| Message ID | 1438834307-26960-12-git-send-email-gwshan@linux.vnet.ibm.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On 08/06/2015 02:11 PM, Gavin Shan wrote: > On P7IOC, the whole DMA32 space is divided evenly to 256MB segments. > Each PE can consume one or multiple DMA32 segments. Current code > doesn't trace the available DMA32 segments and those consumed by > one particular PE. It's conflicting with PCI hotplug. > > The patch introduces one bitmap to PHB to trace the available > DMA32 segments for allocation, more fields to "struct pnv_ioda_pe" > to trace the consumed DMA32 segments by the PE, which is going to > be released when the PE is destroyed at PCI unplugging time. > > Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> > --- > arch/powerpc/platforms/powernv/pci-ioda.c | 40 +++++++++++++++++++++++-------- > arch/powerpc/platforms/powernv/pci.h | 4 +++- > 2 files changed, 33 insertions(+), 11 deletions(-) > > diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c > index cd22002..57ba8fd 100644 > --- a/arch/powerpc/platforms/powernv/pci-ioda.c > +++ b/arch/powerpc/platforms/powernv/pci-ioda.c > @@ -1946,6 +1946,7 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, > > /* Grab a 32-bit TCE table */ > pe->dma32_seg = base; > + pe->dma32_segcount = segs; > pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n", > (base << 28), ((base + segs) << 28) - 1); > > @@ -2006,8 +2007,13 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, > return; > fail: > /* XXX Failure: Try to fallback to 64-bit only ? */ > - if (pe->dma32_seg >= 0) > + if (pe->dma32_seg >= 0) { > + bitmap_clear(phb->ioda.dma32_segmap, > + pe->dma32_seg, pe->dma32_segcount); > pe->dma32_seg = -1; > + pe->dma32_segcount = 0; > + } > + > if (tce_mem) > __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs)); > if (tbl) { > @@ -2443,12 +2449,11 @@ static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, > pnv_ioda_setup_bus_dma(pe, pe->pbus); > } > > -static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, > - struct pnv_ioda_pe *pe, > - unsigned int base) > +static void pnv_ioda1_setup_dma(struct pnv_phb *phb, > + struct pnv_ioda_pe *pe) > { > struct pci_controller *hose = phb->hose; > - unsigned int dma_weight, segs; > + unsigned int dma_weight, base, segs; > > /* Calculate the PHB's DMA weight */ > dma_weight = pnv_ioda_phb_dma_weight(phb); > @@ -2461,11 +2466,28 @@ static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, > else > segs = (pe->dma32_weight * > phb->ioda.dma32_segcount) / dma_weight; > + > + /* > + * Allocate DMA32 segments. We might not have enough > + * resources available. However we expect at least one > + * to be available. > + */ > + do { > + base = bitmap_find_next_zero_area(phb->ioda.dma32_segmap, > + phb->ioda.dma32_segcount, > + 0, segs, 0); > + if (base < phb->ioda.dma32_segcount) { > + bitmap_set(phb->ioda.dma32_segmap, base, segs); > + break; > + } > + } while (--segs); If segs==0 before entering the loop, the loop will execute 0xfffffffe times. Make it for(;segs;--segs){ }. > + > + if (WARN_ON(!segs)) > + return; > + > pe_info(pe, "DMA weight %d, assigned %d segments\n", > pe->dma32_weight, segs); > pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs); > - > - return segs; > } > > #ifdef CONFIG_PCI_MSI > @@ -2933,20 +2955,18 @@ static void pnv_pci_ioda_setup_DMA(void) > struct pci_controller *hose, *tmp; > struct pnv_phb *phb; > struct pnv_ioda_pe *pe; > - unsigned int base; > > list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { > phb = hose->private_data; > pnv_pci_ioda_setup_opal_tce_kill(phb); > > - base = 0; > list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) { > if (!pe->dma32_weight) > continue; > > switch (phb->type) { > case PNV_PHB_IODA1: > - base += pnv_ioda1_setup_dma(phb, pe, base); > + pnv_ioda1_setup_dma(phb, pe); > break; > case PNV_PHB_IODA2: > pnv_pci_ioda2_setup_dma_pe(phb, pe); > diff --git a/arch/powerpc/platforms/powernv/pci.h b/arch/powerpc/platforms/powernv/pci.h > index 574fe43..1dc9578 100644 > --- a/arch/powerpc/platforms/powernv/pci.h > +++ b/arch/powerpc/platforms/powernv/pci.h > @@ -65,6 +65,7 @@ struct pnv_ioda_pe { > > /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */ > int dma32_seg; > + int dma32_segcount; > struct iommu_table_group table_group; > > /* 64-bit TCE bypass region */ > @@ -153,10 +154,11 @@ struct pnv_phb { > unsigned int io_segsize; > unsigned int io_pci_base; > > - /* IO, M32, M64 segment maps */ > + /* IO, M32, M64, DMA32 segment maps */ > unsigned long io_segmap[8]; > unsigned long m32_segmap[8]; > unsigned long m64_segmap[8]; > + unsigned long dma32_segmap[8]; > > /* PE allocation */ > struct mutex pe_alloc_mutex; >
On Mon, Aug 10, 2015 at 07:43:48PM +1000, Alexey Kardashevskiy wrote: >On 08/06/2015 02:11 PM, Gavin Shan wrote: >>On P7IOC, the whole DMA32 space is divided evenly to 256MB segments. >>Each PE can consume one or multiple DMA32 segments. Current code >>doesn't trace the available DMA32 segments and those consumed by >>one particular PE. It's conflicting with PCI hotplug. >> >>The patch introduces one bitmap to PHB to trace the available >>DMA32 segments for allocation, more fields to "struct pnv_ioda_pe" >>to trace the consumed DMA32 segments by the PE, which is going to >>be released when the PE is destroyed at PCI unplugging time. >> >>Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> >>--- >> arch/powerpc/platforms/powernv/pci-ioda.c | 40 +++++++++++++++++++++++-------- >> arch/powerpc/platforms/powernv/pci.h | 4 +++- >> 2 files changed, 33 insertions(+), 11 deletions(-) >> >>diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c >>index cd22002..57ba8fd 100644 >>--- a/arch/powerpc/platforms/powernv/pci-ioda.c >>+++ b/arch/powerpc/platforms/powernv/pci-ioda.c >>@@ -1946,6 +1946,7 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, >> >> /* Grab a 32-bit TCE table */ >> pe->dma32_seg = base; >>+ pe->dma32_segcount = segs; >> pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n", >> (base << 28), ((base + segs) << 28) - 1); >> >>@@ -2006,8 +2007,13 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, >> return; >> fail: >> /* XXX Failure: Try to fallback to 64-bit only ? */ >>- if (pe->dma32_seg >= 0) >>+ if (pe->dma32_seg >= 0) { >>+ bitmap_clear(phb->ioda.dma32_segmap, >>+ pe->dma32_seg, pe->dma32_segcount); >> pe->dma32_seg = -1; >>+ pe->dma32_segcount = 0; >>+ } >>+ >> if (tce_mem) >> __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs)); >> if (tbl) { >>@@ -2443,12 +2449,11 @@ static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, >> pnv_ioda_setup_bus_dma(pe, pe->pbus); >> } >> >>-static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, >>- struct pnv_ioda_pe *pe, >>- unsigned int base) >>+static void pnv_ioda1_setup_dma(struct pnv_phb *phb, >>+ struct pnv_ioda_pe *pe) >> { >> struct pci_controller *hose = phb->hose; >>- unsigned int dma_weight, segs; >>+ unsigned int dma_weight, base, segs; >> >> /* Calculate the PHB's DMA weight */ >> dma_weight = pnv_ioda_phb_dma_weight(phb); >>@@ -2461,11 +2466,28 @@ static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, >> else >> segs = (pe->dma32_weight * >> phb->ioda.dma32_segcount) / dma_weight; >>+ >>+ /* >>+ * Allocate DMA32 segments. We might not have enough >>+ * resources available. However we expect at least one >>+ * to be available. >>+ */ >>+ do { >>+ base = bitmap_find_next_zero_area(phb->ioda.dma32_segmap, >>+ phb->ioda.dma32_segcount, >>+ 0, segs, 0); >>+ if (base < phb->ioda.dma32_segcount) { >>+ bitmap_set(phb->ioda.dma32_segmap, base, segs); >>+ break; >>+ } >>+ } while (--segs); > > >If segs==0 before entering the loop, the loop will execute 0xfffffffe times. >Make it for(;segs;--segs){ }. > The segs is always equal to 1 or more than that. However, "for()" statement seems better and I'll change it. > >>+ >>+ if (WARN_ON(!segs)) >>+ return; >>+ >> pe_info(pe, "DMA weight %d, assigned %d segments\n", >> pe->dma32_weight, segs); >> pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs); >>- >>- return segs; >> } >> >> #ifdef CONFIG_PCI_MSI >>@@ -2933,20 +2955,18 @@ static void pnv_pci_ioda_setup_DMA(void) >> struct pci_controller *hose, *tmp; >> struct pnv_phb *phb; >> struct pnv_ioda_pe *pe; >>- unsigned int base; >> >> list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { >> phb = hose->private_data; >> pnv_pci_ioda_setup_opal_tce_kill(phb); >> >>- base = 0; >> list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) { >> if (!pe->dma32_weight) >> continue; >> >> switch (phb->type) { >> case PNV_PHB_IODA1: >>- base += pnv_ioda1_setup_dma(phb, pe, base); >>+ pnv_ioda1_setup_dma(phb, pe); >> break; >> case PNV_PHB_IODA2: >> pnv_pci_ioda2_setup_dma_pe(phb, pe); >>diff --git a/arch/powerpc/platforms/powernv/pci.h b/arch/powerpc/platforms/powernv/pci.h >>index 574fe43..1dc9578 100644 >>--- a/arch/powerpc/platforms/powernv/pci.h >>+++ b/arch/powerpc/platforms/powernv/pci.h >>@@ -65,6 +65,7 @@ struct pnv_ioda_pe { >> >> /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */ >> int dma32_seg; >>+ int dma32_segcount; >> struct iommu_table_group table_group; >> >> /* 64-bit TCE bypass region */ >>@@ -153,10 +154,11 @@ struct pnv_phb { >> unsigned int io_segsize; >> unsigned int io_pci_base; >> >>- /* IO, M32, M64 segment maps */ >>+ /* IO, M32, M64, DMA32 segment maps */ >> unsigned long io_segmap[8]; >> unsigned long m32_segmap[8]; >> unsigned long m64_segmap[8]; >>+ unsigned long dma32_segmap[8]; >> >> /* PE allocation */ >> struct mutex pe_alloc_mutex; >> Thanks, Gavin
On Mon, Aug 10, 2015 at 07:43:48PM +1000, Alexey Kardashevskiy wrote: >On 08/06/2015 02:11 PM, Gavin Shan wrote: >>On P7IOC, the whole DMA32 space is divided evenly to 256MB segments. >>Each PE can consume one or multiple DMA32 segments. Current code >>doesn't trace the available DMA32 segments and those consumed by >>one particular PE. It's conflicting with PCI hotplug. >> >>The patch introduces one bitmap to PHB to trace the available >>DMA32 segments for allocation, more fields to "struct pnv_ioda_pe" >>to trace the consumed DMA32 segments by the PE, which is going to >>be released when the PE is destroyed at PCI unplugging time. >> >>Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> >>--- >> arch/powerpc/platforms/powernv/pci-ioda.c | 40 +++++++++++++++++++++++-------- >> arch/powerpc/platforms/powernv/pci.h | 4 +++- >> 2 files changed, 33 insertions(+), 11 deletions(-) >> >>diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c >>index cd22002..57ba8fd 100644 >>--- a/arch/powerpc/platforms/powernv/pci-ioda.c >>+++ b/arch/powerpc/platforms/powernv/pci-ioda.c >>@@ -1946,6 +1946,7 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, >> >> /* Grab a 32-bit TCE table */ >> pe->dma32_seg = base; >>+ pe->dma32_segcount = segs; >> pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n", >> (base << 28), ((base + segs) << 28) - 1); >> >>@@ -2006,8 +2007,13 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, >> return; >> fail: >> /* XXX Failure: Try to fallback to 64-bit only ? */ >>- if (pe->dma32_seg >= 0) >>+ if (pe->dma32_seg >= 0) { >>+ bitmap_clear(phb->ioda.dma32_segmap, >>+ pe->dma32_seg, pe->dma32_segcount); >> pe->dma32_seg = -1; >>+ pe->dma32_segcount = 0; >>+ } >>+ >> if (tce_mem) >> __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs)); >> if (tbl) { >>@@ -2443,12 +2449,11 @@ static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, >> pnv_ioda_setup_bus_dma(pe, pe->pbus); >> } >> >>-static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, >>- struct pnv_ioda_pe *pe, >>- unsigned int base) >>+static void pnv_ioda1_setup_dma(struct pnv_phb *phb, >>+ struct pnv_ioda_pe *pe) >> { >> struct pci_controller *hose = phb->hose; >>- unsigned int dma_weight, segs; >>+ unsigned int dma_weight, base, segs; >> >> /* Calculate the PHB's DMA weight */ >> dma_weight = pnv_ioda_phb_dma_weight(phb); >>@@ -2461,11 +2466,28 @@ static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, >> else >> segs = (pe->dma32_weight * >> phb->ioda.dma32_segcount) / dma_weight; >>+ >>+ /* >>+ * Allocate DMA32 segments. We might not have enough >>+ * resources available. However we expect at least one >>+ * to be available. >>+ */ >>+ do { >>+ base = bitmap_find_next_zero_area(phb->ioda.dma32_segmap, >>+ phb->ioda.dma32_segcount, >>+ 0, segs, 0); >>+ if (base < phb->ioda.dma32_segcount) { >>+ bitmap_set(phb->ioda.dma32_segmap, base, segs); >>+ break; >>+ } >>+ } while (--segs); > > >If segs==0 before entering the loop, the loop will execute 0xfffffffe times. >Make it for(;segs;--segs){ }. > "segs" is always equal to or bigger than 1 when entering the loop. >>+ >>+ if (WARN_ON(!segs)) >>+ return; >>+ >> pe_info(pe, "DMA weight %d, assigned %d segments\n", >> pe->dma32_weight, segs); >> pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs); >>- >>- return segs; >> } >> >> #ifdef CONFIG_PCI_MSI >>@@ -2933,20 +2955,18 @@ static void pnv_pci_ioda_setup_DMA(void) >> struct pci_controller *hose, *tmp; >> struct pnv_phb *phb; >> struct pnv_ioda_pe *pe; >>- unsigned int base; >> >> list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { >> phb = hose->private_data; >> pnv_pci_ioda_setup_opal_tce_kill(phb); >> >>- base = 0; >> list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) { >> if (!pe->dma32_weight) >> continue; >> >> switch (phb->type) { >> case PNV_PHB_IODA1: >>- base += pnv_ioda1_setup_dma(phb, pe, base); >>+ pnv_ioda1_setup_dma(phb, pe); >> break; >> case PNV_PHB_IODA2: >> pnv_pci_ioda2_setup_dma_pe(phb, pe); >>diff --git a/arch/powerpc/platforms/powernv/pci.h b/arch/powerpc/platforms/powernv/pci.h >>index 574fe43..1dc9578 100644 >>--- a/arch/powerpc/platforms/powernv/pci.h >>+++ b/arch/powerpc/platforms/powernv/pci.h >>@@ -65,6 +65,7 @@ struct pnv_ioda_pe { >> >> /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */ >> int dma32_seg; >>+ int dma32_segcount; >> struct iommu_table_group table_group; >> >> /* 64-bit TCE bypass region */ >>@@ -153,10 +154,11 @@ struct pnv_phb { >> unsigned int io_segsize; >> unsigned int io_pci_base; >> >>- /* IO, M32, M64 segment maps */ >>+ /* IO, M32, M64, DMA32 segment maps */ >> unsigned long io_segmap[8]; >> unsigned long m32_segmap[8]; >> unsigned long m64_segmap[8]; >>+ unsigned long dma32_segmap[8]; >> >> /* PE allocation */ >> struct mutex pe_alloc_mutex; >> Thanks, Gavin
diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c index cd22002..57ba8fd 100644 --- a/arch/powerpc/platforms/powernv/pci-ioda.c +++ b/arch/powerpc/platforms/powernv/pci-ioda.c @@ -1946,6 +1946,7 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, /* Grab a 32-bit TCE table */ pe->dma32_seg = base; + pe->dma32_segcount = segs; pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n", (base << 28), ((base + segs) << 28) - 1); @@ -2006,8 +2007,13 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, return; fail: /* XXX Failure: Try to fallback to 64-bit only ? */ - if (pe->dma32_seg >= 0) + if (pe->dma32_seg >= 0) { + bitmap_clear(phb->ioda.dma32_segmap, + pe->dma32_seg, pe->dma32_segcount); pe->dma32_seg = -1; + pe->dma32_segcount = 0; + } + if (tce_mem) __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs)); if (tbl) { @@ -2443,12 +2449,11 @@ static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, pnv_ioda_setup_bus_dma(pe, pe->pbus); } -static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, - struct pnv_ioda_pe *pe, - unsigned int base) +static void pnv_ioda1_setup_dma(struct pnv_phb *phb, + struct pnv_ioda_pe *pe) { struct pci_controller *hose = phb->hose; - unsigned int dma_weight, segs; + unsigned int dma_weight, base, segs; /* Calculate the PHB's DMA weight */ dma_weight = pnv_ioda_phb_dma_weight(phb); @@ -2461,11 +2466,28 @@ static unsigned int pnv_ioda1_setup_dma(struct pnv_phb *phb, else segs = (pe->dma32_weight * phb->ioda.dma32_segcount) / dma_weight; + + /* + * Allocate DMA32 segments. We might not have enough + * resources available. However we expect at least one + * to be available. + */ + do { + base = bitmap_find_next_zero_area(phb->ioda.dma32_segmap, + phb->ioda.dma32_segcount, + 0, segs, 0); + if (base < phb->ioda.dma32_segcount) { + bitmap_set(phb->ioda.dma32_segmap, base, segs); + break; + } + } while (--segs); + + if (WARN_ON(!segs)) + return; + pe_info(pe, "DMA weight %d, assigned %d segments\n", pe->dma32_weight, segs); pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs); - - return segs; } #ifdef CONFIG_PCI_MSI @@ -2933,20 +2955,18 @@ static void pnv_pci_ioda_setup_DMA(void) struct pci_controller *hose, *tmp; struct pnv_phb *phb; struct pnv_ioda_pe *pe; - unsigned int base; list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { phb = hose->private_data; pnv_pci_ioda_setup_opal_tce_kill(phb); - base = 0; list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) { if (!pe->dma32_weight) continue; switch (phb->type) { case PNV_PHB_IODA1: - base += pnv_ioda1_setup_dma(phb, pe, base); + pnv_ioda1_setup_dma(phb, pe); break; case PNV_PHB_IODA2: pnv_pci_ioda2_setup_dma_pe(phb, pe); diff --git a/arch/powerpc/platforms/powernv/pci.h b/arch/powerpc/platforms/powernv/pci.h index 574fe43..1dc9578 100644 --- a/arch/powerpc/platforms/powernv/pci.h +++ b/arch/powerpc/platforms/powernv/pci.h @@ -65,6 +65,7 @@ struct pnv_ioda_pe { /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */ int dma32_seg; + int dma32_segcount; struct iommu_table_group table_group; /* 64-bit TCE bypass region */ @@ -153,10 +154,11 @@ struct pnv_phb { unsigned int io_segsize; unsigned int io_pci_base; - /* IO, M32, M64 segment maps */ + /* IO, M32, M64, DMA32 segment maps */ unsigned long io_segmap[8]; unsigned long m32_segmap[8]; unsigned long m64_segmap[8]; + unsigned long dma32_segmap[8]; /* PE allocation */ struct mutex pe_alloc_mutex;
On P7IOC, the whole DMA32 space is divided evenly to 256MB segments. Each PE can consume one or multiple DMA32 segments. Current code doesn't trace the available DMA32 segments and those consumed by one particular PE. It's conflicting with PCI hotplug. The patch introduces one bitmap to PHB to trace the available DMA32 segments for allocation, more fields to "struct pnv_ioda_pe" to trace the consumed DMA32 segments by the PE, which is going to be released when the PE is destroyed at PCI unplugging time. Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> --- arch/powerpc/platforms/powernv/pci-ioda.c | 40 +++++++++++++++++++++++-------- arch/powerpc/platforms/powernv/pci.h | 4 +++- 2 files changed, 33 insertions(+), 11 deletions(-)