Re: [PATCH rdma-next v1 1/2] lib/scatterlist: Fix wrong update of orig_nents
On Wed, Jun 30, 2021 at 07:44:21PM +0200, Thierry Reding wrote: > On Wed, Jun 30, 2021 at 01:12:26PM +0200, Marek Szyprowski wrote: > > Hi Leon, > > > > On 29.06.2021 10:40, Leon Romanovsky wrote: > > > From: Maor Gottlieb > > > > > > orig_nents should represent the number of entries with pages, > > > but __sg_alloc_table_from_pages sets orig_nents as the number of > > > total entries in the table. This is wrong when the API is used for > > > dynamic allocation where not all the table entries are mapped with > > > pages. It wasn't observed until now, since RDMA umem who uses this > > > API in the dynamic form doesn't use orig_nents implicit or explicit > > > by the scatterlist APIs. > > > > > > Fix it by: > > > 1. Set orig_nents as number of entries with pages also in > > > __sg_alloc_table_from_pages. > > > 2. Add a new field total_nents to reflect the total number of entries > > > in the table. This is required for the release flow (sg_free_table). > > > This filed should be used internally only by scatterlist. > > > > > > Fixes: 07da1223ec93 ("lib/scatterlist: Add support in dynamic allocation > > > of SG table from pages") > > > Signed-off-by: Maor Gottlieb > > > Signed-off-by: Leon Romanovsky > > > > This patch landed in linux-next 20210630 as commit a52724456928 > > ("lib/scatterlist: Fix wrong update of orig_nents"). It causes serious > > regression in DMA-IOMMU integration, which can be observed for example > > on ARM Juno board during boot: > > > > Unable to handle kernel paging request at virtual address 00376f42a6e40454 > > Mem abort info: > > ESR = 0x9604 > > EC = 0x25: DABT (current EL), IL = 32 bits > > SET = 0, FnV = 0 > > EA = 0, S1PTW = 0 > > FSC = 0x04: level 0 translation fault > > Data abort info: > > ISV = 0, ISS = 0x0004 > > CM = 0, WnR = 0 > > [00376f42a6e40454] address between user and kernel address ranges > > Internal error: Oops: 9604 [#1] PREEMPT SMP > > Modules linked in: > > CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.13.0-next-20210630+ #3585 > > Hardware name: ARM Juno development board (r1) (DT) > > pstate: 8005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) > > pc : __sg_free_table+0x60/0xa0 > > lr : __sg_free_table+0x7c/0xa0 > > .. > > Call trace: > > __sg_free_table+0x60/0xa0 > > sg_free_table+0x1c/0x28 > > iommu_dma_alloc+0xc8/0x388 > > dma_alloc_attrs+0xcc/0xf0 > > dmam_alloc_attrs+0x68/0xb8 > > sil24_port_start+0x60/0xe0 > > ata_host_start.part.32+0xbc/0x208 > > ata_host_activate+0x64/0x150 > > sil24_init_one+0x1e8/0x268 > > local_pci_probe+0x3c/0xa0 > > pci_device_probe+0x128/0x1c8 > > really_probe+0x138/0x2d0 > > __driver_probe_device+0x78/0xd8 > > driver_probe_device+0x40/0x110 > > __driver_attach+0xcc/0x118 > > bus_for_each_dev+0x68/0xc8 > > driver_attach+0x20/0x28 > > bus_add_driver+0x168/0x1f8 > > driver_register+0x60/0x110 > > __pci_register_driver+0x5c/0x68 > > sil24_pci_driver_init+0x20/0x28 > > do_one_initcall+0x84/0x450 > > kernel_init_freeable+0x31c/0x38c > > kernel_init+0x20/0x120 > > ret_from_fork+0x10/0x18 > > Code: d37be885 6b01007f 5284 54a2 (f8656813) > > ---[ end trace 4ba4f0c9c48711a1 ]--- > > Kernel panic - not syncing: Attempted to kill init! exitcode=0x000b > > > > It looks that some changes to the scatterlist structures are missing > > outside of the lib/scatterlist.c. > > > > For now I would suggest to revert this change. > > I see a very similar crash on Tegra during the HDA driver's probe. > > Leon, let me know if you need a tester for a replacement patch. With a great pleasure, I'll contact you offline when we prepare it. For now, this patch will be dropped. Thanks > > Thanks, > Thierry ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH rdma-next v1 1/2] lib/scatterlist: Fix wrong update of orig_nents
