From 0cf89dcdbc53f2b43e4ce7419b6ff47f4309c2eb Mon Sep 17 00:00:00 2001 From: Hannes Hering Date: Mon, 22 Jun 2009 22:18:51 -0700 Subject: [PATCH 1/8] IB/ehca: Tolerate dynamic memory operations before driver load Implement toleration of dynamic memory operations and 16 GB gigantic pages, where "toleration" means that the driver can cope with dynamic memory operations that happen before the driver is loaded. While the ehca driver is loaded, dynamic memory operations are still prohibited by returning NOTIFY_BAD from the memory notifier. On module load the driver walks through available system memory, checks for available memory ranges and then registers the kernel internal memory region accordingly. The translation of address ranges is implemented via a 3-level busmap. Signed-off-by: Hannes Hering Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ehca/ehca_main.c | 18 +- drivers/infiniband/hw/ehca/ehca_mrmw.c | 508 ++++++++++++++++++++++++- drivers/infiniband/hw/ehca/ehca_mrmw.h | 13 +- 3 files changed, 522 insertions(+), 17 deletions(-) diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c index ce4e6eff4792..14a18b7be245 100644 --- a/drivers/infiniband/hw/ehca/ehca_main.c +++ b/drivers/infiniband/hw/ehca/ehca_main.c @@ -506,6 +506,7 @@ static int ehca_init_device(struct ehca_shca *shca) shca->ib_device.detach_mcast = ehca_detach_mcast; shca->ib_device.process_mad = ehca_process_mad; shca->ib_device.mmap = ehca_mmap; + shca->ib_device.dma_ops = &ehca_dma_mapping_ops; if (EHCA_BMASK_GET(HCA_CAP_SRQ, shca->hca_cap)) { shca->ib_device.uverbs_cmd_mask |= @@ -1028,17 +1029,23 @@ static int __init ehca_module_init(void) goto module_init1; } + ret = ehca_create_busmap(); + if (ret) { + ehca_gen_err("Cannot create busmap."); + goto module_init2; + } + ret = ibmebus_register_driver(&ehca_driver); if (ret) { ehca_gen_err("Cannot register eHCA device driver"); ret = -EINVAL; - goto module_init2; + goto module_init3; } ret = register_memory_notifier(&ehca_mem_nb); if (ret) { ehca_gen_err("Failed registering memory add/remove notifier"); - goto module_init3; + goto module_init4; } if (ehca_poll_all_eqs != 1) { @@ -1053,9 +1060,12 @@ static int __init ehca_module_init(void) return 0; -module_init3: +module_init4: ibmebus_unregister_driver(&ehca_driver); +module_init3: + ehca_destroy_busmap(); + module_init2: ehca_destroy_slab_caches(); @@ -1073,6 +1083,8 @@ static void __exit ehca_module_exit(void) unregister_memory_notifier(&ehca_mem_nb); + ehca_destroy_busmap(); + ehca_destroy_slab_caches(); ehca_destroy_comp_pool(); diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c index 72f83f7df614..7663a2a9f130 100644 --- a/drivers/infiniband/hw/ehca/ehca_mrmw.c +++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c @@ -53,6 +53,38 @@ /* max number of rpages (per hcall register_rpages) */ #define MAX_RPAGES 512 +/* DMEM toleration management */ +#define EHCA_SECTSHIFT SECTION_SIZE_BITS +#define EHCA_SECTSIZE (1UL << EHCA_SECTSHIFT) +#define EHCA_HUGEPAGESHIFT 34 +#define EHCA_HUGEPAGE_SIZE (1UL << EHCA_HUGEPAGESHIFT) +#define EHCA_HUGEPAGE_PFN_MASK ((EHCA_HUGEPAGE_SIZE - 1) >> PAGE_SHIFT) +#define EHCA_INVAL_ADDR 0xFFFFFFFFFFFFFFFFULL +#define EHCA_DIR_INDEX_SHIFT 13 /* 8k Entries in 64k block */ +#define EHCA_TOP_INDEX_SHIFT (EHCA_DIR_INDEX_SHIFT * 2) +#define EHCA_MAP_ENTRIES (1 << EHCA_DIR_INDEX_SHIFT) +#define EHCA_TOP_MAP_SIZE (0x10000) /* currently fixed map size */ +#define EHCA_DIR_MAP_SIZE (0x10000) +#define EHCA_ENT_MAP_SIZE (0x10000) +#define EHCA_INDEX_MASK (EHCA_MAP_ENTRIES - 1) + +static unsigned long ehca_mr_len; + +/* + * Memory map data structures + */ +struct ehca_dir_bmap { + u64 ent[EHCA_MAP_ENTRIES]; +}; +struct ehca_top_bmap { + struct ehca_dir_bmap *dir[EHCA_MAP_ENTRIES]; +}; +struct ehca_bmap { + struct ehca_top_bmap *top[EHCA_MAP_ENTRIES]; +}; + +static struct ehca_bmap *ehca_bmap; + static struct kmem_cache *mr_cache; static struct kmem_cache *mw_cache; @@ -68,6 +100,8 @@ enum ehca_mr_pgsize { #define EHCA_MR_PGSHIFT1M 20 #define EHCA_MR_PGSHIFT16M 24 +static u64 ehca_map_vaddr(void *caddr); + static u32 ehca_encode_hwpage_size(u32 pgsize) { int log = ilog2(pgsize); @@ -135,7 +169,8 @@ struct ib_mr *ehca_get_dma_mr(struct ib_pd *pd, int mr_access_flags) goto get_dma_mr_exit0; } - ret = ehca_reg_maxmr(shca, e_maxmr, (u64 *)KERNELBASE, + ret = ehca_reg_maxmr(shca, e_maxmr, + (void *)ehca_map_vaddr((void *)KERNELBASE), mr_access_flags, e_pd, &e_maxmr->ib.ib_mr.lkey, &e_maxmr->ib.ib_mr.rkey); @@ -251,7 +286,7 @@ struct ib_mr *ehca_reg_phys_mr(struct ib_pd *pd, ret = ehca_reg_mr(shca, e_mr, iova_start, size, mr_access_flags, e_pd, &pginfo, &e_mr->ib.ib_mr.lkey, - &e_mr->ib.ib_mr.rkey); + &e_mr->ib.ib_mr.rkey, EHCA_REG_MR); if (ret) { ib_mr = ERR_PTR(ret); goto reg_phys_mr_exit1; @@ -370,7 +405,7 @@ reg_user_mr_fallback: ret = ehca_reg_mr(shca, e_mr, (u64 *)virt, length, mr_access_flags, e_pd, &pginfo, &e_mr->ib.ib_mr.lkey, - &e_mr->ib.ib_mr.rkey); + &e_mr->ib.ib_mr.rkey, EHCA_REG_MR); if (ret == -EINVAL && pginfo.hwpage_size > PAGE_SIZE) { ehca_warn(pd->device, "failed to register mr " "with hwpage_size=%llx", hwpage_size); @@ -794,7 +829,7 @@ struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd, ret = ehca_reg_mr(shca, e_fmr, NULL, fmr_attr->max_pages * (1 << fmr_attr->page_shift), mr_access_flags, e_pd, &pginfo, - &tmp_lkey, &tmp_rkey); + &tmp_lkey, &tmp_rkey, EHCA_REG_MR); if (ret) { ib_fmr = ERR_PTR(ret); goto alloc_fmr_exit1; @@ -983,6 +1018,10 @@ free_fmr_exit0: /*----------------------------------------------------------------------*/ +static int ehca_reg_bmap_mr_rpages(struct ehca_shca *shca, + struct ehca_mr *e_mr, + struct ehca_mr_pginfo *pginfo); + int ehca_reg_mr(struct ehca_shca *shca, struct ehca_mr *e_mr, u64 *iova_start, @@ -991,7 +1030,8 @@ int ehca_reg_mr(struct ehca_shca *shca, struct ehca_pd *e_pd, struct ehca_mr_pginfo *pginfo, u32 *lkey, /*OUT*/ - u32 *rkey) /*OUT*/ + u32 *rkey, /*OUT*/ + enum ehca_reg_type reg_type) { int ret; u64 h_ret; @@ -1015,7 +1055,13 @@ int ehca_reg_mr(struct ehca_shca *shca, e_mr->ipz_mr_handle = hipzout.handle; - ret = ehca_reg_mr_rpages(shca, e_mr, pginfo); + if (reg_type == EHCA_REG_BUSMAP_MR) + ret = ehca_reg_bmap_mr_rpages(shca, e_mr, pginfo); + else if (reg_type == EHCA_REG_MR) + ret = ehca_reg_mr_rpages(shca, e_mr, pginfo); + else + ret = -EINVAL; + if (ret) goto ehca_reg_mr_exit1; @@ -1316,7 +1362,7 @@ int ehca_rereg_mr(struct ehca_shca *shca, e_mr->fmr_map_cnt = save_mr.fmr_map_cnt; ret = ehca_reg_mr(shca, e_mr, iova_start, size, acl, - e_pd, pginfo, lkey, rkey); + e_pd, pginfo, lkey, rkey, EHCA_REG_MR); if (ret) { u32 offset = (u64)(&e_mr->flags) - (u64)e_mr; memcpy(&e_mr->flags, &(save_mr.flags), @@ -1409,7 +1455,7 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca, ret = ehca_reg_mr(shca, e_fmr, NULL, (e_fmr->fmr_max_pages * e_fmr->fmr_page_size), e_fmr->acl, e_pd, &pginfo, &tmp_lkey, - &tmp_rkey); + &tmp_rkey, EHCA_REG_MR); if (ret) { u32 offset = (u64)(&e_fmr->flags) - (u64)e_fmr; memcpy(&e_fmr->flags, &(save_mr.flags), @@ -1478,6 +1524,90 @@ ehca_reg_smr_exit0: } /* end ehca_reg_smr() */ /*----------------------------------------------------------------------*/ +static inline void *ehca_calc_sectbase(int top, int dir, int idx) +{ + unsigned long ret = idx; + ret |= dir << EHCA_DIR_INDEX_SHIFT; + ret |= top << EHCA_TOP_INDEX_SHIFT; + return abs_to_virt(ret << SECTION_SIZE_BITS); +} + +#define ehca_bmap_valid(entry) \ + ((u64)entry != (u64)EHCA_INVAL_ADDR) + +static u64 ehca_reg_mr_section(int top, int dir, int idx, u64 *kpage, + struct ehca_shca *shca, struct ehca_mr *mr, + struct ehca_mr_pginfo *pginfo) +{ + u64 h_ret = 0; + unsigned long page = 0; + u64 rpage = virt_to_abs(kpage); + int page_count; + + void *sectbase = ehca_calc_sectbase(top, dir, idx); + if ((unsigned long)sectbase & (pginfo->hwpage_size - 1)) { + ehca_err(&shca->ib_device, "reg_mr_section will probably fail:" + "hwpage_size does not fit to " + "section start address"); + } + page_count = EHCA_SECTSIZE / pginfo->hwpage_size; + + while (page < page_count) { + u64 rnum; + for (rnum = 0; (rnum < MAX_RPAGES) && (page < page_count); + rnum++) { + void *pg = sectbase + ((page++) * pginfo->hwpage_size); + kpage[rnum] = virt_to_abs(pg); + } + + h_ret = hipz_h_register_rpage_mr(shca->ipz_hca_handle, mr, + ehca_encode_hwpage_size(pginfo->hwpage_size), + 0, rpage, rnum); + + if ((h_ret != H_SUCCESS) && (h_ret != H_PAGE_REGISTERED)) { + ehca_err(&shca->ib_device, "register_rpage_mr failed"); + return h_ret; + } + } + return h_ret; +} + +static u64 ehca_reg_mr_sections(int top, int dir, u64 *kpage, + struct ehca_shca *shca, struct ehca_mr *mr, + struct ehca_mr_pginfo *pginfo) +{ + u64 hret = H_SUCCESS; + int idx; + + for (idx = 0; idx < EHCA_MAP_ENTRIES; idx++) { + if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]->ent[idx])) + continue; + + hret = ehca_reg_mr_section(top, dir, idx, kpage, shca, mr, + pginfo); + if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED)) + return hret; + } + return hret; +} + +static u64 ehca_reg_mr_dir_sections(int top, u64 *kpage, struct ehca_shca *shca, + struct ehca_mr *mr, + struct ehca_mr_pginfo *pginfo) +{ + u64 hret = H_SUCCESS; + int dir; + + for (dir = 0; dir < EHCA_MAP_ENTRIES; dir++) { + if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir])) + continue; + + hret = ehca_reg_mr_sections(top, dir, kpage, shca, mr, pginfo); + if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED)) + return hret; + } + return hret; +} /* register internal max-MR to internal SHCA */ int ehca_reg_internal_maxmr( @@ -1495,6 +1625,11 @@ int ehca_reg_internal_maxmr( u32 num_hwpages; u64 hw_pgsize; + if (!ehca_bmap) { + ret = -EFAULT; + goto ehca_reg_internal_maxmr_exit0; + } + e_mr = ehca_mr_new(); if (!e_mr) { ehca_err(&shca->ib_device, "out of memory"); @@ -1504,8 +1639,8 @@ int ehca_reg_internal_maxmr( e_mr->flags |= EHCA_MR_FLAG_MAXMR; /* register internal max-MR on HCA */ - size_maxmr = (u64)high_memory - PAGE_OFFSET; - iova_start = (u64 *)KERNELBASE; + size_maxmr = ehca_mr_len; + iova_start = (u64 *)ehca_map_vaddr((void *)KERNELBASE); ib_pbuf.addr = 0; ib_pbuf.size = size_maxmr; num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size_maxmr, @@ -1524,7 +1659,7 @@ int ehca_reg_internal_maxmr( ret = ehca_reg_mr(shca, e_mr, iova_start, size_maxmr, 0, e_pd, &pginfo, &e_mr->ib.ib_mr.lkey, - &e_mr->ib.ib_mr.rkey); + &e_mr->ib.ib_mr.rkey, EHCA_REG_BUSMAP_MR); if (ret) { ehca_err(&shca->ib_device, "reg of internal max MR failed, " "e_mr=%p iova_start=%p size_maxmr=%llx num_kpages=%x " @@ -2077,8 +2212,8 @@ int ehca_mr_is_maxmr(u64 size, u64 *iova_start) { /* a MR is treated as max-MR only if it fits following: */ - if ((size == ((u64)high_memory - PAGE_OFFSET)) && - (iova_start == (void *)KERNELBASE)) { + if ((size == ehca_mr_len) && + (iova_start == (void *)ehca_map_vaddr((void *)KERNELBASE))) { ehca_gen_dbg("this is a max-MR"); return 1; } else @@ -2184,3 +2319,350 @@ void ehca_cleanup_mrmw_cache(void) if (mw_cache) kmem_cache_destroy(mw_cache); } + +static inline int ehca_init_top_bmap(struct ehca_top_bmap *ehca_top_bmap, + int dir) +{ + if (!ehca_bmap_valid(ehca_top_bmap->dir[dir])) { + ehca_top_bmap->dir[dir] = + kmalloc(sizeof(struct ehca_dir_bmap), GFP_KERNEL); + if (!ehca_top_bmap->dir[dir]) + return -ENOMEM; + /* Set map block to 0xFF according to EHCA_INVAL_ADDR */ + memset(ehca_top_bmap->dir[dir], 0xFF, EHCA_ENT_MAP_SIZE); + } + return 