_ON (iova_magazine_full (mag)); +BUG_ON (!iova_magazine_has_space (mag)); mag->pfns[mag->size++] = pfn; }
@@ -905,9 +909,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, cpu_rcache =raw_cpu_ptr (rcache->cpu_rcaches);spin_lock_irqsave (&cpu_rcache->lock, flags); -if (!iova_magazine_full (cpu_rcache->loaded)) { +if (iova_magazine_has_space (cpu_rcache->loaded)) { can_insert = true; - }else if (!iova_magazine_full (cpu_rcache->prev)) { + }else if (iova_magazine_has_space (cpu_rcache->prev)) {swap (cpu_rcache->prev, cpu_rcache->loaded); can_insert = true; }else {
@@ -916,8 +920,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,if (new_mag) {spin_lock (&rcache->lock);if (rcache->depot_size <MAX_GLOBAL_MAGS ) { - rcache->depot[rcache->depot_size++] = - cpu_rcache->loaded; +if (cpu_rcache->loaded) + rcache->depot[rcache->depot_size++] = + cpu_rcache->loaded; }else { mag_to_free = cpu_rcache->loaded; }
@@ -968,9 +973,9 @@ static unsigned long __iova_rcache_get(struct iova_rcache *rcache, cpu_rcache =raw_cpu_ptr (rcache->cpu_rcaches);spin_lock_irqsave (&cpu_rcache->lock, flags); -if (!iova_magazine_empty (cpu_rcache->loaded)) { +if (iova_magazine_has_pfns (cpu_rcache->loaded)) { has_pfn = true; - }else if (!iova_magazine_empty (cpu_rcache->prev)) { + }else if (iova_magazine_has_pfns (cpu_rcache->prev)) {swap (cpu_rcache->prev, cpu_rcache->loaded); has_pfn = true; }else {
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RESEND PATCH v3 2/4] iommu/iova: Avoid double-negatives in magazine helpers
Quoted 115 lines · John Garry
A similar crash to the following could be observed if initial CPU rcache magazine allocations fail in
init_iova_rcaches():Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000 Mem abort info: ESR = 0x96000004 EC = 0x25: DABT (current EL), IL = 32 bits SET = 0, FnV = 0 EA = 0, S1PTW = 0 Data abort info: ISV = 0, ISS = 0x00000004 CM = 0, WnR = 0 [0000000000000000] user address but active_mm is swapper Internal error: Oops: 96000004 [#1] PREEMPT SMP Modules linked in: CPU: 11 PID: 696 Comm: irq/40-hisi_sas Not tainted 5.9.0-rc7-dirty #109 Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 Call trace: free_iova_fast+0xfc/0x280 iommu_dma_free_iova+0x64/0x70 __iommu_dma_unmap+0x9c/0xf8 iommu_dma_unmap_sg+0xa8/0xc8 dma_unmap_sg_attrs+0x28/0x50 cq_thread_v3_hw+0x2dc/0x528 irq_thread_fn+0x2c/0xa0 irq_thread+0x130/0x1e0 kthread+0x154/0x158 ret_from_fork+0x10/0x34 Code: f9400060 f102001f 54000981 d4210000 (f9400043) ---[ end trace 4afcbdfc61b60467 ]---The issue is that expression !iova_magazine_full(NULL) evaluates true; this falls over in in
__iova_rcache_insert()when we attempt to cache a mag andcpu_rcache->loaded== NULL:if (!iova_magazine_full(cpu_rcache->loaded)) { can_insert = true;...
if (
can_insert)iova_magazine_push(cpu_rcache->loaded, iova_pfn);As above,
can_insertis evaluated true, which it shouldn't be, and we try to insert pfns in a NULL mag, which is not safe.To avoid this, stop using double-negatives, like !iova_magazine_full() and !iova_magazine_empty(), and use positive tests, like
iova_magazine_has_space()andiova_magazine_has_pfns(), respectively; these can safely deal with cpu_rcache->{loaded, prev} = NULL.Signed-off-by John Garry <[email protected]>--- drivers/iommu/iova.c | 29 +++++++++++++++++------------ 1 file changed, 17 insertions(+), 12 deletions(-)drivers/iommu/iova.c +12
-8@@ -827,14 +827,18 @@ iova_magazine_free_pfns(struct iova_magazine *mag, struct iova_domain *iovad) mag->size =0 ; } -static bool iova_magazine_full (struct iova_magazine *mag) +static bool iova_magazine_has_space (struct iova_magazine *mag) { -return (mag && mag->size ==IOVA_MAG_SIZE ); +if (!mag) +return false; +return mag->size <IOVA_MAG_SIZE ; } -static bool iova_magazine_empty (struct iova_magazine *mag) +static bool iova_magazine_has_pfns (struct iova_magazine *mag) { -return (!mag || mag->size ==0 ); +if (!mag) +return false; +return mag->size; } static unsigned long iova_magazine_pop(struct iova_magazine *mag,@@ -843,7 +847,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag,int i;unsigned long pfn; -BUG_ON (iova_magazine_empty (mag)); +BUG_ON (!iova_magazine_has_pfns (mag));/* Only fall back to the rbtree if we have no suitable pfns at all */ for (i = mag->size -1 ; mag->pfns[i] > limit_pfn; i--)@@ -859,7 +863,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag,static void iova_magazine_push (struct iova_magazine *mag,unsigned long pfn) { -BUG_ON (iova_magazine_full (mag)); +BUG_ON (!iova_magazine_has_space (mag)); mag->pfns[mag->size++] = pfn; }@@ -905,9 +909,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, cpu_rcache =raw_cpu_ptr (rcache->cpu_rcaches);spin_lock_irqsave (&cpu_rcache->lock, flags); -if (!iova_magazine_full (cpu_rcache->loaded)) { +if (iova_magazine_has_space (cpu_rcache->loaded)) { can_insert = true; - }else if (!iova_magazine_full (cpu_rcache->prev)) { + }else if (iova_magazine_has_space (cpu_rcache->prev)) {swap (cpu_rcache->prev, cpu_rcache->loaded); can_insert = true; }else {
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 923
Quoted 6 lines
drivers/iommu/iova.c +1
-2@@ -916,8 +920,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,if (new_mag) {spin_lock (&rcache->lock);if (rcache->depot_size <MAX_GLOBAL_MAGS ) { - rcache->depot[rcache->depot_size++] = - cpu_rcache->loaded; +if (cpu_rcache->loaded)
Looks like it just needs to change this place. Compiler ensures that mag->size will not be accessed when mag is NULL.
