| Message ID | ZNEyGV0jyI8kOOfz@p100 (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
| Series | io_uring/parisc: Adjust pgoff in io_uring mmap() for parisc | expand |
On 8/7/23 12:04?PM, Helge Deller wrote: > The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by > using architecture-provided get_unmapped_area()") to the parisc > implementation of get_unmapped_area() broke glibc's locale-gen > executable when running on parisc. > > This patch reverts those architecture-specific changes, and instead > adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is > then given to parisc's get_unmapped_area() function. This is much > cleaner than the previous approach, and we still will get a coherent > addresss. > > This patch has no effect on other architectures (SHM_COLOUR is only > defined on parisc), and the liburing testcase stil passes on parisc. What branch is this against? It doesn't apply to my 6.5 io_uring branch, which is odd as that's where the previous commits went through.
On 8/7/23 20:24, Jens Axboe wrote: > On 8/7/23 12:04?PM, Helge Deller wrote: >> The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by >> using architecture-provided get_unmapped_area()") to the parisc >> implementation of get_unmapped_area() broke glibc's locale-gen >> executable when running on parisc. >> >> This patch reverts those architecture-specific changes, and instead >> adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is >> then given to parisc's get_unmapped_area() function. This is much >> cleaner than the previous approach, and we still will get a coherent >> addresss. >> >> This patch has no effect on other architectures (SHM_COLOUR is only >> defined on parisc), and the liburing testcase stil passes on parisc. > > What branch is this against? It doesn't apply to my 6.5 io_uring branch, > which is odd as that's where the previous commits went through. applies for me on git head / Linux 6.5-rc5 Helge
On 8/7/23 12:27?PM, Helge Deller wrote: > On 8/7/23 20:24, Jens Axboe wrote: >> On 8/7/23 12:04?PM, Helge Deller wrote: >>> The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by >>> using architecture-provided get_unmapped_area()") to the parisc >>> implementation of get_unmapped_area() broke glibc's locale-gen >>> executable when running on parisc. >>> >>> This patch reverts those architecture-specific changes, and instead >>> adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is >>> then given to parisc's get_unmapped_area() function. This is much >>> cleaner than the previous approach, and we still will get a coherent >>> addresss. >>> >>> This patch has no effect on other architectures (SHM_COLOUR is only >>> defined on parisc), and the liburing testcase stil passes on parisc. >> >> What branch is this against? It doesn't apply to my 6.5 io_uring branch, >> which is odd as that's where the previous commits went through. > > applies for me on git head / Linux 6.5-rc5 Hmm, maybe something unrelated then. I'll take a look. The patch is garbled fwiw, but that's nothing new. If you download the raw one from lore you can see how.
On Mon, 07 Aug 2023 20:04:09 +0200, Helge Deller wrote: > The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by > using architecture-provided get_unmapped_area()") to the parisc > implementation of get_unmapped_area() broke glibc's locale-gen > executable when running on parisc. > > This patch reverts those architecture-specific changes, and instead > adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is > then given to parisc's get_unmapped_area() function. This is much > cleaner than the previous approach, and we still will get a coherent > addresss. > > [...] Applied, thanks! [1/1] io_uring/parisc: Adjust pgoff in io_uring mmap() for parisc commit: ce11a0688e67aae1e9ba6c8843d7e8b7dd791ead Best regards,
On 8/7/23 20:33, Jens Axboe wrote: > On 8/7/23 12:27?PM, Helge Deller wrote: >> On 8/7/23 20:24, Jens Axboe wrote: >>> On 8/7/23 12:04?PM, Helge Deller wrote: >>>> The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by >>>> using architecture-provided get_unmapped_area()") to the parisc >>>> implementation of get_unmapped_area() broke glibc's locale-gen >>>> executable when running on parisc. >>>> >>>> This patch reverts those architecture-specific changes, and instead >>>> adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is >>>> then given to parisc's get_unmapped_area() function. This is much >>>> cleaner than the previous approach, and we still will get a coherent >>>> addresss. >>>> >>>> This patch has no effect on other architectures (SHM_COLOUR is only >>>> defined on parisc), and the liburing testcase stil passes on parisc. >>> >>> What branch is this against? It doesn't apply to my 6.5 io_uring branch, >>> which is odd as that's where the previous commits went through. >> >> applies for me on git head / Linux 6.5-rc5 > > Hmm, maybe something unrelated then. I'll take a look. The patch is > garbled fwiw, but that's nothing new. If you download the raw one from > lore you can see how. That's strange. I usually send to the mailing list, and then apply from patchwork, in this case from here: https://patchwork.kernel.org/project/linux-parisc/patch/ZNEyGV0jyI8kOOfz@p100/ That worked for me. Helge
