The IOCTL data is in the debugfs header as it is. I could move that to the
amdgpu_drm.h and include it from amdgpu_debugfs.h.
I'll re-write the STATE IOCTL to use a struct and then test against what I
have in umr.
Refactoring the read/write is trivial and I'll do that no problem (with
style fixes).
Cheers,
Tom
On Tue, Aug 24, 2021 at 8:34 AM Christian König <
ckoenig.leichtzumer...@gmail.com> wrote:
> Am 24.08.21 um 14:27 schrieb StDenis, Tom:
> > [AMD Official Use Only]
> >
> > What do you mean a "shared header?" How would they be shared between
> kernel and user?
>
> Somewhere in the include/uapi/drm/ folder I think. Either add that to
> amdgpu_drm.h or maybe amdgpu_debugfs.h?
>
> Or just keep it as a structure in amdgpu_debugfs.h with a comment that
> it is used for an IOCTL.
>
> Just not something hard coded like first byte is this, second that etc....
>
> > As for why not read/write. Jus wanted to keep it simple. It's not
> really performance bound. umr never does reads/writes larger than 32-bits
> anyways. It doesn't have to be this way though but I figured the fewer LOC
> the better.
>
> I think it would be cleaner if we would have separate functions for this.
>
> As far as I can see you also don't need the dance with the power
> managerment etc for the IOCTL. It's just get_user() into your structure.
>
> BTW: In the kernel coding style put "switch" and "case" on the same
> column, otherwise checkpatch.pl might complain.
>
> Christian.
>
> >
> > Tom
> >
> > ________________________________________
> > From: Christian König <ckoenig.leichtzumer...@gmail.com>
> > Sent: Tuesday, August 24, 2021 08:20
> > To: StDenis, Tom; amd-gfx@lists.freedesktop.org
> > Subject: Re: [PATCH] drm/amd/amdgpu: New debugfs interface for MMIO
> registers
> >
> >
> >
> > Am 24.08.21 um 14:16 schrieb Tom St Denis:
> >> This new debugfs interface uses an IOCTL interface in order to pass
> >> along state information like SRBM and GRBM bank switching. This
> >> new interface also allows a full 32-bit MMIO address range which
> >> the previous didn't. With this new design we have room to grow
> >> the flexibility of the file as need be.
> >>
> >> Signed-off-by: Tom St Denis <tom.stde...@amd.com>
> >> ---
> >> drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 177
> ++++++++++++++++++++
> >> drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h | 32 ++++
> >> 2 files changed, 209 insertions(+)
> >>
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> >> index 277128846dd1..ab2d92f84da5 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> >> @@ -279,6 +279,173 @@ static ssize_t amdgpu_debugfs_regs_write(struct
> file *f, const char __user *buf,
> >> return amdgpu_debugfs_process_reg_op(false, f, (char __user
> *)buf, size, pos);
> >> }
> >>
> >> +static int amdgpu_debugfs_regs2_open(struct inode *inode, struct file
> *file)
> >> +{
> >> + struct amdgpu_debugfs_regs2_data *rd;
> >> +
> >> + rd = kzalloc(sizeof *rd, GFP_KERNEL);
> >> + if (!rd)
> >> + return -ENOMEM;
> >> + rd->adev = file_inode(file)->i_private;
> >> + file->private_data = rd;
> >> +
> >> + return 0;
> >> +}
> >> +
> >> +static int amdgpu_debugfs_regs2_release(struct inode *inode, struct
> file *file)
> >> +{
> >> + kfree(file->private_data);
> >> + return 0;
> >> +}
> >> +
> >> +static int amdgpu_debugfs_regs2_op(struct file *f, char __user *buf,
> int write_en)
> >> +{
> >> + struct amdgpu_debugfs_regs2_data *rd = f->private_data;
> >> + struct amdgpu_device *adev = rd->adev;
> >> + int result = 0, r;
> >> + uint32_t value;
> >> +
> >> + if (rd->state.offset & 0x3)
> >> + return -EINVAL;
> >> +
> >> + if (rd->state.use_grbm) {
> >> + if (rd->state.grbm.se == 0x3FF)
> >> + rd->state.grbm.se = 0xFFFFFFFF;
> >> + if (rd->state.grbm.sh == 0x3FF)
> >> + rd->state.grbm.sh = 0xFFFFFFFF;