static bool iova_magazine_full(struct iova_magazine *mag)
{
return (mag && mag->size == IOVA_MAG_SIZE);
}static bool iova_magazine_empty(struct iova_magazine *mag)
{
return (!mag || mag->size == 0);
}Quoted 8 lines
drivers/iommu/iova.c +2
-0+ rcache->depot[rcache->depot_size++] = + cpu_rcache->loaded; }else { mag_to_free = cpu_rcache->loaded; }@@ -968,9 +973,9 @@ static unsigned long __iova_rcache_get(struct iova_rcache *rcache, cpu_rcache =raw_cpu_ptr (rcache->cpu_rcaches);
drivers/iommu/iova.c static unsigned long __iova_rcache_get(struct iova_rcache *rcache, · line 981
Quoted 10 lines
drivers/iommu/iova.c +2
-2spin_lock_irqsave (&cpu_rcache->lock, flags); -if (!iova_magazine_empty (cpu_rcache->loaded)) { +if (iova_magazine_has_pfns (cpu_rcache->loaded)) { has_pfn = true; - }else if (!iova_magazine_empty (cpu_rcache->prev)) { + }else if (iova_magazine_has_pfns (cpu_rcache->prev)) {swap (cpu_rcache->prev, cpu_rcache->loaded); has_pfn = true; }else {
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RESEND PATCH v3 2/4] iommu/iova: Avoid double-negatives in magazine helpers
Quoted 50 lines · Leizhen (ThunderTown)
Quoted 47 lines · John Garry
A similar crash to the following could be observed if initial CPU rcache magazine allocations fail in
init_iova_rcaches():Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000 Mem abort info: ESR = 0x96000004 EC = 0x25: DABT (current EL), IL = 32 bits SET = 0, FnV = 0 EA = 0, S1PTW = 0 Data abort info: ISV = 0, ISS = 0x00000004 CM = 0, WnR = 0 [0000000000000000] user address but active_mm is swapper Internal error: Oops: 96000004 [#1] PREEMPT SMP Modules linked in: CPU: 11 PID: 696 Comm: irq/40-hisi_sas Not tainted 5.9.0-rc7-dirty #109 Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 Call trace: free_iova_fast+0xfc/0x280 iommu_dma_free_iova+0x64/0x70 __iommu_dma_unmap+0x9c/0xf8 iommu_dma_unmap_sg+0xa8/0xc8 dma_unmap_sg_attrs+0x28/0x50 cq_thread_v3_hw+0x2dc/0x528 irq_thread_fn+0x2c/0xa0 irq_thread+0x130/0x1e0 kthread+0x154/0x158 ret_from_fork+0x10/0x34 Code: f9400060 f102001f 54000981 d4210000 (f9400043) ---[ end trace 4afcbdfc61b60467 ]---The issue is that expression !iova_magazine_full(NULL) evaluates true; this falls over in in
__iova_rcache_insert()when we attempt to cache a mag andcpu_rcache->loaded== NULL:if (!iova_magazine_full(cpu_rcache->loaded)) { can_insert = true;...
if (
can_insert)iova_magazine_push(cpu_rcache->loaded, iova_pfn);As above,
can_insertis evaluated true, which it shouldn't be, and we try to insert pfns in a NULL mag, which is not safe.
Quoted 6 lines
Quoted 6 lines
To avoid this, stop using double-negatives, like !iova_magazine_full() and !iova_magazine_empty(), and use positive tests, like
iova_magazine_has_space()andiova_magazine_has_pfns(), respectively; these can safely deal with cpu_rcache->{loaded, prev} = NULL.Signed-off-by John Garry <[email protected]>
Thanks for checking here...
Quoted 47 lines
Quoted 47 lines
--- drivers/iommu/iova.c | 29 +++++++++++++++++------------ 1 file changed, 17 insertions(+), 12 deletions(-)drivers/iommu/iova.c +10
-6@@ -827,14 +827,18 @@ iova_magazine_free_pfns(struct iova_magazine *mag, struct iova_domain *iovad) mag->size =0 ; } -static bool iova_magazine_full (struct iova_magazine *mag) +static bool iova_magazine_has_space (struct iova_magazine *mag) { -return (mag && mag->size ==IOVA_MAG_SIZE ); +if (!mag) +return false; +return mag->size <IOVA_MAG_SIZE ; } -static bool iova_magazine_empty (struct iova_magazine *mag) +static bool iova_magazine_has_pfns (struct iova_magazine *mag) { -return (!mag || mag->size ==0 ); +if (!mag) +return false; +return mag->size; } static unsigned long iova_magazine_pop(struct iova_magazine *mag,@@ -843,7 +847,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag,int i;unsigned long pfn; -BUG_ON (iova_magazine_empty (mag)); +BUG_ON (!iova_magazine_has_pfns (mag));/* Only fall back to the rbtree if we have no suitable pfns at all */ for (i = mag->size -1 ; mag->pfns[i] > limit_pfn; i--)@@ -859,7 +863,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag,static void iova_magazine_push (struct iova_magazine *mag,unsigned long pfn) { -BUG_ON (iova_magazine_full (mag)); +BUG_ON (!iova_magazine_has_space (mag));
Quoted 7 lines
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 908
Quoted 7 lines
drivers/iommu/iova.c +0
-1mag->pfns[mag->size++] = pfn; }@@ -905,9 +909,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, cpu_rcache =raw_cpu_ptr (rcache->cpu_rcaches);spin_lock_irqsave (&cpu_rcache->lock, flags); -if (!iova_magazine_full (cpu_rcache->loaded)) {
*
Quoted 9 lines
Quoted 9 lines
drivers/iommu/iova.c +2
-1+if (iova_magazine_has_space (cpu_rcache->loaded)) { can_insert = true; - }else if (!iova_magazine_full (cpu_rcache->prev)) { + }else if (iova_magazine_has_space (cpu_rcache->prev)) {swap (cpu_rcache->prev, cpu_rcache->loaded); can_insert = true; }else {@@ -916,8 +920,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,if (new_mag) {
Quoted 8 lines
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 923
Quoted 5 lines
drivers/iommu/iova.c +1
-2spin_lock (&rcache->lock);if (rcache->depot_size <MAX_GLOBAL_MAGS ) { - rcache->depot[rcache->depot_size++] = - cpu_rcache->loaded; +if (cpu_rcache->loaded)Looks like it just needs to change this place. Compiler ensures that
mag->sizewill not be accessed when mag is NULL.
