Preserve the flags portion on 1:1 dma-ranges mapping
Subsystems (5)
cover Preserve the flags portion on 1:1 dma-ranges mapping
Empty dma-ranges in DT nodes using 3-cell address spcifier cause the flag portion to be dropped by of_translate_one(), failing the translation chain. This patch aims at fixing this issue.
Part of this patchset was originally preparatory for a bigger patchset (see [1]). It has been split in a standalone one for better management and because it contains a bugfix which is probably of interest to stable branch.
I've also added new tests to unittest to prove it.
Many thanks,
Andrea
References:
[1] - https://lore.kernel.org/all/3029857353c9499659369c1540ba887d7860670f.1730123575.git.andrea.porta@suse.com/
Andrea della Porta (2):
of/unittest: Add empty dma-ranges address translation tests
of: address: Preserve the flags portion on 1:1 dma-ranges mapping
drivers/of/address.c | 3 +- drivers/of/unittest-data/tests-address.dtsi | 2 ++ drivers/of/unittest.c | 39 +++++++++++++++++++++ 3 files changed, 43 insertions(+), 1 deletion(-)
Signature · 1 line
2.35.3
PATCH 1/2 of/unittest: Add empty dma-ranges address translation tests
Intermediate DT PCI nodes dynamically generated by enabling CONFIG_PCI_DYNAMIC_OF_NODES have empty dma-ranges property. PCI address specifiers have 3 cells and when dma-ranges is missing or empty, of_translate_one() is currently dropping the flag portion of PCI addresses which are subnodes of the aforementioned ones, failing the translation. Add new tests covering this case.
With this test, we get 1 new failure which is fixed in subsequent commit:
FAIL of_unittest_pci_empty_dma_ranges():1245 for_each_of_pci_range wrong CPU addr (ffffffffffffffff) on node /testcase-data/address-tests2/pcie@d1070000/pci@0,0/dev@0,0/local-bus@0
--- drivers/of/unittest-data/tests-address.dtsi | 2 ++ drivers/of/unittest.c | 39 +++++++++++++++++++++ 2 files changed, 41 insertions(+)
drivers/of/unittest-data/tests-address.dtsi +2 -0
@@ -114,6 +114,7 @@ pcie@d1070000 { device_type = "pci"; ranges = <0x82000000 0 0xe8000000 0 0xe8000000 0 0x7f00000>, <0x81000000 0 0x00000000 0 0xefff0000 0 0x0010000>; + dma-ranges = <0x43000000 0x10 0x00 0x00 0x00 0x00 0x10000000>; reg = <0x00000000 0xd1070000 0x20000>; pci@0,0 {
@@ -142,6 +143,7 @@ local-bus@0 { #size-cells = <0x01>; ranges = <0xa0000000 0 0 0 0x2000000>, <0xb0000000 1 0 0 0x1000000>; + dma-ranges = <0xc0000000 0x43000000 0x10 0x00 0x10000000>; dev@e0000000 { reg = <0xa0001000 0x1000>,
drivers/of/unittest.c +39 -0
@@ -1213,6 +1213,44 @@ static void __init of_unittest_pci_dma_ranges(void)of_node_put (np); } +static void __init of_unittest_pci_empty_dma_ranges (void ) +{ +struct device_node *np; +struct of_pci_range range; +struct of_pci_range_parser parser; + +if (!IS_ENABLED (CONFIG_PCI )) +return ; + + np =of_find_node_by_path ("/testcase-data/address-tests2/pcie@d1070000/pci@0,0/dev@0,0/local-bus@0" ); +if (!np) { +pr_err ("missing testcase data\n" ); +return ; + } + +if (of_pci_dma_range_parser_init (&parser, np)) { +pr_err ("missing dma-ranges property\n" ); +return ; + } + +/* +* Get the dma-ranges from the device tree +*/ +for_each_of_pci_range (&parser, &range) { +unittest (range.size ==0x10000000 , +"for_each_of_pci_range wrong size on node %pOF size=%llx\n" , + np, range.size); +unittest (range.cpu_addr ==0x00000000 , +"for_each_of_pci_range wrong CPU addr (%llx) on node %pOF" , + range.cpu_addr, np); +unittest (range.pci_addr ==0xc0000000 , +"for_each_of_pci_range wrong DMA addr (%llx) on node %pOF" , + range.pci_addr, np); + } + +of_node_put (np); +} +static void __init of_unittest_bus_ranges (void ) {struct device_node *np;
@@ -4272,6 +4310,7 @@ static int __init of_unittest(void)of_unittest_dma_get_max_cpu_address ();of_unittest_parse_dma_ranges ();of_unittest_pci_dma_ranges (); +of_unittest_pci_empty_dma_ranges ();of_unittest_bus_ranges ();of_unittest_bus_3cell_ranges ();of_unittest_reg ();
Signature · 1 line
2.35.3
PATCH 2/2 of: address: Preserve the flags portion on 1:1 dma-ranges mapping
In reply to Andrea della PortaA missing or empty dma-ranges in a DT node implies a 1:1 mapping for dma translations. In this specific case, the current behaviour is to zero out the entire specifier so that the translation could be carried on as an offset from zero. This includes address specifier that has flags (e.g. PCI ranges).