> >> + if (rd->state.grbm.instance == 0x3FF)
> >> + rd->state.grbm.instance = 0xFFFFFFFF;
> >> + }
> >> +
> >> + r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
> >> + if (r < 0) {
> >> + pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
> >> + return r;
> >> + }
> >> +
> >> + r = amdgpu_virt_enable_access_debugfs(adev);
> >> + if (r < 0) {
> >> + pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
> >> + return r;
> >> + }
> >> +
> >> + if (rd->state.use_grbm) {
> >> + if ((rd->state.grbm.sh != 0xFFFFFFFF && rd->state.grbm.sh
> >= adev->gfx.config.max_sh_per_se) ||
> >> + (rd->state.grbm.se != 0xFFFFFFFF && rd->state.grbm.se
> >= adev->gfx.config.max_shader_engines)) {
> >> + pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
> >> +
> pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
> >> + amdgpu_virt_disable_access_debugfs(adev);
> >> + return -EINVAL;
> >> + }
> >> + mutex_lock(&adev->grbm_idx_mutex);
> >> + amdgpu_gfx_select_se_sh(adev, rd->state.grbm.se,
> >> + rd->
> state.grbm.sh,
> >> +
> rd->state.grbm.instance);
> >> + } else if (rd->state.use_grbm) {
> >> + mutex_lock(&adev->srbm_mutex);
> >> + amdgpu_gfx_select_me_pipe_q(adev, rd->state.srbm.me,
> rd->state.srbm.pipe,
> >> +
> rd->state.srbm.queue, rd->state.srbm.vmid);
> >> + }
> >> +
> >> + if (rd->state.pg_lock)
> >> + mutex_lock(&adev->pm.mutex);
> >> +
> >> + if (!write_en) {
> >> + value = RREG32(rd->state.offset >> 2);
> >> + r = put_user(value, (uint32_t *)buf);
> >> + } else {
> >> + r = get_user(value, (uint32_t *)buf);
> >> + if (!r)
> >> + amdgpu_mm_wreg_mmio_rlc(adev, rd->state.offset >>
> 2, value);
> >> + }
> >> + if (r) {
> >> + result = r;
> >> + goto end;
> >> + }
> >> + result = 0;
> >> +end:
> >> + if (rd->state.use_grbm) {
> >> + amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff,
> 0xffffffff);
> >> + mutex_unlock(&adev->grbm_idx_mutex);
> >> + } else if (rd->state.use_srbm) {
> >> + amdgpu_gfx_select_me_pipe_q(adev, 0, 0, 0, 0);
> >> + mutex_unlock(&adev->srbm_mutex);
> >> + }
> >> +
> >> + if (rd->state.pg_lock)
> >> + mutex_unlock(&adev->pm.mutex);
> >> +
> >> + // in umr (the likely user of this) flags are set per file
> operation
> >> + // which means they're never "unset" explicitly. To avoid
> breaking
> >> + // this convention we unset the flags after each operation
> >> + // flags are for a single call (need to be set for every
> read/write)
> >> + rd->state.use_grbm = 0;
> >> + rd->state.use_srbm = 0;
> >> + rd->state.pg_lock = 0;
> >> +
> >> + pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
> >> + pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
> >> +
> >> + amdgpu_virt_disable_access_debugfs(adev);
> >> + return result;
> >> +}
> >> +
> >> +static long amdgpu_debugfs_regs2_ioctl(struct file *f, unsigned int
> cmd, unsigned long data)
> >> +{
> >> + struct amdgpu_debugfs_regs2_data *rd = f->private_data;
> >> + unsigned char st[32], v;
> >> + int r, x;
> >> +
> >> + // always read first 4 bytes of data
> >> + for (x = 0; x < 4; x++) {
> >> + if ((r = get_user(v, (unsigned char *)data))) {
> >> + return r;
> >> + }
> >> + ++data;
> >> + st[x] = v;
> >> + }
> >> +
> >> + // first 4 bytes are offset
> >> + rd->state.offset = ((u32)st[0]) | ((u32)st[1] << 8) |
> >> + ((u32)st[2] << 16) |
> ((u32)st[3] << 24);
> >> +
> >> + switch (cmd) {
> >> + case AMDGPU_DEBUGFS_REGS2_IOC_SET_STATE:
> >> + for (x = 4; x < 32; x++) {
> >> + if ((r = get_user(v, (unsigned char
> *)data))) {
> >> + return r;
> >> + }
> >> + ++data;
> >> + st[x] = v;
> >> + }
> >> +
> >> + // next byte contains the flag
> >> + // we only consume the remainder of the state if
> bit 1 is set
> >> + // this allows the subsequent read/write
> >> + rd->state.use_grbm = (st[4] & 1) ? 1 : 0;
> >> + rd->state.use_srbm = (st[4] & 2) ? 1 : 0;
> >> + rd->state.pg_lock = (st[4] & 4) ? 1 : 0;