Not sure about that. We would get a crash prior to touching this codepath:
So if cpu_rcache->loaded == NULL at entry, then !iova_magazine_full(cpu_rcache->loaded (==NULL) ) evaluates true, * above, so can_insert is set true, and we then attempt iova_magazine_push(cpu_rcache->loaded (==NULL), iova_pfn), which is not safe.
ok?
Cheers,
John
Quoted 25 lines
static bool
iova_magazine_full(struct iova_magazine *mag){ return (mag && mag->size == IOVA_MAG_SIZE); }static bool
iova_magazine_empty(struct iova_magazine *mag){ return (!mag || mag->size == 0); }Quoted 9 lines
drivers/iommu/iova.c +2
-0+ rcache->depot[rcache->depot_size++] = + cpu_rcache->loaded; }else { mag_to_free = cpu_rcache->loaded; }@@ -968,9 +973,9 @@ static unsigned long __iova_rcache_get(struct iova_rcache *rcache, cpu_rcache =raw_cpu_ptr (rcache->cpu_rcaches);spin_lock_irqsave (&cpu_rcache->lock, flags);
Quoted 10 lines
drivers/iommu/iova.c static unsigned long __iova_rcache_get(struct iova_rcache *rcache, · line 981
Quoted 9 lines
drivers/iommu/iova.c +2
-2-if (!iova_magazine_empty (cpu_rcache->loaded)) { +if (iova_magazine_has_pfns (cpu_rcache->loaded)) { has_pfn = true; - }else if (!iova_magazine_empty (cpu_rcache->prev)) { + }else if (iova_magazine_has_pfns (cpu_rcache->prev)) {swap (cpu_rcache->prev, cpu_rcache->loaded); has_pfn = true; }else {
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Leizhen (ThunderTown)
Reviewed-byRe: [RESEND PATCH v3 2/4] iommu/iova: Avoid double-negatives in magazine helpers
Quoted 127 lines · John Garry
Quoted 50 lines · Leizhen (ThunderTown)
Quoted 47 lines · John Garry
A similar crash to the following could be observed if initial CPU rcache magazine allocations fail in
init_iova_rcaches():Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000Mem abort info:
ESR =0x96000004
EC =0x25: DABT (current EL), IL = 32 bits SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS =0x00000004
CM = 0, WnR = 0
[0000000000000000] user address butactive_mmis swapperInternal error: Oops: 96000004 [#1] PREEMPT SMP Modules linked in: CPU: 11 PID: 696 Comm: irq/40-hisi_sas Not tainted 5.9.0-rc7-dirty #109 Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 Call trace: free_iova_fast+0xfc/0x280 iommu_dma_free_iova+0x64/0x70 __iommu_dma_unmap+0x9c/0xf8 iommu_dma_unmap_sg+0xa8/0xc8 dma_unmap_sg_attrs+0x28/0x50 cq_thread_v3_hw+0x2dc/0x528 irq_thread_fn+0x2c/0xa0 irq_thread+0x130/0x1e0 kthread+0x154/0x158 ret_from_fork+0x10/0x34 Code: f9400060 f102001f 54000981 d4210000 (f9400043) ---[ end trace 4afcbdfc61b60467 ]---The issue is that expression !iova_magazine_full(NULL) evaluates true; this falls over in in
__iova_rcache_insert()when we attempt to cache a mag andcpu_rcache->loaded== NULL:if (!iova_magazine_full(cpu_rcache->loaded)) {
can_insert = true;
...if (
can_insert)iova_magazine_push(cpu_rcache->loaded, iova_pfn);As above,
can_insertis evaluated true, which it shouldn't be, and we try to insert pfns in a NULL mag, which is not safe.Quoted 6 lines
Quoted 6 lines
To avoid this, stop using double-negatives, like !iova_magazine_full() and !iova_magazine_empty(), and use positive tests, like
iova_magazine_has_space()andiova_magazine_has_pfns(), respectively; these can safely deal with cpu_rcache->{loaded, prev} = NULL.Signed-off-by John Garry <[email protected]>Thanks for checking here...
