linux-pci.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Yinghai Lu <yinghai@kernel.org>
To: Bjorn Helgaas <bhelgaas@google.com>,
	David Miller <davem@davemloft.net>,
	Benjamin Herrenschmidt <benh@kernel.crashing.org>,
	Wei Yang <weiyang@linux.vnet.ibm.com>, TJ <linux@iam.tj>,
	Yijing Wang <wangyijing@huawei.com>,
	Khalid Aziz <khalid.aziz@oracle.com>
Cc: linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org,
	Yinghai Lu <yinghai@kernel.org>
Subject: [PATCH v10 27/59] PCI: Optimize bus align/size calculation for optional during sizing
Date: Wed, 24 Feb 2016 18:12:18 -0800	[thread overview]
Message-ID: <1456366370-28995-28-git-send-email-yinghai@kernel.org> (raw)
In-Reply-To: <1456366370-28995-1-git-send-email-yinghai@kernel.org>

Current add_align always use max align, that make required+optional
to get allocated more than needed in some cases.

Now we have new calculate_mem_align(), so we could use it for
add_align calculation.

Need to create separated list for required+optional align/size info.

After that we can get smaller add_align/size, and we have more chance
to make required+optional to get allocated sucessfully.

The result for bridge that have Intel 4x10g card installed:

 pci 0000:20:03.2: bridge window [mem 0x00000000-0x000fffff 64bit pref]
	to [bus 2a-31] calculate_mem for required
 align/size:
   00800000/00800000, 00800000/00800000, 00800000/00800000,
   00800000/00800000, 00008000/00008000, 00008000/00008000,
   00008000/00008000, 00008000/00008000
 original min_align/min_size: 00400000/02400000
 new min_align/min_size: 00400000/02400000

 pci 0000:20:03.2: bridge window [mem 0x00000000-0x000fffff 64bit pref]
	to [bus 2a-31] calculate_mem for required+optional
 align/size:
   00800000/00800000, 00800000/00800000, 00800000/00800000,
   00800000/00800000, 00010000/00200000, 00010000/00200000,
   00010000/00200000, 00010000/00200000, 00008000/00008000,
   00008000/00008000, 00008000/00008000, 00008000/00008000,
   00004000/00080000, 00004000/00080000, 00004000/00080000,
   00004000/00080000
 original code min_align/min_size: 00800000/03000000
 new min_align/min_size: 00100000/02b00000

so required align/size: 0x400000/0x2400000, and
new required+optional align/size: 0x100000/0x2b00000, and it is much better
than original required+optional align/size: 0x800000/0x3000000
and even have smaller min_align than required.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=81431
Reported-by: TJ <linux@iam.tj>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
---
 drivers/pci/setup-bus.c | 82 ++++++++++++++++++++++++++++++-------------------
 1 file changed, 51 insertions(+), 31 deletions(-)

diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
index 4e07a01..6265f4d 100644
--- a/drivers/pci/setup-bus.c
+++ b/drivers/pci/setup-bus.c
@@ -944,7 +944,6 @@ static resource_size_t calculate_iosize(resource_size_t size,
 
 static resource_size_t calculate_memsize(resource_size_t size,
 		resource_size_t min_size,
-		resource_size_t size1,
 		resource_size_t old_size,
 		resource_size_t align)
 {
@@ -954,7 +953,7 @@ static resource_size_t calculate_memsize(resource_size_t size,
 		old_size = 0;
 	if (size < old_size)
 		size = old_size;
-	size = ALIGN(size + size1, align);
+	size = ALIGN(size, align);
 	return size;
 }
 
