| Message ID | 20180510233143.7236-1-qing.huang@oracle.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Headers | show |
On 2018/5/11 7:31, Qing Huang wrote: > When a system is under memory presure (high usage with fragments), > the original 256KB ICM chunk allocations will likely trigger kernel > memory management to enter slow path doing memory compact/migration > ops in order to complete high order memory allocations. > > When that happens, user processes calling uverb APIs may get stuck > for more than 120s easily even though there are a lot of free pages > in smaller chunks available in the system. > > Syslog: > ... > Dec 10 09:04:51 slcc03db02 kernel: [397078.572732] INFO: task > oracle_205573_e:205573 blocked for more than 120 seconds. > ... > > With 4KB ICM chunk size, the above issue is fixed. > > However in order to support 4KB ICM chunk size, we need to fix another > issue in large size kcalloc allocations. > > E.g. > Setting log_num_mtt=30 requires 1G mtt entries. With the 4KB ICM chunk > size, each ICM chunk can only hold 512 mtt entries (8 bytes for each mtt > entry). So we need a 16MB allocation for a table->icm pointer array to > hold 2M pointers which can easily cause kcalloc to fail. > > The solution is to use vzalloc to replace kcalloc. There is no need > for contiguous memory pages for a driver meta data structure (no need Hi, Replace continuous memory pages with virtual memory, is there any performance loss? Zhu Yanjun > of DMA ops). > > Signed-off-by: Qing Huang <qing.huang@oracle.com> > Acked-by: Daniel Jurgens <danielj@mellanox.com> > --- > drivers/net/ethernet/mellanox/mlx4/icm.c | 14 +++++++------- > 1 file changed, 7 insertions(+), 7 deletions(-) > > diff --git a/drivers/net/ethernet/mellanox/mlx4/icm.c b/drivers/net/ethernet/mellanox/mlx4/icm.c > index a822f7a..2b17a4b 100644 > --- a/drivers/net/ethernet/mellanox/mlx4/icm.c > +++ b/drivers/net/ethernet/mellanox/mlx4/icm.c > @@ -43,12 +43,12 @@ > #include "fw.h" > > /* > - * We allocate in as big chunks as we can, up to a maximum of 256 KB > - * per chunk. > + * We allocate in 4KB page size chunks to avoid high order memory > + * allocations in fragmented/high usage memory situation. > */ > enum { > - MLX4_ICM_ALLOC_SIZE = 1 << 18, > - MLX4_TABLE_CHUNK_SIZE = 1 << 18 > + MLX4_ICM_ALLOC_SIZE = 1 << 12, > + MLX4_TABLE_CHUNK_SIZE = 1 << 12 > }; > > static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk) > @@ -400,7 +400,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, > obj_per_chunk = MLX4_TABLE_CHUNK_SIZE / obj_size; > num_icm = (nobj + obj_per_chunk - 1) / obj_per_chunk; > > - table->icm = kcalloc(num_icm, sizeof(*table->icm), GFP_KERNEL); > + table->icm = vzalloc(num_icm * sizeof(*table->icm)); > if (!table->icm) > return -ENOMEM; > table->virt = virt; > @@ -446,7 +446,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, > mlx4_free_icm(dev, table->icm[i], use_coherent); > } > > - kfree(table->icm); > + vfree(table->icm); > > return -ENOMEM; > } > @@ -462,5 +462,5 @@ void mlx4_cleanup_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table) > mlx4_free_icm(dev, table->icm[i], table->coherent); > } > > - kfree(table->icm); > + vfree(table->icm); > } -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Thank you for reviewing it! On 5/10/2018 6:23 PM, Yanjun Zhu wrote: > > > > On 2018/5/11 9:15, Qing Huang wrote: >> >> >> >> On 5/10/2018 5:13 PM, Yanjun Zhu wrote: >>> >>> >>> On 2018/5/11 7:31, Qing Huang wrote: >>>> When a system is under memory presure (high usage with fragments), >>>> the original 256KB ICM chunk allocations will likely trigger kernel >>>> memory management to enter slow path doing memory compact/migration >>>> ops in order to complete high order memory allocations. >>>> >>>> When that happens, user processes calling uverb APIs may get stuck >>>> for more than 120s easily even though there are a lot of free pages >>>> in smaller chunks available in the system. >>>> >>>> Syslog: >>>> ... >>>> Dec 10 09:04:51 slcc03db02 kernel: [397078.572732] INFO: task >>>> oracle_205573_e:205573 blocked for more than 120 seconds. >>>> ... >>>> >>>> With 4KB ICM chunk size, the above issue is fixed. >>>> >>>> However in order to support 4KB ICM chunk size, we need to fix another >>>> issue in large size kcalloc allocations. >>>> >>>> E.g. >>>> Setting log_num_mtt=30 requires 1G mtt entries. With the 4KB ICM chunk >>>> size, each ICM chunk can only hold 512 mtt entries (8 bytes for >>>> each mtt >>>> entry). So we need a 16MB allocation for a table->icm pointer array to >>>> hold 2M pointers which can easily cause kcalloc to fail. >>>> >>>> The solution is to use vzalloc to replace kcalloc. There is no need >>>> for contiguous memory pages for a driver meta data structure (no need >>> Hi, >>> >>> Replace continuous memory pages with virtual memory, is there any >>> performance loss? >> >> Not really. "table->icm" will be accessed as individual pointer >> variables randomly. Kcalloc > > Sure. Thanks. If "table->icm" will be accessed as individual pointer > variables randomly, the performance loss > caused by discontinuous memory will be very trivial. > > Reviewed-by: Zhu Yanjun <yanjun.zhu@oracle.com> > >> also returns a virtual address except its mapped pages are guaranteed >> to be contiguous >> which will provide little advantage over vzalloc for individual >> pointer variable access. >> >> Qing >> >>> >>> Zhu Yanjun >>>> of DMA ops). >>>> >>>> Signed-off-by: Qing Huang <qing.huang@oracle.com> >>>> Acked-by: Daniel Jurgens <danielj@mellanox.com> >>>> --- >>>> drivers/net/ethernet/mellanox/mlx4/icm.c | 14 +++++++------- >>>> 1 file changed, 7 insertions(+), 7 deletions(-) >>>> >>>> diff --git a/drivers/net/ethernet/mellanox/mlx4/icm.c >>>> b/drivers/net/ethernet/mellanox/mlx4/icm.c >>>> index a822f7a..2b17a4b 100644 >>>> --- a/drivers/net/ethernet/mellanox/mlx4/icm.c >>>> +++ b/drivers/net/ethernet/mellanox/mlx4/icm.c >>>> @@ -43,12 +43,12 @@ >>>> #include "fw.h" >>>> /* >>>> - * We allocate in as big chunks as we can, up to a maximum of 256 KB >>>> - * per chunk. >>>> + * We allocate in 4KB page size chunks to avoid high order memory >>>> + * allocations in fragmented/high usage memory situation. >>>> */ >>>> enum { >>>> - MLX4_ICM_ALLOC_SIZE = 1 << 18, >>>> - MLX4_TABLE_CHUNK_SIZE = 1 << 18 >>>> + MLX4_ICM_ALLOC_SIZE = 1 << 12, >>>> + MLX4_TABLE_CHUNK_SIZE = 1 << 12 >>>> }; >>>> static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct >>>> mlx4_icm_chunk *chunk) >>>> @@ -400,7 +400,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, >>>> struct mlx4_icm_table *table, >>>> obj_per_chunk = MLX4_TABLE_CHUNK_SIZE / obj_size; >>>> num_icm = (nobj + obj_per_chunk - 1) / obj_per_chunk; >>>> - table->icm = kcalloc(num_icm, sizeof(*table->icm), >>>> GFP_KERNEL); >>>> + table->icm = vzalloc(num_icm * sizeof(*table->icm)); >>>> if (!table->icm) >>>> return -ENOMEM; >>>> table->virt = virt; >>>> @@ -446,7 +446,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, >>>> struct mlx4_icm_table *table, >>>> mlx4_free_icm(dev, table->icm[i], use_coherent); >>>> } >>>> - kfree(table->icm); >>>> + vfree(table->icm); >>>> return -ENOMEM; >>>> } >>>> @@ -462,5 +462,5 @@ void mlx4_cleanup_icm_table(struct mlx4_dev >>>> *dev, struct mlx4_icm_table *table) >>>> mlx4_free_icm(dev, table->icm[i], table->coherent); >>>> } >>>> - kfree(table->icm); >>>> + vfree(table->icm); >>>> } >>> >> > -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> On 11 May 2018, at 01:31, Qing Huang <qing.huang@oracle.com> wrote: > > When a system is under memory presure (high usage with fragments), > the original 256KB ICM chunk allocations will likely trigger kernel > memory management to enter slow path doing memory compact/migration > ops in order to complete high order memory allocations. > > When that happens, user processes calling uverb APIs may get stuck > for more than 120s easily even though