0; +} + +static inline int ehca_init_bmap(struct ehca_bmap *ehca_bmap, int top, int dir) +{ + if (!ehca_bmap_valid(ehca_bmap->top[top])) { + ehca_bmap->top[top] = + kmalloc(sizeof(struct ehca_top_bmap), GFP_KERNEL); + if (!ehca_bmap->top[top]) + return -ENOMEM; + /* Set map block to 0xFF according to EHCA_INVAL_ADDR */ + memset(ehca_bmap->top[top], 0xFF, EHCA_DIR_MAP_SIZE); + } + return ehca_init_top_bmap(ehca_bmap->top[top], dir); +} + +static inline int ehca_calc_index(unsigned long i, unsigned long s) +{ + return (i >> s) & EHCA_INDEX_MASK; +} + +void ehca_destroy_busmap(void) +{ + int top, dir; + + if (!ehca_bmap) + return; + + for (top = 0; top < EHCA_MAP_ENTRIES; top++) { + if (!ehca_bmap_valid(ehca_bmap->top[top])) + continue; + for (dir = 0; dir < EHCA_MAP_ENTRIES; dir++) { + if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir])) + continue; + + kfree(ehca_bmap->top[top]->dir[dir]); + } + + kfree(ehca_bmap->top[top]); + } + + kfree(ehca_bmap); + ehca_bmap = NULL; +} + +static int ehca_update_busmap(unsigned long pfn, unsigned long nr_pages) +{ + unsigned long i, start_section, end_section; + int top, dir, idx; + + if (!nr_pages) + return 0; + + if (!ehca_bmap) { + ehca_bmap = kmalloc(sizeof(struct ehca_bmap), GFP_KERNEL); + if (!ehca_bmap) + return -ENOMEM; + /* Set map block to 0xFF according to EHCA_INVAL_ADDR */ + memset(ehca_bmap, 0xFF, EHCA_TOP_MAP_SIZE); + } + + start_section = phys_to_abs(pfn * PAGE_SIZE) / EHCA_SECTSIZE; + end_section = phys_to_abs((pfn + nr_pages) * PAGE_SIZE) / EHCA_SECTSIZE; + for (i = start_section; i < end_section; i++) { + int ret; + top = ehca_calc_index(i, EHCA_TOP_INDEX_SHIFT); + dir = ehca_calc_index(i, EHCA_DIR_INDEX_SHIFT); + idx = i & EHCA_INDEX_MASK; + + ret = ehca_init_bmap(ehca_bmap, top, dir); + if (ret) { + ehca_destroy_busmap(); + return ret; + } + ehca_bmap->top[top]->dir[dir]->ent[idx] = ehca_mr_len; + ehca_mr_len += EHCA_SECTSIZE; + } + return 0; +} + +static int ehca_is_hugepage(unsigned long pfn) +{ + int page_order; + + if (pfn & EHCA_HUGEPAGE_PFN_MASK) + return 0; + + page_order = compound_order(pfn_to_page(pfn)); + if (page_order + PAGE_SHIFT != EHCA_HUGEPAGESHIFT) + return 0; + + return 1; +} + +static int ehca_create_busmap_callback(unsigned long initial_pfn, + unsigned long total_nr_pages, void *arg) +{ + int ret; + unsigned long pfn, start_pfn, end_pfn, nr_pages; + + if ((total_nr_pages * PAGE_SIZE) < EHCA_HUGEPAGE_SIZE) + return ehca_update_busmap(initial_pfn, total_nr_pages); + + /* Given chunk is >= 16GB -> check for hugepages */ + start_pfn = initial_pfn; + end_pfn = initial_pfn + total_nr_pages; + pfn = start_pfn; + + while (pfn < end_pfn) { + if (ehca_is_hugepage(pfn)) { + /* Add mem found in front of the hugepage */ + nr_pages = pfn - start_pfn; + ret = ehca_update_busmap(start_pfn, nr_pages); + if (ret) + return ret; + /* Skip the hugepage */ + pfn += (EHCA_HUGEPAGE_SIZE / PAGE_SIZE); + start_pfn = pfn; + } else + pfn += (EHCA_SECTSIZE / PAGE_SIZE); + } + + /* Add mem found behind the hugepage(s) */ + nr_pages = pfn - start_pfn; + return ehca_update_busmap(start_pfn, nr_pages); +} + +int ehca_create_busmap(void) +{ + int ret; + + ehca_mr_len = 0; + ret = walk_memory_resource(0, 1ULL << MAX_PHYSMEM_BITS, NULL, + ehca_create_busmap_callback); + return ret; +} + +static int ehca_reg_bmap_mr_rpages(struct ehca_shca *shca, + struct ehca_mr *e_mr, + struct ehca_mr_pginfo *pginfo) +{ + int top; + u64 hret, *kpage; + + kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL); + if (!kpage) { + ehca_err(&shca->ib_device, "kpage alloc failed"); + return -ENOMEM; + } + for (top = 0; top < EHCA_MAP_ENTRIES; top++) { + if (!ehca_bmap_valid(ehca_bmap->top[top])) + continue; + hret = ehca_reg_mr_dir_sections(top, kpage, shca, e_mr, pginfo); + if ((hret != H_PAGE_REGISTERED) && (hret != H_SUCCESS)) + break; + } + + ehca_free_fw_ctrlblock(kpage); + + if (hret == H_SUCCESS) + return 0; /* Everything is fine */ + else { + ehca_err(&shca->ib_device, "ehca_reg_bmap_mr_rpages failed, " + "h_ret=%lli e_mr=%p top=%x lkey=%x " + "hca_hndl=%llx mr_hndl=%llx", hret, e_mr, top, + e_mr->ib.ib_mr.lkey, + shca->ipz_hca_handle.handle, + e_mr->ipz_mr_handle.handle); + return ehca2ib_return_code(hret); + } +} + +static u64 ehca_map_vaddr(void *caddr) +{ + int top, dir, idx; + unsigned long abs_addr, offset; + u64 entry; + + if (!ehca_bmap) + return EHCA_INVAL_ADDR; + + abs_addr = virt_to_abs(caddr); + top = ehca_calc_index(abs_addr, EHCA_TOP_INDEX_SHIFT + EHCA_SECTSHIFT); + if (!ehca_bmap_valid(ehca_bmap->top[top])) + return EHCA_INVAL_ADDR; + + dir = ehca_calc_index(abs_addr, EHCA_DIR_INDEX_SHIFT + EHCA_SECTSHIFT); + if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir])) + return EHCA_INVAL_ADDR; + + idx = ehca_calc_index(abs_addr, EHCA_SECTSHIFT); + + entry = ehca_bmap->top[top]->dir[dir]->ent[idx]; + if (ehca_bmap_valid(entry)) { + offset = (unsigned long)caddr & (EHCA_SECTSIZE - 1); + return entry | offset; + } else + return EHCA_INVAL_ADDR; +} + +static int ehca_dma_mapping_error(struct ib_device *dev, u64 dma_addr) +{ + return dma_addr == EHCA_INVAL_ADDR; +} + +static u64 ehca_dma_map_single(struct ib_device *dev, void *cpu_addr, + size_t size, enum dma_data_direction direction) +{ + if (cpu_addr) + return ehca_map_vaddr(cpu_addr); + else + return EHCA_INVAL_ADDR; +} + +static void ehca_dma_unmap_single(struct ib_device *dev, u64 addr, size_t size, + enum dma_data_direction direction) +{ + /* This is only a stub; nothing to be done here */ +} + +static u64 ehca_dma_map_page(struct ib_device *dev, struct page *page, + unsigned long offset, size_t size, + enum dma_data_direction direction) +{ + u64 addr; + + if (offset + size > PAGE_SIZE) + return EHCA_INVAL_ADDR; + + addr = ehca_map_vaddr(page_address(page)); + if (!ehca_dma_mapping_error(dev, addr)) + addr += offset; + + return addr; +} + +static void ehca_dma_unmap_page(struct ib_device *dev, u64 addr, size_t size, + enum dma_data_direction direction) +{ + /* This is only a stub; nothing to be done here */ +} + +static int ehca_dma_map_sg(struct ib_device *dev, struct scatterlist *sgl, + int nents, enum dma_data_direction direction) +{ + struct scatterlist *sg; + int i; + + for_each_sg(sgl, sg, nents, i) { + u64 addr; + addr = ehca_map_vaddr(sg_virt(sg)); + if (ehca_dma_mapping_error(dev, addr)) + return 0; + + sg->dma_address = addr; + sg->dma_length = sg->length; + } + return nents; +} + +static void ehca_dma_unmap_sg(struct ib_device *dev, struct scatterlist *sg, + int nents, enum dma_data_direction direction) +{ + /* This is only a stub; nothing to be done here */ +} + +static u64 ehca_dma_address(struct ib_device *dev, struct scatterlist *sg) +{ + return sg->dma_address; +} + +static unsigned int ehca_dma_len(struct ib_device *dev, struct scatterlist *sg) +{ + return sg->length; +} + +static void ehca_dma_sync_single_for_cpu(struct ib_device *dev, u64 addr, + size_t size, + enum dma_data_direction dir) +{ + dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); +} + +static void ehca_dma_sync_single_for_device(struct ib_device *dev, u64 addr, + size_t size, + enum dma_data_direction dir) +{ + dma_sync_single_for_device(dev->dma_device, addr, size, dir); +} + +static void *ehca_dma_alloc_coherent(struct ib_device *dev, size_t size, + u64 *dma_handle, gfp_t flag) +{ + struct page *p; + void *addr = NULL; + u64 dma_addr; + + p = alloc_pages(flag, get_order(size)); + if (p) { + addr = page_address(p); + dma_addr = ehca_map_vaddr(addr); + if (ehca_dma_mapping_error(dev, dma_addr)) { + free_pages((unsigned long)addr, get_order(size)); + return NULL; + } + if (dma_handle) + *dma_handle = dma_addr; + return addr; + } + return NULL; +} + +static void ehca_dma_free_coherent(struct ib_device *dev, size_t size, + void *cpu_addr, u64 dma_handle) +{ + if (cpu_addr && size) + free_pages((unsigned long)cpu_addr, get_order(size)); +} + + +struct ib_dma_mapping_ops ehca_dma_mapping_ops = { + .mapping_error = ehca_dma_mapping_error, + .map_single = ehca_dma_map_single, + .unmap_single = ehca_dma_unmap_single, + .map_page = ehca_dma_map_page, + .unmap_page = ehca_dma_unmap_page, + .map_sg = ehca_dma_map_sg, + .unmap_sg = ehca_dma_unmap_sg, + .dma_address = ehca_dma_address, + .dma_len = ehca_dma_len, + .sync_single_for_cpu = ehca_dma_sync_single_for_cpu, + .sync_single_for_device = ehca_dma_sync_single_for_device, + .alloc_coherent = ehca_dma_alloc_coherent, + .free_coherent = ehca_dma_free_coherent, +}; diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.h b/drivers/infiniband/hw/ehca/ehca_mrmw.h index bc8f4e31c123..50d8b51306dd 100644 --- a/drivers/infiniband/hw/ehca/ehca_mrmw.h +++ b/drivers/infiniband/hw/ehca/ehca_mrmw.h @@ -42,6 +42,11 @@ #ifndef _EHCA_MRMW_H_ #define _EHCA_MRMW_H_ +enum ehca_reg_type { + EHCA_REG_MR, + EHCA_REG_BUSMAP_MR +}; + int ehca_reg_mr(struct ehca_shca *shca, struct ehca_mr *e_mr, u64 *iova_start, @@ -50,7 +55,8 @@ int ehca_reg_mr(struct ehca_shca *shca, struct ehca_pd *e_pd, struct ehca_mr_pginfo *pginfo, u32 *lkey, - u32 *rkey); + u32 *rkey, + enum ehca_reg_type reg_type); int ehca_reg_mr_rpages(struct ehca_shca *shca, struct ehca_mr *e_mr, @@ -118,4 +124,9 @@ void ehca_mrmw_reverse_map_acl(const u32 *hipz_acl, void ehca_mr_deletenew(struct ehca_mr *mr); +int ehca_create_busmap(void); + +void ehca_destroy_busmap(void); + +extern struct ib_dma_mapping_ops ehca_dma_mapping_ops; #endif /*_EHCA_MRMW_H_*/ From af04662b4d80de5797a595bc9855d09ef4fe55cc Mon Sep 17 00:00:00 2001 From: Roel Kluin Date: Mon, 22 Jun 2009 22:23:48 -0700 Subject: [PATCH 2/8] IB/ehca: Ensure that guid_entry index is not negative This prevents the memcpy() of a guid_entries element using a negative index. Signed-off-by: Roel Kluin Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ehca/ehca_hca.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/ehca/ehca_hca.c b/drivers/infiniband/hw/ehca/ehca_hca.c index 9209c5332dfe..8b92f85d4dd0 100644 --- a/drivers/infiniband/hw/ehca/ehca_hca.c +++ b/drivers/infiniband/hw/ehca/ehca_hca.c @@ -319,7 +319,7 @@ int ehca_query_gid(struct ib_device *ibdev, u8 port, ib_device); struct hipz_query_port *rblock; - if (index > 255) { + if (index < 0 || index > 255) { ehca_err(&shca->ib_device, "Invalid index: %x.", index); return -EINVAL; } From 66388d67a0d7bf39735650de54e42064d1af8b62 Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 22 Jun 2009 22:52:30 -0700 Subject: [PATCH 3/8] RDMA/nes: Fix max_qp_init_rd_atom returned from query device In nes_query_device(), max_qp_init_rd_atom is incorrectly set to max_qp_wr. This was found when a test application had a dapl async event error. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_verbs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c index 64d5cfd8f380..21e0fd336cf7 100644 --- a/drivers/infiniband/hw/nes/nes_verbs.c +++ b/drivers/infiniband/hw/nes/nes_verbs.c @@ -654,7 +654,7 @@ static int nes_query_device(struct ib_device *ibdev, struct ib_device_attr *prop default: props->max_qp_rd_atom = 0; } - props->max_qp_init_rd_atom = props->max_qp_wr; + props->max_qp_init_rd_atom = props->max_qp_rd_atom; props->atomic_cap = IB_ATOMIC_NONE; props->max_map_per_fmr = 1; From 68237a0ff84503270373c39229be83e865ea08d4 Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 22 Jun 2009 22:53:28 -0700 Subject: [PATCH 4/8] RDMA/nes: Fix FIN state handling under error conditions During cluster testing, one QP was not closed, as FIN is not handled properly when its rexmit count expires or in some cases when RST is is received after sending FIN. The reason is that the cm_id does not get decremented under these conditions. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 11c7d6642014..114b802771ad 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -472,6 +472,7 @@ int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb, static void nes_retrans_expired(struct nes_cm_node *cm_node) { + struct iw_cm_id *cm_id = cm_node->cm_id; switch (cm_node->state) { case NES_CM_STATE_SYN_RCVD: case NES_CM_STATE_CLOSING: @@ -479,7 +480,9 @@ static void nes_retrans_expired(struct nes_cm_node *cm_node) break; case NES_CM_STATE_LAST_ACK: case NES_CM_STATE_FIN_WAIT1: - case NES_CM_STATE_MPAREJ_RCVD: + if (cm_node->cm_id) + cm_id->rem_ref(cm_id); + cm_node->state = NES_CM_STATE_CLOSED; send_reset(cm_node, NULL); break; default: @@ -1406,6 +1409,7 @@ static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, case NES_CM_STATE_CLOSED: drop_packet(skb); break; + case NES_CM_STATE_FIN_WAIT1: case NES_CM_STATE_LAST_ACK: cm_node->cm_id->rem_ref(cm_node->cm_id); case NES_CM_STATE_TIME_WAIT: @@ -1413,8 +1417,6 @@ static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, rem_ref_cm_node(cm_node->cm_core, cm_node); drop_packet(skb); break; - case NES_CM_STATE_FIN_WAIT1: - nes_debug(NES_DBG_CM, "Bad state %s[%u]\n", __func__, __LINE__); default: drop_packet(skb); break; From 99987bea474ceca8ec6fb05f81d7d188634cdffd Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Mon, 22 Jun 2009 23:04:13 -0700 Subject: [PATCH 5/8] IB/mthca: Replace dma_sync_single() use with proper functions dma_sync_single() is deprecated now, and the use in mthca is wrong: there should be a dma_sync_single_for_cpu() before touching the memory from the CPU, and a dma_sync_single_for_device() afterwards. Fix this, prompted by a kick in the pants from a patch from FUJITA Tomonori . Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mthca/mthca_mr.c | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/mthca/mthca_mr.c b/drivers/infiniband/hw/mthca/mthca_mr.c index d606edf10858..065b20899876 100644 --- a/drivers/infiniband/hw/mthca/mthca_mr.c +++ b/drivers/infiniband/hw/mthca/mthca_mr.c @@ -352,10 +352,14 @@ static void mthca_arbel_write_mtt_seg(struct mthca_dev *dev, BUG_ON(!mtts); + dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle, + list_len * sizeof (u64), DMA_TO_DEVICE); + for (i = 0; i < list_len; ++i) mtts[i] = cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT); - dma_sync_single(&dev->pdev->dev, dma_handle, list_len * sizeof (u64), DMA_TO_DEVICE); + dma_sync_single_for_device(&dev->pdev->dev, dma_handle, + list_len * sizeof (u64), DMA_TO_DEVICE); } int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt, @@ -803,12 +807,15 @@ int mthca_arbel_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, wmb(); + dma_sync_single_for_cpu(&dev->pdev->dev, fmr->mem.arbel.dma_handle, + list_len * sizeof(u64), DMA_TO_DEVICE); + for (i = 0; i < list_len; ++i) fmr->mem.arbel.mtts[i] = cpu_to_be64(page_list[i] | MTHCA_MTT_FLAG_PRESENT); - dma_sync_single(&dev->pdev->dev, fmr->mem.arbel.dma_handle, - list_len * sizeof(u64), DMA_TO_DEVICE); + dma_sync_single_for_device(&dev->pdev->dev, fmr->mem.arbel.dma_handle, + list_len * sizeof(u64), DMA_TO_DEVICE); fmr->mem.arbel.mpt->key = cpu_to_be32(key); fmr->mem.arbel.mpt->lkey = cpu_to_be32(key); From e727f5cde90d5a8b92a1ffa49c636a3790301469 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Mon, 22 Jun 2009 23:07:56 -0700 Subject: [PATCH 6/8] mlx4_core: Fix dma_sync_single_for_cpu() with matching for_device() calls Commit 5d23a1d2 ("net: replace dma_sync_single with dma_sync_single_for_cpu") replaced uses of the deprectated function dma_sync_single() with calls to dma_sync_single_for_cpu(). However, to be correct, the code should do a sync for_cpu() before touching the memory and for_device() after it's done. Signed-off-by: Roland Dreier --- drivers/net/mlx4/mr.