@@ -1217,26 +1216,23 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
 			 struct list_head *realloc_head)
 {
 	struct pci_dev *dev;
-	resource_size_t min_align, align, size, size0, size1;
-	resource_size_t max_align = 0;
+	resource_size_t min_align = 0, min_add_align = 0;
+	resource_size_t max_align = 0, max_add_align = 0;
+	resource_size_t size = 0, size0 = 0, size1 = 0, sum_add_size = 0;
 	struct resource *b_res = find_free_bus_resource(bus,
 					mask | IORESOURCE_PREFETCH, type);
-	resource_size_t children_add_size = 0;
-	resource_size_t children_add_align = 0;
-	resource_size_t add_align = 0;
 	LIST_HEAD(align_test_list);
+	LIST_HEAD(align_test_add_list);
 
 	if (!b_res)
 		return -ENOSPC;
 
-	size = 0;
-
 	list_for_each_entry(dev, &bus->devices, bus_list) {
 		int i;
 
 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
 			struct resource *r = &dev->resource[i];
-			resource_size_t r_size;
+			resource_size_t r_size, align;
 			int flags = pci_resource_pref_compatible(dev, r);
 
 			if (r->parent || (flags & IORESOURCE_PCI_FIXED) ||
@@ -1244,19 +1240,23 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
 			     (flags & mask) != type2 &&
 			     (flags & mask) != type3))
 				continue;
+
 			r_size = resource_size(r);
+			align = pci_resource_alignment(dev, r);
 #ifdef CONFIG_PCI_IOV
 			/* put SRIOV requested res to the optional list */
 			if (realloc_head && i >= PCI_IOV_RESOURCES &&
 					i <= PCI_IOV_RESOURCE_END) {
-				add_align = max(pci_resource_alignment(dev, r), add_align);
+				add_to_align_test_list(&align_test_add_list,
+							align, r_size);
 				r->end = r->start - 1;
 				add_to_list(realloc_head, dev, r, r_size, 0/* don't care */);
-				children_add_size += r_size;
+				sum_add_size += r_size;
+				if (align > max_add_align)
+					max_add_align = align;
 				continue;
 			}
 #endif
-			align = pci_resource_alignment(dev, r);
 			if (align > (1ULL<<37)) { /*128 Gb*/
 				dev_warn(&dev->dev, "disabling BAR %d: %pR (bad alignment %#llx)\n",
 					i, r, (unsigned long long) align);
@@ -1264,33 +1264,52 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
 				continue;
 			}
 
-			if (r_size > 1)
+			if (r_size > 1) {
 				add_to_align_test_list(&align_test_list,
 							align, r_size);
-			size += r_size;
-			if (align > max_align)
-				max_align = align;
+				size += r_size;
+				if (align > max_align)
+					max_align = align;
+			}
 
 			if (realloc_head) {
-				children_add_size += get_res_add_size(realloc_head, r);
-				children_add_align = get_res_add_align(realloc_head, r);
-				add_align = max(add_align, children_add_align);
+				resource_size_t add_r_size, add_align;
+
+				add_r_size = get_res_add_size(realloc_head, r);
+				add_align = get_res_add_align(realloc_head, r);
+				/* no add on ? */
+				if (add_align < align)
+					add_align = align;
+				add_to_align_test_list(&align_test_add_list,
+							add_align,
+							r_size + add_r_size);
+				sum_add_size += r_size + add_r_size;
+				if (add_align > max_add_align)
+					max_add_align = add_align;
 			}
 		}
 	}
 
