* [PATCH 1/5] ehca: Supports large page MRs
@ 2007-07-20 14:01 Hoang-Nam Nguyen
2007-07-21 4:07 ` [ofa-general] " Roland Dreier
0 siblings, 1 reply; 3+ messages in thread
From: Hoang-Nam Nguyen @ 2007-07-20 14:01 UTC (permalink / raw)
To: Roland Dreier, linux-kernel, linuxppc-dev, openib-general
Cc: raisch, stefan.roscher, fenkes
From: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Date: Thu, 19 Jul 2007 20:48:04 +0200
Subject: [PATCH 1/5] IB/ehca: Support large page MRs
Add support for MR pages larger than 4K on eHCA2. This reduces firmware
memory consumption. If enabled via the mr_largepage module parameter, the MR
page size will be determined based on the MR length and the hardware
capabilities - if the MR is >= 16M, 16M pages are used, for example.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
---
drivers/infiniband/hw/ehca/ehca_classes.h | 9 +
drivers/infiniband/hw/ehca/ehca_main.c | 18 ++-
drivers/infiniband/hw/ehca/ehca_mrmw.c | 371 ++++++++++++++++++++++++-----
drivers/infiniband/hw/ehca/ehca_mrmw.h | 2 +-
drivers/infiniband/hw/ehca/hcp_if.c | 20 ++-
5 files changed, 357 insertions(+), 63 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_classes.h b/drivers/infiniband/hw/ehca/ehca_classes.h
index 043e4fb..63b8b9f 100644
--- a/drivers/infiniband/hw/ehca/ehca_classes.h
+++ b/drivers/infiniband/hw/ehca/ehca_classes.h
@@ -100,6 +100,11 @@ struct ehca_sport {
struct ehca_sma_attr saved_attr;
};
+#define HCA_CAP_MR_PGSIZE_4K 1
+#define HCA_CAP_MR_PGSIZE_64K 2
+#define HCA_CAP_MR_PGSIZE_1M 4
+#define HCA_CAP_MR_PGSIZE_16M 8
+
struct ehca_shca {
struct ib_device ib_device;
struct ibmebus_dev *ibmebus_dev;
@@ -115,6 +120,8 @@ struct ehca_shca {
struct h_galpas galpas;
struct mutex modify_mutex;
u64 hca_cap;
+ /* MR pgsize: bit 0-3 means 4K, 64K, 1M, 16M respectively */
+ u32 hca_cap_mr_pgsize;
int max_mtu;
};
@@ -206,6 +213,7 @@ struct ehca_mr {
enum ehca_mr_flag flags;
u32 num_kpages; /* number of kernel pages */
u32 num_hwpages; /* number of hw pages to form MR */
+ u64 hwpage_size; /* hw page size used for this MR */
int acl; /* ACL (stored here for usage in reregister) */
u64 *start; /* virtual start address (stored here for */
/* usage in reregister) */
@@ -240,6 +248,7 @@ struct ehca_mr_pginfo {
enum ehca_mr_pgi_type type;
u64 num_kpages;
u64 kpage_cnt;
+ u64 hwpage_size; /* hw page size used for this MR */
u64 num_hwpages; /* number of hw pages */
u64 hwpage_cnt; /* counter for hw pages */
u64 next_hwpage; /* next hw page in buffer/chunk/listelem */
diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c
index 36377c6..34661c3 100644
--- a/drivers/infiniband/hw/ehca/ehca_main.c
+++ b/drivers/infiniband/hw/ehca/ehca_main.c
@@ -63,6 +63,7 @@ int ehca_port_act_time = 30;
int ehca_poll_all_eqs = 1;
int ehca_static_rate = -1;
int ehca_scaling_code = 0;
+int ehca_mr_largepage = 0;
module_param_named(open_aqp1, ehca_open_aqp1, int, 0);
module_param_named(debug_level, ehca_debug_level, int, 0);
@@ -72,7 +73,8 @@ module_param_named(use_hp_mr, ehca_use_hp_mr, int, 0);
module_param_named(port_act_time, ehca_port_act_time, int, 0);
module_param_named(poll_all_eqs, ehca_poll_all_eqs, int, 0);
module_param_named(static_rate, ehca_static_rate, int, 0);
-module_param_named(scaling_code, ehca_scaling_code, int, 0);
+module_param_named(scaling_code, ehca_scaling_code, int, 0);
+module_param_named(mr_largepage, ehca_mr_largepage, int, 0);
MODULE_PARM_DESC(open_aqp1,
"AQP1 on startup (0: no (default), 1: yes)");
@@ -95,6 +97,9 @@ MODULE_PARM_DESC(static_rate,
"set permanent static rate (default: disabled)");
MODULE_PARM_DESC(scaling_code,
"set scaling code (0: disabled/default, 1: enabled)");
+MODULE_PARM_DESC(mr_largepage,
+ "use large page for MR (0: use PAGE_SIZE (default), "
+ "1: use large page depending on MR size");
DEFINE_RWLOCK(ehca_qp_idr_lock);
DEFINE_RWLOCK(ehca_cq_idr_lock);
@@ -295,6 +300,8 @@ int ehca_sense_attributes(struct ehca_shca *shca)
if (EHCA_BMASK_GET(hca_cap_descr[i].mask, shca->hca_cap))
ehca_gen_dbg(" %s", hca_cap_descr[i].descr);
+ shca->hca_cap_mr_pgsize = rblock->memory_page_size_supported;
+
port = (struct hipz_query_port *)rblock;