> >> +
> >> + // next 6 bytes are grbm data
> >> + rd->state.grbm.se = ((u32)st[5]) |
> ((u32)st[6] << 8);
> >> + rd->state.grbm.sh = ((u32)st[7]) |
> ((u32)st[8] << 8);
> >> + rd->state.grbm.instance = ((u32)st[9]) |
> ((u32)st[10] << 8);
> >> +
> >> + // next 8 are srbm data
> >> + rd->state.srbm.me = ((u32)st[11]) |
> ((u32)st[12] << 8);
> >> + rd->state.srbm.pipe = ((u32)st[13]) |
> ((u32)st[14] << 8);
> >> + rd->state.srbm.queue = ((u32)st[15]) |
> ((u32)st[16] << 8);
> >> + rd->state.srbm.vmid = ((u32)st[17]) |
> ((u32)st[18] << 8);
> >> + break;
> >> + case AMDGPU_DEBUGFS_REGS2_IOC_READ:
> >> + return amdgpu_debugfs_regs2_op(f, (char __user
> *)data, 0);
> >> + case AMDGPU_DEBUGFS_REGS2_IOC_WRITE:
> >> + return amdgpu_debugfs_regs2_op(f, (char __user
> *)data, 1);
> >> + default:
> >> + return -EINVAL;
> >> + }
> >> + return 0;
> >> +}
> >>
> >> /**
> >> * amdgpu_debugfs_regs_pcie_read - Read from a PCIE register
> >> @@ -1091,6 +1258,14 @@ static ssize_t amdgpu_debugfs_gfxoff_read(struct
> file *f, char __user *buf,
> >> return result;
> >> }
> >>
> >> +static const struct file_operations amdgpu_debugfs_regs2_fops = {
> >> + .owner = THIS_MODULE,
> >> + .unlocked_ioctl = amdgpu_debugfs_regs2_ioctl,
> >> + .open = amdgpu_debugfs_regs2_open,
> >> + .release = amdgpu_debugfs_regs2_release,
> >> + .llseek = default_llseek
> >> +};
> >> +
> >> static const struct file_operations amdgpu_debugfs_regs_fops = {
> >> .owner = THIS_MODULE,
> >> .read = amdgpu_debugfs_regs_read,
> >> @@ -1148,6 +1323,7 @@ static const struct file_operations
> amdgpu_debugfs_gfxoff_fops = {
> >>
> >> static const struct file_operations *debugfs_regs[] = {
> >> &amdgpu_debugfs_regs_fops,
> >> + &amdgpu_debugfs_regs2_fops,
> >> &amdgpu_debugfs_regs_didt_fops,
> >> &amdgpu_debugfs_regs_pcie_fops,
> >> &amdgpu_debugfs_regs_smc_fops,
> >> @@ -1160,6 +1336,7 @@ static const struct file_operations
> *debugfs_regs[] = {
> >>
> >> static const char *debugfs_regs_names[] = {
> >> "amdgpu_regs",
> >> + "amdgpu_regs2",
> >> "amdgpu_regs_didt",
> >> "amdgpu_regs_pcie",
> >> "amdgpu_regs_smc",
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
> >> index 141a8474e24f..04c81cd4bcc7 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
> >> @@ -22,6 +22,7 @@
> >> * OTHER DEALINGS IN THE SOFTWARE.
> >> *
> >> */
> >> +#include <linux/ioctl.h>
> >>
> >> /*
> >> * Debugfs
> >> @@ -38,3 +39,34 @@ void amdgpu_debugfs_fence_init(struct amdgpu_device
> *adev);
> >> void amdgpu_debugfs_firmware_init(struct amdgpu_device *adev);
> >> void amdgpu_debugfs_gem_init(struct amdgpu_device *adev);
> >> int amdgpu_debugfs_wait_dump(struct amdgpu_device *adev);
> >> +
> >> +struct amdgpu_debugfs_regs2_data {
> >> + struct amdgpu_device *adev;
> >> + struct {
> >> + // regs state
> >> + int use_srbm,
> >> + use_grbm,
> >> + pg_lock;
> >> + struct {
> >> + u32 se, sh, instance;
> >> + } grbm;
> >> + struct {
> >> + u32 me, pipe, queue, vmid;
> >> + } srbm;
> >> + u32 offset;
> >> + } state;
> >> +};
> > This stuff needs to be in some shared header then.
> >
> >> +
> >> +enum AMDGPU_DEBUGFS_REGS2_CMDS {
> >> + AMDGPU_DEBUGFS_REGS2_CMD_SET_STATE=0,
> >> + AMDGPU_DEBUGFS_REGS2_CMD_READ,
> >> + AMDGPU_DEBUGFS_REGS2_CMD_WRITE,
> > Why not just using the normal read and write functions?
> >
> > Christian.
> >
> >> +};
> >> +
> >> +struct amdgpu_debugfs_regs2_iocdata {
> >> + unsigned char t[32];
> >> +};
> >> +
> >> +#define AMDGPU_DEBUGFS_REGS2_IOC_SET_STATE _IOWR(0x20,
> AMDGPU_DEBUGFS_REGS2_CMD_SET_STATE, struct amdgpu_debugfs_regs2_iocdata)
> >> +#define AMDGPU_DEBUGFS_REGS2_IOC_READ _IOWR(0x20,
> AMDGPU_DEBUGFS_REGS2_CMD_WRITE, struct amdgpu_debugfs_regs2_iocdata)
> >> +#define AMDGPU_DEBUGFS_REGS2_IOC_WRITE _IOWR(0x20,
> AMDGPU_DEBUGFS_REGS2_CMD_READ, struct amdgpu_debugfs_regs2_iocdata)
>
>
