| Message ID | 146861130048.16896.9105590813517990661.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 7bfe97c763fe |
| Headers | show |
Hi Dan, On Fri, Jul 15, 2016 at 12:35:58PM -0700, Dan Williams wrote: > Test the virtual disk ranges that platform firmware like EDK2/OVMF might > emit. > > Cc: "Lee, Chun-Yi" <jlee@suse.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > Changes since v1: > > * Move this test range to nfit_test.1 to not break the spa ordering > expected by the nfit_test.0 tests. > I reviewed and tested this patch by nvdimm testing driver. Please feel free to add Reviewed-by: Lee, Chun-Yi <jlee@suse.com> But I have a question to run nfit testing driver on OVMF. I always got "coherent allocation failed" as below: [ 86.530950] platform nfit_test.0: swiotlb buffer is full (sz: 134217728 bytes) [ 86.530954] swiotlb: coherent allocation failed for device nfit_test.0 size=134217728 [ 86.530956] CPU: 0 PID: 1291 Comm: insmod Tainted: G OE 4.7.0-rc7-default+ #323 [ 86.530958] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 [ 86.530959] 0000000000000000 ffff880234e63bb0 ffffffff81334e5c 0000000008000000 [ 86.530961] ffff880232fa5998 ffff880234e63bf0 ffffffff8135eb79 ffff88020000000f [ 86.530963] 00000000024002c0 ffff880232fa5998 0000000008000000 ffff880234e63c68 [ 86.530965] Call Trace: [ 86.530971] [<ffffffff81334e5c>] dump_stack+0x63/0x87 [ 86.530975] [<ffffffff8135eb79>] swiotlb_alloc_coherent+0x149/0x160 [ 86.530978] [<ffffffff810611d3>] x86_swiotlb_alloc_coherent+0x43/0x50 [ 86.530981] [<ffffffffa054db3a>] dma_alloc_attrs.constprop.10+0x5a/0x80 [nfit_test] [ 86.530984] [<ffffffffa00d618c>] nfit_test_init+0x18c/0x1000 [nfit_test] [ 86.530985] [<ffffffffa00d6000>] ? 0xffffffffa00d6000 [ 86.530988] [<ffffffff81002190>] do_one_initcall+0x50/0x190 [ 86.530992] [<ffffffff811eef00>] ? kmem_cache_alloc_trace+0x170/0x220 [ 86.530995] [<ffffffff8118dbfc>] ? do_init_module+0x27/0x1e2 [ 86.530997] [<ffffffff8118dc35>] do_init_module+0x60/0x1e2 [ 86.531000] [<ffffffff81108fb1>] load_module+0x1411/0x1c20 [ 86.531002] [<ffffffff81105cb0>] ? __symbol_put+0x40/0x40 [ 86.531006] [<ffffffff812dfbbd>] ? ima_post_read_file+0x3d/0x80 [ 86.531009] [<ffffffff811099c9>] SYSC_finit_module+0xa9/0xd0 [ 86.531011] [<ffffffff81109a0e>] SyS_finit_module+0xe/0x10 [ 86.531013] [<ffffffff81003a62>] do_syscall_64+0x62/0x110 [ 86.531016] [<ffffffff81630a61>] entry_SYSCALL64_slow_path+0x25/0x25 [ 86.531018] platform nfit_test.0: need 128M of free cma I tried to add "swiotlb=65536" and "cma=128M" or "cma=256M", even "cma=2G" kernel parameters, but nfit-test.ko still can NOT allocate DMA success. So, I always add a patch to change the DIMM_SIZE to SZ_1M, that means test0 only uses 4M. Then the nfit-test.ko works for testing. Did I miss anything before running the nfit testing driver? Thanks a lot! Joey Lee
On Sun, Jul 17, 2016 at 2:36 AM, joeyli <jlee@suse.com> wrote: > Hi Dan, > > On Fri, Jul 15, 2016 at 12:35:58PM -0700, Dan Williams wrote: >> Test the virtual disk ranges that platform firmware like EDK2/OVMF might >> emit. >> >> Cc: "Lee, Chun-Yi" <jlee@suse.com> >> Signed-off-by: Dan Williams <dan.j.williams@intel.com> >> --- >> Changes since v1: >> >> * Move this test range to nfit_test.1 to not break the spa ordering >> expected by the nfit_test.0 tests. >> > > I reviewed and tested this patch by nvdimm testing driver. > Please feel free to add > Reviewed-by: Lee, Chun-Yi <jlee@suse.com> > > But I have a question to run nfit testing driver on OVMF. I always got > "coherent allocation failed" as below: > > [ 86.530950] platform nfit_test.0: swiotlb buffer is full (sz: 134217728 bytes) > [ 86.530954] swiotlb: coherent allocation failed for device nfit_test.0 