 	max_align = max(max_align, window_alignment(bus, b_res->flags));
-	min_align = calculate_mem_align(&align_test_list, max_align, size,
-					window_alignment(bus, b_res->flags));
-	size0 = calculate_memsize(size, min_size, 0,
+	if (size || min_size) {
+		min_align = calculate_mem_align(&align_test_list, max_align,
+				 size, window_alignment(bus, b_res->flags));
+		size0 = calculate_memsize(size, min_size,
 				  resource_size(b_res), min_align);
+	}
 	free_align_test_list(&align_test_list);
-	add_align = max(min_align, add_align);
-	if (children_add_size > add_size)
-		add_size = children_add_size;
-	size1 = (!realloc_head || (realloc_head && !add_size)) ? size0 :
-		calculate_memsize(size, min_size, add_size,
-				resource_size(b_res), add_align);
+
+	if ((sum_add_size - size) < add_size)
+		sum_add_size = size + add_size;
+	if (sum_add_size > size && realloc_head) {
+		min_add_align = calculate_mem_align(&align_test_add_list,
+					max_add_align, sum_add_size,
+					window_alignment(bus, b_res->flags));
+		size1 = calculate_memsize(sum_add_size, min_size,
+				 resource_size(b_res), min_add_align);
+	}
+	free_align_test_list(&align_test_add_list);
+
 	if (!size0 && !size1) {
 		if (b_res->start || b_res->end)
 			dev_info(&bus->self->dev, "disabling bridge window %pR to %pR (unused)\n",
@@ -1302,11 +1321,12 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
 	b_res->end = size0 + min_align - 1;
 	b_res->flags |= IORESOURCE_STARTALIGN;
 	if (size1 > size0 && realloc_head) {
-		add_to_list(realloc_head, bus->self, b_res, size1-size0, add_align);
+		add_to_list(realloc_head, bus->self, b_res, size1 - size0,
+				min_add_align);
 		dev_printk(KERN_DEBUG, &bus->self->dev, "bridge window %pR to %pR add_size %llx add_align %llx\n",
 			   b_res, &bus->busn_res,
 			   (unsigned long long) (size1 - size0),
-			   (unsigned long long) add_align);
+			   (unsigned long long) min_add_align);
 	}
 	return 0;
 }
-- 
1.8.4.5

  parent reply	other threads:[~2016-02-25  2:12 UTC|newest]