there are a lot of free pages > in smaller chunks available in the system. > > Syslog: > ... > Dec 10 09:04:51 slcc03db02 kernel: [397078.572732] INFO: task > oracle_205573_e:205573 blocked for more than 120 seconds. > ... > > With 4KB ICM chunk size, the above issue is fixed. > > However in order to support 4KB ICM chunk size, we need to fix another > issue in large size kcalloc allocations. > > E.g. > Setting log_num_mtt=30 requires 1G mtt entries. With the 4KB ICM chunk > size, each ICM chunk can only hold 512 mtt entries (8 bytes for each mtt > entry). So we need a 16MB allocation for a table->icm pointer array to > hold 2M pointers which can easily cause kcalloc to fail. > > The solution is to use vzalloc to replace kcalloc. There is no need > for contiguous memory pages for a driver meta data structure (no need > of DMA ops). > > Signed-off-by: Qing Huang <qing.huang@oracle.com> > Acked-by: Daniel Jurgens <danielj@mellanox.com> > --- > drivers/net/ethernet/mellanox/mlx4/icm.c | 14 +++++++------- > 1 file changed, 7 insertions(+), 7 deletions(-) > > diff --git a/drivers/net/ethernet/mellanox/mlx4/icm.c b/drivers/net/ethernet/mellanox/mlx4/icm.c > index a822f7a..2b17a4b 100644 > --- a/drivers/net/ethernet/mellanox/mlx4/icm.c > +++ b/drivers/net/ethernet/mellanox/mlx4/icm.c > @@ -43,12 +43,12 @@ > #include "fw.h" > > /* > - * We allocate in as big chunks as we can, up to a maximum of 256 KB > - * per chunk. > + * We allocate in 4KB page size chunks to avoid high order memory > + * allocations in fragmented/high usage memory situation. > */ > enum { > - MLX4_ICM_ALLOC_SIZE = 1 << 18, > - MLX4_TABLE_CHUNK_SIZE = 1 << 18 > + MLX4_ICM_ALLOC_SIZE = 1 << 12, > + MLX4_TABLE_CHUNK_SIZE = 1 << 12 Shouldn’t these be the arch’s page size order? E.g., if running on SPARC, the hw page size is 8KiB. Thxs, Håkon > }; > > static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk) > @@ -400,7 +400,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, > obj_per_chunk = MLX4_TABLE_CHUNK_SIZE / obj_size; > num_icm = (nobj + obj_per_chunk - 1) / obj_per_chunk; > > - table->icm = kcalloc(num_icm, sizeof(*table->icm), GFP_KERNEL); > + table->icm = vzalloc(num_icm * sizeof(*table->icm)); > if (!table->icm) > return -ENOMEM; > table->virt = virt; > @@ -446,7 +446,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, > mlx4_free_icm(dev, table->icm[i], use_coherent); > } > > - kfree(table->icm); > + vfree(table->icm); > > return -ENOMEM; > } > @@ -462,5 +462,5 @@ void mlx4_cleanup_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table) > mlx4_free_icm(dev, table->icm[i], table->coherent); > } > > - kfree(table->icm); > + vfree(table->icm); > } > -- > 2.9.3 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 5/11/2018 3:27 AM, Håkon Bugge wrote: >> On 11 May 2018, at 01:31, Qing Huang<qing.huang@oracle.com> wrote: >> >> When a system is under memory presure (high usage with fragments), >> the original 256KB ICM chunk allocations will likely trigger kernel >> memory management to enter slow path doing memory compact/migration >> ops in order to complete high order memory allocations. >> >> When that happens, user processes calling uverb APIs may get stuck >> for more than 120s easily even though there are a lot of free pages >> in smaller chunks available in the system. >> >> Syslog: >> ... >> Dec 10 09:04:51 slcc03db02 kernel: [397078.572732] INFO: task >> oracle_205573_e:205573 blocked for more than 120 seconds. >> ... >> >> With 4KB ICM chunk size, the above issue is fixed. >> >> However in order to support 4KB ICM chunk size, we need to fix another >> issue in large size kcalloc allocations. >> >> E.g. >> Setting log_num_mtt=30 requires 1G mtt entries. With the 4KB ICM chunk >> size, each ICM chunk can only hold 512 mtt entries (8 bytes for each mtt >> entry). So we need a 16MB allocation for a table->icm pointer array to >> hold 2M pointers which can easily cause kcalloc to fail. >> >> The solution