size=134217728 > [ 86.530956] CPU: 0 PID: 1291 Comm: insmod Tainted: G OE 4.7.0-rc7-default+ #323 > [ 86.530958] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 > [ 86.530959] 0000000000000000 ffff880234e63bb0 ffffffff81334e5c 0000000008000000 > [ 86.530961] ffff880232fa5998 ffff880234e63bf0 ffffffff8135eb79 ffff88020000000f > [ 86.530963] 00000000024002c0 ffff880232fa5998 0000000008000000 ffff880234e63c68 > [ 86.530965] Call Trace: > [ 86.530971] [<ffffffff81334e5c>] dump_stack+0x63/0x87 > [ 86.530975] [<ffffffff8135eb79>] swiotlb_alloc_coherent+0x149/0x160 > [ 86.530978] [<ffffffff810611d3>] x86_swiotlb_alloc_coherent+0x43/0x50 > [ 86.530981] [<ffffffffa054db3a>] dma_alloc_attrs.constprop.10+0x5a/0x80 [nfit_test] > [ 86.530984] [<ffffffffa00d618c>] nfit_test_init+0x18c/0x1000 [nfit_test] > [ 86.530985] [<ffffffffa00d6000>] ? 0xffffffffa00d6000 > [ 86.530988] [<ffffffff81002190>] do_one_initcall+0x50/0x190 > [ 86.530992] [<ffffffff811eef00>] ? kmem_cache_alloc_trace+0x170/0x220 > [ 86.530995] [<ffffffff8118dbfc>] ? do_init_module+0x27/0x1e2 > [ 86.530997] [<ffffffff8118dc35>] do_init_module+0x60/0x1e2 > [ 86.531000] [<ffffffff81108fb1>] load_module+0x1411/0x1c20 > [ 86.531002] [<ffffffff81105cb0>] ? __symbol_put+0x40/0x40 > [ 86.531006] [<ffffffff812dfbbd>] ? ima_post_read_file+0x3d/0x80 > [ 86.531009] [<ffffffff811099c9>] SYSC_finit_module+0xa9/0xd0 > [ 86.531011] [<ffffffff81109a0e>] SyS_finit_module+0xe/0x10 > [ 86.531013] [<ffffffff81003a62>] do_syscall_64+0x62/0x110 > [ 86.531016] [<ffffffff81630a61>] entry_SYSCALL64_slow_path+0x25/0x25 > [ 86.531018] platform nfit_test.0: need 128M of free cma > > I tried to add "swiotlb=65536" and "cma=128M" or "cma=256M", even "cma=2G" kernel > parameters, but nfit-test.ko still can NOT allocate DMA success. > > So, I always add a patch to change the DIMM_SIZE to SZ_1M, that means test0 > only uses 4M. Then the nfit-test.ko works for testing. > > Did I miss anything before running the nfit testing driver? The CMA dependency is a hassle, so for v4.8 I'm replacing it with vmalloc: https://patchwork.kernel.org/patch/9199761/ https://patchwork.kernel.org/patch/9199763/
On Sun, Jul 17, 2016 at 09:05:19AM -0700, Dan Williams wrote: > On Sun, Jul 17, 2016 at 2:36 AM, joeyli <jlee@suse.com> wrote: > > Hi Dan, > > > > On Fri, Jul 15, 2016 at 12:35:58PM -0700, Dan Williams wrote: > >> Test the virtual disk ranges that platform firmware like EDK2/OVMF might > >> emit. > >> > >> Cc: "Lee, Chun-Yi" <jlee@suse.com> > >> Signed-off-by: Dan Williams <dan.j.williams@intel.com> > >> --- > >> Changes since v1: > >> > >> * Move this test range to nfit_test.1 to not break the spa ordering > >> expected by the nfit_test.0 tests. > >> > > > > I reviewed and tested this patch by nvdimm testing driver. > > Please feel free to add > > Reviewed-by: Lee, Chun-Yi <jlee@suse.com> > > > > But I have a question to run nfit testing driver on OVMF. I always got > > "coherent allocation failed" as below: > > > > [ 86.530950] platform nfit_test.0: swiotlb buffer is full (sz: 134217728 bytes) > > [ 86.530954] swiotlb: coherent allocation failed for device nfit_test.0 size=134217728 > > [ 86.530956] CPU: 0 PID: 1291 Comm: insmod Tainted: G OE 4.7.0-rc7-default+ #323 > > [ 86.530958] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 > > [ 86.530959] 0000000000000000 ffff880234e63bb0 ffffffff81334e5c 0000000008000000 > > [ 86.530961] ffff880232fa5998 ffff880234e63bf0 ffffffff8135eb79 ffff88020000000f > > [ 86.530963] 00000000024002c0 ffff880232fa5998 0000000008000000 ffff880234e63c68 > > [ 86.530965] Call Trace: > > [ 86.530971] [<ffffffff81334e5c>] dump_stack+0x63/0x87 > > [ 86.530975] [<ffffffff8135eb79>] swiotlb_alloc_coherent+0x149/0x160 > > [ 86.530978] [<ffffffff810611d3>] x86_swiotlb_alloc_coherent+0x43/0x50 > > [ 86.530981] [<ffffffffa054db3a>] dma_alloc_attrs.constprop.10+0x5a/0x80 [nfit_test] > > [ 86.530984] [<ffffffffa00d618c>] nfit_test_init+0x18c/0x1000 [nfit_test] > > [ 86.530985] [<ffffffffa00d6000>] ? 