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/drivers/net/mlx4/mr.c b/drivers/net/mlx4/mr.c index 5887e4764d22..f96948be0a44 100644 --- a/drivers/net/mlx4/mr.c +++ b/drivers/net/mlx4/mr.c @@ -399,11 +399,14 @@ static int mlx4_write_mtt_chunk(struct mlx4_dev *dev, struct mlx4_mtt *mtt, if (!mtts) return -ENOMEM; + dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle, + npages * sizeof (u64), DMA_TO_DEVICE); + for (i = 0; i < npages; ++i) mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT); - dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle, - npages * sizeof (u64), DMA_TO_DEVICE); + dma_sync_single_for_device(&dev->pdev->dev, dma_handle, + npages * sizeof (u64), DMA_TO_DEVICE); return 0; } @@ -547,11 +550,14 @@ int mlx4_map_phys_fmr(struct mlx4_dev *dev, struct mlx4_fmr *fmr, u64 *page_list /* Make sure MPT status is visible before writing MTT entries */ wmb(); + dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle, + npages * sizeof(u64), DMA_TO_DEVICE); + for (i = 0; i < npages; ++i) fmr->mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT); - dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle, - npages * sizeof(u64), DMA_TO_DEVICE); + dma_sync_single_for_device(&dev->pdev->dev, fmr->dma_handle, + npages * sizeof(u64), DMA_TO_DEVICE); fmr->mpt->key = cpu_to_be32(key); fmr->mpt->lkey = cpu_to_be32(key); From 1d4d6da535be97b710e87a33c4828c97c36eee21 Mon Sep 17 00:00:00 2001 From: Alexander Schmidt Date: Tue, 23 Jun 2009 10:30:04 -0700 Subject: [PATCH 7/8] IB/ehca: Bump version number Increment version number for DMEM toleration. Signed-off-by: Alexander Schmidt Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ehca/ehca_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c index 14a18b7be245..fab18a2c74a8 100644 --- a/drivers/infiniband/hw/ehca/ehca_main.c +++ b/drivers/infiniband/hw/ehca/ehca_main.c @@ -52,7 +52,7 @@ #include "ehca_tools.h" #include "hcp_if.h" -#define HCAD_VERSION "0027" +#define HCAD_VERSION "0028" MODULE_LICENSE("Dual BSD/GPL"); MODULE_AUTHOR("Christoph Raisch "); From 716abb1fdf3274ac81dc404f3659cc05d8cdf606 Mon Sep 17 00:00:00 2001 From: Peter Huewe Date: Tue, 23 Jun 2009 10:38:42 -0700 Subject: [PATCH 8/8] RDMA: Add __init/__exit macros to addr.c and cma.c Add __init and __exit annotations to the module_init/module_exit functions from drivers/infiniband/core/addr.c and cma.c. Signed-off-by: Peter Huewe Acked-by: Sean Hefty Signed-off-by: Roland Dreier --- drivers/infiniband/core/addr.c | 4 ++-- drivers/infiniband/core/cma.c | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c index ce511d8748ce..5be1bd4fc7ed 100644 --- a/drivers/infiniband/core/addr.c +++ b/drivers/infiniband/core/addr.c @@ -514,7 +514,7 @@ static struct notifier_block nb = { .notifier_call = netevent_callback }; -static int addr_init(void) +static int __init addr_init(void) { addr_wq = create_singlethread_workqueue("ib_addr"); if (!addr_wq) @@ -524,7 +524,7 @@ static int addr_init(void) return 0; } -static void addr_cleanup(void) +static void __exit addr_cleanup(void) { unregister_netevent_notifier(&nb); destroy_workqueue(addr_wq); diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 851de83ff455..075317884b53 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -2960,7 +2960,7 @@ static void cma_remove_one(struct ib_device *device) kfree(cma_dev); } -static int cma_init(void) +static int __init cma_init(void) { int ret, low, high, remaining; @@ -2990,7 +2990,7 @@ err: return ret; } -static void cma_cleanup(void) +static void __exit cma_cleanup(void) { ib_unregister_client(&cma_client); unregister_netdevice_notifier(&cma_nb);