h_ret = hipz_h_query_port(shca->ipz_hca_handle, 1, port);
if (h_ret != H_SUCCESS) {
@@ -590,6 +597,14 @@ static ssize_t ehca_show_adapter_handle(struct device *dev,
}
static DEVICE_ATTR(adapter_handle, S_IRUGO, ehca_show_adapter_handle, NULL);
+static ssize_t ehca_show_mr_largepage(struct device *dev,
+ struct device_attribute *attr,
+ char *buf)
+{
+ return sprintf(buf, "%d\n", ehca_mr_largepage);
+}
+static DEVICE_ATTR(mr_largepage, S_IRUGO, ehca_show_mr_largepage, NULL);
+
static struct attribute *ehca_dev_attrs[] = {
&dev_attr_adapter_handle.attr,
&dev_attr_num_ports.attr,
@@ -606,6 +621,7 @@ static struct attribute *ehca_dev_attrs[] = {
&dev_attr_cur_mw.attr,
&dev_attr_max_pd.attr,
&dev_attr_max_ah.attr,
+ &dev_attr_mr_largepage.attr,
NULL
};
diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c
index 6262c54..ba28783 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.c
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c
@@ -5,6 +5,7 @@
*
* Authors: Dietmar Decker <ddecker@de.ibm.com>
* Christoph Raisch <raisch@de.ibm.com>
+ * Hoang-Nam Nguyen <hnguyen@de.ibm.com>
*
* Copyright (c) 2005 IBM Corporation
*
@@ -56,6 +57,37 @@
static struct kmem_cache *mr_cache;
static struct kmem_cache *mw_cache;
+enum ehca_mr_pgsize {
+ EHCA_MR_PGSIZE4K = 0x1000L,
+ EHCA_MR_PGSIZE64K = 0x10000L,
+ EHCA_MR_PGSIZE1M = 0x100000L,
+ EHCA_MR_PGSIZE16M = 0x1000000L
+};
+
+extern int ehca_mr_largepage;
+
+static u32 ehca_encode_hwpage_size(u32 pgsize)
+{
+ u32 idx = 0;
+ pgsize >>= 12;
+ /*
+ * map mr page size into hw code:
+ * 0, 1, 2, 3 for 4K, 64K, 1M, 64M
+ */
+ while (!(pgsize & 1)) {
+ idx++;
+ pgsize >>= 4;
+ }
+ return idx;
+}
+
+static u64 ehca_get_max_hwpage_size(struct ehca_shca *shca)
+{
+ if (shca->hca_cap_mr_pgsize & HCA_CAP_MR_PGSIZE_16M)
+ return EHCA_MR_PGSIZE16M;
+ return EHCA_MR_PGSIZE4K;
+}
+
static struct ehca_mr *ehca_mr_new(void)
{
struct ehca_mr *me;
@@ -207,19 +239,23 @@ struct ib_mr *ehca_reg_phys_mr(struct ib_pd *pd,
struct ehca_mr_pginfo pginfo;
u32 num_kpages;
u32 num_hwpages;
+ u64 hw_pgsize;
num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size,
PAGE_SIZE);
- num_hwpages = NUM_CHUNKS(((u64)iova_start % EHCA_PAGESIZE) +
- size, EHCA_PAGESIZE);
+ /* for kernel space we try most possible pgsize */
+ hw_pgsize = ehca_get_max_hwpage_size(shca);
+ num_hwpages = NUM_CHUNKS(((u64)iova_start % hw_pgsize) + size,
+ hw_pgsize);
memset(&pginfo, 0, sizeof(pginfo));
pginfo.type = EHCA_MR_PGI_PHYS;
pginfo.num_kpages = num_kpages;
+ pginfo.hwpage_size = hw_pgsize;
pginfo.num_hwpages = num_hwpages;
pginfo.u.phy.num_phys_buf = num_phys_buf;
pginfo.u.phy.phys_buf_array = phys_buf_array;
- pginfo.next_hwpage = (((u64)iova_start & ~PAGE_MASK) /
- EHCA_PAGESIZE);
+ pginfo.next_hwpage =
+ ((u64)iova_start & ~(hw_pgsize - 1)) / hw_pgsize;
ret = ehca_reg_mr(shca, e_mr, iova_start, size, mr_access_flags,
e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
@@ -259,6 +295,7 @@ struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
int ret;
u32 num_kpages;
u32 num_hwpages;
+ u64 hwpage_size;
if (!pd) {
ehca_gen_err("bad pd=%p", pd);
@@ -309,16 +346,32 @@ struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
/* determine number of MR pages */
num_kpages = NUM_CHUNKS((virt % PAGE_SIZE) + length, PAGE_SIZE);
- num_hwpages = NUM_CHUNKS((virt % EHCA_PAGESIZE) + length,
- EHCA_PAGESIZE);
+ /* select proper hw_pgsize */
+ if (ehca_mr_largepage &&
+ (shca->hca_cap_mr_pgsize & HCA_CAP_MR_PGSIZE_16M)) {
+ if (length <= EHCA_MR_PGSIZE4K
+ && PAGE_SIZE == EHCA_MR_PGSIZE4K)
+ hwpage_size = EHCA_MR_PGSIZE4K;
+ else if (length <= EHCA_MR_PGSIZE64K)
+ hwpage_size = EHCA_MR_PGSIZE64K;
+ else if (length <= EHCA_MR_PGSIZE1M)
+ hwpage_size = EHCA_MR_PGSIZE1M;
+ else
+ hwpage_size = EHCA_MR_PGSIZE16M;
+ } else
+ hwpage_size = EHCA_MR_PGSIZE4K;
+ ehca_dbg(pd->device, "hwpage_size=%lx", hwpage_size);
+reg_user_mr_fallback:
+ num_hwpages = NUM_CHUNKS((virt % hwpage_size) + length, hwpage_size);
/* register MR on HCA */
memset(&pginfo, 0, sizeof(pginfo));
pginfo.type = EHCA_MR_PGI_USER;
+ pginfo.hwpage_size = hwpage_size;
pginfo.num_kpages = num_kpages;
pginfo.num_hwpages = num_hwpages;