Once the flags portion has been zeroed, the translation chain is broken since the mapping functions will check the upcoming address specifier against mismatching flags, always failing the 1:1 mapping and its entire purpose of always succeeding.
Set to zero only the address portion while passing the flags through.
--- drivers/of/address.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/of/address.c +2 -1
@@ -455,7 +455,8 @@ static int of_translate_one(struct device_node *parent, struct of_bus *bus, }if (ranges ==NULL || rlen ==0 ) { offset =of_read_number (addr, na); -memset (addr,0 , pna *4 ); +/* copy the address while preserving the flags */ +memset (addr + pbus->flag_cells,0 , (pna - pbus->flag_cells) *4 );pr_debug ("empty ranges; 1:1 translation\n" ); goto finish; }
Signature · 1 line
2.35.3
Re: [PATCH 2/2] of: address: Preserve the flags portion on 1:1 dma-ranges mapping
Quoted 10 lines · Andrea della Porta
A missing or empty dma-ranges in a DT node implies a 1:1 mapping for dma translations. In this specific case, the current behaviour is to zero out the entire specifier so that the translation could be carried on as an offset from zero. This includes address specifier that has flags (e.g. PCI ranges).
Once the flags portion has been zeroed, the translation chain is broken since the mapping functions will check the upcoming address specifier against mismatching flags, always failing the 1:1 mapping and its entire purpose of always succeeding.
Set to zero only the address portion while passing the flags through.
Add blank lines between paragraphs.
Quoted 13 lines
Fixes dbbdee94734b ("of/address: Merge all of the bus translation code")Signed-off-by Andrea della Porta <[email protected]>Tested-by Herve Codina <[email protected]>--- drivers/of/address.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)drivers/of/address.c +0
-0
drivers/of/address.c static int of_translate_one(struct device_node *parent, struct of_bus *bus, · line 458
Quoted 5 lines
drivers/of/address.c +1
-1@@ -455,7 +455,8 @@ static int of_translate_one(struct device_node *parent, struct of_bus *bus, }if (ranges ==NULL || rlen ==0 ) { offset =of_read_number (addr, na); -memset (addr,0 , pna *4 ); +/* copy the address while preserving the flags */
Not knowing the surrounding code, it seems strange to say "copy the address" when the memset() fills with zero and does no copying.
The commit log says "set address to zero, pass flags through," and I could believe *that* matches the memset().
Quoted 7 lines
drivers/of/address.c +1
-0+memset (addr + pbus->flag_cells,0 , (pna - pbus->flag_cells) *4 );pr_debug ("empty ranges; 1:1 translation\n" ); goto finish; }Signature · 1 line
2.35.3
Re: [PATCH 2/2] of: address: Preserve the flags portion on 1:1 dma-ranges mapping
Hi Bjorn,
Quoted 12 lines · Bjorn Helgaas
Quoted 10 lines · Andrea della Porta
A missing or empty dma-ranges in a DT node implies a 1:1 mapping for dma translations. In this specific case, the current behaviour is to zero out the entire specifier so that the translation could be carried on as an offset from zero. This includes address specifier that has flags (e.g. PCI ranges).
Once the flags portion has been zeroed, the translation chain is broken since the mapping functions will check the upcoming address specifier against mismatching flags, always failing the 1:1 mapping and its entire purpose of always succeeding.
Set to zero only the address portion while passing the flags through.
Quoted 1 line
Add blank lines between paragraphs.
Ack.
Quoted 23 lines
Quoted 14 lines
Fixes dbbdee94734b ("of/address: Merge all of the bus translation code")Signed-off-by Andrea della Porta <[email protected]>Tested-by Herve Codina <[email protected]>--- drivers/of/address.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)drivers/of/address.c +0
-0drivers/of/address.c static int of_translate_one(struct device_node *parent, struct of_bus *bus, · line 458
Quoted 5 lines
drivers/of/address.c +1
-1@@ -455,7 +455,8 @@ static int of_translate_one(struct device_node *parent, struct of_bus *bus, }if (ranges ==NULL || rlen ==0 ) { offset =of_read_number (addr, na); -memset (addr,0 , pna *4 ); +/* copy the address while preserving the flags */ Not knowing the surrounding code, it seems strange to say "copy the address" when the
memset()fills with zero and does no copying.
Quoted 2 lines
The commit log says "set address to zero, pass flags through," and I could believe *that* matches the
memset().
Ack.
Many thanks,
Andrea
Quoted 8 lines
Quoted 8 lines
drivers/of/address.c +1
-0+memset (addr + pbus->flag_cells,0 , (pna - pbus->flag_cells) *4 );pr_debug ("empty ranges; 1:1 translation\n" ); goto finish; }Signature · 1 line
2.35.3