On Wed, Jun 30, 2021 at 01:12:26PM +0200, Marek Szyprowski wrote: > Hi Leon, > > On 29.06.2021 10:40, Leon Romanovsky wrote: > > From: Maor Gottlieb > > > > orig_nents should represent the number of entries with pages, > > but __sg_alloc_table_from_pages sets orig_nents as the number of > > total entries in the table. This is wrong when the API is used for > > dynamic allocation where not all the table entries are mapped with > > pages. It wasn't observed until now, since RDMA umem who uses this > > API in the dynamic form doesn't use orig_nents implicit or explicit > > by the scatterlist APIs. > > > > Fix it by: > > 1. Set orig_nents as number of entries with pages also in > > __sg_alloc_table_from_pages. > > 2. Add a new field total_nents to reflect the total number of entries > > in the table. This is required for the release flow (sg_free_table). > > This filed should be used internally only by scatterlist. > > > > Fixes: 07da1223ec93 ("lib/scatterlist: Add support in dynamic allocation of > > SG table from pages") > > Signed-off-by: Maor Gottlieb > > Signed-off-by: Leon Romanovsky > > This patch landed in linux-next 20210630 as commit a52724456928 > ("lib/scatterlist: Fix wrong update of orig_nents"). It causes serious > regression in DMA-IOMMU integration, which can be observed for example > on ARM Juno board during boot: > > Unable to handle kernel paging request at virtual address 00376f42a6e40454 > Mem abort info: > ESR = 0x9604 > EC = 0x25: DABT (current EL), IL = 32 bits > SET = 0, FnV = 0 > EA = 0, S1PTW = 0 > FSC = 0x04: level 0 translation fault > Data abort info: > ISV = 0, ISS = 0x0004 > CM = 0, WnR = 0 > [00376f42a6e40454] address between user and kernel address ranges > Internal error: Oops: 9604 [#1] PREEMPT SMP > Modules linked in: > CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.13.0-next-20210630+ #3585 > Hardware name: ARM Juno development board (r1) (DT) > pstate: 8005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) > pc : __sg_free_table+0x60/0xa0 > lr : __sg_free_table+0x7c/0xa0 > .. > Call trace: > __sg_free_table+0x60/0xa0 > sg_free_table+0x1c/0x28 > iommu_dma_alloc+0xc8/0x388 > dma_alloc_attrs+0xcc/0xf0 > dmam_alloc_attrs+0x68/0xb8 > sil24_port_start+0x60/0xe0 > ata_host_start.part.32+0xbc/0x208 > ata_host_activate+0x64/0x150 > sil24_init_one+0x1e8/0x268 > local_pci_probe+0x3c/0xa0 > pci_device_probe+0x128/0x1c8 > really_probe+0x138/0x2d0 > __driver_probe_device+0x78/0xd8 > driver_probe_device+0x40/0x110 > __driver_attach+0xcc/0x118 > bus_for_each_dev+0x68/0xc8 > driver_attach+0x20/0x28 > bus_add_driver+0x168/0x1f8 > driver_register+0x60/0x110 > __pci_register_driver+0x5c/0x68 > sil24_pci_driver_init+0x20/0x28 > do_one_initcall+0x84/0x450 > kernel_init_freeable+0x31c/0x38c > kernel_init+0x20/0x120 > ret_from_fork+0x10/0x18 > Code: d37be885 6b01007f 5284 54a2 (f8656813) > ---[ end trace 4ba4f0c9c48711a1 ]--- > Kernel panic - not syncing: Attempted to kill init! exitcode=0x000b > > It looks that some changes to the scatterlist structures are missing > outside of the lib/scatterlist.c. > > For now I would suggest to revert this change. I see a very similar crash on Tegra during the HDA driver's probe. Leon, let me know if you need a tester for a replacement patch. Thanks, Thierry signature.asc Description: PGP signature ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH rdma-next v1 1/2] lib/scatterlist: Fix wrong update of orig_nents
On Wed, Jun 30, 2021 at 01:12:26PM +0200, Marek Szyprowski wrote: > Hi Leon, > > On 29.06.2021 10:40, Leon Romanovsky wrote: > > From: Maor Gottlieb > > > > orig_nents should represent the number of entries with pages, > > but __sg_alloc_table_from_pages sets orig_nents as the number of > > total entries in the table. This is wrong when the API is used for > > dynamic allocation where not all the table entries are mapped with > > pages. It wasn't observed until now, since RDMA umem who uses this > > API in the dynamic form doesn't use orig_nents implicit or explicit > > by the scatterlist APIs. > > > > Fix it by: > > 1. Set orig_nents as number of entries with pages also in > > __sg_alloc_table_from_pages. > > 2. Add a new field total_nents to reflect the total number of entries > > in the table. This is required for the release flow (sg_free_table). > > This filed should be used internally only by scatterlist. > > > > Fixes: 07da1223ec93 ("lib/scatterlist: Add support in dynamic allocation of > > SG table from pages") > > Signed-off-by: Maor Gottlieb > > Signed-off-by: Leon Romanovsky <...> > For now I would suggest to revert this change. Thanks for the report, we will drop this patch. ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH rdma-next v1 1/2] lib/scatterlist: Fix wrong update of orig_nents