Quoted 38 lines
Quoted 38 lines
---
drivers/iommu/iova.c| 29 +++++++++++++++++------------1 file changed, 17 insertions(+), 12 deletions(-)drivers/iommu/iova.c +9
-2@@ -827,14 +827,18 @@ iova_magazine_free_pfns(struct iova_magazine *mag, struct iova_domain *iovad) mag->size = 0; } -static bool iova_magazine_full(struct iova_magazine *mag) +static bool iova_magazine_has_space(struct iova_magazine *mag) { - return (mag && mag->size == IOVA_MAG_SIZE); + if (!mag) + return false; + return mag->size < IOVA_MAG_SIZE; } -static bool iova_magazine_empty(struct iova_magazine *mag) +static bool iova_magazine_has_pfns(struct iova_magazine *mag) { - return (!mag || mag->size == 0); + if (!mag) + return false; + return mag->size; } static unsigned long iova_magazine_pop(struct iova_magazine *mag,@@ -843,7 +847,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag, int i; unsigned long pfn; - BUG_ON(iova_magazine_empty(mag)); + BUG_ON(!iova_magazine_has_pfns(mag)); /* Only fall back to the rbtree if we have no suitable pfns at all */ for (i = mag->size - 1; mag->pfns[i] > limit_pfn; i--)@@ -859,7 +863,7 @@ static unsigned long iova_magazine_pop(struct iova_magazine *mag, static void iova_magazine_push(struct iova_magazine *mag, unsigned long pfn) {Quoted 8 lines
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 911
Quoted 8 lines
drivers/iommu/iova.c +1
-1- BUG_ON(iova_magazine_full(mag)); + BUG_ON(!iova_magazine_has_space(mag)); mag->pfns[mag->size++] = pfn; }@@ -905,9 +909,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, cpu_rcache = raw_cpu_ptr(rcache->cpu_rcaches); spin_lock_irqsave(&cpu_rcache->lock, flags); - if (!iova_magazine_full(cpu_rcache->loaded)) {*
Quoted 9 lines
Quoted 9 lines
drivers/iommu/iova.c +2
-1+ if (iova_magazine_has_space(cpu_rcache->loaded)) { can_insert = true; - } else if (!iova_magazine_full(cpu_rcache->prev)) { + } else if (iova_magazine_has_space(cpu_rcache->prev)) { swap(cpu_rcache->prev, cpu_rcache->loaded); can_insert = true; } else {@@ -916,8 +920,9 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, if (new_mag) {Quoted 8 lines
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 923
Quoted 5 lines
drivers/iommu/iova.c +1
-2spin_lock(&rcache->lock); if (rcache->depot_size < MAX_GLOBAL_MAGS) { - rcache->depot[rcache->depot_size++] = - cpu_rcache->loaded; + if (cpu_rcache->loaded)Looks like it just needs to change this place. Compiler ensures that
mag->sizewill not be accessed when mag is NULL.
Quoted 5 lines
Not sure about that. We would get a crash prior to touching this codepath:
So if
cpu_rcache->loaded== NULL at entry, then !iova_magazine_full(cpu_rcache->loaded (==NULL) ) evaluates true, * above, socan_insertis set true, and we then attemptiova_magazine_push(cpu_rcache->loaded (==NULL), iova_pfn), which is not safe.ok?
OK, I got it.
Quoted 40 lines
Cheers,
JohnQuoted 25 lines
static bool
iova_magazine_full(struct iova_magazine *mag){
return (mag &&mag->size==IOVA_MAG_SIZE); }static bool
iova_magazine_empty(struct iova_magazine *mag){
return (!mag ||mag->size== 0);
}Quoted 9 lines
drivers/iommu/iova.c +2
-0+ rcache->depot[rcache->depot_size++] = + cpu_rcache->loaded; } else { mag_to_free = cpu_rcache->loaded; }@@ -968,9 +973,9 @@ static unsigned long __iova_rcache_get(struct iova_rcache *rcache, cpu_rcache = raw_cpu_ptr(rcache->cpu_rcaches); spin_lock_irqsave(&cpu_rcache->lock, flags); - if (!iova_magazine_empty(cpu_rcache->loaded)) {Quoted 9 lines
drivers/iommu/iova.c static unsigned long __iova_rcache_get(struct iova_rcache *rcache, · line 981
Quoted 8 lines
drivers/iommu/iova.c +2
-1+ if (iova_magazine_has_pfns(cpu_rcache->loaded)) { has_pfn = true; - } else if (!iova_magazine_empty(cpu_rcache->prev)) { + } else if (iova_magazine_has_pfns(cpu_rcache->prev)) { swap(cpu_rcache->prev, cpu_rcache->loaded); has_pfn = true; } else {
Quoted 2 lines
.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
PATCH RESEND v3 3/4 iommu/iova: Flush CPU rcache for when a depot fills
In reply to John GarryLeizhen reported some time ago that IOVA performance may degrade over time [0], but unfortunately his solution to fix this problem was not given attention.
To summarize, the issue is that as time goes by, the CPU rcache and depot rcache continue to grow. As such, IOVA RB tree access time also continues to grow.
At a certain point, a depot may become full, and also some CPU rcaches may also be full when inserting another IOVA is attempted. For this scenario, currently the "loaded" CPU rcache is freed and a new one is created. This freeing means that many IOVAs in the RB tree need to be freed, which makes IO throughput performance fall off a cliff in some storage scenarios:
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6314MB/0KB/0KB /s] [1616K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [5669MB/0KB/0KB /s] [1451K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6673MB/0KB/0KB /s] [1708K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6761MB/0KB/0KB /s] [1731K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6685MB/0KB/0KB /s] [1711K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6178MB/0KB/0KB /s] [1582K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6731MB/0KB/0KB /s] [1723K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2387MB/0KB/0KB /s] [611K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2689MB/0KB/0KB /s] [688K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2278MB/0KB/0KB /s] [583K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1288MB/0KB/0KB /s] [330K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1632MB/0KB/0KB /s] [418K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1765MB/0KB/0KB /s] [452K/0/0 iops]
And continue in this fashion, without recovering. Note that in this example it was required to wait 16 hours for this to occur. Also note that IO throughput also becomes gradually becomes more unstable leading up to this point.
This problem is only seen for non-strict mode. For strict mode, the rcaches stay quite compact.
As a solution to this issue, judge that the IOVA caches have grown too big when cached magazines need to be free, and just flush all the CPUs rcaches instead.
The depot rcaches, however, are not flushed, as they can be used to immediately replenish active CPUs.
In future, some IOVA compaction could be implemented to solve the instabilty issue, which I figure could be quite complex to implement.