On 8/7/23 20:34, Jens Axboe wrote: > > On Mon, 07 Aug 2023 20:04:09 +0200, Helge Deller wrote: >> The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by >> using architecture-provided get_unmapped_area()") to the parisc >> implementation of get_unmapped_area() broke glibc's locale-gen >> executable when running on parisc. >> >> This patch reverts those architecture-specific changes, and instead >> adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is >> then given to parisc's get_unmapped_area() function. This is much >> cleaner than the previous approach, and we still will get a coherent >> addresss. >> >> [...] > > Applied, thanks! That was fast :-) Actually I had hoped for some more testing from Christoph and other parisc guys first. Anyway, since you have a parisc machine in your test ring, you will notice if something breaks, What's important: Please add to the patch: Cc: stable@vger.kernel.org # 6.4 Thank you! Helge > [1/1] io_uring/parisc: Adjust pgoff in io_uring mmap() for parisc > commit: ce11a0688e67aae1e9ba6c8843d7e8b7dd791ead
On Aug 7, 2023, at 12:33 PM, Helge Deller <deller@gmx.de> wrote: > > On 8/7/23 20:34, Jens Axboe wrote: >> >>> On Mon, 07 Aug 2023 20:04:09 +0200, Helge Deller wrote: >>> The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by >>> using architecture-provided get_unmapped_area()") to the parisc >>> implementation of get_unmapped_area() broke glibc's locale-gen >>> executable when running on parisc. >>> >>> This patch reverts those architecture-specific changes, and instead >>> adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is >>> then given to parisc's get_unmapped_area() function. This is much >>> cleaner than the previous approach, and we still will get a coherent >>> addresss. >>> >>> [...] >> >> Applied, thanks! > > That was fast :-) > Actually I had hoped for some more testing from Christoph and other > parisc guys first. > Anyway, since you have a parisc machine in your test ring, you will > notice if something breaks, > > What's important: > Please add to the patch: > Cc: stable@vger.kernel.org # 6.4 It’s not going upstream just yet, just easiser to apply for testing on my end. I will test it locally too. — Jens Axboe
diff --git a/arch/parisc/kernel/sys_parisc.c b/arch/parisc/kernel/sys_parisc.c index ca2d537e25b1..9915062d5243 100644 --- a/arch/parisc/kernel/sys_parisc.c +++ b/arch/parisc/kernel/sys_parisc.c @@ -27,17 +27,12 @@ #include <linux/elf-randomize.h> /* - * Construct an artificial page offset for the mapping based on the virtual + * Construct an artificial page offset for the mapping based on the physical * address of the kernel file mapping variable. - * If filp is zero the calculated pgoff value aliases the memory of the given - * address. This is useful for io_uring where the mapping shall alias a kernel - * address and a userspace adress where both the kernel and the userspace - * access the same memory region. */ -#define GET_FILP_PGOFF(filp, addr) \ - ((filp ? (((unsigned long) filp->f_mapping) >> 8) \ - & ((SHM_COLOUR-1) >> PAGE_SHIFT) : 0UL) \ - + (addr >> PAGE_SHIFT)) +#define GET_FILP_PGOFF(filp) \ + (filp ? (((unsigned long) filp->f_mapping) >> 8) \ + & ((SHM_COLOUR-1) >> PAGE_SHIFT) : 0UL) static unsigned long shared_align_offset(unsigned long filp_pgoff, unsigned long pgoff) @@ -117,7 +112,7 @@ static unsigned long arch_get_unmapped_area_common(struct file *filp, do_color_align = 0; if (filp || (flags & MAP_SHARED)) do_color_align = 1; - filp_pgoff = GET_FILP_PGOFF(filp, addr); + filp_pgoff = GET_FILP_PGOFF(filp); if (flags & MAP_FIXED) { /* Even MAP_FIXED mappings must reside within TASK_SIZE */ diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c index f4591b912ea8..93db3e4e7b68 100644 --- a/io_uring/io_uring.c +++ b/io_uring/io_uring.c @@ -3470,6 +3470,8 @@ static unsigned long io_uring_mmu_get_unmapped_area(struct file *filp, * - use the kernel virtual address of the shared io_uring context * (instead of the userspace-provided address, which has to be 0UL * anyway). + * - use the same pgoff which the get_unmapped_area() uses to + * calculate the page colouring. * For architectures without such aliasing requirements, the * architecture will return any suitable mapping because addr is 0. */ @@ -3478,6 +3480,7 @@ static unsigned long io_uring_mmu_get_unmapped_area(struct file *filp, pgoff = 0; /* has been translated to ptr above */ #ifdef SHM_COLOUR addr = (uintptr_t) ptr; + pgoff = addr >> PAGE_SHIFT; #else addr = 0UL; #endif
The changes from commit 32832a407a71 ("io_uring: Fix io_uring mmap() by using architecture-provided get_unmapped_area()") to the parisc implementation of get_unmapped_area() broke glibc's locale-gen executable when running on parisc. This patch reverts those architecture-specific changes, and instead adjusts in io_uring_mmu_get_unmapped_area() the pgoff offset which is then given to parisc's get_unmapped_area() function. This is much cleaner than the previous approach, and we still will get a coherent addresss. This patch has no effect on other architectures (SHM_COLOUR is only defined on parisc), and the liburing testcase stil passes on parisc. Signed-off-by: Helge Deller <deller@gmx.de> Reported-by: Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de> Fixes: 32832a407a71 ("io_uring: Fix io_uring mmap() by using architecture-provided get_unmapped_area()") Fixes: d808459b2e31 ("io_uring: Adjust mapping wrt architecture aliasing requirements")