pginfo.u.usr.region = e_mr->umem;
- pginfo.next_hwpage = e_mr->umem->offset / EHCA_PAGESIZE;
+ pginfo.next_hwpage = e_mr->umem->offset / hwpage_size;
pginfo.u.usr.next_chunk = list_prepare_entry(pginfo.u.usr.next_chunk,
(&e_mr->umem->chunk_list),
list);
@@ -326,6 +379,18 @@ struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
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);
+ if (ret == -EINVAL && pginfo.hwpage_size > PAGE_SIZE) {
+ ehca_warn(pd->device, "failed to register mr "
+ "with hwpage_size=%lx", hwpage_size);
+ ehca_info(pd->device, "try to register mr with "
+ "kpage_size=%lx", PAGE_SIZE);
+ /*
+ * this means kpages are not contiguous for a hw page
+ * try kernel page size as fallback solution
+ */
+ hwpage_size = PAGE_SIZE;
+ goto reg_user_mr_fallback;
+ }
if (ret) {
ib_mr = ERR_PTR(ret);
goto reg_user_mr_exit2;
@@ -452,6 +517,8 @@ int ehca_rereg_phys_mr(struct ib_mr *mr,
new_pd = container_of(mr->pd, struct ehca_pd, ib_pd);
if (mr_rereg_mask & IB_MR_REREG_TRANS) {
+ u64 hw_pgsize = ehca_get_max_hwpage_size(shca);
+
new_start = iova_start; /* change address */
/* check physical buffer list and calculate size */
ret = ehca_mr_chk_buf_and_calc_size(phys_buf_array,
@@ -468,16 +535,17 @@ int ehca_rereg_phys_mr(struct ib_mr *mr,
}
num_kpages = NUM_CHUNKS(((u64)new_start % PAGE_SIZE) +
new_size, PAGE_SIZE);
- num_hwpages = NUM_CHUNKS(((u64)new_start % EHCA_PAGESIZE) +
- new_size, EHCA_PAGESIZE);
+ num_hwpages = NUM_CHUNKS(((u64)new_start % hw_pgsize) +
+ new_size, hw_pgsize);
memset(&pginfo, 0, sizeof(pginfo));
pginfo.type = EHCA_MR_PGI_PHYS;
pginfo.num_kpages = num_kpages;
+ pginfo.hwpage_size = hw_pgsize;
pginfo.num_hwpages = num_hwpages;
pginfo.u.phy.num_phys_buf = num_phys_buf;
pginfo.u.phy.phys_buf_array = phys_buf_array;
- pginfo.next_hwpage = (((u64)iova_start & ~PAGE_MASK) /
- EHCA_PAGESIZE);
+ pginfo.next_hwpage =
+ ((u64)iova_start & ~(hw_pgsize - 1)) / hw_pgsize;
}
if (mr_rereg_mask & IB_MR_REREG_ACCESS)
new_acl = mr_access_flags;
@@ -709,6 +777,7 @@ struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
int ret;
u32 tmp_lkey, tmp_rkey;
struct ehca_mr_pginfo pginfo;
+ u64 hw_pgsize;
/* check other parameters */
if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
@@ -738,8 +807,8 @@ struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
ib_fmr = ERR_PTR(-EINVAL);
goto alloc_fmr_exit0;
}
- if (((1 << fmr_attr->page_shift) != EHCA_PAGESIZE) &&
- ((1 << fmr_attr->page_shift) != PAGE_SIZE)) {
+ hw_pgsize = ehca_get_max_hwpage_size(shca);
+ if ((1 << fmr_attr->page_shift) != hw_pgsize) {
ehca_err(pd->device, "unsupported fmr_attr->page_shift=%x",
fmr_attr->page_shift);
ib_fmr = ERR_PTR(-EINVAL);
@@ -755,6 +824,10 @@ struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
/* register MR on HCA */
memset(&pginfo, 0, sizeof(pginfo));
+ /*
+ * pginfo.num_hwpages==0, ie register_rpages() will not be called
+ * but deferred to map_phys_fmr()
+ */
ret = ehca_reg_mr(shca, e_fmr, NULL,
fmr_attr->max_pages * (1 << fmr_attr->page_shift),
mr_access_flags, e_pd, &pginfo,
@@ -765,6 +838,7 @@ struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
}
/* successful */
+ e_fmr->hwpage_size = hw_pgsize;
e_fmr->fmr_page_size = 1 << fmr_attr->page_shift;
e_fmr->fmr_max_pages = fmr_attr->max_pages;
e_fmr->fmr_max_maps = fmr_attr->max_maps;
@@ -822,10 +896,12 @@ int ehca_map_phys_fmr(struct ib_fmr *fmr,
memset(&pginfo, 0, sizeof(pginfo));
pginfo.type = EHCA_MR_PGI_FMR;
pginfo.num_kpages = list_len;
- pginfo.num_hwpages = list_len * (e_fmr->fmr_page_size / EHCA_PAGESIZE);
+ pginfo.hwpage_size = e_fmr->hwpage_size;
+ pginfo.num_hwpages =
+ list_len * e_fmr->fmr_page_size / pginfo.hwpage_size;
pginfo.u.fmr.page_list = page_list;
- pginfo.next_hwpage = ((iova & (e_fmr->fmr_page_size-1)) /
- EHCA_PAGESIZE);
+ pginfo.next_hwpage =
+ (iova & (e_fmr->fmr_page_size-1)) / pginfo.hwpage_size;
pginfo.u.fmr.fmr_pgsize = e_fmr->fmr_page_size;
ret = ehca_rereg_mr(shca, e_fmr, (u64 *)iova,
@@ -964,7 +1040,7 @@ int ehca_reg_mr(struct ehca_shca *shca,
struct ehca_mr_hipzout_parms hipzout;
ehca_mrmw_map_acl(acl, &hipz_acl);
- ehca_mrmw_set_pgsize_hipz_acl(&hipz_acl);