[0] https://lore.kernel.org/linux-iommu/[email protected]/
--- drivers/iommu/iova.c | 16 ++++++---------- 1 file changed, 6 insertions(+), 10 deletions(-)
drivers/iommu/iova.c +6 -10
@@ -901,7 +901,6 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, struct iova_rcache *rcache, unsigned long iova_pfn) { -struct iova_magazine *mag_to_free =NULL ;struct iova_cpu_rcache *cpu_rcache;bool can_insert = false;unsigned long flags;
@@ -923,13 +922,12 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,if (cpu_rcache->loaded) rcache->depot[rcache->depot_size++] = cpu_rcache->loaded; - }else { - mag_to_free = cpu_rcache->loaded; + can_insert = true; + cpu_rcache->loaded = new_mag; }spin_unlock (&rcache->lock); - - cpu_rcache->loaded = new_mag; - can_insert = true; +if (!can_insert) +iova_magazine_free (new_mag); } }
@@ -938,10 +936,8 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,spin_unlock_irqrestore (&cpu_rcache->lock, flags); -if (mag_to_free) { -iova_magazine_free_pfns (mag_to_free, iovad); -iova_magazine_free (mag_to_free); - } +if (!can_insert) +free_all_cpu_cached_iovas (iovad);return can_insert; }
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RESEND PATCH v3 3/4] iommu/iova: Flush CPU rcache for when a depot fills
Quoted 90 lines · John Garry
Leizhen reported some time ago that IOVA performance may degrade over time [0], but unfortunately his solution to fix this problem was not given attention.
To summarize, the issue is that as time goes by, the CPU rcache and depot rcache continue to grow. As such, IOVA RB tree access time also continues to grow.
At a certain point, a depot may become full, and also some CPU rcaches may also be full when inserting another IOVA is attempted. For this scenario, currently the "loaded" CPU rcache is freed and a new one is created. This freeing means that many IOVAs in the RB tree need to be freed, which makes IO throughput performance fall off a cliff in some storage scenarios:
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6314MB/0KB/0KB /s] [1616K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [5669MB/0KB/0KB /s] [1451K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6673MB/0KB/0KB /s] [1708K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6761MB/0KB/0KB /s] [1731K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6685MB/0KB/0KB /s] [1711K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6178MB/0KB/0KB /s] [1582K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6731MB/0KB/0KB /s] [1723K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2387MB/0KB/0KB /s] [611K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2689MB/0KB/0KB /s] [688K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2278MB/0KB/0KB /s] [583K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1288MB/0KB/0KB /s] [330K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1632MB/0KB/0KB /s] [418K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1765MB/0KB/0KB /s] [452K/0/0 iops]And continue in this fashion, without recovering. Note that in this example it was required to wait 16 hours for this to occur. Also note that IO throughput also becomes gradually becomes more unstable leading up to this point.
This problem is only seen for non-strict mode. For strict mode, the rcaches stay quite compact.
As a solution to this issue, judge that the IOVA caches have grown too big when cached magazines need to be free, and just flush all the CPUs rcaches instead.
The depot rcaches, however, are not flushed, as they can be used to immediately replenish active CPUs.
In future, some IOVA compaction could be implemented to solve the instabilty issue, which I figure could be quite complex to implement.
[0] https://lore.kernel.org/linux-iommu/[email protected]/
Analyzed-by Zhen Lei <[email protected]>Reported-by Xiang Chen <[email protected]>Signed-off-by John Garry <[email protected]>--- drivers/iommu/iova.c | 16 ++++++---------- 1 file changed, 6 insertions(+), 10 deletions(-)drivers/iommu/iova.c +4
-6@@ -901,7 +901,6 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, struct iova_rcache *rcache, unsigned long iova_pfn) { -struct iova_magazine *mag_to_free =NULL ;struct iova_cpu_rcache *cpu_rcache;bool can_insert = false;unsigned long flags;@@ -923,13 +922,12 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,if (cpu_rcache->loaded) rcache->depot[rcache->depot_size++] = cpu_rcache->loaded; - }else { - mag_to_free = cpu_rcache->loaded; + can_insert = true; + cpu_rcache->loaded = new_mag; }spin_unlock (&rcache->lock); - - cpu_rcache->loaded = new_mag; - can_insert = true; +if (!can_insert) +iova_magazine_free (new_mag); } }
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 943
Quoted 6 lines
drivers/iommu/iova.c +0
-3@@ -938,10 +936,8 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,spin_unlock_irqrestore (&cpu_rcache->lock, flags); -if (mag_to_free) { -iova_magazine_free_pfns (mag_to_free, iovad); -iova_magazine_free (mag_to_free);
mag_to_free has been stripped out, that's why lock protection is not required here.
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 940
Quoted 3 lines
drivers/iommu/iova.c +2
-1- } +if (!can_insert) +free_all_cpu_cached_iovas (iovad);
Lock protection required.
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad,
Quoted 4 lines
drivers/iommu/iova.c +0
-0return can_insert; }
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RESEND PATCH v3 3/4] iommu/iova: Flush CPU rcache for when a depot fills
Quoted 50 lines · Leizhen (ThunderTown)
Quoted 47 lines · John Garry
Leizhen reported some time ago that IOVA performance may degrade over time [0], but unfortunately his solution to fix this problem was not given attention.
To summarize, the issue is that as time goes by, the CPU rcache and depot rcache continue to grow. As such, IOVA RB tree access time also continues to grow.