Hi Leon, On 29.06.2021 10:40, Leon Romanovsky wrote: > From: Maor Gottlieb > > orig_nents should represent the number of entries with pages, > but __sg_alloc_table_from_pages sets orig_nents as the number of > total entries in the table. This is wrong when the API is used for > dynamic allocation where not all the table entries are mapped with > pages. It wasn't observed until now, since RDMA umem who uses this > API in the dynamic form doesn't use orig_nents implicit or explicit > by the scatterlist APIs. > > Fix it by: > 1. Set orig_nents as number of entries with pages also in > __sg_alloc_table_from_pages. > 2. Add a new field total_nents to reflect the total number of entries > in the table. This is required for the release flow (sg_free_table). > This filed should be used internally only by scatterlist. > > Fixes: 07da1223ec93 ("lib/scatterlist: Add support in dynamic allocation of > SG table from pages") > Signed-off-by: Maor Gottlieb > Signed-off-by: Leon Romanovsky This patch landed in linux-next 20210630 as commit a52724456928 ("lib/scatterlist: Fix wrong update of orig_nents"). It causes serious regression in DMA-IOMMU integration, which can be observed for example on ARM Juno board during boot: Unable to handle kernel paging request at virtual address 00376f42a6e40454 Mem abort info: ESR = 0x9604 EC = 0x25: DABT (current EL), IL = 32 bits SET = 0, FnV = 0 EA = 0, S1PTW = 0 FSC = 0x04: level 0 translation fault Data abort info: ISV = 0, ISS = 0x0004 CM = 0, WnR = 0 [00376f42a6e40454] address between user and kernel address ranges Internal error: Oops: 9604 [#1] PREEMPT SMP Modules linked in: CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.13.0-next-20210630+ #3585 Hardware name: ARM Juno development board (r1) (DT) pstate: 8005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) pc : __sg_free_table+0x60/0xa0 lr : __sg_free_table+0x7c/0xa0 .. Call trace: __sg_free_table+0x60/0xa0 sg_free_table+0x1c/0x28 iommu_dma_alloc+0xc8/0x388 dma_alloc_attrs+0xcc/0xf0 dmam_alloc_attrs+0x68/0xb8 sil24_port_start+0x60/0xe0 ata_host_start.part.32+0xbc/0x208 ata_host_activate+0x64/0x150 sil24_init_one+0x1e8/0x268 local_pci_probe+0x3c/0xa0 pci_device_probe+0x128/0x1c8 really_probe+0x138/0x2d0 __driver_probe_device+0x78/0xd8 driver_probe_device+0x40/0x110 __driver_attach+0xcc/0x118 bus_for_each_dev+0x68/0xc8 driver_attach+0x20/0x28 bus_add_driver+0x168/0x1f8 driver_register+0x60/0x110 __pci_register_driver+0x5c/0x68 sil24_pci_driver_init+0x20/0x28 do_one_initcall+0x84/0x450 kernel_init_freeable+0x31c/0x38c kernel_init+0x20/0x120 ret_from_fork+0x10/0x18 Code: d37be885 6b01007f 5284 54a2 (f8656813) ---[ end trace 4ba4f0c9c48711a1 ]--- Kernel panic - not syncing: Attempted to kill init! exitcode=0x000b It looks that some changes to the scatterlist structures are missing outside of the lib/scatterlist.c. For now I would suggest to revert this change. > --- > include/linux/scatterlist.h | 8 ++-- > lib/scatterlist.c | 32 > 2 files changed, 14 insertions(+), 26 deletions(-) > > diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h > index 6f70572b2938..1c889141eb91 100644 > --- a/include/linux/scatterlist.h > +++ b/include/linux/scatterlist.h > @@ -35,8 +35,12 @@ struct scatterlist { > > struct sg_table { > struct scatterlist *sgl;/* the list */ > - unsigned int nents; /* number of mapped entries */ > - unsigned int orig_nents;/* original size of list */ > + unsigned int nents; /* number of DMA mapped entries */ > + unsigned int orig_nents;/* number of CPU mapped entries */ > + /* The fields below should be used internally only by > + * scatterlist implementation. > + */ > + unsigned int total_nents; /* number of total entries in the table > */ > }; > > /* > diff --git a/lib/scatterlist.c b/lib/scatterlist.c > index a59778946404..6db70a1e7dd0 100644 > --- a/lib/scatterlist.c > +++ b/lib/scatterlist.c > @@ -192,33 +192,26 @@ static void sg_kfree(struct scatterlist *sg, unsigned > int nents) > void __sg_free_table(struct sg_table *table, unsigned int max_ents, >unsigned int nents_first_chunk, sg_free_fn *free_fn) > { > - struct scatterlist *sgl, *next; > + struct scatterlist *sgl, *next = NULL; > unsigned curr_max_ents = nents_first_chunk ?: max_ents; > > if (unlikely(!table->sgl)) > return; > > sgl = table->sgl; > - while (table->orig_nents) { > - unsigned int alloc_size = table->orig_nents; > - unsigned int sg_size; > + while (table->total_nents) { > + unsigned int alloc_size = table->total_nents; > > /* >* If we have more than max_ents segments left, >* then assign 'next' to the