is to use vzalloc to replace kcalloc. There is no need >> for contiguous memory pages for a driver meta data structure (no need >> of DMA ops). >> >> Signed-off-by: Qing Huang<qing.huang@oracle.com> >> Acked-by: Daniel Jurgens<danielj@mellanox.com> >> --- >> drivers/net/ethernet/mellanox/mlx4/icm.c | 14 +++++++------- >> 1 file changed, 7 insertions(+), 7 deletions(-) >> >> diff --git a/drivers/net/ethernet/mellanox/mlx4/icm.c b/drivers/net/ethernet/mellanox/mlx4/icm.c >> index a822f7a..2b17a4b 100644 >> --- a/drivers/net/ethernet/mellanox/mlx4/icm.c >> +++ b/drivers/net/ethernet/mellanox/mlx4/icm.c >> @@ -43,12 +43,12 @@ >> #include "fw.h" >> >> /* >> - * We allocate in as big chunks as we can, up to a maximum of 256 KB >> - * per chunk. >> + * We allocate in 4KB page size chunks to avoid high order memory >> + * allocations in fragmented/high usage memory situation. >> */ >> enum { >> - MLX4_ICM_ALLOC_SIZE = 1 << 18, >> - MLX4_TABLE_CHUNK_SIZE = 1 << 18 >> + MLX4_ICM_ALLOC_SIZE = 1 << 12, >> + MLX4_TABLE_CHUNK_SIZE = 1 << 12 > Shouldn’t these be the arch’s page size order? E.g., if running on SPARC, the hw page size is 8KiB. Good point on supporting wider range of architectures. I got tunnel vision when fixing this on our x64 lab machines. Will send an v2 patch. Thanks, Qing > Thxs, Håkon > >> }; >> >> static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk) >> @@ -400,7 +400,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, >> obj_per_chunk = MLX4_TABLE_CHUNK_SIZE / obj_size; >> num_icm = (nobj + obj_per_chunk - 1) / obj_per_chunk; >> >> - table->icm = kcalloc(num_icm, sizeof(*table->icm), GFP_KERNEL); >> + table->icm = vzalloc(num_icm * sizeof(*table->icm)); >> if (!table->icm) >> return -ENOMEM; >> table->virt = virt; >> @@ -446,7 +446,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, >> mlx4_free_icm(dev, table->icm[i], use_coherent); >> } >> >> - kfree(table->icm); >> + vfree(table->icm); >> >> return -ENOMEM; >> } >> @@ -462,5 +462,5 @@ void mlx4_cleanup_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table) >> mlx4_free_icm(dev, table->icm[i], table->coherent); >> } >> >> - kfree(table->icm); >> + vfree(table->icm); >> } >> -- >> 2.9.3 >> >> -- >> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in >> the body of a message tomajordomo@vger.kernel.org >> More majordomo info athttp://vger.kernel.org/majordomo-info.html > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message tomajordomo@vger.kernel.org > More majordomo info athttp://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/net/ethernet/mellanox/mlx4/icm.c b/drivers/net/ethernet/mellanox/mlx4/icm.c index a822f7a..2b17a4b 100644 --- a/drivers/net/ethernet/mellanox/mlx4/icm.c +++ b/drivers/net/ethernet/mellanox/mlx4/icm.c @@ -43,12 +43,12 @@ #include "fw.h" /* - * We allocate in as big chunks as we can, up to a maximum of 256 KB - * per chunk. + * We allocate in 4KB page size chunks to avoid high order memory + * allocations in fragmented/high usage memory situation. */ enum { - MLX4_ICM_ALLOC_SIZE = 1 << 18, - MLX4_TABLE_CHUNK_SIZE = 1 << 18 + MLX4_ICM_ALLOC_SIZE = 1 << 12, + MLX4_TABLE_CHUNK_SIZE = 1 << 12 }; static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk) @@ -400,7 +400,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, obj_per_chunk = MLX4_TABLE_CHUNK_SIZE / obj_size; num_icm = (nobj + obj_per_chunk - 1) / obj_per_chunk; - table->icm = kcalloc(num_icm, sizeof(*table->icm), GFP_KERNEL); + table->icm = vzalloc(num_icm * sizeof(*table->icm)); if (!table->icm) return -ENOMEM; table->virt = virt; @@ -446,7 +446,7 @@ int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, mlx4_free_icm(dev, table->icm[i], use_coherent); } - kfree(table->icm); + vfree(table->icm); return -ENOMEM; } @@ -462,5 +462,5 @@ void mlx4_cleanup_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table) mlx4_free_icm(dev, table->icm[i], table->coherent); } - kfree(table->icm); + vfree(table->icm); }