At a certain point, a depot may become full, and also some CPU rcaches may also be full when inserting another IOVA is attempted. For this scenario, currently the "loaded" CPU rcache is freed and a new one is created. This freeing means that many IOVAs in the RB tree need to be freed, which makes IO throughput performance fall off a cliff in some storage scenarios:
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6314MB/0KB/0KB /s] [1616K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [5669MB/0KB/0KB /s] [1451K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6673MB/0KB/0KB /s] [1708K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6761MB/0KB/0KB /s] [1731K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6685MB/0KB/0KB /s] [1711K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6178MB/0KB/0KB /s] [1582K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6731MB/0KB/0KB /s] [1723K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2387MB/0KB/0KB /s] [611K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2689MB/0KB/0KB /s] [688K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2278MB/0KB/0KB /s] [583K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1288MB/0KB/0KB /s] [330K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1632MB/0KB/0KB /s] [418K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1765MB/0KB/0KB /s] [452K/0/0 iops]And continue in this fashion, without recovering. Note that in this example it was required to wait 16 hours for this to occur. Also note that IO throughput also becomes gradually becomes more unstable leading up to this point.
This problem is only seen for non-strict mode. For strict mode, the rcaches stay quite compact.
As a solution to this issue, judge that the IOVA caches have grown too big when cached magazines need to be free, and just flush all the CPUs rcaches instead.
The depot rcaches, however, are not flushed, as they can be used to immediately replenish active CPUs.
Quoted 8 lines
Quoted 8 lines
In future, some IOVA compaction could be implemented to solve the instabilty issue, which I figure could be quite complex to implement.
[0] https://lore.kernel.org/linux-iommu/[email protected]/
Analyzed-by Zhen Lei <[email protected]>Reported-by Xiang Chen <[email protected]>Signed-off-by John Garry <[email protected]>
Thanks for having a look
Quoted 43 lines
Quoted 35 lines
--- drivers/iommu/iova.c | 16 ++++++---------- 1 file changed, 6 insertions(+), 10 deletions(-)drivers/iommu/iova.c +4
-6@@ -901,7 +901,6 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, struct iova_rcache *rcache, unsigned long iova_pfn) { -struct iova_magazine *mag_to_free =NULL ;struct iova_cpu_rcache *cpu_rcache;bool can_insert = false;unsigned long flags;@@ -923,13 +922,12 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,if (cpu_rcache->loaded) rcache->depot[rcache->depot_size++] = cpu_rcache->loaded; - }else { - mag_to_free = cpu_rcache->loaded; + can_insert = true; + cpu_rcache->loaded = new_mag; }spin_unlock (&rcache->lock); - - cpu_rcache->loaded = new_mag; - can_insert = true; +if (!can_insert) +iova_magazine_free (new_mag); } }drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 943
Quoted 6 lines
drivers/iommu/iova.c +0
-3@@ -938,10 +936,8 @@ static bool __iova_rcache_insert(struct iova_domain *iovad,spin_unlock_irqrestore (&cpu_rcache->lock, flags); -if (mag_to_free) { -iova_magazine_free_pfns (mag_to_free, iovad); -iova_magazine_free (mag_to_free);
mag_to_freehas been stripped out, that's why lock protection is not required here.
Quoted 4 lines
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad, · line 940
Quoted 3 lines
drivers/iommu/iova.c +2
-1- } +if (!can_insert) +free_all_cpu_cached_iovas (iovad);Lock protection required.
But we have the per-CPU rcache locking again in free_cpu_cached_iovas() (which is called per-CPU from free_all_cpu_cached_iovas()).
ok? Or some other lock you mean?
Cheers,
John
Quoted 8 lines
drivers/iommu/iova.c static bool __iova_rcache_insert(struct iova_domain *iovad,
Quoted 5 lines
drivers/iommu/iova.c +0
-0return can_insert; }.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Leizhen (ThunderTown)
Reviewed-byRe: [RESEND PATCH v3 3/4] iommu/iova: Flush CPU rcache for when a depot fills
Quoted 102 lines · John Garry
Quoted 50 lines · Leizhen (ThunderTown)
Quoted 47 lines · John Garry
Leizhen reported some time ago that IOVA performance may degrade over time [0], but unfortunately his solution to fix this problem was not given attention.
To summarize, the issue is that as time goes by, the CPU rcache and depot rcache continue to grow. As such, IOVA RB tree access time also continues to grow.
At a certain point, a depot may become full, and also some CPU rcaches may also be full when inserting another IOVA is attempted. For this scenario, currently the "loaded" CPU rcache is freed and a new one is created. This freeing means that many IOVAs in the RB tree need to be freed, which makes IO throughput performance fall off a cliff in some storage scenarios:
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6314MB/0KB/0KB /s] [1616K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [5669MB/0KB/0KB /s] [1451K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6673MB/0KB/0KB /s] [1708K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6031MB/0KB/0KB /s] [1544K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6761MB/0KB/0KB /s] [1731K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6705MB/0KB/0KB /s] [1717K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6685MB/0KB/0KB /s] [1711K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6178MB/0KB/0KB /s] [1582K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [6731MB/0KB/0KB /s] [1723K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2387MB/0KB/0KB /s] [611K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2689MB/0KB/0KB /s] [688K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [2278MB/0KB/0KB /s] [583K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1288MB/0KB/0KB /s] [330K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1632MB/0KB/0KB /s] [418K/0/0 iops]
Jobs: 12 (f=12): [RRRRRRRRRRRR] [0.0% done] [1765MB/0KB/0KB /s] [452K/0/0 iops]And continue in this fashion, without recovering. Note that in this example it was required to wait 16 hours for this to occur. Also note that IO throughput also becomes gradually becomes more unstable leading up to this point.
This problem is only seen for non-strict mode. For strict mode, the rcaches stay quite compact.
As a solution to this issue, judge that the IOVA caches have grown too big when cached magazines need to be free, and just flush all the CPUs rcaches instead.
The depot rcaches, however, are not flushed, as they can be used to immediately replenish active CPUs.
Quoted 8 lines
Quoted 8 lines
In future, some IOVA compaction could be implemented to solve the instabilty issue, which I figure could be quite complex to implement.