0xffffffffa00d6000 > > [ 86.530988] [<ffffffff81002190>] do_one_initcall+0x50/0x190 > > [ 86.530992] [<ffffffff811eef00>] ? kmem_cache_alloc_trace+0x170/0x220 > > [ 86.530995] [<ffffffff8118dbfc>] ? do_init_module+0x27/0x1e2 > > [ 86.530997] [<ffffffff8118dc35>] do_init_module+0x60/0x1e2 > > [ 86.531000] [<ffffffff81108fb1>] load_module+0x1411/0x1c20 > > [ 86.531002] [<ffffffff81105cb0>] ? __symbol_put+0x40/0x40 > > [ 86.531006] [<ffffffff812dfbbd>] ? ima_post_read_file+0x3d/0x80 > > [ 86.531009] [<ffffffff811099c9>] SYSC_finit_module+0xa9/0xd0 > > [ 86.531011] [<ffffffff81109a0e>] SyS_finit_module+0xe/0x10 > > [ 86.531013] [<ffffffff81003a62>] do_syscall_64+0x62/0x110 > > [ 86.531016] [<ffffffff81630a61>] entry_SYSCALL64_slow_path+0x25/0x25 > > [ 86.531018] platform nfit_test.0: need 128M of free cma > > > > I tried to add "swiotlb=65536" and "cma=128M" or "cma=256M", even "cma=2G" kernel > > parameters, but nfit-test.ko still can NOT allocate DMA success. > > > > So, I always add a patch to change the DIMM_SIZE to SZ_1M, that means test0 > > only uses 4M. Then the nfit-test.ko works for testing. > > > > Did I miss anything before running the nfit testing driver? > > The CMA dependency is a hassle, so for v4.8 I'm replacing it with vmalloc: > > https://patchwork.kernel.org/patch/9199761/ > https://patchwork.kernel.org/patch/9199763/ Those patches work great! Thanks a lot! Joey Lee
diff --git a/tools/testing/nvdimm/test/nfit.c b/tools/testing/nvdimm/test/nfit.c index 52df3c20231d..d4a4635c1186 100644 --- a/tools/testing/nvdimm/test/nfit.c +++ b/tools/testing/nvdimm/test/nfit.c @@ -104,6 +104,7 @@ enum { NUM_MEM = NUM_DCR + NUM_BDW + 2 /* spa0 iset */ + 4 /* spa1 iset */, DIMM_SIZE = SZ_32M, LABEL_SIZE = SZ_128K, + SPA_VCD_SIZE = SZ_4M, SPA0_SIZE = DIMM_SIZE, SPA1_SIZE = DIMM_SIZE*2, SPA2_SIZE = DIMM_SIZE, @@ -618,7 +619,7 @@ static int nfit_test0_alloc(struct nfit_test *t) static int nfit_test1_alloc(struct nfit_test *t) { - size_t nfit_size = sizeof(struct acpi_nfit_system_address) + size_t nfit_size = sizeof(struct acpi_nfit_system_address) * 2 + sizeof(struct acpi_nfit_memory_map) + offsetof(struct acpi_nfit_control_region, window_size); @@ -631,6 +632,10 @@ static int nfit_test1_alloc(struct nfit_test *t) if (!t->spa_set[0]) return -ENOMEM; + t->spa_set[1] = test_alloc(t, SPA_VCD_SIZE, &t->spa_set_dma[1]); + if (!t->spa_set[1]) + return -ENOMEM; + return ars_state_init(&t->pdev.dev, &t->ars_state); } @@ -1335,7 +1340,16 @@ static void nfit_test1_setup(struct nfit_test *t) spa->address = t->spa_set_dma[0]; spa->length = SPA2_SIZE; - offset += sizeof(*spa); + /* virtual cd region */ + spa = nfit_buf + sizeof(*spa); + spa->header.type = ACPI_NFIT_TYPE_SYSTEM_ADDRESS; + spa->header.length = sizeof(*spa); + memcpy(spa->range_guid, to_nfit_uuid(NFIT_SPA_VCD), 16); + spa->range_index = 0; + spa->address = t->spa_set_dma[1]; + spa->length = SPA_VCD_SIZE; + + offset += sizeof(*spa) * 2; /* mem-region0 (spa0, dimm0) */ memdev = nfit_buf + offset; memdev->header.type = ACPI_NFIT_TYPE_MEMORY_MAP;
Test the virtual disk ranges that platform firmware like EDK2/OVMF might emit. Cc: "Lee, Chun-Yi" <jlee@suse.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- Changes since v1: * Move this test range to nfit_test.1 to not break the spa ordering expected by the nfit_test.0 tests. tools/testing/nvdimm/test/nfit.c | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-)