+ ehca_mrmw_set_pgsize_hipz_acl(pginfo->hwpage_size, &hipz_acl);
if (ehca_use_hp_mr == 1)
hipz_acl |= 0x00000001;
@@ -987,6 +1063,7 @@ int ehca_reg_mr(struct ehca_shca *shca,
/* successful registration */
e_mr->num_kpages = pginfo->num_kpages;
e_mr->num_hwpages = pginfo->num_hwpages;
+ e_mr->hwpage_size = pginfo->hwpage_size;
e_mr->start = iova_start;
e_mr->size = size;
e_mr->acl = acl;
@@ -1029,6 +1106,9 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
u32 i;
u64 *kpage;
+ if (!pginfo->num_hwpages) /* in case of fmr */
+ return 0;
+
kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
if (!kpage) {
ehca_err(&shca->ib_device, "kpage alloc failed");
@@ -1036,7 +1116,7 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
goto ehca_reg_mr_rpages_exit0;
}
- /* max 512 pages per shot */
+ /* max MAX_RPAGES ehca mr pages per register call */
for (i = 0; i < NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES); i++) {
if (i == NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES) - 1) {
@@ -1049,8 +1129,8 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
ret = ehca_set_pagebuf(pginfo, rnum, kpage);
if (ret) {
ehca_err(&shca->ib_device, "ehca_set_pagebuf "
- "bad rc, ret=%x rnum=%x kpage=%p",
- ret, rnum, kpage);
+ "bad rc, ret=%x rnum=%x kpage=%p",
+ ret, rnum, kpage);
goto ehca_reg_mr_rpages_exit1;
}
@@ -1065,9 +1145,10 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
} else
rpage = *kpage;
- h_ret = hipz_h_register_rpage_mr(shca->ipz_hca_handle, e_mr,
- 0, /* pagesize 4k */
- 0, rpage, rnum);
+ h_ret = hipz_h_register_rpage_mr(
+ shca->ipz_hca_handle, e_mr,
+ ehca_encode_hwpage_size(pginfo->hwpage_size),
+ 0, rpage, rnum);
if (i == NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES) - 1) {
/*
@@ -1131,7 +1212,7 @@ inline int ehca_rereg_mr_rereg1(struct ehca_shca *shca,
struct ehca_mr_hipzout_parms hipzout;
ehca_mrmw_map_acl(acl, &hipz_acl);
- ehca_mrmw_set_pgsize_hipz_acl(&hipz_acl);
+ ehca_mrmw_set_pgsize_hipz_acl(pginfo->hwpage_size, &hipz_acl);
kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
if (!kpage) {
@@ -1182,6 +1263,7 @@ inline int ehca_rereg_mr_rereg1(struct ehca_shca *shca,
*/
e_mr->num_kpages = pginfo->num_kpages;
e_mr->num_hwpages = pginfo->num_hwpages;
+ e_mr->hwpage_size = pginfo->hwpage_size;
e_mr->start = iova_start;
e_mr->size = size;
e_mr->acl = acl;
@@ -1268,13 +1350,14 @@ int ehca_rereg_mr(struct ehca_shca *shca,
/* set some MR values */
e_mr->flags = save_mr.flags;
+ e_mr->hwpage_size = save_mr.hwpage_size;
e_mr->fmr_page_size = save_mr.fmr_page_size;
e_mr->fmr_max_pages = save_mr.fmr_max_pages;
e_mr->fmr_max_maps = save_mr.fmr_max_maps;
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);
if (ret) {
u32 offset = (u64)(&e_mr->flags) - (u64)e_mr;
memcpy(&e_mr->flags, &(save_mr.flags),
@@ -1355,6 +1438,7 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca,
/* set some MR values */
e_fmr->flags = save_fmr.flags;
+ e_fmr->hwpage_size = save_fmr.hwpage_size;
e_fmr->fmr_page_size = save_fmr.fmr_page_size;
e_fmr->fmr_max_pages = save_fmr.fmr_max_pages;
e_fmr->fmr_max_maps = save_fmr.fmr_max_maps;
@@ -1363,8 +1447,6 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca,
memset(&pginfo, 0, sizeof(pginfo));
pginfo.type = EHCA_MR_PGI_FMR;
- pginfo.num_kpages = 0;
- pginfo.num_hwpages = 0;
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,
@@ -1373,7 +1455,6 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca,
u32 offset = (u64)(&e_fmr->flags) - (u64)e_fmr;
memcpy(&e_fmr->flags, &(save_mr.flags),
sizeof(struct ehca_mr) - offset);
- goto ehca_unmap_one_fmr_exit0;
}
ehca_unmap_one_fmr_exit0:
@@ -1401,7 +1482,7 @@ int ehca_reg_smr(struct ehca_shca *shca,
struct ehca_mr_hipzout_parms hipzout;
ehca_mrmw_map_acl(acl, &hipz_acl);
- ehca_mrmw_set_pgsize_hipz_acl(&hipz_acl);
+ ehca_mrmw_set_pgsize_hipz_acl(e_origmr->hwpage_size, &hipz_acl);
h_ret = hipz_h_register_smr(shca->ipz_hca_handle, e_newmr, e_origmr,
(u64)iova_start, hipz_acl, e_pd->fw_pd,
@@ -1420,6 +1501,7 @@ int ehca_reg_smr(struct ehca_shca *shca,
/* successful registration */
e_newmr->num_kpages = e_origmr->num_kpages;
e_newmr->num_hwpages = e_origmr->num_hwpages;
+ e_newmr->hwpage_size = e_origmr->hwpage_size;
e_newmr->start = iova_start;
e_newmr->size = e_origmr->size;
e_newmr->acl = acl;
@@ -1452,6 +1534,7 @@ int ehca_reg_internal_maxmr(
struct ib_phys_buf ib_pbuf;
u32 num_kpages;
u32 num_hwpages;
+ u64 hw_pgsize;
e_mr = ehca_mr_new();
if (!e_mr) {
@@ -1468,13 +1551,15 @@ int ehca_reg_internal_maxmr(
ib_pbuf.size = size_maxmr;
num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size_maxmr,
PAGE_SIZE);
- num_hwpages = NUM_CHUNKS(((u64)iova_start % EHCA_PAGESIZE) + size_maxmr,
- EHCA_PAGESIZE);
+ hw_pgsize = ehca_get_max_hwpage_size(shca);
+ num_hwpages = NUM_CHUNKS(((u64)iova_start % hw_pgsize) + size_maxmr,
+ hw_pgsize);
memset(&pginfo, 0, sizeof(pginfo));
pginfo.type = EHCA_MR_PGI_PHYS;
pginfo.num_kpages = num_kpages;
pginfo.num_hwpages = num_hwpages;
+ pginfo.hwpage_size = hw_pgsize;
pginfo.u.phy.num_phys_buf = 1;
pginfo.u.phy.phys_buf_array = &ib_pbuf;
@@ -1523,7 +1608,7 @@ int ehca_reg_maxmr(struct ehca_shca *shca,
struct ehca_mr_hipzout_parms hipzout;
ehca_mrmw_map_acl(acl, &hipz_acl);
- ehca_mrmw_set_pgsize_hipz_acl(&hipz_acl);
+ ehca_mrmw_set_pgsize_hipz_acl(e_origmr->hwpage_size, &hipz_acl);
h_ret = hipz_h_register_smr(shca->ipz_hca_handle, e_newmr, e_origmr,
(u64)iova_start, hipz_acl, e_pd->fw_pd,
@@ -1539,6 +1624,7 @@ int ehca_reg_maxmr(struct ehca_shca *shca,
/* successful registration */
e_newmr->num_kpages = e_origmr->num_kpages;
e_newmr->num_hwpages = e_origmr->num_hwpages;
+ e_newmr->hwpage_size = e_origmr->hwpage_size;
e_newmr->start = iova_start;
e_newmr->size = e_origmr->size;
e_newmr->acl = acl;
@@ -1684,6 +1770,7 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
u64 pgaddr;
u32 i = 0;
u32 j = 0;
+ int hwpages_per_kpage = PAGE_SIZE / pginfo->hwpage_size;
/* loop over desired chunk entries */
chunk = pginfo->u.usr.next_chunk;
@@ -1695,7 +1782,7 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
<< PAGE_SHIFT ;
*kpage = phys_to_abs(pgaddr +
(pginfo->next_hwpage *
- EHCA_PAGESIZE));
+ pginfo->hwpage_size));
if ( !(*kpage) ) {
ehca_gen_err("pgaddr=%lx "
"chunk->page_list[i]=%lx "
@@ -1708,8 +1795,7 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
(pginfo->hwpage_cnt)++;
(pginfo->next_hwpage)++;
kpage++;
- if (pginfo->next_hwpage %
- (PAGE_SIZE / EHCA_PAGESIZE) == 0) {
+ if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
(pginfo->kpage_cnt)++;
(pginfo->u.usr.next_nmap)++;
pginfo->next_hwpage = 0;
@@ -1738,6 +1824,143 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
return ret;
}
+/*
+ * check given pages for contiguous layout
+ * last page addr is returned in prev_pgaddr for further check
+ */
+static int ehca_check_kpages_per_ate(struct scatterlist *page_list,
+ int start_idx, int end_idx,
+ u64 *prev_pgaddr)
+{
+ int t;
+ for (t = start_idx; t <= end_idx; t++) {
+ u64 pgaddr = page_to_pfn(page_list[t].page) << PAGE_SHIFT;
+ ehca_gen_dbg("chunk_page=%lx value=%016lx", pgaddr,
+ *(u64 *)abs_to_virt(phys_to_abs(pgaddr)));
+ if (pgaddr - PAGE_SIZE != *prev_pgaddr) {
+ ehca_gen_err("uncontiguous page found pgaddr=%lx "
+ "prev_pgaddr=%lx page_list_i=%x",
+ pgaddr, *prev_pgaddr, t);
+ return -EINVAL;
+ }
+ *prev_pgaddr = pgaddr;
+ }
+ return 0;
+}
+
+/* PAGE_SIZE < pginfo->hwpage_size */
+static int ehca_set_pagebuf_user2(struct ehca_mr_pginfo *pginfo,
+ u32 number,
+ u64 *kpage)
+{
+ int ret = 0;
+ struct ib_umem_chunk *prev_chunk;
+ struct ib_umem_chunk *chunk;
+ u64 pgaddr, prev_pgaddr;
+ u32 i = 0;
+ u32 j = 0;
+ int kpages_per_hwpage = pginfo->hwpage_size / PAGE_SIZE;
+ int nr_kpages = kpages_per_hwpage;
+
+ /* loop over desired chunk entries */
+ chunk = pginfo->u.usr.next_chunk;
+ prev_chunk = pginfo->u.usr.next_chunk;
+ list_for_each_entry_continue(
+ chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
+ for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
+ if (nr_kpages == kpages_per_hwpage) {
+ pgaddr = ( page_to_pfn(chunk->page_list[i].page)
+ << PAGE_SHIFT );
+ *kpage = phys_to_abs(pgaddr);
+ if ( !(*kpage) ) {
+ ehca_gen_err("pgaddr=%lx i=%x",
+ pgaddr, i);
+ ret = -EFAULT;
+ return ret;
+ }
+ /*
+ * The first page in a hwpage must be aligned;
+ * the first MR page is exempt from this rule.
+ */
+ if (pgaddr & (pginfo->hwpage_size - 1)) {
+ if (pginfo->hwpage_cnt) {
+ ehca_gen_err(
+ "invalid alignment "
+ "pgaddr=%lx i=%x "
+ "mr_pgsize=%lx",
+ pgaddr, i,
+ pginfo->hwpage_size);
+ ret = -EFAULT;
+ return ret;
+ }
+ /* first MR page */
+ pginfo->kpage_cnt =
+ (pgaddr &
+ (pginfo->hwpage_size - 1)) >>
+ PAGE_SHIFT;
+ nr_kpages -= pginfo->kpage_cnt;
+ *kpage = phys_to_abs(
+ pgaddr &
+ ~(pginfo->hwpage_size - 1));
+ }
+ ehca_gen_dbg("kpage=%lx chunk_page=%lx "
+ "value=%016lx", *kpage, pgaddr,
+ *(u64 *)abs_to_virt(
+ phys_to_abs(pgaddr)));
+ prev_pgaddr = pgaddr;
+ i++;
+ pginfo->kpage_cnt++;
+ pginfo->u.usr.next_nmap++;
+ nr_kpages--;
+ if (!nr_kpages)
+ goto next_kpage;
+ continue;
+ }
+ if (i + nr_kpages > chunk->nmap) {
+ ret = ehca_check_kpages_per_ate(
+ chunk->page_list, i,
+ chunk->nmap - 1, &prev_pgaddr);
+ if (ret) return ret;
+ pginfo->kpage_cnt += chunk->nmap - i;
+ pginfo->u.usr.next_nmap += chunk->nmap - i;
+ nr_kpages -= chunk->nmap - i;
+ break;
+ }
+
+ ret = ehca_check_kpages_per_ate(chunk->page_list, i,
+ i + nr_kpages - 1,
+ &prev_pgaddr);
+ if (ret) return ret;
+ i += nr_kpages;
+ pginfo->kpage_cnt += nr_kpages;
+ pginfo->u.usr.next_nmap += nr_kpages;
+next_kpage:
+ nr_kpages = kpages_per_hwpage;
+ (pginfo->hwpage_cnt)++;
+ kpage++;
+ j++;
+ if (j >= number) break;
+ }
+ if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
+ (j >= number)) {
+ pginfo->u.usr.next_nmap = 0;
+ prev_chunk = chunk;
+ break;
+ } else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
+ pginfo->u.usr.next_nmap = 0;
+ prev_chunk = chunk;
+ } else if (j >= number)
+ break;
+ else
+ prev_chunk = chunk;
+ }
+ pginfo->u.usr.next_chunk =
+ list_prepare_entry(prev_chunk,
+ (&(pginfo->u.usr.region->chunk_list)),
+ list);
+ return ret;
+}
+
int ehca_set_pagebuf_phys(struct ehca_mr_pginfo *pginfo,
u32 number,
u64 *kpage)
@@ -1750,9 +1973,10 @@ int ehca_set_pagebuf_phys(struct ehca_mr_pginfo *pginfo,
/* loop over desired phys_buf_array entries */
while (i < number) {
pbuf = pginfo->u.phy.phys_buf_array + pginfo->u.phy.next_buf;
- num_hw = NUM_CHUNKS((pbuf->addr % EHCA_PAGESIZE) +
- pbuf->size, EHCA_PAGESIZE);
- offs_hw = (pbuf->addr & ~PAGE_MASK) / EHCA_PAGESIZE;
+ num_hw = NUM_CHUNKS((pbuf->addr % pginfo->hwpage_size) +
+ pbuf->size, pginfo->hwpage_size);
+ offs_hw = (pbuf->addr & ~(pginfo->hwpage_size - 1)) /
+ pginfo->hwpage_size;
while (pginfo->next_hwpage < offs_hw + num_hw) {
/* sanity check */
if ((pginfo->kpage_cnt >= pginfo->num_kpages) ||
@@ -1768,21 +1992,23 @@ int ehca_set_pagebuf_phys(struct ehca_mr_pginfo *pginfo,
return -EFAULT;
}
*kpage = phys_to_abs(
- (pbuf->addr & EHCA_PAGEMASK)
- + (pginfo->next_hwpage * EHCA_PAGESIZE));
+ (pbuf->addr & ~(pginfo->hwpage_size - 1)) +
+ (pginfo->next_hwpage * pginfo->hwpage_size));
if ( !(*kpage) && pbuf->addr ) {
- ehca_gen_err("pbuf->addr=%lx "
- "pbuf->size=%lx "
+ ehca_gen_err("pbuf->addr=%lx pbuf->size=%lx "
"next_hwpage=%lx", pbuf->addr,
- pbuf->size,
- pginfo->next_hwpage);
+ pbuf->size, pginfo->next_hwpage);
return -EFAULT;
}
(pginfo->hwpage_cnt)++;
(pginfo->next_hwpage)++;
- if (pginfo->next_hwpage %
- (PAGE_SIZE / EHCA_PAGESIZE) == 0)
- (pginfo->kpage_cnt)++;
+ if (PAGE_SIZE >= pginfo->hwpage_size) {
+ if (pginfo->next_hwpage %
+ (PAGE_SIZE / pginfo->hwpage_size) == 0)
+ (pginfo->kpage_cnt)++;
+ } else
+ pginfo->kpage_cnt += pginfo->hwpage_size /
+ PAGE_SIZE;
kpage++;
i++;
if (i >= number) break;
@@ -1806,8 +2032,8 @@ int ehca_set_pagebuf_fmr(struct ehca_mr_pginfo *pginfo,
/* loop over desired page_list entries */
fmrlist = pginfo->u.fmr.page_list + pginfo->u.fmr.next_listelem;
for (i = 0; i < number; i++) {
- *kpage = phys_to_abs((*fmrlist & EHCA_PAGEMASK) +
- pginfo->next_hwpage * EHCA_PAGESIZE);
+ *kpage = phys_to_abs((*fmrlist & ~(pginfo->hwpage_size - 1)) +
+ pginfo->next_hwpage * pginfo->hwpage_size);
if ( !(*kpage) ) {
ehca_gen_err("*fmrlist=%lx fmrlist=%p "
"next_listelem=%lx next_hwpage=%lx",
@@ -1817,15 +2043,38 @@ int ehca_set_pagebuf_fmr(struct ehca_mr_pginfo *pginfo,
return -EFAULT;
}
(pginfo->hwpage_cnt)++;
- (pginfo->next_hwpage)++;
- kpage++;
- if (pginfo->next_hwpage %
- (pginfo->u.fmr.fmr_pgsize / EHCA_PAGESIZE) == 0) {
- (pginfo->kpage_cnt)++;
- (pginfo->u.fmr.next_listelem)++;
- fmrlist++;
- pginfo->next_hwpage = 0;
+ if (pginfo->u.fmr.fmr_pgsize >= pginfo->hwpage_size) {
+ if (pginfo->next_hwpage %
+ (pginfo->u.fmr.fmr_pgsize /
+ pginfo->hwpage_size) == 0) {
+ (pginfo->kpage_cnt)++;
+ (pginfo->u.fmr.next_listelem)++;
+ fmrlist++;
+ pginfo->next_hwpage = 0;
+ } else
+ (pginfo->next_hwpage)++;
+ } else {
+ unsigned int cnt_per_hwpage = pginfo->hwpage_size /
+ pginfo->u.fmr.fmr_pgsize;
+ unsigned int j;
+ u64 prev = *kpage;
+ /* check if adrs are contiguous */
+ for (j = 1; j < cnt_per_hwpage; j++) {
+ u64 p = phys_to_abs(fmrlist[j] &
+ ~(pginfo->hwpage_size - 1));
+ if (prev + pginfo->u.fmr.fmr_pgsize != p) {
+ ehca_gen_err("uncontiguous fmr pages "
+ "found prev=%lx p=%lx "
+ "idx=%x", prev, p, i + j);
+ return -EINVAL;
+ }
+ prev = p;
+ }
+ pginfo->kpage_cnt += cnt_per_hwpage;
+ pginfo->u.fmr.next_listelem += cnt_per_hwpage;
+ fmrlist += cnt_per_hwpage;
}
+ kpage++;
}
return ret;
}
@@ -1842,7 +2091,9 @@ int ehca_set_pagebuf(struct ehca_mr_pginfo *pginfo,
ret = ehca_set_pagebuf_phys(pginfo, number, kpage);
break;
case EHCA_MR_PGI_USER:
- ret = ehca_set_pagebuf_user1(pginfo, number, kpage);
+ ret = PAGE_SIZE >= pginfo->hwpage_size ?
+ ehca_set_pagebuf_user1(pginfo, number, kpage) :
+ ehca_set_pagebuf_user2(pginfo, number, kpage);
break;
case EHCA_MR_PGI_FMR:
ret = ehca_set_pagebuf_fmr(pginfo, number, kpage);
@@ -1895,9 +2146,9 @@ void ehca_mrmw_map_acl(int ib_acl,
/*----------------------------------------------------------------------*/
/* sets page size in hipz access control for MR/MW. */
-void ehca_mrmw_set_pgsize_hipz_acl(u32 *hipz_acl) /*INOUT*/
+void ehca_mrmw_set_pgsize_hipz_acl(u32 pgsize, u32 *hipz_acl) /*INOUT*/
{
- return; /* HCA supports only 4k */
+ *hipz_acl |= (ehca_encode_hwpage_size(pgsize) << 24);
} /* end ehca_mrmw_set_pgsize_hipz_acl() */
/*----------------------------------------------------------------------*/
diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.h b/drivers/infiniband/hw/ehca/ehca_mrmw.h
index 24f13fe..bc8f4e3 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.h
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.h
@@ -111,7 +111,7 @@ int ehca_mr_is_maxmr(u64 size,
void ehca_mrmw_map_acl(int ib_acl,
u32 *hipz_acl);
-void ehca_mrmw_set_pgsize_hipz_acl(u32 *hipz_acl);
+void ehca_mrmw_set_pgsize_hipz_acl(u32 pgsize, u32 *hipz_acl);
void ehca_mrmw_reverse_map_acl(const u32 *hipz_acl,
int *ib_acl);
diff --git a/drivers/infiniband/hw/ehca/hcp_if.c b/drivers/infiniband/hw/ehca/hcp_if.c
index 3394e05..358796c 100644
--- a/drivers/infiniband/hw/ehca/hcp_if.c
+++ b/drivers/infiniband/hw/ehca/hcp_if.c
@@ -427,7 +427,8 @@ u64 hipz_h_register_rpage(const struct ipz_adapter_handle adapter_handle,
{
return ehca_plpar_hcall_norets(H_REGISTER_RPAGES,
adapter_handle.handle, /* r4 */
- queue_type | pagesize << 8, /* r5 */
+ (u64)queue_type | ((u64)pagesize) << 8,
+ /* r5 */
resource_handle, /* r6 */
logical_address_of_page, /* r7 */
count, /* r8 */
@@ -724,6 +725,9 @@ u64 hipz_h_alloc_resource_mr(const struct ipz_adapter_handle adapter_handle,
u64 ret;
u64 outs[PLPAR_HCALL9_BUFSIZE];
+ ehca_gen_dbg("kernel PAGE_SIZE=%x access_ctrl=%016x "
+ "vaddr=%lx length=%lx",
+ (u32)PAGE_SIZE, access_ctrl, vaddr, length);
ret = ehca_plpar_hcall9(H_ALLOC_RESOURCE, outs,
adapter_handle.handle, /* r4 */
5, /* r5 */
@@ -746,8 +750,22 @@ u64 hipz_h_register_rpage_mr(const struct ipz_adapter_handle adapter_handle,
const u64 logical_address_of_page,
const u64 count)
{
+ extern int ehca_debug_level;
u64 ret;
+ if (unlikely(ehca_debug_level >= 2)) {
+ if (count > 1) {
+ u64 *kpage;
+ int i;
+ kpage = (u64 *)abs_to_virt(logical_address_of_page);
+ for (i = 0; i < count; i++)
+ ehca_gen_dbg("kpage[%d]=%p",
+ i, (void *)kpage[i]);
+ } else
+ ehca_gen_dbg("kpage=%p",
+ (void *)logical_address_of_page);
+ }
+
if ((count > 1) && (logical_address_of_page & (EHCA_PAGESIZE-1))) {
ehca_gen_err("logical_address_of_page not on a 4k boundary "
"adapter_handle=%lx mr=%p mr_handle=%lx "
--
1.5.2
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [ofa-general] [PATCH 1/5] ehca: Supports large page MRs
2007-07-20 14:01 [PATCH 1/5] ehca: Supports large page MRs Hoang-Nam Nguyen
@ 2007-07-21 4:07 ` Roland Dreier
2007-07-21 8:22 ` Hoang-Nam Nguyen
0 siblings, 1 reply; 3+ messages in thread
From: Roland Dreier @ 2007-07-21 4:07 UTC (permalink / raw)
To: Hoang-Nam Nguyen
Cc: linux-kernel, linuxppc-dev, openib-general, fenkes, stefan.roscher
I applied this, but I agree with checkpatch.pl:
> WARNING: externs should be avoided in .c files
> #227: FILE: drivers/infiniband/hw/ehca/ehca_mrmw.c:67:
> +extern int ehca_mr_largepage;
>
> WARNING: externs should be avoided in .c files
> #949: FILE: drivers/infiniband/hw/ehca/hcp_if.c:753:
> + extern int ehca_debug_level;
if you need to use a variable in more than one .c file, put the extern
declaration in a common header that's included everywhere you use the
variable, including the .c file that it is defined in. That way the
compiler can see if you get confused about the type of the variable.
When you get a chance, please post a follow-on patch to fix this.
- R.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [ofa-general] [PATCH 1/5] ehca: Supports large page MRs
2007-07-21 4:07 ` [ofa-general] " Roland Dreier
@ 2007-07-21 8:22 ` Hoang-Nam Nguyen
0 siblings, 0 replies; 3+ messages in thread
From: Hoang-Nam Nguyen @ 2007-07-21 8:22 UTC (permalink / raw)
To: Roland Dreier
Cc: general-bounces, Hoang-Nam Nguyen, Joachim Fenkes, linux-kernel,
linuxppc-dev, openib-general, Stefan Roscher
Hi Roland!
> I applied this, but I agree with checkpatch.pl:
>
> > WARNING: externs should be avoided in .c files
> > #227: FILE: drivers/infiniband/hw/ehca/ehca_mrmw.c:67:
> > +extern int ehca_mr_largepage;
> >
> > WARNING: externs should be avoided in .c files
> > #949: FILE: drivers/infiniband/hw/ehca/hcp_if.c:753:
> > + extern int ehca_debug_level;
>
> if you need to use a variable in more than one .c file, put the extern
> declaration in a common header that's included everywhere you use the
> variable, including the .c file that it is defined in. That way the
> compiler can see if you get confused about the type of the variable.
That's true.
> When you get a chance, please post a follow-on patch to fix this.
Sure thing. Will do that for rc2.
Thanks!
Nam
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2007-07-21 8:23 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-07-20 14:01 [PATCH 1/5] ehca: Supports large page MRs Hoang-Nam Nguyen
2007-07-21 4:07 ` [ofa-general] " Roland Dreier
2007-07-21 8:22 ` Hoang-Nam Nguyen
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).