[0] https://lore.kernel.org/linux-iommu/[email protected]/
Analyzed-by Zhen Lei <[email protected]>Reported-by Xiang Chen <[email protected]>Signed-off-by John Garry <[email protected]>Thanks for having a look
Quoted 40 lines
Quoted 30 lines
---
drivers/iommu/iova.c| 16 ++++++----------1 file changed, 6 insertions(+), 10 deletions(-)drivers/iommu/iova.c +0
-1@@ -901,7 +901,6 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, struct iova_rcache *rcache, unsigned long iova_pfn) { - struct iova_magazine *mag_to_free = NULL; struct iova_cpu_rcache *cpu_rcache; bool can_insert = false; unsigned long flags;@@ -923,13 +922,12 @@ static bool __iova_rcache_insert(struct iova_domain *iovad, if (cpu_rcache->loaded)rcache->depot[rcache->depot_size++] =
cpu_rcache->loaded; - } else {
-mag_to_free=cpu_rcache->loaded; +can_insert= true;
+cpu_rcache->loaded=new_mag;
}
spin_unlock(&rcache->lock);
-
-cpu_rcache->loaded=new_mag;
-can_insert= true;
+ if (!can_insert)Quoted 8 lines
+
iova_magazine_free(new_mag);
}
}
@@ -938,10 +936,8 @@ static bool __iova_rcache_insert(structiova_domain*iovad,spin_unlock_irqrestore(&cpu_rcache->lock, flags);
- if (mag_to_free) {
-iova_magazine_free_pfns(mag_to_free, iovad);
-iova_magazine_free(mag_to_free);
mag_to_freehas been stripped out, that's why lock protection is not required here.
Quoted 8 lines
Quoted 4 lines
Quoted 3 lines
- }
+ if (!can_insert)
+free_all_cpu_cached_iovas(iovad);Lock protection required.
But we have the per-CPU rcache locking again in
free_cpu_cached_iovas()(which is called per-CPU fromfree_all_cpu_cached_iovas()).ok? Or some other lock you mean?
Oh, Sorry, think of function free_cpu_cached_iovas() as function free_iova_rcaches().
Quoted 9 lines
Cheers,
JohnQuoted 5 lines
Quoted 4 lines
return
can_insert;
}
Quoted 6 lines
Quoted 2 lines
.
.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
PATCH RESEND v3 4/4 iommu: avoid taking iova_rbtree_lock twice
In reply to John GarryFrom: Cong Wang <[email protected]>
Both find_iova() and __free_iova() take iova_rbtree_lock, there is no reason to take and release it twice inside free_iova().
Fold them into one critical section by calling the unlock versions instead.
--- drivers/iommu/iova.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/iommu/iova.c +6 -2
@@ -398,10 +398,14 @@ EXPORT_SYMBOL_GPL(__free_iova);void free_iova (struct iova_domain *iovad,unsigned long pfn) { -struct iova *iova =find_iova (iovad, pfn); +unsigned long flags; +struct iova *iova; +spin_lock_irqsave (&iovad->iova_rbtree_lock, flags); + iova =private_find_iova (iovad, pfn);if (iova) -__free_iova (iovad, iova); +private_free_iova (iovad, iova); +spin_unlock_irqrestore (&iovad->iova_rbtree_lock, flags); }EXPORT_SYMBOL_GPL (free_iova);
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
On 17/11/2020 10:25, John Garry wrote:
Hi Will,
Is there any chance that we can get these picked up for 5.11? We've seen this issue solved here for a long time.
Or, @Robin, let me know if not happy with this since v1.
BTW, patch #4 has been on the go for ~1 year now, and is a nice small optimisation from Cong, which I picked up and already had a RB tag.
Thanks,
John
Quoted 34 lines
This series contains a patch to solve the longterm IOVA issue which leizhen originally tried to address at [0].
A sieved kernel log is at the following, showing periodic dumps of IOVA sizes, per CPU and per depot bin, per IOVA size granule:
https://raw.githubusercontent.com/hisilicon/kernel-dev/topic-iommu-5.10-iova-debug-v3/aging_testNotice, for example, the following logs:
[13175.355584] print_iova1 cpu_total=40135 depot_total=3866 total=44001 [83483.457858] print_iova1 cpu_total=62532 depot_total=24476 total=87008Where total IOVA rcache size has grown from 44K->87K over a long time.
Along with this patch, I included the following:
- A smaller helper to clear all IOVAs for a domain
- Change polarity of the IOVA magazine helpers
- Small optimisation from Cong Wang included, which was never applied [1]. There was some debate of the other patches in that series, but this one is quite straightforward.Differnces to v2:
- Update commit message for patch 3/4Differences to v1:
- Add IOVA clearing helper
- Add patch to change polarity of mag helpers
- Avoid logically-redundant extra variable in__iova_rcache_insert()[0] https://lore.kernel.org/linux-iommu/[email protected]/
[1] https://lore.kernel.org/linux-iommu/[email protected]/Cong Wang (1):
iommu: avoid takingiova_rbtree_locktwice
Quoted 8 lines
John Garry (3):
iommu/iova: Addfree_all_cpu_cached_iovas()iommu/iova: Avoid double-negatives in magazine helpers iommu/iova: Flush CPU rcache for when a depot fillsdrivers/iommu/iova.c | 66 +++++++++++++++++++++++++------------------- 1 file changed, 38 insertions(+), 28 deletions(-)
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Will Deacon
maintainerQuoted 8 lines · John Garry
On 17/11/2020 10:25, John Garry wrote:
Is there any chance that we can get these picked up for 5.11? We've seen this issue solved here for a long time.Or, @Robin, let me know if not happy with this since v1.
BTW, patch #4 has been on the go for ~1 year now, and is a nice small optimisation from Cong, which I picked up and already had a RB tag.
I can pick the last patch up, but I'd really like some reviewed/tested-bys on the others.
Will
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
John Garry
Tested-by Xiang ChenOn 01/12/2020 21:02, Will Deacon wrote:
cc'ing some more people who have touched iova code recently
Quoted 5 lines · John Garry
Quoted 4 lines · John Garry
Is there any chance that we can get these picked up for 5.11? We've seen this issue solved here for a long time.