Thread overview: 77+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-02-25  2:11 [PATCH v10 00/59] PCI: Resource allocation cleanup for v4.6 Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 01/59] PCI: Add pci_find_root_bus_resource() Yinghai Lu
2016-03-10  3:54   ` Bjorn Helgaas
2016-03-11 21:21     ` Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 02/59] sparc/PCI: Use correct bus address to resource offset Yinghai Lu
2016-03-10 18:24   ` Bjorn Helgaas
2016-03-10 19:47     ` David Miller
2016-03-12  8:22     ` Yinghai Lu
2016-03-12 11:26       ` Bjorn Helgaas
2016-03-19  6:01         ` Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 03/59] sparc/PCI: Reserve legacy mmio after PCI mmio Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 04/59] sparc/PCI: Add IORESOURCE_MEM_64 for 64-bit resource in OF parsing Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 05/59] sparc/PCI: Keep resource idx order with bridge register number Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 06/59] PCI: Kill wrong quirk about M7101 Yinghai Lu
2016-03-10 17:40   ` Bjorn Helgaas
2016-03-11 22:08     ` Yinghai Lu
2016-03-12  1:06       ` Linus Torvalds
2016-03-12  7:52       ` Meelis Roos
2016-03-12  8:26         ` Yinghai Lu
2016-03-12  8:39           ` Meelis Roos
2016-04-25 20:57             ` Bjorn Helgaas
2016-02-25  2:11 ` [PATCH v10 07/59] PCI: Ignore BAR for ALi M1533 PCI-ISA bridge Yinghai Lu
2016-03-10 17:54   ` Bjorn Helgaas
2016-03-11 22:21     ` Yinghai Lu
2016-02-25  2:11 ` [PATCH v10 08/59] powerpc/PCI: Keep resource idx order with bridge register number Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 09/59] powerpc/PCI: Add IORESOURCE_MEM_64 for 64-bit resource in OF parsing Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 10/59] OF/PCI: Add IORESOURCE_MEM_64 for 64-bit resource Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 11/59] PCI: Check pref compatible bit for mem64 resource of PCIe device Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 12/59] PCI: Only treat non-pref mmio64 as pref if all bridges have MEM_64 Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 13/59] PCI: Add has_mem64 for struct host_bridge Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 14/59] PCI: Only treat non-pref mmio64 as pref if host bridge has mmio64 Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 15/59] PCI: Restore pref MMIO allocation logic for host bridge without mmio64 Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 16/59] PCI: Don't release fixed resource for realloc Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 17/59] PCI: Claim fixed resource during remove/rescan path Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 18/59] PCI: Set resource to FIXED for LSI devices Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 19/59] PCI: Separate realloc list checking after allocation Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 20/59] PCI: Treat optional as required in first try for bridge rescan Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 21/59] PCI: Get new realloc size for bridge for last try Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 22/59] PCI: Don't release sibling bridge resources during hotplug Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 23/59] PCI: Cleanup res_to_dev_res() printout Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 24/59] PCI: Reuse res_to_dev_res() in reassign_resources_sorted() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 25/59] PCI: Use correct align for optional only resources during sorting Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 26/59] PCI: Optimize bus min_align/size calculation during sizing Yinghai Lu
2016-02-25  2:12 ` Yinghai Lu [this message]
2016-02-25  2:12 ` [PATCH v10 28/59] PCI: Don't add too much optional size for hotplug bridge MMIO Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 29/59] PCI: Reorder resources list for required/optional resources Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 30/59] PCI: Remove duplicated code for resource sorting Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 31/59] PCI: Rename pdev_sort_resources() to pdev_assign_resources_prepare() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 32/59] PCI: Treat ROM resource as optional during realloc Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 33/59] PCI: Add debug printout during releasing partial assigned resources Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 34/59] PCI: Simplify res reference using in __assign_resources_sorted() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 35/59] PCI: Add __add_to_list() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 36/59] PCI: Cache window alignment value during bus sizing Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 37/59] PCI: Check if resource is allocated before trying to assign one Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 38/59] PCI: Separate out save_resources()/restore_resources() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 39/59] PCI: Move comment to pci_need_to_release() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 40/59] PCI: Separate required+optional assigning to another function Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 41/59] PCI: Skip required+optional if there is no optional Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 42/59] PCI: Move saved required resource list out of required+optional assigning Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 43/59] PCI: Add alt_size ressource allocation support Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 44/59] PCI: Add support for more than two alt_size entries under same bridge Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 45/59] PCI: Fix size calculation with old_size on rescan path Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 46/59] PCI: Don't add too much optional size for hotplug bridge io Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 47/59] PCI: Move ISA io port align out of calculate_iosize() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 48/59] PCI: Don't add too much io port for hotplug bridge with old size Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 49/59] PCI: Unify calculate_size() for io port and MMIO Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 50/59] PCI: Allow bridge optional only io port resource required size to be 0 Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 51/59] PCI: Unify skip_ioresource_align() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 52/59] PCI: Kill macro checking for bus io port sizing Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 53/59] resources: Make allocate_resource() return best fit resource Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 54/59] PCI, x86: Allocate from high in available window for MMIO Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 55/59] PCI: Add debug print out for min_align and alt_size Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 56/59] PCI, x86: Add pci=assign_pref_bars to reallocate pref BARs Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 57/59] PCI: Introduce resource_disabled() Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 58/59] PCI: Don't set flags to 0 when assign resource fail Yinghai Lu
2016-02-25  2:12 ` [PATCH v10 59/59] PCI: Only try to assign io port only for root bus that support it Yinghai Lu
2016-03-10 18:30 ` [PATCH v10 00/59] PCI: Resource allocation cleanup for v4.6 Bjorn Helgaas

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1456366370-28995-28-git-send-email-yinghai@kernel.org \
    --to=yinghai@kernel.org \
    --cc=benh@kernel.crashing.org \
    --cc=bhelgaas@google.com \
    --cc=davem@davemloft.net \
    --cc=khalid.aziz@oracle.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pci@vger.kernel.org \
    --cc=linux@iam.tj \
    --cc=wangyijing@huawei.com \
    --cc=weiyang@linux.vnet.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).