Quoted 7 lines
Quoted 4 lines
Or, @Robin, let me know if not happy with this since v1.
BTW, patch #4 has been on the go for ~1 year now, and is a nice small optimisation from Cong, which I picked up and already had a RB tag.
I can pick the last patch up, but I'd really like some reviewed/tested-bys on the others.
ok, fair enough.
Considering the extremes required to unearth the main problem, it'll be hard to get testers, but, fwiw, I can provide a tested-by from the reporter:
@Robin, You originally had some interest in this topic - are you now satisfied with the changes I am proposing?
Please let me know.
Thanks,
John
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Will Deacon
maintainerQuoted 8 lines · John Garry
This series contains a patch to solve the longterm IOVA issue which leizhen originally tried to address at [0].
A sieved kernel log is at the following, showing periodic dumps of IOVA sizes, per CPU and per depot bin, per IOVA size granule:
https://raw.githubusercontent.com/hisilicon/kernel-dev/topic-iommu-5.10-iova-debug-v3/aging_test[...]
Applied the final patch to arm64 (for-next/iommu/iova), thanks!
[4/4] iommu: avoid taking iova_rbtree_lock twice
https://git.kernel.org/arm64/c/3a651b3a27a1
Cheers,
Signature · 9 lines
Will https://fixes.arm64.dev https://next.arm64.dev https://will.arm64.dev _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Quoted 9 lines · Will Deacon
Quoted 7 lines · John Garry
This series contains a patch to solve the longterm IOVA issue which leizhen originally tried to address at [0].
A sieved kernel log is at the following, showing periodic dumps of IOVA sizes, per CPU and per depot bin, per IOVA size granule:
https://raw.githubusercontent.com/hisilicon/kernel-dev/topic-iommu-5.10-iova-debug-v3/aging_test
Quoted 6 lines
Quoted 1 line
[...]
Applied the final patch to arm64 (for-next/iommu/iova), thanks!
[4/4] iommu: avoid taking
iova_rbtree_locktwicehttps://git.kernel.org/arm64/c/3a651b3a27a1
Glad it made in next, 2 years ago I couldn't convince iommu maintainer it's worth it (but with a different justification):
https://lore.kernel.org/linux-iommu/[email protected]/
Thanks,
Dmitry
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Quoted 9 lines · Dmitry Safonov
Quoted 8 lines · Will Deacon
Quoted 7 lines · John Garry
This series contains a patch to solve the longterm IOVA issue which leizhen originally tried to address at [0].
A sieved kernel log is at the following, showing periodic dumps of IOVA sizes, per CPU and per depot bin, per IOVA size granule:
https://raw.githubusercontent.com/hisilicon/kernel-dev/topic-iommu-5.10-iova-debug-v3/aging_test
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
[...]
Applied the final patch to arm64 (for-next/iommu/iova), thanks!
[4/4] iommu: avoid taking
iova_rbtree_locktwicehttps://git.kernel.org/arm64/c/3a651b3a27a1Glad it made in next, 2 years ago I couldn't convince iommu maintainer it's worth it (but with a different justification):
https://lore.kernel.org/linux-iommu/[email protected]/
Hi Dmitry,
I was unaware of your series, and it’s unfortunate that your optimization never made it. However I was having a quick look there, and, in case you did not notice, that the code which you were proposing changing in patch #1 for intel-iommu.c was removed in e70b081c6f37 ("iommu/vt-d: Remove IOVA handling code from the non-dma_ops path").
BTW, split_and_remove_iova() has no in-tree users anymore, so I can send a patch to delete if nobody else wants to.
BTW2, there's some more patches in my series which could use a review if you're feeling very helpful :)
Cheers,
John
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
+ linux-scsi (see https://lore.kernel.org/linux-iommu/[email protected]/)
Quoted 7 lines · John Garry
This series contains a patch to solve the longterm IOVA issue which leizhen originally tried to address at [0].
A sieved kernel log is at the following, showing periodic dumps of IOVA sizes, per CPU and per depot bin, per IOVA size granule:
https://raw.githubusercontent.com/hisilicon/kernel-dev/topic-iommu-5.10-iova-debug-v3/aging_test
Quoted 6 lines
Notice, for example, the following logs:
[13175.355584] print_iova1 cpu_total=40135 depot_total=3866 total=44001 [83483.457858] print_iova1 cpu_total=62532 depot_total=24476 total=87008Where total IOVA rcache size has grown from 44K->87K over a long time.
JFYI, I am able to reproduce this aging issue on another storage card, an LSI SAS 3008, so now it's harder to say it's an issue specific to a (buggy) single driver.
A log of the IOVA size dumps is here:
https://raw.githubusercontent.com/hisilicon/kernel-dev/064c4dc8869b3f2ad07edffceafde0b129f276b0/lsi3008_dmesg
Notice again how the total IOVA size goes up over time, like:
[ 68.176914] print_iova1 cpu_total=23663 depot_total=256 total=23919 [ 2337.008194] print_iova1 cpu_total=67361 depot_total=9088 total=76449 [17141.860078] print_iova1 cpu_total=73397 depot_total=10368 total=83765 [27087.850830] print_iova1 cpu_total=73386 depot_total=10624 total=84010 [10434.042877] print_iova1 cpu_total=90652 depot_total=12928 total=103580
I had to change some settings for that storage card to reproduce, though [0]. Could explain why no other reports.
So please consider this issue again...
Thanks,
john
[0] https://lore.kernel.org/linux-scsi/[email protected]/T/#m953d21446a5756981412c92d0924ca65c8d2f3a5
Quoted 9 lines
Along with this patch, I included the following:
- A smaller helper to clear all IOVAs for a domain
- Change polarity of the IOVA magazine helpers
- Small optimisation from Cong Wang included, which was never applied [1]. There was some debate of the other patches in that series, but this one is quite straightforward.Differnces to v2:
- Update commit message for patch 3/4
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu